paper_id
stringlengths 9
16
| version
stringclasses 26
values | yymm
stringclasses 311
values | created
timestamp[s] | title
stringlengths 6
335
| secondary_subfield
sequencelengths 1
8
| abstract
stringlengths 25
3.93k
| primary_subfield
stringclasses 124
values | field
stringclasses 20
values | fulltext
stringlengths 0
2.84M
|
|---|---|---|---|---|---|---|---|---|---|
1108.0390 | 2 | 1108 | 2013-03-01T20:02:54 | Tensile material properties of human rib cortical bone under quasi-static and dynamic failure loading and influence of the bone microstucture on failure characteristics | [
"physics.bio-ph",
"physics.med-ph"
] | Finite element models of the thorax are under development to assist vehicle safety researchers with the design of countermeasures such as advanced restrain systems. Computational models have become more refined with increasing geometrical complexity as element size decreases. These finite element models can now capture small geometrical features with an attempt to predict fracture. However, the bone material properties currently available, and in particular the rate sensitivity, have been mainly determined from compression tests or tests on long bones. There is a need for a new set of material properties for the human rib cortical bone. With this objective, a new clamping technique was developed to test small bone coupons under tensile loading. Ten coupons were harvested from the cortical shell of the sixth and seventh left ribs from three cadavers. The coupons were tested to fracture under quasi-static (target strain rate of 0.07 %/s) and dynamic loading (target strain rate of 170 %/s). Prior to testing, each coupon was imaged with a computed micro-tomograph to document the bone microstructure. An optical method was used to determine the strain field in the coupon for the quasi-static tests. The rib bone coupons were found to be elastic, with brittle fracture. No plastic behavior was observed in this test series. The bone coupons were assumed isotropic, homogeneous and elastic linear, and the average Young's modulus for the quasi-static tests (13.5 GPa) and the failure stresses (quasi-static: 112 MPa, dynamic: 124.6 MPa) were in line with published data. Fracture however did not always occur in the gage area where the cross-sectional area was the smallest, which contradicted the assumption of isotropy and homogeneity. The comparison with material properties obtained for long bones suggests that the effective cross-section has an effect on the calculated material properties. | physics.bio-ph | physics |
Tensile material properties of human rib cortical bone
under quasi-static and dynamic failure loading and
influence of the bone microstucture on failure
characteristics
Damien Subita,∗, Eduardo del Pozo de Diosb, Juan Vel´azquez-Ameijidec,
Carlos Arregui-Dalmasesa,d, Jeff Crandalla
aUniversity of Virginia, Center for Applied Biomechanics, 4040 Lewis and Clark drive,
Charlottesville, VA 22902, USA
bUniversidad de Navarra, European Center for Injury Prevention, Irunlarrea 1 (ed. Los
Castanos), 31008 Pamplona, Navarra, Spain
cUniversitat Polit`ecnica de Catalunya, Dept. Resist`encia de Materials i Estructures a
l’Enginyeria, Comte d’Urgell 187, 08036 Barcelona, Spain
dUniversidad de Navarra, Preventive and Public Health Dpt., School of Medicine,
Irunlarrea 1, Ed. investigaci´on (despacho 2440), 31008 Pamplona, Navarra, Spain
Abstract
Finite element models of the thorax are being developed to assist engineers
and vehicle safety researchers with the design and validation of counter-
measures such as advanced restrain systems. Computational models have
become more refined with increasing geometrical complexity as element size
decreases. These finite element models can now capture small geometri-
cal features with an attempt to predict fracture. However, the bone ma-
terial properties currently available, and in particular the rate sensitivity,
have been mainly determined from compression tests or tests on long bones.
There is a need for a new set of material properties for the human rib cor-
tical bone. With this objective, a new clamping technique was developed
∗Corresponding author
Email addresses: [email protected] (Damien Subit), [email protected]
(Eduardo del Pozo de Dios), [email protected] (Juan Vel´azquez-Ameijide),
[email protected] (Carlos Arregui-Dalmases), [email protected] (Jeff
Crandall)
URL: www.centerforappliedbiomechanics.org (Damien Subit), www.unav.es/ecip
(Eduardo del Pozo de Dios), www.upc.edu (Juan Vel´azquez-Ameijide),
www.unav.es/ecip (Carlos Arregui-Dalmases),
www.centerforappliedbiomechanics.org (Jeff Crandall)
May 20, 2014
to test small bone coupons under tensile loading. This technique allows for
applying minimal constraints to the coupon during clamping and ensures
that the main type of loading is tension. Ten coupons were harvested from
the cortical shell of the sixth and seventh left ribs from three cadavers. The
coupons were tested to fracture under quasi-static (target strain rate of rate
of 0.07 %/s) and dynamic loading (target strain rate of 170 %/s). Prior
to testing, each coupon was imaged with a computed micro-tomograph to
document the bone microstructure. An optical method was used to deter-
mine the strain field in the coupon for the quasi-static tests. The rib bone
coupons were found to be elastic, with brittle fracture. No plastic behavior
was observed in this test series. The bone coupons were assumed isotropic,
homogeneous and elastic linear, and the average Young’s modulus for the
quasi-static tests (13.5 GPa) and the failure stresses (quasi-static: 112 MPa,
dynamic: 124.6 MPa) were in line with published data. Fracture however
did not always occur in the gage area where the cross-sectional area was the
smallest, which contradicted the assumption of isotropy and homogeneity.
The comparison of the results obtained in the current study with published
results on tibia and femur bone coupons suggests that the effective cross-
section has an effect on the calculated material properties, and that further
analysis of the bone microstructure is required to establish the rib cortical
bone fracture mechanism.
Keywords: bone, cadaver, rib, strain, porosity, fracture, quasi-static,
dynamic, micro-computed tomography
1. Introduction
Rib fractures are a good indicator of the severity of an impact to the tho-
rax as the protection to the internal organs such as the lungs and the heart
is greatly reduced with the increasing number of fractured ribs (abbreviated
injury scale, AAAM (2008)). Injury mechanisms for the ribs and the whole
rib cage have been widely studied, either through experiments (Kent et al.,
2004; Vezin and Berthet, 2009; Kuppa and Eppinger, 1998; Trosseille et al.,
2008; Hallman et al., 2010; Petitjean et al., 2003; Lessley et al., 2010b) or
computational simulations (Murakami et al., 2006; Liz´ee et al., 1998; Song
et al., 2009; Robin, 2001; Vezin and Verriest, 2005; Shigeta et al., 2009;
Kimpara et al., 2005; Plank and Eppinger, 1989; Ruan et al., 2003; Li et al.,
2010a,b; Kimpara et al., 2006) to determine injury mechanisms and thresh-
olds under diverse load conditions. A significant milestone was achieved in
the characterization of the strength of the thorax by accounting for the geo-
2
metrical variations in the rib cage and the rib themselves, and for the effects
of biological variations such as aging (Berthet et al., 2005; Ito et al., 2009;
Gayzik et al., 2008; Kent et al., 2004). Ribs were shown to have a complex
geometry that includes variation in the shape of the cross-section along the
rib axis (Kindig, 2009), an increase of the twist from the posterior to the
anterior aspect (Mohr et al., 2007), as well as a non uniform distribution
of the cortical thickness (Choi and Lee, 2009). In a recent study, Li et al.
(2010b) investigated the sensitivity of the rib structural response obtained
from a computational model to (a) the accuracy in the reconstruction of the
rib (quantified by the mesh density), (b) the cortical thickness distribution
and (c) the material properties. The response of the rib finite element model
was found to be little sensitive to the choice of material properties, whereas
clear trends were observed in the effect of the mesh density and cortical
thickness distribution. In finite element modeling, the cortical shell is de-
fined as a continuum made of pure cortical bone (Li et al., 2011). However,
rib cortical bone has voids and is not homogeneous (figure 1), and therefore
the bone material properties documented for homogeneous bone are likely
to inadequately represent the mechanical behavior of rib cortical bone.
Figure 1: Cross-section of the left 6th from the antero-lateral aspect, from micro-computed
tomography. The arrows indicate voids. The bar scale is 1 mm long.
Bone tensile material properties have been reported extensively for bovine
bones (Pithioux et al., 2004; Ferreira et al., 2006; Adharapurapu et al., 2006;
Saha and Hayes, 1974; Wright et al., 1976; Wright and Hayes, 1977; Crown-
inshield and Pope, 1974), but less often for human bone (Saha and Hayes,
1976; Kemper et al., 2005; Keller et al., 1990; Hansen et al., 2008). The
3
experimental tests by Kemper et al. (2005) are the only study targeted at
human rib bones. Delicate ”dog bone” coupons (length: 30 mm, width:
9 mm at the ends, 2.5 mm in the gage area, thickness: 0.5 mm) were ma-
chined from cortical rib bones, and a slack adaptor system was designed to
apply a constant strain rate, with a target of 50%/s. Special care was taken
to prevent the breaking of the samples during handling. However, for more
than 80 % of the tests (95 out of 117 tests) the bone samples fractured in the
grip area. For the other tests, the actual fracture location was not reported
and therefore one can only assume that the fracture occurred in the gage
area. Dog bone samples have a weak point by design – the gage area where
the width is much smaller than that at the ends of the sample – so as to
generate greater tensile stress in the gage area and ensure that fracture is
caused by tension, at least under the assumption that the sample is homo-
geneous. The fact that fracture occurred in the grip area so often suggests
either that the test apparatus caused the fracture, for instance because of
inadequate clamping force or misalignment of the top and bottom grips,
or that the hypothesis of homogeneity did not hold. In addition, Kemper
et al. (2005) used an extensometer to measure the average strain for the gage
area that may have interfered with the mechanical response of the samples.
The questions raised in Kemper et al. (2005) and the paucity of rib bone
material data highlighted the need for new experimental data. The goal of
the current study was to perform new tensile tests of rib bone coupons to
link the fracture characteristics to the coupon bone microstructure. To do
so, an experimental protocol was developed to machine and test rib cortical
bone coupons of constant thickness under quasi-static and dynamic load-
ing. All the samples were imaged prior to testing using a micro-computed
tomograph. This protocol was evaluated on coupons machined from the left
sixth and seventh ribs harvested from three post mortem human subjects.
2. Materials and methods
2.1. Coupon preparation
The ribs were first cut into five 60-mm long pieces named A (most an-
terior piece) to E (most posterior piece). Isotropic images similar to figure
1 were obtained with a micro-computed tomograph (Scanco Medical Viva
CT40, Bruttisellen, Switzerland) at a resolution of 30 µm/pixel and used to
determine where the bone coupons could be machined based on the curva-
ture and the thickness of the cortical shell (Figure 2-a). Next, the periosteum
was removed, and the sample was shortened with a small abrasive band
saw equipped with a diamond blade (Diamond bone band saw, Mar-med
4
inc., Cleveland, OH, USA). The next cuts were performed with a low speed
diamond saw (IsoMet low speed saw, Buehler Ltd, Lake Bluff, IL, USA,
equipped with a 15HC Diamond blade) that can execute parallel cuts with
micrometric accuracy. The sample was glued to a piece of hard foam that
was then clamped in the sample holder in the desired orientation thanks to
screws with spear tips (Figure 2-b). This step was critical in the machining
process: the orientation of the bone slab had to be adjusted so that the cut
performed with the low speed diamond saw would be entirely in the cortical
shell. A first cut was made to generate the outermost flat surface. From
there, an other cut was made parallel to the first one to create a flat piece
of cortical bone (referred to as wafer) that was 0.5 mm thick (Figure 2-c).
Next the two holes used to reference the position of the bone coupon (figure
3) were drilled in the wafer.
Figure 2: Machining of bone coupons.
5
Figure 3: Dimensions of the bone coupons. The 2 holes are used only for references for
machining and testing, and do not carry any load.
The next step consisted in machining the wafer to give it its final shape.
This happened in two successive actions. First, the wafer was shortened
to the desired length, and second, the narrow section of the coupon was
machined. These two actions utilized the same principle: the wafer was
sandwiched between two pieces of plastic (Figure 2-d), and attached thanks
to two screws to a template made of brass. A router equipped with a bearing
flush trim bit was used to trim the bone wafer to the desired shape: the
bearing was kept in contact with the brass template that was used as a
guide (Figure 2-e). To make the narrow section of the coupon, a template
with two grooves was used. After the first side was machined, the sample
was rotated about one screw and set along the other groove to machine the
other side (Figures 2-f and g). During the entire procedure, the orientation
of the wafer relative to the original bone was tracked and the bone was kept
hydrated (0.9 % saline solution). The coupons were then stored in a tube
filled with saline solution and kept in a fridge until the test day (between
1 and 4 days after preparation) to ensure proper conservation, and imaged
with the micro-tomograph. The thickness and width at the center of each
coupon were measured prior to each test.
2.2. Test fixture and procedure
A hydraulic tensile machine was used (Model 8874, Instron Inc, Nor-
wood, MA, USA). A clamping system made of aluminum was designed to
avoid misalignment between the top and the bottom ends of the coupon
(figure 4). Because of their fragility, it was critical to be able to install the
coupons on the test machine without applying any load that could gener-
ate bending, shear or too much tension. Conventional fixed wedge clamps
would not ensure the required control of the load during the installation of
6
the coupon. Therefore small low-mass clamps (10.9 grams each) were de-
signed to be attached to the coupon (figure 5) prior to mounting the sample
to the test machine. A torque of 1 Nm was applied to the screws used to
clamp the samples. Preliminary tests showed that this torque value was
adequate to prevent slippage of the sample, while avoiding bone crushing.
Figure 4: Clamping system for the coupons.
Pins that go through the rod-end ball joints were used to connect the
clamps to the slitted blocks affixed to the base of the tensile machine (bottom
clamp) and to the end of the piston (top clamp, Figure 4).
The top clamp was first installed and the coupon equipped with the
clamps was let to hang. The position of the sample was adjusted by moving
the piston so that the bottom pin could go through the bottom ball joint
without applying any load. The piston was then slowly moved up until
the bottom pin touches the clevace, resulting in a preload of about 2 N.
The bone coupons were stored in saline solution until the paint pattern was
applied, a few minutes prior to testing (see section 2.4).
Figure 5: Close-up view of the clamp.
2.3. Test matrix
The sixth and seventh right ribs were harvested from the three sub-
jects included in Lessley et al. (2010a) (Table 1). Ten bone coupons were
7
machined, and tested up to fracture with a constant velocity displacement
under quasi static (0.01 mm/s and 0.02 mm/s) or dynamic (24 mm/s) rates
(Table 2). The velocity of the applied displacement was determined to gen-
erate a strain rate similar to that reported on in Lessley et al. (2010a) for
the dynamic tests.
Subject Age at time of death
Cause of death
Body mass (kg) Stature (cm)
468
473
480
67
54
71
Stroke
Brain aneurysm
Laryngeal cancer
64
73
70
166
182
182
Table 1: Subjects characteristics.
Subject Rib level Location Aspect Velocity (mm/s) Strain rate (%/s)
468
473
480
6
7
7
6
7
7
7
6
7
7
A
B
C
B
B
B
C
A
B
C
Lateral
Lateral
Lateral
Medial
Lateral
Medial
Lateral
Lateral
Lateral
Lateral
0.02
0.01
24
0.01
0.01
24
24
0.01
24
0.01
0.089
0.055
170
0.046
0.043
170
170
0.069
170
0.055
Table 2: Test matrix. The actual strain rates are provided for the quasi-static tests,
whereas only the target strain rate is provided for the dynamic tests as the strain time
history data could not be determined (see section 2.4).
2.4. Data acquisition and processing
The tensile load was measured by a three axis loadcell located under the
bottom clamp (model 6085, Denton Inc, Plymouth, MI, USA), connected to
a standard data acquisition system (DEWE-2010, Dewetron GmbH, Graz,
Austria). The tensile and shear loads were measured. The sampling rate
was 500 Hz for the quasi-static tests, and 100 kHz for the dynamic tests,
and the signals were filtered using a low-pass 2nd Butterworth filter with
a cut-off frequency of 10 Hz for the quasi-static tests and 600 Hz for the
dynamic tests. These filters were chosen to remove the noise of the signals
while keeping their shapes and introducing negligeable time-shift.
8
For the quasi-static tests, the strain was measured by performing image
analysis of pictures taken during the tests with a 12-megapixel single-lens
reflex camera equipped with a 100-mm macro lens and triggered by a pro-
grammable controller (The Time Machine, Mumford Micro Systems, Santa
Barbara, CA, USA) that allows for tripping the camera shutter at a constant
frequency of 3 frames per second. The controller signal was also sampled by
the data acquisition system to synchronize strain and force measurements.
The outermost surface of the coupon (with respect to its orientation rela-
tive to the rib before machining) was painted with a black and white pattern
(figure 6) to allow for strain measurement based on contrast analysis (Frank
and Spolenak, 2008). Strain could not be measured for dynamic tests due to
the unavailability of a continuous (non-flickering) light source. The tensile
stress was defined as the ratio of the tensile force by the cross-sectional area
in the middle of the coupon. The effective Young’s modulus (referred to as
Young’s modulus) was defined as the slope of the strain-stress curve between
0 and 0.5 %. For the quasi-static tests, stress and strain could be measured.
For the dynamic tests, only failure stress could be determined. In addition,
fracture location was documented by measuring the distance between the
fracture line and the anterior end of the coupon.
Figure 6: Coupon with high contrast pattern prior to test.
2.5. Analysis of the images from micro-computed tomography
The images obtained with the micro-tomograph (DICOM format) were
analyzed to determine the actual cross-sectional area that accounts for the
presence of voids in the cortical microstructure and for the cross-sections
not being perfectly rectangular. A Matlab script (2010a, The Mathworks,
Natick, MA, USA) was written to perform the segmentation of the DICOM
images: a threshold of 3000 Housfield Unit (HU) was used, and all the pixels
9
with a HU value greater than the threshold were considered as bone. The
position along the coupon was normalized, with 0 and 1 being respectively
the anterior and posterior aspects of the coupon (relative to the rib).
3. Results
3.1. Material properties
Strain-stress curves were plotted for all the quasi-static tests (figure 7).
The material properties were determined by assuming a linear elastic model
between 0 and 0.5 % of tensile strain (table 3). Young’s modulus ranged
from 11.4 to 18.5 GPa, failure stress from 83.4 to 143.9 MPa, and failure
strain from 0.71 to 1.49 % in quasi-static, and failure stress ranged from
94.7 to 155.9 MPa in dynamic.
Figure 7: Strain-stress curves for the quasi-static tests. *: sample fractured in the clamp.
3.2. Fracture location
Fracture occurred outside the grip area in nine of the ten tests. Fracture
occurred first within the clamping area for one sample (473, 7, B) because
the sample slid of the clamp, causing the screw to bear the tension load. The
posterior end (right, figure 3) for this sample was too thin (about 0.43 mm)
to provide sufficient clamping load, and the sample slid relative to the clamp
10
Coupon
Quasi-static
468, 6, A
468, 7, B
473, 6, B
473, 7, B
480, 6, A
480, 7, C
Mean values
± standard deviation
Dynamic
468, 7, C
473, 7, B
473, 7, C
480, 7, B
Mean values
± standard deviation
E (GPa)
σf ailure (MPa)
εf ailure (%) Fracture location
18.5
14
12.2
12.8
11.4
129.9
143.9
83.4
(*)
106
0.71
1.19
0.9
(*)
1.49
12.2
13.5 ± 2.6
97.3
112.1 ± 24.5
1.02
1.06 ± 0.29
†
140.8
155.9
107
94.7
124.6 ± 28.5
Table 3: Rib cortical bone material properties for the quasi-static and dynamic tests, and
images of the fractured coupons (posterior surface). The anterior end is the left side. (*)
sample fractured in the clamp. (†) the fracture by the right hole occurred after the test
during handling.
after a certain tensile load was applied. The analysis of the images collected
during the test for strain measurement confirms that sliding occurred after
0.5 % deformation; therefore only the Young’s modulus could be determined
for this test. The variation of the cross-section along the length of the
coupons shows that the gage area cross-sectional areas was close to the
target (1.25 mm) for all the coupons (figures 8 and 9), whereas the cross-
sectional area of the ends varied greatly from sample to sample. The images
obtained from micro-computed tomography are provided for each coupon
(Appendix A).
11
Figure 8: Variation of the cross-section (quasi-static tests). The fracture location is indi-
cated by the dot. Zero is the anterior end of the coupon, and one is its posterior end.
4. Discussion
4.1. Material properties
The rib bone mechanical properties were calculated based on the as-
sumption that the rib cortical bone was homogeneous and isotropic. The
average value and standard deviation for the Young’s moduli determined for
the quasi-static tests (13.5 ± 2.6 GPa) were similar to the 13.9 ± 3.7 GPa
reported in Kemper et al. (2005). The failure stresses were also in the same
range for both the quasi-static (121 ± 24.5 MPa) and dynamic tests (124.6
± 28.5 MPa) in the current study compared to Kemper et al. (124.3 ± 35.4
MPa). Although the target strain rates were different in these two studies
(0.07 %/s for the quasi-static and 170 %/s for the dynamic tests in the cur-
rent study, 50 %/s in Kemper et al.), the measured strain rates are actually
wide spread (0.043 %/s to 0.089 %/s for the current study, and from 9 %/s
to 90 %/s in Kemper et al.). Based on Hansen et al. (2008) analysis of sev-
eral studies reporting bone material data, the Young’s modulus is little rate
sensitive for strain rates below 200 %/s. Therefore the comparison of the
results reported in the current study with that reported in Kemper et al. is
justified. A major difference between these two studies is the non linearity,
as Kemper et al. reported on a substantial yield behavior, and much greater
12
Figure 9: Variation of the cross-section (dynamic tests). The fracture location is indicated
by the dot. Zero is the anterior end of the coupon, and 1 is its posterior end.
failure strains (2.68 ± 1.4 %) than what was determined here (1.05 ± 0.29
%). In addition, the failure strains reported on in Kemper et al. range from
0.52 to 6.56%/s. This range is high compared to what is commonly used for
bone. It is not clear whether the discrepancies are artifacts because of the
experimental methods or due to age differences, as the subjects in the cur-
rent study are a little bit older than in Kemper et al. (2005). The coupons
machined in Kemper et al. had a wide range of cross-sectional areas (from
0.38 to 2.08 mm2), whereas the cross-sectional areas of coupons used for the
present tests were more uniform (1.25 mm2). The effect of the size of the
coupons on the measured mechanical properties is not well understood, in
particular how the loading varies with the varying cross-sectional area. In
the analysis performed in the two studies, tensile is assumed to be the main
type of loading. Thanks to the image analysis performed to measure strain
in the current study, it was possible to ensure that there was no slipping in
the clamps (except for the one test identified in the results section), whereas
Kemper et al. reported that most of the coupons broke in the clamp area
which suggests that these fractures were artifactual of the experimental set-
up. Too few samples were tested in this study to determine whether bone
properties vary as a function of the location along the rib.
13
Study
Yamada and Evans (1970) - Dry Human long bones
Yamada and Evans (1970) - Wet Human long bones
Hansen et al. (2008)
Kemper et al. (2005)
Current study - Quasi-static
Current study - Dynamic
Human ribs
Human ribs
Human ribs
Bone type
Human femora
Young’s modulus (GPa) Failure stress (MPa) Failure strain (%)
18.3
21.1
16.1 ± 2.1
13.9 ± 3.7
13.5 ± 2.6
140
172
119.8 ± 20.7
124.3 ± 35.4
112.1 ± 24.5
124.6 ± 28.5
1.49
1.3
2.5 ± 0.8
2.68 ± 1.4
1.06 ± 0.29
Table 4: Summaries of the cortical bone material properties reported by various authors.
The failure stress was found to be on average higher for the dynamic
tests than for the quasi-static tests (however fewer samples were tested un-
der dynamic loading). There is very little data in the published literature
regarding rib cortical bone rate sensitivity under tensile loading. The data
reported on in McElhaney (1966) are often mistakenly used to describe the
rate sensitivity of the bone material properties in tension, although the re-
sults established by McElhaney are based on compression tests. Other stud-
ies such as Crowninshield and Pope (1974) used animal tissue and the bone
samples were machined in the lower extremity; consequently the coupons
are bigger. In addition, the mechanical function of a a bovine leg is different
from that of a human rib, and therefore the microstructure is likely to be
different. It is indeed true that the coupons machined in the human cortical
femur and tibia bone are denser than that machined in the rib bone (on-
going analysis of unpublished test data). Therefore, the effective modulus
is higher for the lower extremity coupons than for the rib ones (table 4).
The coupons machined in the human cortical bone are heterogeneous
(Appendix A), and the actual cross-sections (determined by taking into
account the presence of the voids) were smaller than the target of 1.25 mm2
(figures 8 and 9). When compared to the material parameters determined
for the human femur (table 4), rib bone exhibits a lower Young’s modulus,
but the failure stress and failure strain are in the same range. Although
the strains were not measured using the same methods for all these studies
(extensometer or non contact optical measurements), it makes sense that
the effective modulus appears smaller for the rib bones. The voids in the rib
bone microstructure are not accounted for in the measurement of the cross-
sectional area as the coupon cross-section is assumed to be rectangular and
its area is approximated by measuring its thickness and width. With the
femur cortical bone being denser than the rib cortical bone, coupons made
of femur bone or rib bone with the same external dimensions would appear
to have the same cross-sectional area, whereas the actual cross-section of
the rib bone coupon will be smaller because of the voids.
14
The analysis performed in the present study relies on the assumption
that bone is homogeneous and isotropic, and therefore that the stress is
greater in the gage area. However, the fracture location was not systemati-
cally in the gage area (table 3), which suggest that the continuum mechanics
approach has its limitations. This has several implications in terms of frac-
ture prediction, as the microstructure - in addition to the geometry - alters
the stress field in the cortical shell and therefore the fracture threshold and
location. Li et al. (2010b) performed an extensive computational analysis
to determine which features of the rib geometry and structure needed to
be included in a finite element model in addition to the adequate material
properties to predict fracture under antero-posterior loading. Load and dis-
placement at fracture, could be successfully predicted, whereas the fracture
location predicted by the model did not match the experimental results.
Both the cortical and trabecular bones were assumed elastic homogeneous
and isotropic, and this may be why the fracture location was not correct,
even with the cortical thickness distribution mapped from the actual ribs
used in the experiments.
4.2. Fracture prediction
None of the published studies on bone coupons reported the fracture
locations. However this would provide a valuable piece of information re-
garding the modeling approach that should be used to adequately predict
bone fracture. As reported in the current study, fracture did not always took
place in the gage area. Preliminary tests were performed on plastic homoge-
neous and isotropic plastic coupons prepared following the same procedure,
and they all fractured in the gage area because of tension (the fracture line
was perpendicular to the long axis of the coupons). Besides, the shear forces
recorded with the three axis load cell were small compared to the load mea-
sured in the tensile direction (less than 5 %). This confirmed that the test
apparatus itself did not generate artifactual fractures, and therefore the frac-
ture characteristics are the results of the coupons material and structural
properties. Therefore, although bone material might be adequately assumed
to be homogeneous and elastic linear for sub-fracture behavior, the fracture
properties seem to be dependent on the bone microstructure which is not
properly accounted for with an elastic linear model. Researchers have had
limited success to predict fracture location in finite element models (Li et al.,
2010b). Current finite element models of the ribs include the rib as structure
composed of two distinct materials (trabecular and cortical bone). The re-
sults of the current study suggest that the delineation between cortical and
trabecular bone has become less clear with improving imaging capabilities
15
(figure 1): the cortical shell in the rib has pores, and the transition from the
cortical to the trabecular bone can be somewhat blurred. Rather than using
a two-part model for bone, a gradient approach may prove more accurate to
predict the actual deformation modes and fracture locations under dynamic
loading. Rib bone material properties need to be refined to include a more
precise model for fracture prediction, such as stress concentration caused by
the presence of pores or by the connection between the trabeculae and the
cortical shell. Micromechanics approaches such as the cohesive zone models
(Subit et al., 2009) would be worth evaluating as they can predict the onset
of a crack based on the local microstructure.
4.3. Experimental set-up
The test apparatus designed for this study is promising: by design no
load is applied when the coupon is affixed to the machine, and only tension
is applied during loading. This led to a very high success rate during testing
(no coupons were broken during handling or connection to the machine).
However, it is not clear how the ”loose” boundary conditions (compared to
the traditional wedge clamps) affect the fracture outcome. When a crack
initiates, the load distribution is the coupons changed suddenly and the
clamps are likely to reorient themselves, which could lead to premature
fracture. The same phenomenon could occur with plasticity. None of the
data collected in the current study allow for checking whether the clamps
accelerate the fracture process.
A non contact strain measurement system provided a non invasive method
to estimate strain in the coupon. With the current method, only the av-
erage strain was estimated, similar to what a strain gauge would measure.
This prevents from determining the link between the microstructure and
the strain field, and therefore the fracture mechanism. Finally, the strain
measurement procedure is 2D only, and consequently any relative motion of
one of the clamps towards or away from the camera was seen as a change in
strain, and therefore the measured strain could be overestimated (and the
Young’s modulus underestimated).
5. Conclusion
Predicting rib fracture based on the computational model of the thorax
remains an elusive challenge: although the contribution of the geometry to
the fracture mechanism has been demonstrated and included to some extent
in finite models of the thorax (Song et al., 2009; Li et al., 2011), the con-
tribution of the bone material properties has to be better described. The
16
intrinsic bone features that contribute to the onset of fracture are not well
comprised by the commonly used material models (isotropic and homoge-
neous). The paradigm that describes bone as two entities — trabecular or
cortical — needs to be revised to include a finer description of the bone
microstructure. The results presented in this paper supplemented with past
research highlight that the bone fracture mechanism is not well accounted
for with a linear elastic model, at least for the rib bone. The attempt made
in the current study to capture the bone microstructure by measuring the
effective cross-sectional area along the coupons length proved unsuccessful.
A more in-depth analysis of the bone microstructure (such as the direction
of the voids and pores in the bone cortical shell) and how it modifies the
strain field compared to the simplified approach that considers bone as elas-
tic linear is required. It would allow to determine how the microstructure
generates areas of weakness and strength to establish whether local strains
or stresses play a role in the fracture mechanism.
6. Acknowledgment
The authors would like to acknowledge the contribution of James Bolton,
Brian Overby amd Thomas Gochenour at the Center for Applied Biome-
chanics - University of Virginia for the design and fabrication of the test
fixture, and of Stacy Hollins for the microstructure analysis, and thank
Andrew Kemper and Stefan Duma from the Center for Injury Biomechan-
ics (Virginia Tech - Wake Forest, USA) for sharing their expertise in bone
coupon testing.
7. Conflict of interest
The authors declare that they do not have any conflict of interest in this
study.
Appendix A. Cross-sectional images from computed micro-tomography
17
Figure A.10:
(coupons tested under quasi-static loading). Only one every ten images is included.
Images of the cross-section obtained from micro-computed tomography
Figure A.11:
(coupons tested under dynamic loading). Only one every ten images is included.
Images of the cross-section obtained from micro-computed tomography
References
AAAM, 2008. Abbreviated Injury Scale (AIS) 2005 - Update 2008. Associ-
ation for the Advancement of Automotive Medicine.
18
Adharapurapu, R.R., Jiang, F., Vecchio, K.S., 2006. Dynamic fracture of
bovine bone. Materials Science and Engineering C 26, 1325–1332.
Berthet, F., Vezin, P., Ch`eze, L., Verriest, J.P., 2005. Assessment and anal-
ysis of the human rib lateral slopes. Computer Methods in Biomechanics
and Biomedical Engineering 8, 35–36.
Choi, H., Lee, I., 2009. Thorax FE model for older population, in: Japan
Society of Mechanical Engineering conference.
Crowninshield, R., Pope, M., 1974. The response of compact bone in tension
at various strain rates. Annals of Biomedical Engineering 2, 217–225.
Ferreira, F., Vaz, M., Simoes, J., 2006. Mechanical properties of bovine
cortical bone at high strain rate. Materials Characterization 57, 71–79.
Frank, S., Spolenak, R., 2008. Optical strain measurement by digital im-
age analysis. Matlab script developed at the Laboratory for Nanomet-
allurgy, Dept. of Materials, Zurich, Switzerland, available from Matlab
central (http://www.mathworks.com/matlabcentral/fileexchange/20438-
optical-strain-measurement-by-digital-image-analysis).
Gayzik, F.S., Yu, M.M., Danelson, K.a., Slice, D.E., Stitzel, J.D., 2008.
Quantification of age-related shape change of the human rib cage through
geometric morphometrics. Journal of biomechanics 41, 1545–54.
Hallman, J.J., Yoganandan, N., Pintar, F.A., 2010. Biomechanical and
Injury Response to Posterolateral Loading and Response to Posterolateral
Loading from from Torso Side Airbags Torso Side Airbags. Stapp car crash
journal 54, 227–257.
Hansen, U., Zioupos, P., Simpson, R., Currey, J., Hynd, D., 2008. The
effect of strain rate on the mechanical properties of human cortical bone.
J Biomech Eng 130.
Ito, O., Ohhashi, K., Dokko, Y., 2009. Development of Adult and Elderly
FE Thorax Skeletal Models. SAE Word Congress .
Keller, T., Mao, Z., Spengler, D., 1990. Young’ modulus, bending strength,
and tissu physical properties of human compact bone. Journal of or-
thopaedic research 8, 592–603.
19
Kemper, A.R., McNally, C., Kennedy, E.A., Manoogian, S.J., Rath, A.L.,
Ng, T.P., Stitzel, J.D., Smith, E.P., Duma, S.M., Matsuoka, F., 2005. Ma-
terial properties of human rib cortical bone from dynamic tension coupon
testing. Stapp Car Crash Journal 49, 199–230.
Kent, R., Lessley, D., Sherwood, C., 2004. Thoracic response to dynamic,
non-impact loading from a hub, distributed belt, diagonal belt, and double
diagonal belts. Stapp car crash journal 48, 495–519.
Kimpara, H., Iwamoto, M., Watanabe, I., Miki, K., Lee, J.B., Yang, K.H.,
King, A.I., 2006. Effect of assumed stiffness and mass density on the
impact response of the human chest using a three-dimensional FE model
of the human body. Journal of biomechanical engineering 128, 772–6.
Kimpara, H., Lee, J., Yang, K., King, A., Iwamoto, M., Watanabe, I., Miki,
K., 2005. Development of a three-dimensional finite element chest model
for the 5th percentile female. Stapp Car Crash Journal 49, 251–269.
Kindig, M., 2009. Tolerance to failure and geometric influences on stiffness
of human ribs under anteriorposterior loads. Master’s thesis. University
of Virginia.
Kuppa, S., Eppinger, R., 1998. Development of an improved thoracic injury
criterion, in: 42nd Stapp car crash conference proceedings, SAE. pp. 139–
154.
Lessley, D., Shaw, G., Parent, D., Arregui-Dalmases, C., Kindig, M., Riley,
P., Purtsezov, S., Sochor, M., Gochenour, T., Bolton, J., Subit, D., Cran-
dall, J., Takayama, S., Ono, K., Kamiji, K., Yasuki, T., 2010a. Whole-
body response to pure lateral impacts. Stapp car crash journal 54, 289–
336.
Lessley, D.J., Salzar, R., Crandall, J., Kent, R., Bolton, J.R., Bass, C.R.,
Guillemot, H., Forman, J.L., 2010b. Kinematics of the thorax under
dynamic belt loading conditions. International Journal of Crashworthiness
15, 175–190.
Li, Z., Kindig, M., Kerrigan, J., Untaroiu, C., Subit, D., Crandall, J., Kent,
R., 2010a. Rib fractures under anterior-posterior dynamic loads: Experi-
mental and finite-element study. Journal of Biomechanics 43, 228 – 234.
Li, Z., Kindig, M., Subit, D., Kent, R., 2010b. Influence of mesh density, cor-
tical thickness and material properties on human rib fracture prediction.
Medical Engineering & Physics 32, 998–1008.
20
Li, Z., Subit, D., Kindig, M., kent, R., 2011. Development and assessment
of a thorax finite element model of the 50th percentile male. Stapp Car
Crash Conference under review.
Liz´ee, E., Robin, S., Song, E., Bertholon, N., Le Coz, J., Besnault, B.,
Lavaste, F., 1998. Development of a 3d Finite Element Model of the
Human Body. 42nd Stapp Car Crash Conference Proceedings .
McElhaney, J.H., 1966. Dynamic response of bone and muscle tissue. J
Appl Physiol 21, 1231–1236.
Mohr, M., Abrams, E., Engel, C., Long, W.B., Bottlang, M., 2007. Ge-
ometry of human ribs pertinent to orthopedic chest-wall reconstruction.
Journal of Biomechanics 40, 1310 – 1317.
Murakami, D., Kobayashi, S., Torigaki, T., Kent, R., 2006. Finite element
analysis of hard and soft tissue contributions to thoracic response: sen-
sitivity analysis of fluctuations in boundary conditions. Stapp car crash
journal 50, 169–89.
Petitjean, A., Lebarb´e, M., Potier, P., Trosseille, X., Lassau, J.P., 2003.
Laboratory reconstructions of real world frontal crash configurations using
the hybrid III and THOR dummies and PMHS. Stapp Car Crash Journal
46, 27–54.
Pithioux, M., Subit, D., Chabrand, P., 2004. Comparison of compact bone
failure under two different loading rates: experimental and modelling ap-
proaches. Medical Engineering & Physics 26, 647 – 653.
Plank, G., Eppinger, R., 1989. Computed dynamic response of the human
thorax from a finite element model, in: Proc. of the 12th Int. Tech. Conf.
on the Enhanced Safety of Vehicles.
Robin, S., 2001. Human model for safetya joint effort towards the develop-
ment of refined human-like car occupant models, in: Proceedings of the
17th international technical conference on the enhanced safety vehicle.
Paper 297.
Ruan, J., El-Jawahri, R., Chai, L., Barbat, S., Prasad, P., 2003. Prediction
and analysis of human thoracic impact responses and injuries in cadaver
impacts using a full human body finite element model. Stapp Car Crash
Journal 47.
21
Saha, S., Hayes, W., 1974. Instrumented tensile-impact tests of bone. Ex-
perimental Mechanics 14, 473–478. 10.1007/BF02323147.
Saha, S., Hayes, W., 1976. Tensile impact properties of human compact
bone. Journal of biomechanics 9, 243–251.
Shigeta, K., Kitagawa, Y., Yasuki, T., 2009. Development of next generation
human fe model capable of organ injury prediction, in: Proc. 21st ESV
Conference. Paper 09-0111.
Song, E., Trosseille, X., Baudrit, P., 2009. Evaluation of thoracic deflection
as an injury criterion for side impact using a finite elements thorax model.
Stapp car crash journal 53, 155–91.
Subit, D., Chabrand, P., Masson, C., 2009. A micromechanical model to pre-
dict damage and failure in biological tissues. application to the ligament-
to-bone attachment in the human knee joint. Journal of Biomechanics 42,
261 – 265.
Trosseille, X., Baudrit, P., Leport, T., Vallancien, G., 2008. Rib cage strain
pattern as a function of chest loading configuration. Stapp car crash
journal 52, 205.
Vezin, P., Berthet, F., 2009. Structural characterization of human rib cage
behavior under dynamic loading. Stapp car crash journal 53, 93–125.
Vezin, P., Verriest, J.P., 2005. Development of a set of numerical human
models for safety, in: Proc. 19th ESV Conference. Paper 05-0163.
Wright, T., Hayes, W., 1977. Fracture mechanics parameters for compact
bone-effects of density and specimen thickness. J. Biomechanics 10, 419–
430.
Wright, T.M., , Hayes, W.C., 1976. Tensile testing of bone over a wide range
of strain rates: effects of strain rate, microstructure and density. Med.
Biol. Eng. 14, 671–680.
Yamada, H., Evans, F.G., 1970. Strength of biological materials. Edited by
F. Gaynor Evans. Williams & Wilkins, Baltimore,.
22
|
1504.00131 | 1 | 1504 | 2015-04-01T08:03:35 | Entrainment of heterogeneous glycolytic oscillations in single cells | [
"physics.bio-ph",
"q-bio.MN"
] | Cell signaling, gene expression, and metabolism are affected by cell-cell heterogeneity and random changes in the environment. The effects of such fluctuations on cell signaling and gene expression have recently been studied intensively using single-cell experiments. In metabolism heterogeneity may be particularly important because it may affect synchronisation of metabolic oscillations, an important example of cell-cell communication. This synchronisation is notoriously difficult to describe theoretically as the example of glycolytic oscillations shows: neither is the mechanism of glycolytic synchronisation understood nor the role of cell-cell heterogeneity. To pin down the mechanism and to assess its robustness and universality we have experimentally investigated the entrainment of glycolytic oscillations in individual yeast cells by periodic external perturbations. We find that oscillatory cells synchronise through phase shifts and that the mechanism is insensitive to cell heterogeneity (robustness) and similar for different types of external perturbations (universality). | physics.bio-ph | physics | Entrainment of heterogeneous
glycolytic oscillations in single cells
Anna-Karin Gustavsson, Caroline B. Adiels, Bernhard Mehlig, and Mattias Goksör*
Department of Physics, University of Gothenburg, SE-41296 Gothenburg, Sweden
*[email protected]
Cell signaling, gene expression, and metabolism are affected by cell-cell heterogeneity and random
changes in the environment. The effects of such fluctuations on cell signaling and gene expression
have recently been studied intensively using single-cell experiments. In metabolism heterogeneity
may be particularly important because it may affect synchronisation of metabolic oscillations, an
important example of cell-cell communication. This synchronisation is notoriously difficult to
describe theoretically as the example of glycolytic oscillations shows: neither is the mechanism of
glycolytic synchronisation understood nor the role of cell-cell heterogeneity. To pin down the
mechanism and to assess its robustness and universality we have experimentally investigated the
entrainment of glycolytic oscillations in individual yeast cells by periodic external perturbations. We
find that oscillatory cells synchronise through phase shifts and that the mechanism is insensitive to
cell heterogeneity (robustness) and similar for different types of external perturbations (universality).
Introduction
Cellular signaling, gene expression, and metabolism are determined by chemical reactions within the cell. The
discrete nature of molecular reactions as well as environmental fluctuations and heterogeneity cause fluctuations
in these processes. The effect of such noise on cell signaling and gene expression has recently been studied
intensively using single-cell analysis1-5. However, despite its importance the role of noise and heterogeneity in
metabolism6 is not yet well understood. Heterogeneity is very important in systems where cell-cell
communication may cause the cells to synchronise their metabolic oscillations. Cell-cell communication is
important because it is a prerequisite for organisation of cell communities and is necessary for evolution to
proceed from unicellular to multicellular behaviour.
One of the most intensively studied metabolic dynamics is that of glycolytic oscillations in yeast cells
Saccharomyces cerevisiae7-10, for a review see11. Remarkably, yeast cells in dense populations appear to be able
to synchronise their glycolytic oscillations9, 12-15 and macroscopic oscillations have been observed in yeast
populations7-10. This effect appears quite universal in that different signaling molecules/receptors can cause
synchronisation9, 12-19. It is thus expected that similar synchronisation mechanisms are at work in other oscillatory
cell types, possibly via different metabolic species. However, the empirical data pertain to dense cell cultures of
millions of cells and previous attempts to investigate the oscillatory behaviour in individual cells have proven
challenging8, 20, 21. Since observations of macroscopic oscillations do not distinguish between oscillation and
synchronisation such measurements neither allow to deduce the microscopic mechanism of synchronisation nor
to investigate the role of cell-cell variations.
There is a long history of theoretical investigations of oscillations in glycolytic networks22, 23 and it is now
well understood that sustained oscillations may occur under a wide range of conditions24-29. Synchronisation, by
contrast, appears to be much more difficult to achieve in computer simulations of model systems24-27. Whether or
not synchronisation occurs appears to very sensitively depend upon the model parameters24. This is at variance
with the apparent robustness of the phenomenon observed in dense cell cultures.
In order to understand the synchronisation mechanism and to elucidate the role of heterogeneity it is
necessary to quantify how individual cells respond to periodic perturbations of their environments, either due to
an externally applied perturbation or through interactions with other cells. To achieve a qualitative and
quantitative understanding of the phenomenon requires us to answer three fundamental questions.
First, what is the mechanism of synchronisation? Experimental studies of macroscopic oscillations indicate
that phase synchronisation may play a role15: Figure 3(A) in15 shows that the macroscopic phase shift due to a
1
sudden change in environmental conditions depends upon the value of the macroscopic phase just before the
perturbation, see also 9. In order to quantify the effect and to unequivocally establish whether the synchronisation
can be achieved by phase changes alone it is necessary to follow how an individual cell is entrained by a periodic
perturbation. One must determine whether and at which frequencies cells may continue to oscillate after the
perturbation has been switched off, and, most importantly, whether the amplitude changes during entrainment.
Last but not least an important question is how the frequency and amplitude of oscillation in the absence of
perturbations affect the propensity of the cell to be entrained when the periodic perturbation is switched on.
Existing experiments9, 14, 15 do not answer these questions because (i) only the macroscopic response is measured,
and (ii) the response is only measured at a single perturbation pulse. Some model calculations show that mainly
non-oscillatory cells become entrained by perturbations30. A very important open question is how the phase of an
entrained cell relates to the phase of the perturbation. Do cells typically oscillate in phase with the perturbation
or not? Theoretical models can show in-phase or out-of-phase entrainment, sensitively depending on model
parameters26. Macroscopic experiments do not allow resolving this question, because subpopulations oscillating
out-of-phase will only lead to a lowering of the amplitude of the macroscopic signal. In order to determine the
mechanism of synchronisation, these experiments must be performed on individual cells.
Second, how robust is the synchronisation mechanism? In a theoretical model for phase synchronisation, the
efficiency of the mechanism is determined by the heterogeneity of the cells as well as the strength of the
entrainment31. This is very important because no two cells are alike, and different cells respond differently to
external perturbations. In order to quantify how this heterogeneity may prevent synchronisation, we must study
the response of an ensemble of independent individual cells with different properties. Can a periodic perturbation
entrain cells that initially oscillate at substantially different frequencies? Theoretical analysis using a recent
model has shown that perturbations with the synchronising metabolite acetaldehyde (ACA) are not sufficient to
entrain cells when they have different frequencies32. Most importantly, in an ensemble of many cells, how large
fraction of the individual cells becomes entrained? To answer this question single-cell studies are required,
because it is impossible to distinguish between full and partial synchronisation in experiments that only measure
macroscopic properties. In order to determine how quickly the mechanism entrains it is necessary to investigate
responses of individual cells over many subsequent periods of the periodic perturbation. The degree of
synchronisation is quantified by computing an order parameter31.
Third, how universal is the synchronisation mechanism? Entrainment involves the entire glycolytic network,
and not just a single reaction or chemical species. The effect might in fact be the result of the combined response
to several different chemical species32. The kinetics leading to synchronisation is thus very complicated, but
entrainment appears to occur for a wide range of different conditions and types of perturbations9, 12-19. The
question is whether the mechanism leading to synchronisation is the same for all species.
In order to answer the questions raised above we report on the results of novel experiments subjecting
individual yeast cells to periodic perturbations. Using an experimental setup that combines an optical trap for cell
positioning with microfluidics for precise spatial and temporal control of the extracellular environment we
induce glycolytic oscillations in individual cells and investigate their response to periodic perturbations. These
experiments allow us to address the questions raised above.
To enforce strict periodicity of the perturbations, a microfluidic device is used to achieve complete and
reversible changes of the extracellular milieu. This makes it possible to experimentally investigate how the
phase, the frequency and the amplitude of an individual-cell oscillation are affected by periodic external
perturbations. Analysing these parameters from individual cells allows us to determine the mechanism of
entrainment, and thus to answer the first question raised above.
To answer the second question, we calculate the order parameter of the phases of the oscillations of the
individual cells. This reveals the degree of synchronisation. Studying individual cells allows us to determine
which fraction of cells becomes entrained by the perturbation. To ensure that no additional perturbations from
cell-cell interactions complicate the analysis, we position the cells far apart from each other by means of an
optical trap. Cell-cell interactions are further reduced by flushing out intercellular signaling agents by the flow in
the microfluidic chamber28.
To answer the third question we investigate through which biochemical reactions in the glycolytic reaction
network the perturbation causes entrainment. Starting out, we investigated the individual cell responses to
periodic addition of ACA to gain information about the single-cell responses at conditions previously studied in
populations13, 15 (see Supplementary Fig. S1 online). Strikingly, we found that removal of cyanide without any
2
addition of ACA caused a similar response, in contrast to the results shown in previous population studies9. In
this study we therefore chose to investigate the effect of periodic cyanide perturbations. Induction of oscillations
has mostly been achieved by addition of glucose and subsequent addition of 2-8 mM cyanide26, 33, 34. The most
common explanation for macroscopic oscillations to be detected only for this cyanide concentration range is
twofold. Firstly, in this range respiration is inhibited by cyanide binding to cytochrome c oxidase. Secondly, in
this range the ACA concentration is lowered to levels where the cells become sensitive to ACA secretion by
other cells9, 13, 15, 33-35. The reason is that cyanide binds ACA and forms lactonitrile36.
Results
Entrainment by cyanide removal. Alternating between a solution containing 20 mM glucose + 5 mM cyanide
and a solution containing only glucose, we show that periodic removal of cyanide entrains the oscillations of
individual cells in a similar way as addition of ACA (Fig. 1a and Supplementary Fig. S1 online). In control
experiments where only the flow rates in the microfluidic flow chamber are changed but the concentration of
chemicals in the solutions are kept constant, no entrainment can be seen (Fig. 2a). The results of the control
experiment are in agreement with a previous study, which showed that the oscillatory behaviour is affected only
by changes of chemicals and not by changes of flow rates in the microfluidic flow chamber 37.
Mechanism of synchronisation. To investigate the mechanism responsible for the observed entrainment, the
instantaneous phases were calculated (Figs. 1b and 2b) by means of a standard procedure based upon evaluating
the Hilbert transform of the signal38. To reveal temporal information about the synchronisation mechanism, the
phases before and 9 s after cyanide removal were plotted for perturbations 1, 2, 4 and 8 (Fig. 3a-d). Data from
experiments where entrainment is induced by cyanide removal is shown in red, while data from control
experiments is shown in black. The corrected phase shifts at the first perturbation were fitted with second-degree
polynomials (Fig. 4 and Table 1). The cells in the control experiments show the expected, constant phase shift,
independent of the phase of the oscillations at the perturbation. The cells exposed to the periodic cyanide
removal, on the other hand, show a significant phase dependence of the phase shift. The results show that cells
that have phases close to -π or π become negatively phase shifted, while cells that have phases close to that of the
perturbation remain relatively unaffected by the perturbation. Three important conclusions can thus be drawn
from these graphs; (1) there is a phase dependence on the phase shifts in the experiment where cells become
entrained (2) most of the entraining phase shifts occur during the first perturbations and at the consequent
perturbations the cells have already become entrained (3) there is no phase dependence on the phase shift in the
control experiment.
Phase synchronisation assumes that the physical properties of the entrained oscillator remain essentially
unchanged and that entrainment is due to phase shifts caused by the perturbation. To test whether the frequencies
of the oscillations become affected by the perturbations, the mean and standard deviation of the frequencies were
calculated before and after the perturbation interval (Fig. 6a). No significant difference could be seen between
the experiments where entrainment was induced and the control experiment, neither with regard to the actual
frequencies of the oscillations, nor with regard to the distribution of the frequencies. The amplitudes of the
oscillations also remained relatively unaffected by the perturbation for oscillatory cells. However, also non-
oscillatory cells became entrained by the perturbation (Fig. 5). For this response to be detectable, also the
amplitude must become affected. This indicates that phase changes are not sufficient to entrain non-oscillatory
cells. In contrast to results of model calculations26, under no circumstances could entrainment out-of-phase with
the periodic perturbation be observed, neither for oscillatory nor non-oscillatory cells.
Robustness of the synchronisation mechanism. The degree of synchronisation is quantified by the so-called
order parameter 𝑟(𝑡) (see Methods). It increases very rapidly in the perturbation interval and already within the
first minute of perturbations, the order parameter reaches a value close to unity, corresponding to complete
synchrony (Fig. 1c). In the absence of external perturbation the order parameter is expected to tend to zero,
assuming that there are no cell-cell interactions that could stabilise synchronisation. However, this is true only
provided that the number of cells 𝑁 used in the analysis is large. In general, we expect the order parameter to be
of the order of 1 √𝑁⁄
2c).
, corresponding well with the measured value in the absence of perturbations (Figs. 1c and
3
How quickly the order parameter decays after the periodic perturbation is switched off depends upon the
distribution of the frequencies of the individual cells. We expect that the order parameter decays as
𝑟(𝑡) = 𝑟0𝑒−∆𝜔𝑡, see Methods. This expectation is borne out by our experiments (Fig. 6b-d). Note also that since
the oscillations are initially induced simultaneously in all cells, it is expected that the order parameter should be
high at the start of the experiment and then decay if there are no cell-cell interactions (Figs. 1c and 2c). The
decay of the order parameter at the start of the experiment cannot, however, be well described simply by the
distribution of frequencies at the start of the experiment. It can instead be explained by a measured difference in
duration of the initial spike of NADH, causing cells with the same frequency to become out of phase already
within the first oscillation period.
Universality of the synchronisation mechanism. It remains to determine how cyanide can work as a
synchronising agent and whether this mechanism is likely to be universal. There are several possible
explanations as to why removal of cyanide may have a synchronising effect. One reason could be that respiration
is no longer inhibited when cyanide is removed. To cause the observed response, which occurs within a few
seconds, it would be necessary that cyanide be released very rapidly from its interaction with cytochrome c
oxidase. To investigate whether respiratory reactions are responsible for the synchronisation, the experiment was
repeated by changing from glucose/cyanide to glucose using a hypoxic glucose solution. However, also here
entrainment could be seen (data not shown). This indicates that the main cause of entrainment at cyanide
removal is not due to respiratory reactions. However, care has to be taken when performing experiments using
hypoxic media, since yeast cells can respire also in very oxygen poor environments. Even low concentrations of
oxygen in the hypoxic medium could cause cell responses when cyanide is removed.
A second reason for the response could be that the pH is altered when cyanide is added. In previous studies,
Poulsen et al. induced oscillations using 4 mM KOH and hypoxic solutions instead of cyanide35. In their studies,
the addition of KOH caused a similar pH increase as the addition of cyanide, raising the pH value 0.06 units.
Measurements of the pH of our solutions showed that addition of 5 mM cyanide increased the pH value with
0.09 units. To determine whether the cells respond to such a change in pH, the measurement was repeated with
perturbations with a hypoxic glucose solution without cyanide, where the pH was adjusted using KOH. This
experiment showed that the cells still became entrained by a change from glucose/cyanide to glucose at hypoxic
conditions, also when the pH was kept constant (data not shown). It thus seems unlikely that the pH difference is
causing the response at cyanide removal.
A third reason for the observed synchronisation could be cyanide binding to intracellular ACA. The phase
dependence of the phase shift differs for different chemical species13. Our data reveals that the phase response to
cyanide removal is comparable to previous studies of synchronisation by ACA addition in populations15 and in
single cells (Supplementary Fig. S1 online), while population studies of perturbations with e.g. glucose reveals a
different phase relation9. This supports the hypothesis that the removal of cyanide causes a similar response as
addition of ACA.
Discussion
The experiments reported here show that the mechanism behind the synchronisation of individual oscillatory
yeast cells is phase synchronization rather than frequency or amplitude modulation. Our data show that the
strongest phase response is achieved when the concentration of NADH is at a minimum, where the removal of
cyanide prolongs the subsequent period roughly by one third. Oscillatory cells continue to oscillate with
frequencies and frequency distributions largely unaffected by the periodic perturbation. Also the amplitude of the
oscillations remains relatively unaffected by the perturbations. These results support the paradigm originally put
forward by Kuramoto and his collaborators (for a review see39): that periodic perturbations can entrain oscillators
by affecting just the phases of the oscillators, resulting in a robust and universal mechanism. Theoretical models
tend to be very sensitive to changes in parameter values and using different parameters can result in in- or out-of-
phase synchronisation of the oscillations. In contrast to some model results26, our results show no cases of
entrainment out-of-phase, neither for oscillatory nor non-oscillatory cells.
Our experiments show that non-oscillatory cells also become entrained by external periodic perturbations, so
these cells must also contribute to macroscopic oscillations observed in e.g. Ref. 40. However, in contrast to the
4
entrainment of oscillatory cells, which relies solely on phase synchronisation, entrainment of non-oscillatory
cells requires a change of the amplitude (as well as the frequency). In this case the entrainment cannot be
explained solely in terms of phase shifts.
The graphs in Figure 3 reveal temporal information about the entrainment of the individual cells, where the
clustering of phases can be seen to occur already during the first few perturbations. This is consistent with the
rapid increase of the order parameter (Fig. 1c). Heterogeneity is present in the phase response of individual cells,
but despite the heterogeneity the majority of oscillatory cells eventually become entrained by the perturbation.
The final phase distribution is within 20% of the original distribution. This shows that the synchronisation
mechanism is very robust with regard to cell heterogeneity. This result is in stark contrast with results of model
calculations24, 25, 30.
Our results indicate that cyanide can work as a synchronising agent due to its ability to bind intracellular
ACA. Removal of cyanide will then have the same effect as addition of ACA, where the response is transduced
to the first part of glycolysis via NADH/NAD+ and ATP/ADP9, 40, 41. Here it is thus impossible to completely
separate the cell responses due to cyanide removal and ACA addition. Different externally applied chemical
species thus appear to synchronise individual cells by similar mechanisms, giving further support to the
hypothesis that the synchronisation mechanism is universal. Intracellular cyanide reactions are rather complex.
That cyanide can act in more ways than by binding ACA and inhibit respiration has been indicated in previous
studies, where the oscillatory tendency was stronger when cyanide was added than under hypoxic conditions34
and under hypoxic conditions where ACA was flushed away by a flow42. And indeed, recently Hald et al.
showed that cyanide also reacts with other metabolites, namely pyruvate and dihydroxyacetone phosphate
(DHAP), and that cyanide might affect the behavior of glycolytic oscillations in more ways than just by binding
ACA and inhibiting respiration36. Previous studies have also shown that cyanide causes longer trains of
oscillations than other inhibitors of respiration, such as antimycin A and azide43, 44 and that oscillations disappear
if both cyanide and azide are present35. The role of cyanide inhibiting respiration by binding to cytochrome c
oxidase and the contribution of respiratory reactions to the oscillatory behavior have recently been discussed by
Schrøder et al., who found that oscillations disappear for strains with deletions of a gene coding for subunit VI of
cytochrome c oxidase45. This finding was unexpected, since the general assumption is that cytochrome c oxidase
is completely inhibited at cyanide concentrations used for oscillation studies and should not contribute at all to
the oscillatory behavior of the cells. Cyanide reactions with glucose has, however, been shown to be negligible
in the pH range of our experiments36. These results highlight the need for further studies to fully understand
intracellular cyanide reactions.
In summary, we report on experimental results that allow us to determine the mechanism, the robustness, and
the universality of synchronisation of glycolytic oscillations in yeast cells. Our results pose challenges to
theoretical modeling, namely to reproduce our results by simulations of the dynamics of glycolytic reaction
networks, and to explain the robustness and universality of the synchronisation mechanism. A successful model
must show (1) the phase responses to periodic perturbations reported in this work, (2) the robustness of phase
synchronisation also for the cell heterogeneity reported here, and (3) the universality of the mechanism; a wide
range of periodic perturbations entrain the oscillations. Our experiments are the first that show entrainment of
individual yeast cells by periodic perturbations. They support the paradigm originally put forward by Kuramoto:
that periodic perturbations can entrain oscillators by affecting just the phases of the oscillators, resulting in a
robust and universal mechanism.
Methods
Cell preparation. Yeast cells (Saccharomyces cerevisiae, X2180), were grown in a medium comprising 10 g/l
glucose, 6.7 g/l yeast nitrogen base and 100 mM potassium phthalate (pH 5.0) on a rotary shaker at 30°C until
glucose was exhausted, as described in46. The cells were harvested by centrifugation (3500 g for 5 min), washed
twice in 100 mM potassium phosphate buffer (pH 6.8) and subsequently glucose starved in the buffer solution on
a rotary shaker for 3 h at 30°C. The cells were then washed in the buffer solution and stored at 4°C until use.
Experimental procedure. The experimental setup and procedure was described in detail in42. The cells were
introduced into a microfluidic flow chamber with four inlet channels42 and positioned at the bottom of the
chamber in arrays of 5x5 cells with a cell-cell distance of 10 µm using optical tweezers42, 47, 48. By positioning the
5
cells far apart, cell-cell interactions by e.g. ACA was avoided28. The flows in the channels were laminar and
solutions from different channels could only mix by diffusion. The extracellular environment in the cell array
area could thus be controlled both spatially and temporally by adjusting the flow rates in the different inlet
channels, causing the cell array area to be covered by the chemicals from the intended channel. A complete
change of environment was achieved within 4 s of a change in flow rates. All channels contained 100 mM
potassium phosphate buffer (pH 6.8) and the cells were introduced in the lowest channel. Glycolytic oscillations
were induced by a 20 mM glucose + 5 mM KCN solution, introduced in the second highest channel for 10 min
before the periodic perturbations started and for 15 min after the perturbations had ended. Entrainment was
investigated by periodically introducing the solution of interest in the top channel for 20 s and the
glucose/cyanide solution for 20 s, giving a total period time of the perturbation of 40 s. The perturbations
continued for 15 periods (10 min). For perturbations with ACA, a solution comprising 20 mM glucose + 5 mM
KCN + 1 mM ACA was used. For experiments performed with hypoxic solutions, the solutions were bubbled
through with N2-gas for 10 min before the experiment. The cell responses were studied by imaging the NADH
autofluorescence from the individual cells using an epi-fluorescence microscope (DMI6000B, Leica
Microsystems) and an EM-CCD camera (C9100-12, Hamamatsu Photonics) , where images were acquired every
other second42.
Data analysis. The acquired images were processed and analysed as described previously42. The resulting
signals of the NADH fluorescence intensity of individual cells were then analysed using MATLAB (The
MathWorks, Inc.). Running averages of the NADH signals were calculated using a window of approximately
two periods (55 data points) and were subtracted from the NADH signals to reduce spurious trends and drift
(Supplementary Fig. S2 online). This facilitated extraction of the phases of the oscillations. The instantaneous
phases of the oscillations of the individual cells, 𝜙𝑛(𝑡), were extracted from the NADH signals by means of a
standard procedure that is based upon evaluating the Hilbert transform of the signals in MATLAB 38. The phases
are defined in the interval [-π, π] and a phase of 0 represents a maximum in the concentration of NADH (Figs. 1b
and 2b). In Figures 1 and 2, the ten cells that showed the most stable oscillations before the perturbation where
chosen. The phases before, 𝜙𝑏𝑒𝑓𝑜𝑟𝑒 , and 9 s after cyanide removal, 𝜙𝑎𝑓𝑡𝑒𝑟, were plotted for perturbations 1, 2, 4
and 8, for the experiments where entrainment was induced (Fig. 3a-d, 𝑁=20, red dots) and for control
experiments where only the flow rates were changed but not the chemicals in the solutions (Fig. 3a-d, 𝑁=32,
black dots). The 9 s delay was chosen to allow the cells time to respond to the perturbation.
The oscillation frequency of the individual cells,
, was calculated in time intervals 0-5, 5-10 min and 20-
25 min as the frequency at the maximum peak of the Fourier transforms of the NADH signals multiplied by 2π,
and is given as mean ± standard deviation for experiments where entrainment was induced (Fig. 6a, 𝑁=24, red
bars) and for the control experiments (Fig. 6a, 𝑁=20, black bars).
The phase shifts of the data sets at the first perturbation were calculated as ∆𝜙 = 𝜙𝑎𝑓𝑡𝑒𝑟 − 𝜙𝑏𝑒𝑓𝑜𝑟𝑒. If
∆𝜙 > 𝜋, ∆𝜙 was adjusted with ±2π to move it into the interval [-π, π]. The phase shifts were then corrected by
subtracting the expected phase shifts of each cell, 𝜔𝑛∆𝑡, where ∆𝑡 is the time between the measurement of the
phase before and after the perturbation. This corrected phase shift was fitted by a second-degree polynomial on
the form:
∆𝜙 − 𝜔𝑛∆𝑡 = 𝑐2𝜙2 + 𝑐1𝜙 + 𝑐0, (1)
where the parameter values are estimated as mean values and their uncertainty (due to the spread of data points
and the finite sample size) is expressed in terms of their 95% confidence interval (Fig. 4 and Table 1).
The degree of synchronisation was characterised by the order parameter 𝑟(𝑡), defined as31, 49
𝑟(𝑡) = ∑
𝑁
𝑛=1
𝑒𝑖𝜙𝑛(𝑡)
, (2)
where 𝑁 is the total number of cells in the experiment (𝑁=10 in Figs. 1, 2 and 6c-d, and 𝑁=14 in Fig. 6b). An
order parameter close to unity indicates a high degree of synchronisation, while an order parameter close to zero
indicates large heterogeneity in phases among the individual cells and thus low entrainment by the external,
periodic perturbation.
6
n When the cells are independent and there is no external perturbation, the order parameter is expected to decay
as
𝑟(𝑡) = 𝑟0𝑒−∆𝜔𝑡, (3)
where 𝑟0is the order at 𝑡 = 0 and ∆𝜔 is the standard deviation of 𝜔𝑛. The frequencies were calculated in time
interval 20-25 min and the decay was set to start 18 s after the end of the last perturbation (Fig. 6b-d).
References
Paulsson, J. Summing up the noise in gene networks. Nature 427, 415-418 (2004).
Newman, J.R.S. et al. Single-cell proteomic analysis of S. cerevisiae reveals the architecture of
biological noise. Nature 441, 840-846 (2006).
Coulon, A., Chow, C.C., Singer, R.H. & Larson, D.R. Eukaryotic transcriptional dynamics: from single
molecules to cell populations. Nat. Rev. Genet. 14, 572-584 (2013).
Ribrault, C., Sekimoto, K. & Triller, A. From the stochasticity of molecular processes to the variability
of synaptic transmission. Nat. Rev. Neurosci. 12, 375-387 (2011).
Frank, S.A. Evolution of robustness and cellular stochasticity of gene expression. PLOS Biol. 11,
e1001578 (2013).
Kiviet, D.J. et al. Stochasticity of metabolism and growth at the single-cell level. Nature 514, 376-379
(2014).
Aldridge, J. & Pye, E.K. Cell density dependence of oscillatory metabolism. Nature 259, 670-671
(1976).
Chandra, F.A., Buzi, G. & Doyle, J.C. Glycolytic oscillations and limits on robust efficiency. Science
333, 187-192 (2011).
Danø, S., Sørensen, P.G. & Hynne, F. Sustained oscillations in living cells. Nature 402, 320-322
(1999).
Goldbeter, A. Computational approaches to cellular rhythms. Nature 420, 238-245 (2002).
Richard, P. The rhythm of yeast. FEMS Microbiol. Rev. 27, 547-557 (2003).
Bier, M., Bakker, B.M. & Westerhoff, H.V. How yeast cells synchronize their glycolytic oscillations: a
perturbation analytic treatment. Biophys. J. 78, 1087-1093 (2000).
Danø, S. et al. Synchronization of glycolytic oscillations in a yeast cell population. Faraday Discuss.
120, 261-276 (2001).
Danø, S., Madsen, M.F. & Sørensen, P.G. Quantitative characterization of cell synchronization in yeast.
Proc. Natl. Acad. Sci. USA 104, 12732-12736 (2007).
Richard, P., Bakker, B.M., Teusink, B., van Dam, K. & Westerhoff, H.V. Acetaldehyde mediates the
synchronization of sustained glycolytic oscillations in populations of yeast cells. Eur. J. Biochem. 235,
238-241 (1996).
Boiteux, A., Goldbeter, A. & Hess, B. Control of oscillating glycolysis of yeast by stochastic, periodic,
and steady source of substrate: a model and experimental study. Proc. Natl. Acad. Sci. USA 72, 3829-
3833 (1975).
Markus, M., Kuschmitz, D. & Hess, B. Chaotic dynamics in yeast glycolysis under periodic substrate
input flux. FEBS Lett. 172, 235-238 (1984).
Reijenga, K.A. et al. Control of glycolytic dynamics by hexose transport in Saccharomyces cerevisiae.
Biophys. J. 80, 626-634 (2001).
7
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Winfree, A.T. Oscillatory glycolysis in yeast - pattern of phase resetting by oxygen. Arch. Biochem.
Biophys. 149, 388-401 (1972).
Aon, M.A., Cortassa, S., Westerhoff, H.V. & van Dam, K. Synchrony and mutual stimulation of yeast
cells during fast glycolytic oscillations. J. Gen. Microbiol. 138, 2219-2227 (1992).
Poulsen, A.K., Petersen, M.O. & Olsen, L.F. Single cell studies and simulation of cell-cell interactions
using oscillating glycolysis in yeast cells. Biophys. Chem. 125, 275-280 (2007).
Goldbeter, A. & Lefever, R. Dissipative structures for an allosteric model. Application to glycolytic
oscillations. Biophys. J. 12, 1302-1315 (1972).
Selkov, E.E. Self-oscillations in glycolysis. Eur. J. Biochem. 4, 79-86 (1968).
du Preez, F.B., van Niekerk, D.D., Kooi, B., Rohwer, J.M. & Snoep, J.L. From steady-state to
synchronized yeast glycolytic oscillations I: model construction. FEBS J. 279, 2810-2822 (2012).
Hald, B.O. & Sorensen, P.G. Modeling diauxic glycolytic oscillations in yeast. Biophys. J. 99, 3191-
3199 (2010).
Hynne, F., Danø, S. & Sørensen, P.G. Full-scale model of glycolysis in Saccharomyces cerevisiae.
Biophys. Chem. 94, 121-163 (2001).
Wolf, J. & Heinrich, R. Effect of cellular interaction on glycolytic oscillations in yeast: a theoretical
investigation. Biochem. J. 345, 321-334 (2000).
Gustavsson, A.-K. et al. Allosteric regulation of phosphofructokinase controls the emergence of
glycolytic oscillations in isolated yeast cells. FEBS J. 281, 2784-2793 (2014).
Teusink, B., Bakker, B.M. & Westerhoff, H.V. Control of frequency and amplitudes is shared by all
enzymes in three models for yeast glycolytic oscillations. Biochim. Biophys. Acta. 1275, 204-212
(1996).
Hald, B.O., Hendriksen, M.G. & Sorensen, P.G. Programming strategy for efficient modeling of
dynamics in a population of heterogeneous cells. Bioinformatics 29, 1292-1298 (2013).
Shinomoto, S. & Kuramoto, Y. Phase-transitions in active rotator systems. Prog. Theor. Phys. 75, 1105-
1110 (1986).
Serizawa, H., Amemiya, T. & Itoh, K. Glycolytic synchronization in yeast cells via ATP and other
metabolites: mathematical analyses by two-dimensional reaction-diffusion models. Natural Science 6,
719-732 (2014).
Kloster, A. & Olsen, L.F. Oscillations in glycolysis in Saccharomyces cerevisiae: the role of
autocatalysis and intracellular ATPase activity. Biophys. Chem. 165, 39-47 (2012).
Richard, P. et al. Yeast-cells with a specific cellular make-up and an environment that removes
acetaldehyde are prone to sustained glycolytic oscillations. FEBS Lett. 341, 223-226 (1994).
Poulsen, A.K., Lauritsen, F.R. & Olsen, L.F. Sustained glycolytic oscillations - no need for cyanide.
FEMS Microbiol. Lett. 236, 261-266 (2004).
Hald, B.O., Smrcinova, M. & Sørensen, P.G. Influence of cyanide on diauxic oscillations in yeast.
FEBS J. 279, 4410-4420 (2012).
Gustavsson, A.-K., Adiels, C.B. & Goksör, M. Induction of sustained glycolytic oscillations in single
yeast cells using microfluidics and optical tweezers. Proc. SPIE 8458, 84580Y-1-84580Y-7 (2012).
Rosenblum, M.G., Pikovsky, A.S. & Kurths, J. Phase synchronization of chaotic oscillators. Phys. Rev.
Lett. 76, 1804-1807 (1996).
Strogatz, S.H. From Kuramoto to Crawford: exploring the onset of synchronization in populations of
coupled oscillators. Physica D 143, 1-20 (2000).
8
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
De Monte, S., d'Ovidio, F., Danø, S. & Sørensen, P.G. Dynamical quorum sensing: population density
encoded in cellular dynamics. Proc. Natl. Acad. Sci. USA 104, 18377-18381 (2007).
Wolf, J. et al. Transduction of intracellular and intercellular dynamics in yeast glycolytic oscillations.
Biophys. J. 78, 1145-1153 (2000).
Gustavsson, A.-K. et al. Sustained glycolytic oscillations in individual isolated yeast cells. FEBS J. 279,
2837-2847 (2012).
Aon, M.A. et al. Dynamic regulation of yeast glycolytic oscillations by mitochondrial functions. J. Cell.
Sci. 99, 325-334 (1991).
Ghosh, A.K., Chance, B. & Pye, E.K. Metabolic coupling and synchronization of NADH oscillations in
yeast cell populations. Arch. Biochem. Biophys. 145, 319-331 (1971).
Schrøder, T.D., Özalp, V.C., Lunding, A., Jernshøj, K.D. & Olsen, L.F. An experimental study of the
regulation of glycolytic oscillations in yeast. FEBS J. 280, 6033-6044 (2013).
Richard, P., Teusink, B., Westerhoff, H.V. & van Dam, K. Around the growth phase transition S.
cerevisiae's make-up favours sustained oscillations of intracellular metabolites. FEBS Lett. 318, 80-82
(1993).
Ashkin, A., Dziedzic, J.M., Bjorkholm, J.E. & Chu, S. Observation of a single-beam gradient force
optical trap for dielectric particles. Opt. Lett. 11, 288 (1986).
Fällman, E. & Axner, O. Design for fully steerable dual-trap optical tweezers. Appl. Opt. 36, 2107-2113
(1997).
Pikovsky, A., Rosenblum, M. & Kurths, J. Synchronization: A Universal Concept In Nonlinear
Sciences. (Cambridge University Press, Cambridge; 2001).
Acknowledgements
We acknowledge the financial support from the Swedish Research Council to M.G. and to B.M., the Seventh
Framework Programme UNICELLSYS to M.G., Stiftelsen Längmanska kulturfonden to A.-K.G, and from the
Göran Gustafsson Foundation for Research in Natural Science and Medicine to B.M..
Author contributions
A.-K.G planned and performed the experiments, analysed the data and wrote the manuscript. All authors
discussed the results and commented on the manuscript. C.B.A, B.M. and M.G provided guidance throughout.
Additional information
Supplementary information accompanies this paper at http://www.nature.com/scientificreports
Competing financial interests: The authors declare no competing financial interests.
How to cite this article: Gustavsson, A.-K., Adiels, C.B., Mehlig, B. & Goksör, M. Entrainment of
heterogeneous glycolytic oscillations in single cells. Sci. Rep. 5, 9404; DOI:10.1038/srep09404 (2015).
9
Figure 1 Entrainment of oscillations by removal of cyanide. (a) Graph showing the NADH fluorescence
intensity from individual cells (cell index in box), where each trace is offset 90 units for clarity. Oscillations are
induced at time 0 of the experiment by exposing the cells to a solution containing 20 mM glucose + 5 mM
cyanide. After 10 min, cell entrainment by removal of cyanide is investigated by periodically changing the flow
rates in the inlet channels of the microfluidic flow chamber, causing a periodic change between the
glucose/cyanide solution and a solution containing only 20 mM glucose with a period time of 40 s for a total of
15 periods. The time intervals during which the cells are exposed to cyanide are marked in red. As can be seen,
the amplitude of the oscillations remains relatively unaffected by the perturbation. (b) Graph showing the phases
of the oscillations of the individual cells shown in (a). Here it can be seen that the phases are shifted as the
oscillations become entrained by the perturbation. (c) Graph showing the time dependent order parameter,
calculated from the phases shown in (b). An order parameter close to unity indicates synchronisation, while a
low value indicates a large heterogeneity of the phases. Here it is thus clear that the cells become entrained by
the periodic removal of cyanide through phase shifts.
10
Figure 2 No entrainment of oscillations by flow rate changes. (a) Graph showing the NADH fluorescence
intensity from individual cells (cell index in box), where each trace is offset 90 units for clarity. Oscillations are
induced at time 0 of the experiment by exposing the cells to a solution containing 20 mM glucose + 5 mM
cyanide. After 10 min, cell entrainment is investigated by periodically changing the flow rates in the inlet
channels of the microfluidic flow chamber with a period time of 40 s for a total of 15 periods, while keeping the
concentration of the chemicals in the solutions constant. The time intervals during which the flow rates are
changed are marked in red. As can be seen, the amplitude of the oscillations remains unaffected by the
perturbation. (b) Graph showing the phases of the oscillations of the individual cells shown in (a). Here no shift
of the phases can be seen when the flow rates are changed. (c) Graph showing the time dependent order
parameter, calculated from the phases shown in (b). An order parameter close to unity indicates synchronisation,
while a low value, as shown here, indicates large heterogeneity of the phases.
11
Figure 3 Phases before and after perturbations. Graphs showing the phases 9 s after cyanide removal/flow
change as a function of the phases just before the removal/flow change for perturbations 1, 2, 4 and 8 in panel
(a)-(d) respectively. Data from experiments where cells were entrained by cyanide removal is shown in red
(𝑁=20), while data from control experiments where only the flow rates were changed is shown in black (𝑁=32).
In panel (a) it can be seen that the phases before the first perturbation are evenly distributed for both data sets. In
panels (b)-(d), the phases of cells in the control experiments remain evenly distributed, while the distribution of
phases of cells experiencing cyanide removal narrows to about 20% of the original distribution after perturbation
8. The graphs show that synchronising phase shifts are induced in the experiment where cyanide is removed and
that most of the synchronising shifts occur at the first perturbations. For the control experiment, on the other
hand, no phase shift dependence on the phase can be seen.
12
Figure 4 Fits of corrected phase shifts. Graph showing the phase shifts calculated from the phases shown in
Figure 3a, corrected with the expected phase shift if no entrainment by the perturbations occur, 𝜔𝑛∆𝑡, where 𝜔𝑛
is the frequency of the individual cells and ∆𝑡 is the time between the measurements of the phases before and
after the perturbation. The two data sets are fitted to second-degree polynomials (see equation (1) and Table 1 for
parameter values), which reveals a significant difference between the values of the quadratic terms. Cells that
have phases close to -π or π become negatively phase shifted when cyanide is removed, while cells which have
phases close to that of the perturbation remain relatively unaffected by the perturbation.
13
Figure 5 Entrainment of non-oscillatory cells. Graph showing the NADH fluorescence intensity of a single
cell during the same experimental conditions as shown in Figure 1. Even though the cell is non-oscillatory for 20
mM Glc + 5 mM KCN, it still becomes entrained when cyanide is periodically removed (shown in red), and can
thus contribute to a synchronised macroscopic response to perturbations. However, in contrast to the entrainment
of oscillatory cells, which relies on phase synchronisation, entrainment of non-oscillatory cells requires a change
also of the amplitude.
14
Figure 6 Heterogeneity of the frequency of the oscillations and its influence on the decay of the order
parameter. (a) Graph showing the frequency of the oscillations expressed as mean ± standard deviation, where
data from experiments where cells were entrained is shown in red (𝑁=24), while data from control experiments
is shown in black (𝑁=20). No significant difference between the two data sets can be seen, neither with respect
to the mean values of the frequency nor to their distribution. That entrainment occurs despite the large
heterogeneity of phases indicate that the synchronisation mechanism is robust. In (b)-(d) the decay of the order
parameter is shown for three different experiments where cells were entrained (solid lines, 𝑁=14, 10 and 10
respectively), together with exponential equations of the decay, where the rates are described by the distribution
of frequencies ∆𝜔 of the oscillations in the respective experiments (dashed lines, see equation (3)). The decay
was set to start 18 s after the end of the experimental perturbation at 20 min. It can be seen that the decay rate of
the order parameter well can be estimated by the frequency distribution among the individual cells in the
experiments.
15
Table 1 Results of polynomial fit of the corrected phase shifts: ∆𝜙 − 𝜔𝑛∆𝑡 = 𝑐2𝜙2 + 𝑐1𝜙 + 𝑐0 at cyanide
removal (20 cells) and for control experiments without cyanide removal (32 cells) see also Figure 4 and
Methods. The constant term (𝑐0) and the linear slope (𝑐1) do not significantly differ from zero for either of the
data sets. However, the most interesting parameter is the quadratic term (𝑐2), which does not significantly differ
from zero for the control experiments, but has a significant non-zero value for the experiments where cells were
entrained.
Parameter estimates: mean value (95% confidence interval)
𝑐2
𝑐1
𝑐0
Cyanide removal
-0.18 (-0.27, -0.09)
0.07 (-0.07, 0.21)
0.34 (-0.01, 0.78)
Control
-0.02 (-0.06, 0.02)
0.04 (-0.02, 0.12)
0.08 (-0.12, 0.27)
16
|
1012.2444 | 1 | 1012 | 2010-12-11T10:41:56 | The effects of bio-fluid on the internal motion of DNA | [
"physics.bio-ph",
"q-bio.BM"
] | The internal motions of DNA immersed in bio-fluid are investigated. The interactions between the fragments of DNA and the surrounding bio-fluid are modeled using the gauge fluid lagrangian. In the model, the bio-fluid is coupled to the standard gauge invariant bosonic lagrangian describing the DNA. It is shown that at non-relativistic limit various equation of motions, from the well-known Sine-Gordon equation to the simultaneous nonlinear equations, can be constructed within a single framework. The effects of bio-fluid are investigated for two cases : single and double stranded DNA. It is argued that the small and large amplitudes of a single stranded DNA motion immersed in bio-fluid can be explained in a natural way within the model as a solitonic wave regardless with the fluid velocity. In contrary the double stranded DNA behaves as regular or damped harmonic oscillator and is highly depending on the fluid velocity. | physics.bio-ph | physics |
The effects of bio-fluid on the internal motion of
DNA
A. Sulaimana,b∗ and
L.T. Handokoc,d†
a)Department of Physics, Bandung Institute of Technology1, Jl. Ganesha 10, Bandung
b)P3 TISDA BPPT2, BPPT Bld. II (19th floor), Jl. M.H. Thamrin 8, Jakarta 10340,
40132, Indonesia
c)Group for Theoretical and Computational Physics, Research Center for Physics,
Indonesian Institute of Sciences3, Kompleks Puspiptek Serpong, Tangerang 15310,
Indonesia
d)Department of Physics, University of Indonesia4, Kampus UI Depok, Depok 16424,
Indonesia
Indonesia
Abstract
The internal motions of DNA immersed in bio-fluid are investigated. The
interactions between the fragments of DNA and the surrounding bio-fluid are
modeled using the gauge fluid lagrangian. In the model, the bio-fluid is cou-
pled to the standard gauge invariant bosonic lagrangian describing the DNA.
It is shown that at non-relativistic limit various equation of motions, from the
well-known Sine-Gordon equation to the simultaneous nonlinear equations,
can be constructed within a single framework. The effects of bio-fluid are in-
vestigated for two cases : single and double stranded DNA. It is argued that
the small and large amplitudes of a single stranded DNA motion immersed
in bio-fluid can be explained in a natural way within the model as a solitonic
wave regardless with the fluid velocity. In contrary the double stranded DNA
behaves as regular or damped harmonic oscillator and is highly depending on
the fluid velocity.
Keywords : elementary biomatter; biomatter structure; biomatter interaction; DNA;
modeling
∗Email : [email protected], [email protected]
†Email : [email protected], [email protected]
1http://www.fi.itb.ac.id
2http://tisda.bppt.go.id
3http://teori.fisika.lipi.go.id
4http://www.fisika.ui.ac.id
1
Introduction
Both deoxyribo- and ribo-nucleic acid (DNA and RNA) have been recognized as
the most important biomolecules. Especially DNA helical structures undergo a
very complex dynamics which plays several important roles in various biological
phenomena such as storage of information, inheritance (replication, etc) and the
usage of genetic information (transcription, etc). The importance of biopolymers
like DNA/RNA is motivated by established observations that the homologous re-
combination is preceded by recognition and local pairing of intact double stranded
DNA fragments, rather than involving known recombination proteins. Therefore, it
should be attributed to direct DNA-DNA interactions whose physical origin has not
been understood [1, 2]. Experimentally, the physical properties of DNA/RNA have
been measured in many works, for example : the DNA single-molecule [3, 4, 5], dou-
ble stranded DNA forming bubbles [6], the DNA/RNA nucleoside and nucleotides
[7], the structural transitions of DNA through torques measurements [8], the ther-
modynamic fluctuations of DNA in a reacting system [9], the stretching DNA with
a receding meniscus [10], the electrical transport through single DNA molecules [11]
and so forth.
From physical point of view, a biopolymer like DNA molecule is considered as
a system consisting of many interacting matters in a particular configuration of
space-time. Some models treats this kind of DNA dynamics as the phenomena
of nonlinear excitations like soliton. This type of models has been pioneered by
Englander et.al. using nonlinear dynamics relevant to the transcription process in
terms of coupled pendulum chain which generates the sine-Gordon equation and
its classical solitons [12]. Further, Davydov described the alpha helices in quantum
solitons [13]. Following these suggestions, a number of models for the nonlinear DNA
have been elaborated in the last decades, in both classical or quantum approaches
[14, 15, 16]. A typical classical approach is the so-called PDB model which takes into
account twisted DNA molecules [17, 18, 19]. On the other hand, there are several
models based on the particle interactions [20, 21, 22]. Also, the polyelectrolyte
model which treats DNA molecule as a cylinder with a net charge homogeneously
distributed along its surface, and has further been modified to be the electrostatic
zipper motif for DNA aggregation [23], to solve high dependency of the electrostatic
interaction between DNA duplexer on surface charge patterns [24].
It has also been shown that under particular external conditions the DNA molecules
form a double helix, and its (transverse, longitudinal and torsional) motions can be
divided into two main regions : the small and large amplitude of internal motions
[25]. The small amplitude of motion can be described by the hamiltonian of harmonic
oscillator. On the other hand, the large amplitude is described by a non-harmonic
one [26]. Recently, many works have discussed and arrived at the conclusion that
the large amplitude of internal motion can be considered as a nonlinear dynamical
system where solitary conformational waves can be excited [14]. Then nonlinear in-
teraction between molecules in DNA gives rise to a very stable excitation as soliton
[26, 27].
As mentioned above, DNA is not motionless. It is in a constantly wriggling dy-
2
namics state in a medium of bio-organic fluid in the nucleus cell [28]. However, the
motion of DNA surrounded by fluid is rarely studied. Previous studies are usually
done by solving the fluid equations and its wave equations simultaneously using
appropriate boundary conditions. On the other hand, in the Hamiltonian formula-
tion the viscous force is considered to be comparable with other forces arising from
Hamiltonian [29, 30]. The solution is then obtained by expansion and performing
order-by-order calculation. In these approaches, anyway the picture of interaction
between DNA and its surrounding fluid is not clear. Also, in most models the over-
damped DNA dynamics are treated by putting some additional terms by hand in
the differential equation to obtain the non-homogenous ones [31]. The stochastic
simulations of DNA in flow has been done for a fully parametrized beadspring chain
model by taking into account the fluctuating hydrodynamic interactions [32].
In this paper, a new model to describe various internal motions of DNA inspired
by gauge fluid theory is proposed. The DNA dynamics is modeled as the result
of interactions among matters in a fluid medium using the relativistic and gauge
invariant fluid lagrangian. Although the theory is a relativistic one, we take its
non-relativistic limit at the final stage to deal with problems in DNA as done in
some previous works, for instance in some models using the ideal gas approximation
[33]. Moreover, the lagrangian is intended for physics at scale of order transport
mean free paths, that is the transition region where neither a hydrodynamics nor
kinetic theory is valid. Therefore it fits the current interest of modeling "elementary"
biomatters like DNA. Just to mention, the lagrangian is originally devoted to model
the quark gluon plasma (QGP) as a relativistic fluid system [34, 35, 36, 37], inspired
by the similarity between the dynamical properties of fluid and electromagnetic field
[38, 39]. The DNA is treated as strongly coupled system like non-Abelian plasmas
where neither a hydrodynamics nor kinetic theory is really valid. Within the model,
a single and double stranded biopolymers are described in a general way as the
results of interactions among the fluid and matter fields.
We show in two specific cases how to derive the equation of motion (EOM) and
investigate the internal motions through its solutions and behaviors. From the EOM
of a DNA as a single bulk, we argue that small and large amplitude regions of the
internal motion of DNA are determined by its internal dynamics and interactions
with surrounding fluid. On the other hand, in the case of double stranded DNA the
EOM is solved analytically to investigate the effects of fluid velocity to its internal
motion.
The paper is organized as follows. First we briefly introduce the theory of gauge
invariant fluid lagrangian, and then provide the allowed interactions within the
model. After explaining how to model DNA using the interactions in the lagrangian,
we provide two typical examples : 1) the Abelian U(1) case to model the dynamics
of a single bulk of DNA, and 2) the non-Abelian SU(2) case to describe the internal
motion of double stranded DNA. Finally, the paper is ended with summary and
discussion.
3
2 Theoretical background
Here, a new approach to investigate the interaction between biopolymer and its
surrounding bio-fluid is discussed using the lagrangian method. Rather putting it
by hand, the interaction is described in a more natural way from first principle, i.e.
by introducing some symmetries in the lagrangian under consideration.
2.1 The lagrangian
The model is an extension of the original model based on the U(1) gauge theory
devoted for QGP as a magnetofluid system [34, 35, 36]. Thereafter it has been
extended to the non-Abelian case to accommodate a system with many matters,
either bosonic or fermionic ones [37]. Concerning the fact that an (elementary)
matter has no intrinsic degree of freedom like spin, it is considerable to represent its
elementary constituents as scalar (boson) fields governed by the bosonic lagrangian,
Lmatter = (∂µΦ)† (∂µΦ) + V (Φ) ,
where V (Φ) is the potential. For example in the typical Φ4−theory,
V (Φ) = −
1
2
m2
Φ Φ†Φ −
1
4!
λ (Φ†Φ)2 ,
(1)
(2)
where mΦ and λ are the mass of matter and the dimensionless coupling constant of
matter self-interaction. The hermite conjugate is Φ† ≡ (Φ∗)T for a general complex
field Φ.
We impose the above bosonic lagrangian to be gauge invariant under local
(in general non-Abelian) gauge transformation [40, 41], U ≡ exp[−iT aθa(x)] ≈
1 − iT aθa(x) with θa ≪ 1. T a's are generators belong to a particular Lie group
and satisfy certain commutation relation [T a, T b] = if abcT c with f abc is the anti-
symmetric structure constant [42]. The matter field is then transformed as Φ U−→
Φ′ ≡ exp[−iT aθa(x)] Φ, with T a are n × n matrices while Φ is an n × 1 multiplet
containing n elements, i.e.
and ΦT = (Φ1 Φ2 · · · Φn) ,
(3)
U−→ U a
µ
for n dimension Lie groups as SU(n), O(n + 1), etc. It is well-known that the sym-
metry in Eq. (1) is revealed by introducing gauge fields Aa
µ which are transformed
as U a
µ, and replacing the derivative with the
µ
covariant one, Dµ ≡ ∂µ + ig T aU a
µ. Anyway, the number of generators, and also
gauge bosons, is determined by the dimension of group under consideration. For an
SU(n) group one has n2 − 1 generators and the index a runs over 1, 2, · · · , n2 − 1.
g (∂µθa) + f abcθbU c
′ ≡ U a
µ − 1
Φ1
Φ2
...
Φn
Φ =
4
For example the SU(2) group is realized by 2 × 2 matrices T a ≡ 1
Pauli matrices [42],
2σa with σa are the
σ1 =(cid:18) 0 1
1 0 (cid:19) , σ2 =(cid:18) 0 −i
0 (cid:19) , σ3 =(cid:18) 1
0 −1 (cid:19) ,
0
i
(4)
In particular, the Abelian U(1) case is revealed by putting T aθa(x) → θ(x), i.e. the
phase transformation, respectively.
Finally, the gauge invariance leads to the total lagrangian with some additional
terms in the lagrangian to keep its gauge invariance,
L = Lmatter + Lgauge + Lint ,
where,
Lgauge = −
1
4
µνSaµν ,
Sa
(5)
(6)
(7)
.
The strength tensor is Sa
is,
µ U bµ
Lint = −gJ a
ν − ∂νU a
µν ≡ ∂µU a
µU aµ + g2(cid:0)Φ†T aT bΦ(cid:1) U a
µ = −i(cid:2)(∂µΦ)†T aΦ − Φ†T a(∂µΦ)(cid:3) .
µ + gf abcU b
µU c
J a
ν , while the 4-vector current
(8)
The coupling constant g then represents the interaction strength between gauge
field and matter. We should note that, the current conservation is realized by the
covariant current ∂µJ a
µ = 0 with J a
The gauge boson Uµ is interpreted as a "fluid field" with velocity uµ, and takes
µ ≡ −i(cid:2)(DµΦ)†T aΦ − Φ†T a(DµΦ)(cid:3) [37].
U a
µ = (U a
0 , Ua) ≡ ua
µ φ ,
(9)
the form [34, 35, 36, 37],
with,
uµ ≡ γa(1, −va) ,
(10)
where φ is an auxiliary boson field, while γa ≡ (1 − va2)−1/2. Here we adopt the
natural unit, i.e. the light speed c = 1. Eq. (10) is nothing else than rewriting a
gauge field in terms of its polarization vector and wave function which represents
the fluid distribution in a system. It has further been shown that the non-relativistic
fluid equation can be reproduced using Eq. (9) [35, 37]. This fact actually justifies
us to model the DNA dynamics in a fluid medium using the total lagrangian in Eq.
(5).
Now we are ready to model the DNA using the above lagrangian. First, we
should investigate the allowed interactions in the present theory.
2.2 The interactions
In order to be specific, let us consider the Φ4−potential in Eq.
(2) for matter
lagrangian in Eq. (1). With a complete lagrangian at hand, we can extract m−point
interactions for fluid and matter with m is the number of relevant legs involved in
an interaction. We list all allowed interactions below for each element in the matter
multiplet denoted by the indices i, j.
5
• 2−point interactions :
The interactions arise through the kinetic and mass terms of matter in Eqs.
(1) and (2), and the fluid kinetic term in Eq. (6),
ΦΦ :
(∂µΦ∗
i ) (∂µΦi) −
m2
Φ Φ∗
i Φi .
1
2
ν − ∂νU a
1
4(cid:0)∂µU a
µ(cid:1) (∂µU aν − ∂νU aµ) .
UU : −
• 3−point interactions :
These interactions are induced by the fluid self-interaction in Eq. (6) and the
fluid-matter interaction in Eq. (7),
ΦΦU :
UUU :
• 4−point interactions :
ig T a
ijh(∂µΦ∗)i Φj − Φ∗
U cν(cid:0)∂µU a
g f abcU bµ
1
2
i (∂µΦ)ji U aµ .
ν − ∂νU a
µ(cid:1) ,
(13)
(14)
These interactions are induced through the matter self-interaction in Eq. (1),
the fluid kinetic term in Eq. (6) and the fluid-matter interaction in Eq. (7),
(11)
(12)
(15)
(16)
(17)
ΦΦΦΦ : −
UUUU : −
λ (Φ∗
i Φi)2 ,
1
4!
1
g2 f abcf adeU bµ
4
ΦΦUU : g2 Φ∗
i (cid:0)T aT b(cid:1)ij ΦjU a
µU e
ν ,
U cνU d
µ U bµ
.
All of these constitute the so-called Feynman diagrams and its order of mag-
nitudes that will be used soon in the subsequent section. Now we are ready to
construct the models relevant for biopolymers.
3 Modeling the DNA
Here, we consider two typical examples on how to describe various dynamics of DNA
within the present model. First we present a model for a single bulk of DNA or a
fragment of DNA molecule like nucleotide or nucleoside. Further we construct a
more complicated picture for the double stranded DNA. The model is a new type
of the mesoscale model of DNA that reduces the complexity of a nucleotide to three
interactions sites [21].
3.1 Single bulk of DNA : the Abelian U(1) model
The Abelian U(1) lagrangian involves only a single matter and a fluid field. In this
case, the 3−point interaction in Eq. (14) and the 4−point interaction in Eq. (16)
vanish.
It is also clear that we are not able to construct a realistic model for a
6
biopolymer composed by several different matters in this case [22]. However, we
can model the dynamics of a single bulk of DNA or its fragment like nucleoside
which could be considered as a composite field of sugar and base. This means we
investigate the internal dynamics of namely DNA molecules through the EOM of
its fragments and study the basic behaviors.
The total lagrangian in this case becomes,
L = (∂µΦ∗) (∂µΦ) −
1
2
m2
Φ Φ∗Φ −
1
4!
λ (Φ∗Φ)2 + g2 UµU µ Φ∗Φ
−
1
4
(∂µUν − ∂νUµ) (∂µU ν − ∂νU µ) + ig U µ [(∂µΦ∗) Φ − Φ∗ (∂µΦ)] , (18)
using Eqs. (2) and (5)∼(8). Imposing the variational principle of action and the
Euler-Lagrange equation in term of Φ [42],
∂L
∂Φ
− ∂µ
∂L
∂ (∂µΦ)
= 0 ,
we find the EOM for a single matter as follow,
(cid:0)∂2 + m2
Φ + 2g2 U 2(cid:1) Φ +
1
3!
λ Φ3 = 0 .
for a real Φ field.
(19)
(20)
This result leads to a solitonic wave equation for λ 6= 0 described by the well-
known nonlinear Klein-Gordon equation,
(cid:0)∂2 + ¯m2
Φ(cid:1) Φ +
1
3!
λ Φ3 = 0 ,
(21)
Φ ≡ m2
Φ sin Φ = 0 using sin Φ ≈ Φ− 1
with ¯m2
the 'level of non-linearity' for the Klein-Gordon equation. If one puts λ ≈ ¯m2
arrive at the sine-Gordon equation in 4−dimensional space-time (t, x), ∂2
t Φ − ∂2
¯m2
models based on the coupled pendulum chains pioneered by Englander et.al.
However, we should note that the equality λ ≈ ¯m2
since λ and ¯m2
that special case, let us solve Eq. (21) in a general way.
Φ + 2g2 U 2, and U 2 = φ2 from Eqs. (9) and (10). Here λ determines
Φ, we
xΦ −
3! Φ3 +· · ·. This kind of equation often appears in the
[12].
Φ in this model doesn't make sense
Φ have different dimensions. In this paper, rather than considering
For the sake of simplicity, we consider a traveling wave in 2−dimensional space-
x′Φ
time (t, x), i.e. Φ(x′) ≡ Φ(x − Ct), where C is a phase velocity. Since ∂2
and ∂2
x′Φ, Eq. (21) can be rewritten as,
t Φ = C 2∂2
xΦ = ∂2
x′Φ + m2
∂2
Φ Φ + λ Φ3 = 0 ,
(22)
Φ ≡ ¯m2
Φ/(C 2 − 1) and λ ≡ λ/[3!(C 2 − 1)]. Assuming that vx = v is a
with m2
constant makes mΦ to also be a constant. Hence we can multiply both sides of Eq.
(22) with ∂x′Φ to obtain,
∂x′(cid:20)(∂x′Φ)2 + m2
Φ Φ2 +
1
2
λ Φ4(cid:21) = 0 .
7
(23)
4
3
Φ
2
1
0
0
2
x'
4
6
Figure 1: The solitonic wave function for a 2-dimensional DNA as a function of x′
with the coupling constants g = 0.1 (solid line) and g = 0 (dashed line) for a fixed
parameter set (mΦ, φ, C, λ) = (1, 1, 2, 4).
Concerning that the quantum wave function Φ has the Gaussian distribution, it is
integrable and then leads to the following differential equation,
(∂x′Φ)2 + m2
Φ Φ2 +
λ Φ4 = 0 .
1
2
(24)
Through standard mathematical procedures, we can straightforwardly get the solu-
tion,
Φ(x′) = mΦr 2
λ
sech ( mΦ x′) ,
(25)
for λ > 0, or C > 1.
The non-relativistic limit can be obtained by performing a transformation t →
τ ≡ it in Eq. (21), and putting γ → 1 respectively. This leads to the same result as
Eq. (25), but t is replaced with −iτ . The behavior of this solitonic wave function
is depicted in Fig. 1 as a function of x′ with (solid line) and without (dashed
line) surrounding fluid for a fixed parameter set. Anyway, the fluid contribution is
independent on its velocity v, since the effective mass m2
Φ is shifted by U 2 = φ2. From
the figure, we can conclude that the large and small amplitudes can be considered
as the effects of fluid surrounding the DNA.
3.2 Double stranded DNA : the non-Abelian SU(2) model
Now let us apply the present lagrangian in a more realistic case of double stranded
DNA. Concerning the smallest group beyond U(1), we take the SU(2) group to
8
construct the model. In this group, we have 2 sub-matters in a doublet of matter
field as Eq. (3) with n = 2.
Since we have only 2 different states of matter, Φ1 and Φ2, it is convincing to split
the nucleotide to be a phosphate and a nucleoside consisting of sugar and base. So,
the interaction between two nucleotides, which further form the backbone of DNA
molecule, is attributed to the interaction of two different matters, i.e. phosphate and
nucleoside. On the other hand, the base pair is revealed as the interaction between
two identical matters, i.e. two neighboring nucleosides belonging to different strands.
The model is schematically illustrated in Fig. 2 where we have assigned Φ1 for
the nucleosides and Φ2 for the phosphates. Following the allowed interactions in
Eqs. (11)∼(17), we can easily estimate the order of magnitudes for each interaction
relevant to Fig. 2 as listed in Fig. 3.
From Fig. 3, it is straightforward to deduce that I1 bound is materialized by
vertex A, while vertex B is responsible for I2 and I3 bounds. Anyway, we should note
that there are another diagrams with spring loops in the vertices A and B, however
they would be vanishing due to the anti-symmetric f abc. Now, we unfortunately
face a problem on distinguishing I2 with I3 in Fig. 2. It is quite natural to consider
I3 must be larger than I2, since the backbone should be rather strongly tied and
rigid than the nucleotide. Therefore in order to resolve this problem we propose an
additional contribution to I3 coming from interacting fluid (either fluid absorption or
emission) with matters depicted in vertex D of Fig. 3. Of course, so I1 could receive
additional contribution from vertex C too. This scenario could be understood in
the following way. Since the backbone is more open to surrounding fluid than the
phosphate−nucleoside encaged in the nucleotide bound-state, its interaction with
surrounding fluid would contributes more significantly, and then should be taken
into account.
We might remark that in the present case the nucleoside, consisting of sugar and
base, should be considered as a well-confined bound-state. So we are not going into
insight to investigate its structure. In consequence of this, we can not distinguish the
A−T (adenine−thymine) with the G−C (guanine−cytosine) base pairs. Although
in principle, these might be explained using multi-loops gauge boson exchanges
inside nucleosides, and two different base pairs are attributed to the fluid velocities
in the fluid loops (the second diagram of vertex A in Fig. 3) with opposite sign.
However we postpone this point in this paper since it would require larger group
like SU(3) containing more matter fields. Anyway, the opposite torsional motions
of neighboring strands forming a DNA molecule can be explained, at time being,
qualitatively by considering the surrounding fluid in both strands have the same
velocities (v) but with opposite sign each other.
Now, we investigate the EOM in SU(2) as done in Sec. 3.1. Substituting the full
lagrangian, Eqs. (5)∼(7), into Eq. (19), we obtain for each element of matter,
∂2Φ1 + m2
Φ Φ1 − 2g (∂µU µ
2 ) Φ2 +
∂2Φ2 + m2
Φ Φ2 + 2g (∂µU µ
2 ) Φ1 +
1
3!
1
3!
2 (∂µΦ2) = 0 , (26)
2 (∂µΦ1) = 0 , (27)
for real fields Φi (i : 1, 2). Eqs. (4), (26) and (27) immediately yield the following
1 + Φ2
1 + Φ2
2(cid:1) Φ1 − 4g U µ
2(cid:1) Φ2 + 4g U µ
λ (cid:0)Φ2
λ (cid:0)Φ2
9
2Φ
I2
I3
I3
2Φ
I2
2Φ
I2
1Φ
1Φ
1Φ
I1
I1
I1
1Φ
1Φ
1Φ
2Φ
I2
2Φ
I2
2Φ
I2
I3
I3
sugar
phosphate
base
nucleoside
Figure 2: The double stranded DNA in the non-Abelian SU(2) model with nucle-
osides and phosphates are represented by Φ1 and Φ2. The vertices I1, I2 and I3
denote different types of interactions connecting nucleosides (Φ1 − Φ1) manifest-
ing base pairs in neighboring strands, nucleoside−phosphate (Φ1 − Φ2) within a
nucleotide, and nucleoside−phosphate (Φ1 − Φ2) between nucleotides in a strand.
EOM,
(28)
Φ − 4ig σ2U µ
(cid:0)∂2 + m2
1
3!
2 ∂µ(cid:1) Φ +
λ (cid:0)ΦT ΦΦ(cid:1) = 0 ,
for constant fluid velocity and φ. Comparing this result with Eq. (21), contribution
from the interacting fluid medium also appears in the third term but it contributes
differently. Using Eqs. (9) and (10) we arrive at non-relativistic limit,
∂2
τ Φ + ∂2
xΦ − m2
ΦΦ + 4gσ2 φ (∂τ Φ ∓ i v · ∂xΦ) −
1
3!
λ (cid:0)ΦΦT Φ(cid:1) = 0 ,
(29)
for v2 = v. This is the nonlinear EOM governing the double stranded DNA dynam-
ics with surrounding fluid medium in the present theory. The plus and minus signs
show the dynamics of a strand and its counterpart surrounded by the fluids with
opposite velocities.
Obviously, in contrast with the U(1) case it is hard to solve Eq. (29) exactly.
For the sake of simplification, let us consider 2-dimensional (t, x) case of Eq. (28),
∂2Φ
∂t2 −
∂2Φ
∂x2 + αt
∂Φ
∂t
− αx
∂Φ
∂x
+ m2
ΦΦ +
λ
3!
Φ3 = 0 ,
(30)
where αt ≡ −4igσ2γ φ and αx = 4igσ2γ vx φ. Borrowing the traveling wave Φ(x′) ≡
Φ(x − Ct) as before we obtain,
d2Φ
dx′2 − α
dΦ
dx′ + m2
ΦΦ + λΦ3 = 0 ,
(31)
10
i
+
2
O(g )
i
i
+
i
λO( )
i
i
i
i
i
A
B
C
D
i
j
i
j
=
=
=
=
i
i
i
i
2O(m )
2O(g )
O(g)
O(g)
i
j
i
j
Figure 3: The 2−point interactions and its first order contents relevant for double
stranded DNA in Fig. 2 and its order of magnitudes. The plain and spring lines
indicate matter and fluid fields.
with α ≡ (Cαt + αx)/(C 2 − 1), m2
λ = 0 it coincides with the equation of inharmonic oscillator, Eq. (25).
Φ/(C 2 − 1) and λ ≡ λ/(3!(C 2 − 1)). For
Φ ≡ m2
For further simplification, we assume that λ is small enough such that the last
term in Eq. (31) can be treated perturbatively. Then, we can expand the mass mΦ
in term of λ, i.e. mΦ ≃ mΦ0 + λ mΦ1, and Φ ≃ Φ0 + λΦ1 up to O(λ) accuracy. Now
we are ready to solve Eq. (31) order by order.
The lowest order with respect to λ, i.e. O(1), satisfies the following equation,
d2Φ0
dx′2 − α
dΦ0
dx′ + m2
Φ0Φ0 = 0 .
(32)
Following the standard mathematical procedures the solution can in general be
expressed in the form of Φ0 = N +
0 ek+x′ + N −
Therefore, the solution of Eq. (32) is simply,
2(cid:16)α ±q α2 − 4 m2
Φ0(cid:17).
0 ek−x′ with k± = 1
Φ0 = 2N0 e
α
2 x′
×
,
(33)
for α2 = 4 m2
Φ0
− m2
Φ0x′! for α2 > 4 m2
x′! for α2 < 4 m2
α2
4
Φ0
Φ0
1
4
cosh r α2
sin r m2
Φ0 −
11
by putting the normalization factor to be N +
0 ≡ N0. Each solution is corre-
sponding to the over-damped, damped and regular harmonic oscillators respectively.
0 = N −
Subsequently, the next order, i.e. O(λ), is governed by the equation,
d2Φ1
dx′2 − α
dΦ1
dx′ + m2
Φ0Φ1 + 2 mΦ0 mΦ1Φ0 + Φ3
0 = 0 .
The over-damped Φ0 in Eq. (33) yields the general solution for Φ1 should be,
Φ1 = N +
1 e
α
2 x′
+ N −
1 e
3 α
2 x′
.
(34)
(35)
Substituting the over-damped Φ0 and Eq.
−(32N 3
for any N +
Finally,
0 )/(cid:0)3 α2 + 4 m2
1 since α2 = 4 m2
Φ0(cid:1). In the present case the first term in Eq. (35) is vanishing
Φ0. This also leads to the result mΦ1 = 0 since mΦ0, N0 6= 0.
(35) into Eq.
(34), we obtain N −
1 =
Φ1 = −
32N 3
0
3 α2 + 4 m2
Φ0
3 α
2 x′
e
,
(36)
and mΦ = mΦ0.
In the second case of damped Φ0, we make use of the equality cosh3 x = 1/2 cosh(3x)+
cosh x to obtain,
d2Φ1
dx′2 − α
dΦ1
dx′ + m2
2 x′(cid:18)2 +
3 α
+4N 3
0 e
Φ0Φ1 + 4N 3
0 e
mΦ0 mΦ1
N 2
0
3 α
2 x′
cosh 3r α2
e− αx′(cid:19) cosh r α2
4
4
− m2
− m2
Φ0x′!
Φ0x′! = 0 .
(37)
Since cosh x < cosh(3x) and the last term is enhanced only by a factor of as small
as 2, Eq. (37) can be approximately reduced to be,
d2Φ1
dx′2 − α
dΦ1
dx′ + m2
Φ0Φ1 + 4N 3
0 e
3 α
2 x′
cosh 3r α2
4
− m2
Φ0x′! ≃ 0 .
(38)
The solution is given by,
Φ1 = e
3 α
2 x′"N +
1 cosh 3r α2
4
− m2
Φ0x′! + N −
1 sinh 3r α2
4
− m2
Φ0x′!# .
(39)
Again substituting it into Eq. (38) yields,
N +
1 = −
N −
1 = −
9
4
9
8
N 3
0
N 3
0
α2 − 3 m2
Φ0
3 α4 + 36 m4
Φ0 − 20 α2 m2
Φ0
αq α2 − 4 m2
Φ0
3 α4 + 36 m4
Φ0 − 20 α2 m2
Φ0
,
.
(40)
(41)
From these results, for x′ > 0 we can safely omit the sub-dominant sinh term in Eq.
(39), also because N +
1 since α2 > 4 m2
1 > 2N −
Φ0. Hence,
Φ1 = −
9
4
N 3
0
α2 − 3 m2
Φ0
3 α4 + 36 m4
Φ0 − 20 α2 m2
Φ0
3 α
2 x′
e
cosh 3r α2
4
− m2
Φ0x′! .
(42)
12
2
1
Φ
0
-1
-2
0
10
20
x'
30
40
Figure 4: The wave function for 2-dimensional double stranded DNA as functions
of x′ with vx = 1.9 (solid line) and vx = 1.7 (dashed line) for a fixed parameter sets
(mΦ0, φ, C, λ, g, N0) = (1, 1, 2, 4, 1, 1).
The last case of regular harmonic oscillator is governed by the following equation,
d2Φ1
dx′2 − α
dΦ1
dx′ + m2
2 x′(cid:18)3 +
3 α
+2N 3
0 e
3 α
2 x′
sin 3r m2
e− αx′(cid:19) sin r m2
Φ0 −
Φ0 −
α2
4
α2
4
x′!
x′! = 0 ,
(43)
Φ0Φ1 − 2N 3
0 e
mΦ0 mΦ1
N 2
0
using the relation 4 sin3 x = 3 sin x − sin(3x). In contrary with previous cases the
general solution for Eq. (43) is complicated. So, let us assume here that the more
rapid oscillation term, i.e. the fourth term, is dominant than the last one which
reduces the equation to be,
d2Φ1
dx′2 − α
dΦ1
dx′ + m2
Φ0Φ1 − 2N 3
0 e
3 α
2 x′
sin 3r m2
Φ0 −
α2
4
x′! = 0 ,
Hence the general solution is simply,
Φ1 = e
3 α
2 x′"N +
1 cos 3r m2
Φ0 −
α2
4
x′! + N −
1 sin 3r m2
Φ0 −
α2
4
x′!# .
Following similar procedures as before,
N +
1 =
N −
1 =
4
3
8
3
N 3
0
N 3
0
αq4 m2
Φ0 − α2
Φ0 − 16 α2 m2
Φ0
3 α4 + 36 m2
α2 − 3 m2
Φ0
3 α4 + 36 m2
Φ0 − 16 α2 m2
Φ0
13
,
.
(44)
(45)
(46)
(47)
In non-relativistic case, by definition the condition mΦ0 > 2g(C − vx)φ should be
fulfilled. Obviously, for large enough mΦ0 ( α2 ≪ 4 m2
Φ0 or mΦ0 ≫ 2g(C − vx)φ) the
solution is dominated by N −
1 term, while both terms are comparable for α2 → 4 m2
Φ0
or mΦ0 → 2g(C − vx)φ. These arguments lead to the result,
Φ1 =
4
3
N 3
0
αq4 m2
Φ0 − α2
Φ0 − 16 α2 m2
Φ0
3 α4 + 36 m2
3 α
2 x′
e
cos 3r m2
Φ0 −
α2
4
x′! .
(48)
We should remark that up to the current accuracy there is no need in all cases to
calculate the leading order of mass, mΦ1. As a typical example, the wave function
for the harmonically oscillating, i.e. the sum of Eqs. (33) and (48), double stranded
DNA is depicted in Fig. 4 as a function of x′ for certain velocities. It can also be
seen that the oscillation is sensitive to the fluid velocity.
4 Summary and discussion
We have introduced a new type of model to describe DNA using the gauge invariant
fluid lagrangian. The lagrangian is able to accomodate various internal motions of
DNA, from the single bulk to the double stranded of DNA as done in the preceeding
section. The EOM's and its solutions for two typical cases using the Abelian U(1)
and non-Abelian SU(2) lagrangians have been derived and investigated.
In the case of Abelian U(1) lagrangian, we have seen from Eq. (25) that the
interacting fluid medium characterized by the coupling constant g influences the
magnitude and also the width (associated to the dispersion or steppening rate) of
solitonic wave equation as well, but regardless with the fluid velocity. On the other
hand, obviously the matter self-interaction represented by its coupling constant λ
could change only the magnitude and not the dispersion or steppening rate of soliton.
Actually, this provides a natural explanation for small and large amplitude regions
of the internal motion of a single bulk of DNA immersed in bio-fluid without adding
any new terms by hand as done in some previous works [43]. Furthermore, that
contribution shifts the matter mass mΦ to be ¯mΦ. This is the so-called running
mass induced by the dynamical fluctuation of internal kinematics in the system
as a result of interaction between matter and fluid. However, the result is again
independent on the fluid velocity.
In the second case, using the non-Abelian SU(2) lagrangian we have constructed a
model for double stranded DNA in detail up to the level of its constituents, except for
sugar and base composing the nucleoside. It has been shown that the EOM follows
a similar form as in the U(1) model, but the interacting fluid medium contributes
in different way. The model requires that the DNA polymer would exist if and only
if it resides in a fluid medium, represented by I2 and I3 bounds realized by fluid-
matter interactions. Otherwise, the binding interactions I2 and I3 would vanish and
the strands are broken. These results could be used to explain the deformation of
DNA molecules associated with vanishing interactions in I1,2,3. In contrast with the
previous case, the fluid velocity plays an important role and changes the dynamics
drastically, namely the highly damped, damped and regular harmonic oscillators.
14
This supports a conclusion obtained in [30], that is the effect of hydrodynamic
interactions on the dynamics of DNA translocation depends on the fluid velocity.
As mentioned earlier, both strands in a double stranded molecule are considered
to follow the same EOM as Eq.
In
contrary, the single fragments of DNA belong to those strands are governed by Eq.
(21) independent on the fluid velocity, and should behave identically no matter with
the directions of its surrounding fluid velocities.
(29) with opposite sign of fluid velocities.
Further studies can be done using the lattice gauge simulation to calculate
numerically, for instance the finite temperature partition function density Z =
exp(1/TR d3xL). This kind of macroscopic ensemble provides direct relation be-
tween the internal dynamics of DNA and some physical observables like tempera-
ture and so on. Actually this is the main advantage of deploying the gauge invariant
lagrangian like the present one. Such numerical calculations would be able to sim-
ulate quantitatively some phenomena in DNA like critical temperature or pressure
related to the deformation of DNA molecules, etc. For example, one can investigate
the critical temperature as a double stranded DNA is splitted into single strands
[44], i.e. I1 → 0 in the present model. Such studies are in the progress.
Acknowledgment
We greatly appreciate fruitful discussion with T.P. Djun throughout the work. AS
thanks the Group for Theoretical and Computational Physics LIPI for warm hos-
pitality during the work. This work is partially funded by the Indonesia Ministry
of Research and Technology and the Riset Kompetitif LIPI in fiscal years 2009 and
2010 (Contract no. 11.04/SK/KPPI/II/2009 and 11.04/SK/KPPI/II/2010).
References
[1] S.M. Burges, N. Kleckner and B.M. Weiner, Genes Dev. 13, 1627-1641 (1999).
[2] B.M. Weiner and N. Kleckner, Cell 77, 977-991 (1994).
[3] S.B. Smith, L. Finzi and C. Bustamante, Science 258, 1122-1126 (1992).
[4] R. Lavery, A. Lebrun, J.F. Allemand, D. Bensimon and V. Croquette, J. Phys.:
Condensed Matter 14, R383-R414 (2002).
[5] T.R. Strick, M.N. Dessinges, G. Charvin, N.H. Dekker, J.F. Allemand, D. Ben-
simon and V. Croquette, Rep. Prog. Phys. 66, 1-45 (2003).
[6] G. Altan-Bonnet, A. Libchaber and O. Krichevsky, Phys. Rev. Lett. 90, 138101
(2003).
[7] J. Peon and A.H. Zewall, Chem. Phys. Lett. 348, 255-262 (2001).
[8] Z. Bryant, M.D. Stone, S.B. Smith. N.R. Cozzarelli and C. Bustamante, Nature
424, 338-341 (2003).
15
[9] D. Magde, E. Elson and W.W. Webb, Phys. Rev. Lett. 29, 705-708 (1972).
[10] D. Bensimon, A.J. Simon, V. Croquette and A. Bensimon, Phys. Rev. Lett. 74,
4754-4757 (1995).
[11] D. Porath, A. Bezryadin, S. de Vries and C. Dekker, Nature 403, 635 (2000).
[12] S.W. Englander, N.R. Kallenbach, A.J. Heeger, J.A. Krumhansl and A. Litwin,
Proc. Natl. Acad. Sci. USA 77, 72227226 (1980).
[13] A.S. Davydov, Solitons in Molecular Systems , Kluwer (1981).
[14] L.V. Yakushevich, Nonlinear Physics of DNA , Wiley & Sons (1998).
[15] M. Peyrard, Nonlinearity 17, R1-R40 (2004).
[16] M. Cardoni, R. de Leo and G. Gaeta, Phys. Rev. E75, 021919 (2007).
[17] M. Peyrard and A.R. Bishop, Phys. Rev. Lett. 62, 2755-2758 (1989).
[18] T. Dauxois, Phys. Lett. A159, 390395 (1991).
[19] T. Dauxois and M. Peyrard, Lecture Notes in Physics 393 , 79 (1991).
[20] D.J. Lee, A. Wynveen and A.A. Kornyshev, Phys. Rev. E70, 051913 (2004).
[21] T. A. Knotts, N. Rathore, D. C. Schwartz and J. J. de Pablo, J. Chem. Phys.
126, 084901 (2007).
[22] A. Sulaiman, J. Theor. Comput. Stud. 4, 0109 (2005).
[23] A.A. Kornyshev and S. Leikin, Phys. Rev. Lett. 82, 4138-4141 (1999).
[24] A.A. Kornyshev and S. Leikin, J. Chem. Phys. 107, 3656-3674 (1997);
J. Chem. Phys. 108, 7035(E) (1998).
[25] L.V. Yakushevich, J. Biosci. 26, 305-313 (2001).
[26] S.F. Mingalev, P.L. Christiansen, Y.B. Gaididei, M. Johansson and K.O. Ras-
mussen, J. Bio. Phys. 25, 41-63 (1999).
[27] M. Cardoni, R. de Leo and G. Gaeta, J. Nonlinear Math. Phys. 14, 128-146
(2007).
[28] J. A. Berashevich and T. Chakraborty, J. Chem. Phys. 126, 035104 (2007).
[29] S. Zdrakovici, J.A. Tuszynski and M.V. Sataric, J. Comput. Theor. Nanosci.
21, 1 (2005).
[30] A. Izmitli, D. C. Schwartz, M. D. Graham and J. J. de Pablo, J. Chem. Phys.
128, 085102 (2008).
[31] T. P. Westcott, I. Tobias and W. K. Olson, J. Chem. Phys. 107, 3967 (1997).
16
[32] R. M. Jendrejack, J. J. de Pablo and M. D. Graham, J. Chem. Phys. 116, 7752
(2002).
[33] V.K. Fedyanin and L.V. Yakushevich, Stud. Biophys. 103, 171-178 (1984).
[34] S.M. Mahajan, Phys. Rev. Lett. 90, 035001 (2003).
[35] A. Sulaiman and L.T. Handoko, Proc. Intl. Conf. on Applied Mathematics 2005
, Bandung (arXiv:physics/0508092).
[36] B.A. Bambah, S.M. Mahajan and C. Mukku, Phys. Rev. Lett. 97, 072301 (2006).
[37] A. Sulaiman, A. Fajarudin, T.P. Djun and L.T. Handoko, Intl. J. Mod. Phys.
A24, 3630-3637 (2009).
[38] Marmanis, Phys. of Fluid 10, 1428-1437 (1998).
[39] Marmanis, Phys. of Fluid 10, 3031 (1998).
[40] C. N. Yang, Proc. 6th Hawaii Topical Conf. Part. Phys , (1975).
[41] R. Mills, Phys. Rev. 96, 191-195 (1954).
[42] T.P. Cheng and L.F. Li, Gauge Theory of Elementary Particle Physics , Oxford
Univ. Press (1991).
[43] C.W. Lim and J-J. Shu, Proc. 2nd Intl. Conf. on Computational Nanoscience
and Nanotechnology , 387 (2002).
[44] R. Bundschuh and U. Gerland, Euro. Phys. J. E19, 347-349 (2006).
17
|
1807.04546 | 1 | 1807 | 2018-07-12T11:27:12 | Ionic-heterogeneity-induced spiral- and scroll-wave turbulence in mathematical models of cardiac tissue | [
"physics.bio-ph",
"q-bio.TO"
] | Spatial variations in the electrical properties of cardiac tissue can occur because of cardiac diseases. We introduce such gradients into mathematical models for cardiac tissue and then study, by extensive numerical simulations, their effects on reentrant electrical waves and their stability in both two and three dimensions. We explain the mechanism of spiral- and scroll-wave instability, which entails anisotropic thinning in the wavelength of the waves because of anisotropic variation in its electrical properties. | physics.bio-ph | physics | Ionic-heterogeneity-induced spiral- and scroll-wave turbulence in mathematical
models of cardiac tissue
APS/123-QED
Soling Zimik1, Rupamanjari Majumder2,∗ and Rahul Pandit1†
1Centre for Condensed Matter Theory,
Department of Physics, Indian Institute of Science,
Bangalore, 560012, India.
2Laboratory for Fluid Physics,
Pattern Formation and Biocomplexity,
Max Planck Institute for Dynamics and Self-Organization,
37077 Gottingen, Germany
(Dated: July 13, 2018)
Spatial variations in the electrical properties of cardiac tissue can occur because of cardiac diseases.
We introduce such gradients into mathematical models for cardiac tissue and then study, by extensive
numerical simulations, their effects on reentrant electrical waves and their stability in both two and
three dimensions. We explain the mechanism of spiral- and scroll-wave instability, which entails
anisotropic thinning in the wavelength of the waves because of anisotropic variation in its electrical
properties.
8
1
0
2
l
u
J
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
6
4
5
4
0
.
7
0
8
1
:
v
i
X
r
a
PACS numbers: 87.19.Hh,89.75.-k
Nonlinear waves in the form of spirals occur in many
excitable media, examples of which include Belousov-
Zhabotinsky-type systems [1], calcium-ion waves in
Xenopus oocytes [2], the aggregation of Dictyostelium
discoideum by cyclic-AMP signaling [3], the oxidation of
carbon monoxide on a platinum surface [4], and, most
important of all, cardiac tissue [5]. Understanding the
development of such spiral waves and their spatiotem-
poral evolution is an important challenge in the study of
extended dynamical systems, in general, and especially in
cardiac tissue, where these waves are associated with ab-
normal rhythm disorders, which are also called arrhyth-
mias. Cardiac tissue can support many patterns of non-
linear waves of electrical activation, like traveling waves,
target waves, and spiral and scroll waves [6]. The occur-
rence of spiral- and scroll-wave turbulence of electrical
activation in cardiac tissue has been implicated in the
precipitation of life-threatening cardiac arrhythmias like
ventricular tachycardia (VT) and ventricular fibrillation
(VF), which destroy the regular rhythm of a mammalian
heart and render it incapable of pumping blood. These
arrhythmias are the leading cause of death in the indus-
trialized world [7–11].
Biologically, VF can arise because of many complex
mechanisms. Some of these are associated with the devel-
opment of instability-induced spiral- or scroll-wave tur-
bulence [12]. One such instability-inducing factor is ionic
heterogeneity [13, 14], which arises from variations in
the electrophysiological properties of cardiac cells (my-
ocytes), like the morphology and duration of their action-
potentials (AP s) [15–18]. Such variations may appear
in cardiac tissue because of electrical remodeling [19–
21], induced by alterations in ion-channel expression and
activity, which arise, in turn, from diseases [22] like is-
chemia [23, 24], some forms of cardiomyopathy [25], and
the long-QT syndrome [26]. To a certain extent, some
heterogeneity is normal in healthy hearts; and it has
an underlying physiological purpose [16, 27–31]; but, if
the degree of heterogeneity is more than is physiologi-
cally normal, it can be arrhythmogenic [20, 24, 32]. It
is important, therefore, to explore ionic-heterogeneity-
induced spiral- or scroll-wave turbulence in mathemat-
ical models of cardiac tissue, which allow us to control
this heterogeneity precisely, in order to be able to iden-
tify the nonlinear-wave instability that leads to such tur-
bulence. We initiate such a study by examining the
effects of this type of heterogeneity in three cardiac-
tissue models, which are, in order of increasing complex-
ity and biological realism, (a) the two-variable Aliev-
Panfilov model [33], (b) the ionically realistic O'Hara-
Rudy (ORd) model [34] in two dimensions (2D), and (c)
the ORd model in an anatomically realistic simulation
domain.
In each one of these models, we control pa-
rameters (see below) in such a way that the ion-channel
properties change anisotropically in our simulation do-
mains, thereby inducing an anisotropic spatial variation
in the local action potential duration AP D. We show
that this variation in the AP D leads, in all these mod-
els, to an anisotropic reduction of the wavelength of the
spiral or scroll waves; and this anisotropic reduction of
the wavelength paves the way for an instability that pre-
cipitates turbulence, the mathematical analog of VF, in
these models.
The Aliev-Panfilov model provides a simplified descrip-
tion of an excitable cardiac cell [33]. It comprises a set of
coupled ordinary differential equations (ODEs), for the
normalized representations of the transmembrane poten-
tial V and the generalized conductance r of the slow,
∗ [email protected]
† [email protected]; also at Jawaharlal Nehru Centre for Advanced Scientific Research, Bangalore, Karnataka, India
fast inward Na+ current
transient outward K+ current
sarcolemmal Ca2+ pump current
rapid delayed rectifier K+ current
slow delayed rectifier K+ current
inward rectifier K+ current
INa
Ito
ICaL L-type Ca2+ current
IKr
IKs
IK1
INaCa Na+/Ca2+ exchange current
INaK Na+/K+ ATPase current
INab Na+ background current
ICab Ca2+ background current
IpCa
IKb K+ background current
ICaNa Na+ current through the L-type Ca2+ channel
ICaK K+ current through the L-type Ca2+ channel
2
ORd model, and g is a multiplication factor. We model
gradients in Gx as follows:
Gx(y) = [gmin +
y(gmax − gmin)
L
]Gxo, 0 ≤ y ≤ L;
(5)
here, L is the length of the side of the square simulation
domain, and gmax and gmin are, respectively, the maxi-
mal and minimal values of g; we can impose gradients in
k in the Aliev-Panfilov model in the same manner. For
simplicity, we induce the gradient along one spatial di-
rection only: the vertical axis in 2D; and the apico-basal
(apex-to-base) direction in 3D. The spatiotemporal evo-
lution of V in both models is governed by the following
reaction-diffusion equation:
TABLE I. The various ionic currents incorporated in the ORd
model are tabulated above. The symbols used for the currents
follow Ref. [34].
∂V
∂t
+ I = ∇.(D∇V ),
(6)
repolarizing current:
dV
dt
= −kV (V − a)(V − 1) − V r;
(1)
dr
dt
= [ǫ +
µ1r
µ2 + V
][−r − kV (V − b − 1)];
(2)
fast processes are governed by the first term in Eq.(1),
whereas, the slow, recovery phase of the AP is deter-
mined by the function ǫ + µ1r
µ2+V in Eq.(2). The param-
eter a represents the threshold of activation and k con-
trols the magnitude of the transmembrane current. We
use the standard values for all parameters [33], except
for the parameter k. We write k=g × ko, where g is a
multiplication factor and ko is the control value of k. In
2D simulations we introduce a spatial gradient (a linear
variation) in the value of k along the vertical direction
of the domain. To mimic the electrophysiology of a hu-
man ventricular cell, we perform similar studies using a
slightly modified version of the ionically-realistic O'Hara-
Rudy model (ORd) [34, 35]. Here, the transmembrane
potential V is governed by the ODE
dV
dt
= −
Iion
Cm
,
Iion = ΣxIx,
(3)
where Ix, the membrane ionic current, for a generic ion
channel x, of a cardiac cell, is
Ix = Gxf1(pact)f2(pinact)(Vm − Ex),
(4)
where Cm=1 µF is the membrane capacitance, f1(pact)
and f2(pinact) are, respectively, functions of probabilities
of activation (pact) and inactivation (pinact) of the ion
channel x, and Ex is its Nernst potential. We give a list
of all the ionic currents in the ORd model in Table I.
We write Gx = g × Gxo, where Gxo is the original value
of the maximal conductance of the ion channel x in the
where D is the diffusion tensor, and I = Iion
and
Cm
kV (V − a)(V − 1) + V r for ORd and Aliev-Panfilov mod-
els, respectively. For the numerical implementation of
the diffusion term in Eq. (6), we follow Refs. [35, 36].
We construct our anatomically realistic simulation do-
main with processed human-ventricular data, obtained
by using Diffusion Tensor Magnetic Resonance Imaging
(DTMRI) [37].
FIG. 1. (Color online) Variation of (a) AP D and (b) ω (see
text) with k. AP D = AP D/AP Do; here, AP Do is the con-
trol value of AP D at g = 1 (so AP D = 1 at g = 1); we
also use other combinations of (AP D, g) in our numerical
simulations. We find that AP D decreases with increasing k;
however, ω increases with increasing k. (c) Pseudocolor plots
of V , at three representative times (time increases from left
to right), illustrating the precipitation of the spiral-wave in-
stability in the Aliev-Panfilov model with a linear gradient in
k; the video S1 in the Supplemental Material [38] shows the
complete spatiotemporal evolution of this instability.
In Fig. 1(a) we show the variation, with the parameter
g, of AP D = AP D/AP Do, where AP Do is the control
AP D value for g = 1. We find that AP D decreases with
increasing g. Changes in the AP D at the single-cell level
influence electrical-wave dynamics at the tissue level. In
particular, such changes affect the rotation frequency ω
of reentrant activity (spiral waves). If θ and λ denote,
respectively, the conduction velocity and wavelength of a
plane electrical wave in tissue, then ω ≃ θ
λ , λ ≃ θ ×AP D.
Therefore, if we neglect curvature effects [39], the spiral-
wave frequency
ω ≃
1
AP D
.
(7)
We find, in agreement with this simple, analytical esti-
mate, that ω decreases as the AP D increases. We show
this in Fig. 1(b) by plotting ω = ω/ω0 versus g; here, ω0
is the frequency for g = 1 [40].
Similarly, in the ionically realistic ORd model, changes
in the ion-channel conductances Gx alter the AP D of the
cell and, therefore, the spiral-wave frequency ω. In Figs. 2
(a1) and (a2) we present a family of plots to illustrate
the variation in AP D with changes in Gx. We find that
AP D decreases with an increase in g for most currents
(IKr, IKs, IK1, IN a and IN aK ); but it increases for some
other currents (ICa, IN aCa and Ito). The rate of change
of AP D is most significant when we change GKr; by
contrast, it is most insensitive to changes in GN a and Gto.
In Figs. 2 (b1) and (b2) we show the variation of ω with
g for different ion channels x. We find that changes in
Gx, which increase AP D, decrease ω and vice versa; this
follows from Eq. (2). The sensitivity of ω, with respect
to changes in Gx, is most for Gx = GKr and least for
Gx = Gto: ω increases by ∆ω ≃ 1.23, as g goes from 0.2
to 5; for Gto, the same variation in g decreases the value
of ω by ∆ω ≃ 0.04.
We now investigate the effects, on spiral-wave dynam-
ics, of spatial gradients in k, in the 2D Aliev-Panfilov
model, and in Gx, in the 2D ORd model. A linear gradi-
ent in k, in the Aliev-Panfilov model, induces a gradient
in ω (see Fig. 1(b)); and such a spatial gradient in ω
induces a spiral-wave instability in the low-ω region. In
Fig. 1(c) we demonstrate how a gradient in k (gmax = 1.5
and gmin = 0.5) leads to the precipitation of this instabil-
ity (also see video S1 of the Supplemental Material [38]).
Similarly, for each current listed in Table I for the ORd
model, we find wave breaks in a medium with a gradi-
ent in Gx. We illustrate, in Fig. 3, such wave breaks
in our 2D simulation domain, with a gradient (∇Gx) in
any Gx, for 3 representative currents; we select IKr, be-
cause it has the maximal impact on the single-cell AP D,
and also on ω in tissue simulations; and we choose IK1
and IN aCa, because they have moderate and contrary ef-
fects on AP D and ω (Figs. 2). Our results indicate that
gradient-induced wave breaks are generic, insofar as they
occur in both the simple two-variable (Aliev-Panfilov)
and the ionically realistic (ORd) models of cardiac tis-
sue.
In Figs. 3 (d-f), we present power spectra of the
time series of V , recorded from a representative point of
the simulation domain; these spectra show broad-band
backgrounds, which are signatures of chaos, for the gra-
dients ∇GKr and ∇GK1; however, the gradient ∇GN aCa
3
induces wave breaks while preserving the periodicity of
the resultant, reentrant electrical activity, at least at the
points from which we have recorded V .
The instability in spiral waves occurs because spatial
gradients in k (Aliev-Panfilov) or in Gx (ORd) induce
spatial variations in both AP D and ω: In our simulation
domain, the local value of ω (AP D) decreases (increases)
from the top to the bottom. In the presence of a single
spiral wave (left panel of Fig. 4), the domain is paced, in
effect, at the frequency ω of the spiral, i.e., with a fixed
time period T = 1/ω = AP D + DI, where DI is the
diastolic interval (the time between the repolarization of
one AP and the initiation of the next AP ). Thus, the
bottom region, with a long AP D, has a short DI and
vice versa. The restitution of the conduction velocity θ
implies that a small DI leads to a low value of θ and vice
versa [41] (see Fig. S1 in the Supplemental Material [38]).
To compensate for this reduction of θ, the spiral wave
must reduce its wavelength λ, in the bottom, large-AP D
(small-DI) region, so that its rotation frequency ω ≃ θ
λ
remains unchanged, as shown in Fig. 4 (also see video S2
in the Supplemental Material [38]), where the thinning
of the spiral arms is indicated by the variation of λ along
the spiral arm (λ2 > λ1, in the pseudocolor plot of Vm
in the top-left panel t= 1.46 s). Clearly, this thinning is
anisotropic, because of the uni-directional variation in k
or Gx; this anisotropy creates functional heterogeneity in
wave propagation, which leads in turn to the spiral-wave
instability we have discussed above (Fig. 4).
In the ORd model, we find that gradients in GKr eas-
ily induce instabilities of the spiral for small values of
∆g ≡ gmax − gmin ≃ 0.5; by contrast, in a medium with
gradients in Gto, the spiral remains stable for values of
∆g as large as 4.8. This implies that the stability of the
spiral depends on the magnitude of the gradient in ω that
is induced in the medium.
In Fig. 5 (also see video S3 in the Supplemental Ma-
terial [38]), we extend our study to illustrate the onset
of scroll-wave instabilities in a 3D, anatomically realis-
tic human-ventricular domain, in the presence of spatial
gradients in GKr.
In mammalian hearts, the AP D is
typically lower in the apical region as compared to that
in the basal region [16]. Therefore, we use values of the
AP D that increase from the apex to the base (and, hence,
ω decreases from the apex to base). With gmax(GKr)=
6 and ∆g= 4, we observe breakup in a scroll wave that
is otherwise stable in the absence of this spatial gradi-
ent. We note that the mechanism for the onset of such
scroll-wave instabilities is the same as in 2D, and it relies
on the gradient-induced anisotropic thinning of the scroll
wavelength.
We have shown that gradients in parameters that affect
the AP D of the constituent cells induce spatial gradients
in the local value of ω. This gradient in the value of ω
leads to an anisotropic reduction in the wavelength of
the waves, because of the conduction-velocity restitution
property of the tissue, and it paves the way for spiral-
and scroll-wave instability in the domain. This gradient-
4
FIG. 4. (Color Online) Pseudocolor plots of Vm illustrating
the development of a spiral-wave instability, with the passage
of time t, in the 2D ORd model, with a spatial gradient in
GN aCa. The left frame shows the thinning of the spiral arm
(λ varies along the spiral arm, and λ2 > λ1) indicated just
before the spiral wave breaks (see the middle and the right
frames).
FIG. 5.
(Color Online) Scroll-wave instabilities in our
anatomically realistic human-ventricular domain, in the pres-
ence of an apico-basal gradient in GKr
has the conduction-velocity-restitution property. We find
that the spiral or scroll waves always break up in the low-
ω region. This finding is in line with that of the experi-
mental study by Campbell, et al., [15] who observe spiral-
wave break-up in regions with a large AP D in neonatal-
rat-ventricular-myocyte cell culture. We find that the
stability of the spiral is determined by the magnitude of
the gradient in ω; the larger the magnitude of the gradi-
ent in the local value of ω, the more likely is the break
up of the spiral or scroll wave. By using the ORd model,
we find that ω varies most when we change GKr (as com-
pared to other ion-channel conductances) and, therefore,
spiral waves are most unstable in the presence of a gra-
dient of GKr. By contrast, we find that ω varies most
gradually with Gto, and hence the spiral wave is most
stable in the presence of a gradient in Gto (as compared
to gradients in other conductances).
Earlier studies have investigated the effects of ionic-
heterogeneity on spiral-wave dynamics. The existence of
regional ionic heterogeneities have been found to initiate
spiral waves [42], attract spiral waves to the heterogene-
ity [43], and destabilize spiral waves [44]. The presence of
AP D gradients in cardiac tissue has been shown to drive
spirals towards large-AP D (low ω) regions [45]. A study
by Zimik, et al., [35] finds that spatial gradients in ω, in-
duced by gradients in the density of fibroblasts, can pre-
cipitate a spiral-wave instability. However, none of these
studies provides a clear understanding of the mechanisms
FIG. 2. (Color online) Plots of AP D and ω versus g; here,
AP D = AP D/AP Do and ω = ω/ω0, where AP Do= 250 ms,
and ω0= 4.38 Hz are, respectively, the control values of AP D
and ω; (a1) and (a2) show, respectively, that AP D decreases
with the conductances Gx, for the currents IKr, IKs, IK1,
IN a and IN aK; however, it increases with increasing Gx, for
the currents ICaL, IN aCa and Ito; (b1) and (b2) show that
the variation of ω, with the various channel conductances, is
consistent with Eq. (2).
FIG. 3. (Color Online) Pseudocolor plots of the transmem-
brane potential Vm illustrating spiral-wave instabilities from
our numerical simulations of the 2D ORd model for human
ventricular tissue, with spatial gradients in (a) GKr, (b) GK1,
and (c) GN aCa (because GN aCa decreases with g (Fig. 2), the
gradient in GN aCa must be chosen to be the negative of that
in Eq. (5)); in (a)-(c) the local value of ω decreases from the
top of the simulation domain to its bottom. Power spectra
of the time series of Vm, from representative points in our
simulation domain, are shown for gradients in (d) GKr, (e)
GK1, and (f) GN aCa; the spectra in (d) and (e) are consistent
with the onset of spiral-wave turbulence; the power spectrum
in (f) shows the continuation of periodic electrical activity, in
spite of wave breaks.
induced instability is a generic phenomenon because we
obtain this instability in the simple Aliev-Panfilov and
the detailed ORd model for cardiac tissue. Such an insta-
bility should be observable in any excitable medium that
underlying the onset of spiral- and scroll-wave instabil-
ities, from a fundamental standpoint. Moreover, none
of these studies has carried out a detailed calculation
of the pristine effects of each individual major ionic cur-
rents, present in a myocyte, on the spiral-wave frequency;
nor have they investigated, in a controlled manner, how
gradients in ion-channel conductances lead to spiral- or
scroll-wave instabilities. Our work makes up for these
lacunae and leads to specific predictions that should be
tested experimentally.
ACKNOWLEDGMENTS
5
We thank the Department of Science and Technology
(DST), India, and the Council for Scientific and Indus-
trial Research (CSIR), India, for financial support, and
Supercomputer Education and Research Centre (SERC,
IISc) for computational resources.
[1] A. Zaikin and A. Zhabotinsky, Nature 225, 535 (1970).
[2] D. E. Clapham, Cell 80, 259 (1995).
[3] J. J. Tyson and J. Murray, Development 106, 421 (1989).
[4] R. Imbihl and G. Ertl, Chemical Reviews 95, 697 (1995).
[5] J. M. Davidenko, A. V. Pertsov, R. Salomonsz, W. Bax-
ter, and J. Jalife, Nature 355, 349 (1992).
(1978).
[24] X. Jie and N. A. Trayanova, Heart Rhythm 7, 379 (2010).
[25] G. Sivagangabalan, H. Nazzari, O. Bignolais, A. Maguy,
P. Naud, T. Farid, S. Mass´e, N. Gaborit, A. Varro,
K. Nair, P. Backx, E. Vigmond, S. Nattel, S. Demolombe,
and K. Nanthakumar, PLoS ONE 9, e82179 (2014).
[6] J. J. Tyson and J. P. Keener, Physica D: Nonlinear Phe-
[26] P. C. Viswanathan and Y. Rudy, Circulation 101, 1192
nomena 32, 327 (1988).
(2000).
[7] P. Bayly, B. KenKnight, J. Rogers, E. Johnson, R. Ideker,
and W. Smith, Chaos: An Interdisciplinary Journal of
Nonlinear Science 8, 103 (1998).
[8] F. X. Witkowski, L. J. Leon, P. A. Penkoske, W. R. Giles,
M. L. Spano, W. L. Ditto, and A. T. Winfree, Nature
392, 78 (1998).
[9] G. P. Walcott, G. N. Kay, V. J. Plumb, W. M. Smith,
J. M. Rogers, A. E. Epstein, and R. E. Ideker, Journal
of the American College of Cardiology 39, 109 (2002).
[10] I. R. Efimov, V. Sidorov, Y. Cheng,
and B. Wollen-
zier, Journal of cardiovascular electrophysiology 10, 1452
(1999).
[11] J. De Bakker, F. Van Capelle, M. Janse, A. Wilde,
R. Coronel, A. Becker, K. Dingemans, N. Van Hemel,
and R. Hauer, Circulation 77, 589 (1988).
[12] F. H. Fenton, E. M. Cherry, H. M. Hastings, and S. J.
Evans, Chaos: An Interdisciplinary Journal of Nonlinear
Science 12, 852 (2002).
[27] C. Antzelevitch, S. Sicouri, S. H. Litovsky, A. Lukas,
S. C. Krishnan, J. M. Di Diego, G. A. Gintant, and
D.-W. Liu, Circ Res 69, 1427 (1991).
[28] T. Furukawa, R. J. Myerburg, N. Furukawa, A. L. Bas-
and S. Kimura, Circulation Research 67, 1287
sett,
(1990).
[29] D. Fedida and W. Giles, The Journal of Physiology 442,
191 (1991).
[30] S. Zicha, L. Xiao, S. Stafford, T. J. Cha, W. Han,
A. Varro, and S. Nattel, The Journal of physiology 561,
735 (2004).
[31] F. H. Samie, O. Berenfeld, J. Anumonwo, S. F. Mironov,
S. Udassi, J. Beaumont, S. Taffet, A. M. Pertsov, and
J. Jalife, Circulation research 89, 1216 (2001).
[32] M. J. Janse, Cardiovascular research 61, 208 (2004).
[33] R. R. Aliev and A. V. Panfilov, Chaos, Solitons & Frac-
tals 7, 293 (1996).
[34] T. O'Hara, L. Vir´ag, A. Varr´o, and Y. Rudy, PLoS Com-
[13] G. K. Moe, W. C. Rheinboldt, and J. Abildskov, Amer-
put Biol 7, e1002061 (2011).
ican heart journal 67, 200 (1964).
[14] J. Jalife, Annual review of physiology 62, 25 (2000).
[15] K. Campbell, C. J. Calvo, S. Mironov, T. Herron,
O. Berenfeld, and J. Jalife, The Journal of physiology
590, 6363 (2012).
[16] N. Szentadrassy, T. Banyasz, T. Biro, G. Szabo, B. I.
Toth, J. Magyar, J. Lazar, A. Varro, L. Kovacs, and
P. P. Nanasi, Cardiovascular Research 65, 851 (2005).
[17] M. Stoll, M. Quentin, A. Molojavyi, V. Thamer, and
U. K. Decking, Cardiovascular research (2007).
[18] D.-W. Liu and C. Antzelevitch, Circulation research 76,
351 (1995).
[19] M. M. Elshrif, P. Shi, and E. M. Cherry, IEEE journal
of biomedical and health informatics 19, 1308 (2015).
[35] S. Zimik and R. Pandit, New J. Phys 18, 123014 (2016).
[36] R. Majumder, R. Pandit,
and A. V. Panfilov, JETP
letters 104, 796 (2016).
[37] "Dtmri data from ex-vivo canine and human hearts, the cardiovascular research grid (http://cvrgrid.org/data/ex-vivo)." .
[38] Supplementary Information.
[39] Z. Qu, F. Xie, A. Garfinkel, and J. N. Weiss, Annals of
biomedical engineering 28, 755 (2000).
[40] For the parameter a this simple relation between ω and
AP D is not observed, because change in a affects not
only the APD but also other quantities like θ, which has
effects on the value of ω.
[41] E. M. Cherry and F. H. Fenton, American Journal
of Physiology-Heart and Circulatory Physiology 286,
H2332 (2004).
[20] S. Nattel, A. Maguy, S. Le Bouter, and Y.-H. Yeh, Phys-
[42] A. Defauw, P. Dawyndt, and A. V. Panfilov, Physical
iological reviews 87, 425 (2007).
Review E 88, 062703 (2013).
[21] M. J. Cutler, D. Jeyaraj, and D. S. Rosenbaum, Trends
in pharmacological sciences 32, 174 (2011).
[22] A. S. Amin, H. L. Tan, and A. A. Wilde, Heart Rhythm
7, 117 (2010).
[43] A. Defauw, N. Vandersickel, P. Dawyndt, and A. V. Pan-
filov, American Journal of Physiology-Heart and Circu-
latory Physiology 307, H1456 (2014).
[44] A. Xu and M. R. Guevara, Chaos: An Interdisciplinary
[23] A. H. Harken, C. H. Barlow, W. R. Harden,
and
B. Chance, The American journal of cardiology 42, 954
Journal of Nonlinear Science 8, 157 (1998).
[45] K. Ten Tusscher and A. V. Panfilov, American Journal of
Physiology-Heart and Circulatory Physiology 284, H542
(2003).
6
8
1
0
2
l
u
J
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
6
4
5
4
0
.
7
0
8
1
:
v
i
X
r
a
Supplemental Material
Electrical-wave turbulence in anisotropic cardiac
tissue
Soling Zimik1, Rupamanjari Majumder2, and Rahul Pandit1
1Centre for Condensed Matter Theory, Department of Physics, Indian
Institute of Science, Bangalore, 560012, India
2Laboratory for Fluid Physics, Pattern Formation and Biocomplexity,
Max Planck Institute for Dynamics and Self-Organization, 37077
Göttingen, Germany
Video Captions:
Video S1: Spiral-wave instability in the Aliev-Panfilov model. Video
of pseudocolor plots of transmembrane potential V showing the formation of
spiral-wave instability in a medium with gradient in k: gmin=0.5 and gmax=
1.5. For the video, we use 10 frames per second with each frame separated
from the succeeding frame by 20ms in real time.
Video S2: Spiral-wave instability in the ORd model. Video pseudo-
color plots of transmembrane potential Vm showing the formation of spiral-
wave instability in a medium with a gradient in GN aca (gmin=0.2 and gmax=
2). For the video, we use 10 frames per second with each frame separated
from the succeeding frame by 20ms in real time.
Video S3: Scroll-wave instability. Video pseudocolor plots of transmem-
brane potential Vm showing the formation of scroll-wave instability in an
anatomically-realistic model for human ventricles. A linear gradient in GK r
is applied along the apico-basal direction: gmin = 2 in the apex and gmax=6
in the base. For the video, we use 10 frames per second with each frame
separated from the succeeding frame by 20ms in real time.
1
Figures:
65
60
55
50
0
200
400
600
800
Figure showing a conduction-velocity restitution curve, generated by using
the ORd model. The value of conduction velocity θ initially increases with
the increase of diastolic interval DI and saturates at large values of DI.
2
|
1611.02921 | 1 | 1611 | 2016-11-09T13:31:57 | High Spatio-Temporal-Resolution Detection of Chlorophyll Fluorescence Dynamics from a Single Chloroplast with Confocal Imaging Fluorometer | [
"physics.bio-ph"
] | Chlorophyll fluorescence (CF) is a key indicator to study plant physiology or photosynthesis efficiency. Conventionally, CF is characterized by fluorometers, which only allows ensemble measurement through wide-field detection. For imaging fluorometers, the typical spatial and temporal resolutions are on the order of millimeter and second, far from enough to study cellular/sub-cellular CF dynamics. In addition, due to the lack of optical sectioning capability, conventional imaging fluorometers cannot identify CF from a single cell or even a single chloroplast. Here we demonstrated a novel fluorometer based on confocal imaging, that not only provides high contrast images, but also allows CF measurement with spatiotemporal resolution as high as micrometer and millisecond. CF transient (the Kautsky curve) from a single chloroplast is successfully obtained, with both the temporal dynamics and the intensity dependences corresponding well to the ensemble measurement from conventional studies. The significance of confocal imaging fluorometer is to identify the variation among individual chloroplasts, e.g. the half-life period of the slow decay in the Kautsky curve, that is not possible to analyze with wide-field techniques. A linear relationship is found between excitation Intensity and the temporal positions of peaks/valleys in the Kautsky curve. In addition, an interesting 6-order increase in excitation intensity is found between wide-field and confocal fluorometers, whose pixel integration time and optical sectioning may account for this substantial difference. Confocal imaging fluorometers provide micrometer and millisecond CF characterization, opening up unprecedented possibilities toward detailed spatiotemporal analysis of CF transients and its propagation dynamics, as well as photosynthesis efficiency analysis, on the scale of organelles, in a living plant. | physics.bio-ph | physics | High Spatio-Temporal-Resolution Detection of Chlorophyll
Fluorescence Dynamics from a Single Chloroplast with
Confocal Imaging Fluorometer
Yi-Chin Tseng1 and Shi-Wei Chu1,2*
(1 Department of Physics, National Taiwan University, Taipei, Taiwan;
2 Molecular Imaging Center, National Taiwan University;
Correspondence*: [email protected])
Abstract
Background: Chlorophyll fluorescence (CF) is a key indicator to study plant
physiology or photosynthesis efficiency. Conventionally, CF is characterized by
fluorometers, which only allows ensemble measurement through wide-field detection.
For imaging fluorometers, the typical spatial and temporal resolutions are on the order
of millimeter and second, far from enough to study cellular/sub-cellular CF dynamics.
In addition, due to the lack of optical sectioning capability, conventional imaging
fluorometers cannot identify CF from a single cell or even a single chloroplast.
Result and Discussion: Here we demonstrated a novel fluorometer based on confocal
imaging, that not only provides high contrast images, but also allows CF measurement
with spatiotemporal resolution as high as micrometer and millisecond. CF transient
(the Kautsky curve) from a single chloroplast is successfully obtained, with both the
temporal dynamics and the intensity dependences corresponding well to the ensemble
measurement from conventional studies. The significance of confocal imaging
fluorometer is to identify the variation among individual chloroplasts, e.g. the half-life
period of the slow decay in the Kautsky curve, that is not possible to analyze with
wide-field techniques. A linear relationship is found between excitation Intensity and
the temporal positions of peaks/valleys in the Kautsky curve. In addition, an
interesting 6-order increase in excitation intensity is found between wide-field and
confocal fluorometers, whose pixel integration time and optical sectioning may
account for this substantial difference.
Conclusion:
Confocal imaging fluorometers provide micrometer and millisecond CF
characterization, opening up unprecedented possibilities toward detailed
spatiotemporal analysis of CF transients and its propagation dynamics, as well as
photosynthesis efficiency analysis, on the scale of organelles, in a living plant.
Key words: chlorophyll fluorescence; confocal microscopy; Kautsky effect;
photosynthesis efficiency
Background
Chlorophyll fluorescence (CF) has been proven to be one of the most powerful and
widely used techniques for plant physiologists [1-7]. Despite of its low quantum
efficiency (2% to 10% of absorbed light [8]), CF detections are meaningful due to its
intricate connection with numerous processes taking place during photosynthesis,
such as reduction of photosystem reaction centers, non-photochemical quenching, etc.
[9, 10]. It is well known that the efficiency of photosynthesis can be derived from CF
dynamics, thus providing noninvasive, fast and accurate characterization for
photosynthesis. It has been widely adopted to study plant physiology, including stress
tolerance, nitrogen balance, carbon fixation efficiency, etc. [11]. It is not too
exaggerated to say that nowadays, no investigation about photosynthetic process
would be complete without CF analysis.
Conventionally, the tool of choice to study CF is a fluorometer. There are many
different fluorometry techniques, such as plant efficiency analyzer (PEA) [12], pulse
amplitude modulation (PAM) [13], the pump and probe (P&P) [14, 15] and the fast
repetition rate (FRR) [16]. It is interesting to note that these various detection
approaches are all based on the same principle, i.e. the Kautsky effect [7], or
equivalent, CF transient when moving photosynthetic material from dark adaption to
light environment.
Conventional imaging fluorometers (e.g., PAM and P&P fluorometers) are based
on wide-field detection, and are routinely adopted to study ensemble of CF transients
from a large area of a leaf, significantly limiting its spatiotemporal resolution. For
example, to study stress propagation in a plant leaf [17], current imaging fluorometers
only provide spatial resolution on the order of sub-millimeter, with temporal
resolution on the order of second. To unravel the more detailed propagation dynamics,
the required spatial resolution should be at least on single cell or sub-cellular level,
while the temporal resolution should be enhanced to millisecond scale.
One additional drawback of the conventional imaging fluorometers is lack of
optical section capability due to their wide-field nature, and thus prevents study of CF
transient on a single cell or even a single chloroplast level. In this work, we introduce
a novel concept of confocal imaging fluorometer, which is the combination of
confocal microscopy and CF transient detection, where the former provides optical
sectioning with exceptionally high axial contrast. The technique not only detects CF
signals with microsecond temporal resolution, but also attains micrometer spatial
resolution in all three dimensions. We have successfully reproduced the CF transient
(Kaustky curve) within a single chloroplast. We found that the CF transients of a
group of palisade cells and the ensemble of single chloroplasts are similar to each
other, and both correspond well to the result of conventional imaging fluorometers,
showing the reliability of our result. Nevertheless, the CF transient of individual
chloroplast can be substantially different, manifesting the value of the unusual
capability to study plant cell organelles. Furthermore, we found that the shape of
transients is highly intensity-dependent, which was also shown in the previous
research [18]. Finally, short integration time and optical section characteristic of
confocal image fluorometer make a significant difference of illumination intensity
comparing to that of conventional fluorometers. Given CF transient from a single
chloroplast, it is possible to investigate degree of influence from external or internal
plant-stress with scale of organelle, and confocal imaging fluorometer has paved the
way for this high spatio-temporal resolution CF detection.
Principle
Basic Concept of Confocal Imaging Fluorometer
The optical principle of confocal imaging fluorometer is basically the same as
confocal laser-scanning microscopy [19], which is an optical imaging technique for
increasing contrast and resolution. The essential components of a confocal imaging
fluorometer is shown in Figure 1, including a laser system, a dichroic mirror, a
scanning mirror system, an objective lens, a pinhole and a photomultiplier tube
(PMT).
The laser system in a confocal imaging fluorometer provides strong and
monochromatic illumination, whose wavelength can be selected to meet sample
request. The laser beam is sent to the objective after the scanning mirror system to
achieve two-dimensional raster scanning at the focal plane. The backward
fluorescence signal is collected by the same objective, de-scanned through the
scanning mirrors, and separated from residual laser by the dichroic mirror. The
fluorescence signal then is focused onto the pinhole, which is placed at the conjugate
plane of objective focus, to achieve optical sectioning by excluding out-of-focus
signals. One or more PMTs are placed behind the pinhole to collect the in-focus
fluorescence signals, which are reconstructed into images by synchronization with the
scanning mirrors [19].
Figure 1 Principle and basic components of a confocal imaging fluorometer. Laser
beam is reflected by a dichroic mirror and goes through a set of scanning mirrors,
then focused by an objective lens onto the specimen. Fluorescence signals is epi-
collected in the same path, and filtered out by the dichroic mirror. A confocal
pinhole is used to allow only fluorescence emitted from the focal plane being
detected by the PMT.
In general, a confocal imaging system is capable of collecting signal with a well-
defined optical section on the order of 1 µm [20]. This high axial resolution makes
confocal system an invaluable tool to observe single cell or sub-cellular organelles
[21-23]
The objective lens is characterized by magnification and numerical aperture (NA).
To enable large field-of-view observation, low magnification objectives are typically
required. However, please note that resolution is determined by NA, which can be
independent from magnification. NA describes the light acceptance cone of an
objective lens and hence light gathering ability and resolution. The definition of NA
is:
(1)
In the definition of NA, n is the index of refraction of the immersion medium,
while θ is the half-angle of the maximum light acceptance cone. Both lateral (xy-
direction) and axial (z-direction) resolutions for fluorescence imaging mode are
defined by NA and the wavelength (λ) [24].
sinnNA (2)
(3)
Kautsky Effect
Kautsky effect, first discovered in 1931, describes the dynamics of CF when dark-
adapted photosynthetic chlorophyll suddenly exposes to continuous light illumination
[25]. After initial light absorption, chlorophyll becomes excited and soon releases its
energy into one of the three internal decay pathways, including photosynthesis
(photochemical quenching, qP), heat (non-photochemical quenching, NPQ) and light
emission (CF). Owing to energy conservation, the sum of quantum efficiencies for
these three pathways should be unity. Therefore, the yield of CF is strongly related to
the efficiency of both qP and NPQ [26].
To be more specific, when transferring a photosynthetic material from dark
adaption into light illumination, CF yield typically exhibits a fast rising phase (within
1 second) and a slow decay phase (few minute duration), as shown by the green curve
in Figure 2. The fast rising phase is labeled as OP, where O is for origin, and P is the
peak [18]. It is mainly caused by the reduction of qP; that is, depletion of electron
acceptors, quinine (Qa) in the electron transport chain [27]. The slow decay phase is
labeled as PSMT, where S stands for semisteady state, M for a local maximum, and T
for a terminal steady state level [18]. One very interesting phenomenon is the shape of
this decay phase depends strongly on illumination intensity. At low intensity (32
μmol/m2/s), the Kaustky curve is the green one. When the intensity grows one order
larger, the amplitude of SM-rise in the transient is smaller, as shown by the red curve.
At one more order higher intensity, the SM section disappears completely, leaving an
exponential decay in the PT section, as shown by the blue curve. Such intensity-
dependent curve transition is known to be the result of photosynthetic state transition,
on which more detailed discussion can be found in the references [1, 10, 13, 28-30].
NArlateral43.02267.0NAnnraxial
Figure 2 The Kautsky effect, showing the CF transient as well as its intensity
dependence. Wavelength of excitation: 650nm. Excitation light intensity for
curves labeled 1, 2 and 3 was 32, 320 and 3200 μmol/m2/s, respectively. For
definition of OPSMT, O is the origin, P is the peak, S stands for semi-steady state,
M for a local maximum, and T for a terminal steady state level. (Modified figure
from [1], with copyright permission)
Material and Method
Plant Material
Brugmansia suaveolens (solanaceae), also known as Angel's Trumpet, is a woody
plant usually 3 m to 4 m in height with pendulous flowers and furry leaves distributed
widely in Taiwan, especially in wet areas. Being interested in spatial temporal
dynamics of CF, we selected B.suaveolens as our target material since the CF of its
cousin Datura wrightii, also known as Devil's Trumpet, had been studied in depth
[17]. B.suaveolens leaves were collected from the Botanical Garden of National
Taiwan University, Taipei, Taiwan (25◦1' N, 121◦31' E, 9 m a.s.l.). All sample leaves
are picked as fully expanded leaves that had neither experienced detectable physical
damage nor herbivory. In order to minimize the sampling error, 3 leaves are chosen
within plants that grow in similar micro-climate. Furthermore, all the measurements
are completed no longer than two hours after disleaving. Fresh leaves are sealed in
slide glass (76 × 26 mm), and slide samples are dark-adapted under constant
temperature and constant humidity dark environment (20 oC, 70 %RH) for 20 min.
Experimental Setup
A confocal microscope (Leica TCS SP5) in the Molecular Imaging Center of
National Taiwan University was adopted. CF was excited by a HeNe laser, whose
wavelength (633 nm) is the same as that used in popular conventional fluorometers,
such as LI-6400 from LI-COR. A relatively low-NA objective (HC PL Apo 10x/0.4
CS) was selected to allow not only large field of view over a few millimeters, but also
resolution much better than a single chloroplast. From Eq. (2) and (3), The lateral and
axial resolutions are about 1 µm and 5 µm, respectively.
The initial step is to bring the sample to focus by weak excitation (~1 kW /cm2),
and then the leaf is left in dark again for 5 minutes. To observe the Kautsky effect, the
633-nm laser was focused on the sample, and the fluorescence emission was recorded
in the spectral range of 670 to 690 nm. The intensity-dependent CF transient curves
were obtained by taking time-lapsed images while varying the 633-nm excitation
intensity from 1 kW/cm2 to 50 kW/cm2, at different sample regions. With different
number of total pixels, the temporal resolution of the CF transient varies from 10
milliseconds (16x16 pixels) to about 200 milliseconds (256x256 pixels). No
significant photobleaching of CF is expected at this intensity range [31].
Results
Fluorescence dynamics from a single chloroplast
Conventional fluorometers observe CF dynamics over a large area on a leaf, and
here we demonstrate that our confocal imaging fluorometer allows us to obtain CF
transients from a precisely chosen cells or even a single chloroplast. Figure 3(a)
shows the confocal images of the leaf sample. Figure 3(a1) is the large-area view,
showing the distribution of vascular bundles. (a2) gives a zoom-in view of a group of
palisade cells, showing clear distribution of chloroplasts in each cell. By further
zooming in, the field of view is focused onto a single chloroplast, as given in (a3),
showing the distribution of chlorophyll density inside the organelle [32].
Fig. 3(b) presents the CF transients from a group of palisade cells (b1) and a single
chloroplast (b2). Both curves are normalized with its own maximal fluorescence
intensity. The characteristic slow decay of Kautsky curve are obvious in both curves,
but the fast rising phase is only visible in (b1), because (b2) is noisier due to less
pixels involved. By fitting the curves with exponential decay function, the time
constants are found to be 28 ± 0.5 s for (b1) and 25 ± 1.8 s for (b2). The numbers
correspond well to the large-area measurement with conventional fluorometers [33],
manifesting the reliability of the confocal fluorometer. However, this result also
demonstrates that the CF transients can be slightly different among individual
chloroplasts and cells, as shown in Fig. 4. Such detailed characterization is not
possible with conventional fluorometers, which only deliver the ensemble response
from a lot of cells.
Figure 3 Confocal images and CF transients on different spatial scales inside a
living leaf. A 633-nm laser, with 40 kW/cm2 intensity, is adopted for 50 s
continuous confocal imaging. The sample leaf was kept in darkness for ~20 min
before imaging. (a1) is the image over a large area of the leaf, (a2) is zoomed in to
show a group of palisade cells, and (a3) is further zoomed in to focus onto a single
chloroplast. (b1) and (b2) show normalized CF transient from a group of palisade
cells and a single chloroplast, respectively. (b2) is noisier since less pixels are
involved.
Figure 4 Fluorescence transients for individual chloroplasts within a living leaf,
which is confocally imaged at 40 kW/cm2 with a HeNe laser (633nm).
Intensity dependent fluorescence transient
As we have mentioned in Figure 2 of the Principle section, it is well known that
the Kautsky curve changes with intensity. Figure 5 shows the intensity-dependent
Kautsky curves from a single chloroplast (gray lines) and from a group of cells
(colored lines), obtained by the confocal fluorometer. Fig. 5(a) is acquired with low
laser intensity (3 kW/cm2), and a temporal variation similar to curve 1 of Fig. 2 is
found, i.e. a complete OPSMT curve. The CF intensity rises to its first peak within 1
s, quickly decreasing to a local minimum (i.e., PS-decrease), rising again to a second
peak (i.e., SM-rise) then slowly falling as exponential decay. At slightly higher
intensity (10 kW/cm2), a temporal variation similar to curve 2 of Fig. 2 is observed.
The PS-fall and SM-rise still exist, but become much smaller, while the position of P,
S, and M appear earlier in the curve. At high intensity (55 kW/cm2), the SM part
disappears completely, leaving a single exponential decay (the PT section), similar to
curve 3 of Fig. 2. This result matches very well to the conventional wide-field
fluorometer [1, 13, 29], but with much higher spatio-temporal resolution, manifesting
again the reliability and usefulness of the confocal technique.
Figure 5 CF transients of a single chloroplast (grey) and average among a group of
cells (colored) under excitation intensity at (a) 3, (b) 10, and (c) 55 kW/cm2,
respectively, showing clearly the intensity-dependent Kautsky curves.
An interesting observation in Fig. 5 is not only the curve shape is intensity-
dependent, but also the positions of local maxima and minimum P, S, M points are
also strongly dependent on excitation intensity. Fig. 6(a) shows the detailed curve
variation relative to intensity, in the range of 3- 55 kW/cm2, and the corresponding
temporal position of local maximum of induced transients, i.e., point M, is given in
Fig. 6(b). Surprisingly, an almost perfect linear trend is observed. Similar linear
results are found for the semisteady state point S in Fig. 6(c), and for the peak point P
in Fig. 6(d). Due to the limitation of temporal resolution (200 ms for 256x256 pixels),
S and P points are analyzed with intensity range 3- 40 kW/cm2 and 3- 20 kW/cm2,
respectively. The linear trends indicate that the state transition rate increases with
higher excitation intensity. The underlying mechanism relies more investigation in the
future.
Figure 6 (a) Detailed Kautsky curve variation in the intensity range of 3 – 55
kW/cm2 The temporal positions of (b) the local maximum (M), (c) semisteady
state (S), and (d) peak (P), all change linearly with excitation intensity. The Grey
area represents 95% confidence region.
Figure 6(c)~(d) An analysis of semisteady state (S) and peak (P) in induced
transients with different illumination intensity. The laser intensity was applied
from 3- 40 kW/cm2 for S analysis while 3- 40 kW/cm2 for P analysis. The
responding time ranges from 1- 6 seconds for S state; 0.3- 1.2 second for P state.
Upper panel: linear regression of illuminating intensity and time for semisteady
state in transients (S). Lower panel: linear regression of illuminating intensity and
time for peak in transients (P). Grey area represents 95% confidence region.
Discussion
We have successfully obtained the Kautsky curve, as well as its intensity
dependence, with the confocal imaging fluorometer. Comparing to conventional wide-
field imaging fluorometers, the confocal technique allows much better spatial
confinement due to optical sectioning capability, and thus observation from a single
chloroplast becomes possible. It's very interesting to notice that the Kautsky curves
acquired by the confocal fluorometer are similar to those acquired by the wide-field
fluorometers. In terms of the half-life period of the slow decay, the ensemble results
of the confocal fluorometer agree very well to the wide-field ones, but only the
confocal fluorometer is capable to identify the difference among individual plant cells
or chloroplasts.
In terms of the temporal resolution performance, the confocal and wide-field
fluorometers should be similar in terms of a single pixel detection, which takes about
1 – 10 μs in both cases. As we mentioned in [17], the wide-field fluorometer takes
about 1 second to record one image. Nevertheless, the advantage of the confocal
scheme is the freedom to select number of pixels, as well as the position of these
pixels, significantly enhancing the temporal responses. By using more advanced
scanning approaches, such as random-access microscopy [34], high-speed CF
detection among distant chloroplasts is possible. In addition, by adopting a multi-
focus scanning approach, such as being demonstrated by spinning disk confocal
microscopy in 2009 [35], the frame rate of confocal fluorometer can be significantly
improved. Please note that when using the spinning disk technique, the illumination
intensity has to be calculated by the focus spot size, not by the imaging area, so the
intensity description in [35] should be multiplied by at least 103.
Another important aspect to notice is that the illumination intensity of the confocal
fluorometer is much higher than that of the wide-field fluorometers. As shown in Fig.
5 and Fig. 6, to eliminate the semi-steady state S in the CF transient, about 55
kW/cm2 is required for the confocal fluorometer. However, in the case of wide-field
fluorometer, as shown in the example of Fig. 2 [1], to eliminate S, 3200 μmol/m2/s is
required. Considering the wavelength to be 650 nm in [1], the photon energy is
1240/650 = 1.9 eV = 3 10-19 J. Therefore, the intensity unit (μmol/m2/s) is
equivalent to [10-6 6 1023 (# of photons)] [3 10-19 (J/photon)] / 104 cm2 / s = 18
10-9 kW/cm2. As a result, in the wide-field fluorometer, the required illumination
intensity is 3200 18 10-9 kW/cm2 = 5.76 10-5 kW/cm2, six orders of magnitude
smaller than that in the confocal one.
To explain this 6-order intensity difference, optical sectioning and illumination
time of the confocal imaging fluorometer have to be considered. In a conventional
fluorometer (wide-field detection), CF signals are emitted throughout the whole leaf
in the axial direction, so the depth of field (i.e. signal collection depth) is equivalent to
the thickness of a leaf, which is usually 100-1000 µm. On the other hand, for a
confocal fluorometer, a pinhole is inserted before the detector to reject most out-of-
focus fluorescence, and thus the total signal strength is significantly reduced. The
typical depth of field in a confocal fluorometer is about 1-10 µm, which is two orders
less than that of the wide-field one. Hence, the signal strength of the confocal
fluorometer is expected to be two orders weaker than the wide-field counterpart.
In terms of the illumination time, in a conventional wide-field imaging
fluorometer, the whole leaf sample is illuminated continuously, so the illumination
time for each pixel is the same as the frame acquisition time. On the other hand, a
small laser focus scans across the sample in the confocal scheme, making the
illumination time for each pixel much shorter than the frame time. For example, in the
case of Fig. 4(a2), one frame takes about 1 second, and the frame is composed of 256
× 256 pixels, so the illuminating time for each pixel (1 pixel is roughly 1 µm2 in this
case) of the confocal imaging fluorometer is about four orders shorter than that of
conventional wide-field imaging fluorometer.
Combining the above two reasons, it is reasonable that the illumination intensity in
the confocal imaging fluorometer needs to be much higher than that in the wide-field
fluorometer to achieve similar CF signal strength, as well as the Kautsky curves. The
latter is somewhat surprising since it indicates that the physiological response of the
chlorophyll remains the same with such high-intensity, yet short-period, illumination.
One possible reason is that there is a slow reaction during photosynthesis and CF
generation, so the chlorophyll only "feels" the average intensity, not the instantaneous
intensity. By looking into the electron transport chains in the photosystem, the
bottleneck reaction might be the reduction of plastoquinone (PQ), which has a
relatively slow reaction rate (100 molChl mmol−1 s−1) [36]. Further studies are
necessary to identify the underlying photochemical mechanism.
Conclusion
In this work, we demonstrated a novel confocal imaging fluorometer that can
provide high spatiotemporal characterization of CF inside a living leaf. The three-
dimensional spatial resolution is on the order of micrometer, and the temporal
resolution reaches tens of milliseconds, allowing us to study CF transient, i.e. the
Kautsky effect, from even a single chloroplast. Although the ensemble behavior of CF
transient, as well as the intensity-dependent Kautsky curves, agree well with the
results of conventional wide-field fluorometers, confocal imaging fluorometer
provides valuable information toward the difference of CF decay rate among
individual chloroplasts. The features of optical sectioning and laser focus scanning in
the confocal fluorometer result in much higher illumination intensity compared to
conventional techniques, while maintaining normal cellular physiological responses.
Our work not only opens up new possibilities to study CF dynamics on the level of
organelles, but also is promising to unravel more spatial/temporal details in the
associated photosynthetic processes.
Competing interests
The authors declare that they have no competing interests.
Author's contribution
YCT designed the experiment, carried out signal analysis, and wrote most of the
manuscript. SWC envisioned the idea, provided the experimental hardware, and
helped to polish the manuscript.
Acknowledge
The authors appreciate the inspirational discussion with Prof. Govindjee from
University of Illinois at Urbana-Champaign. This work is supported by the Molecular
Imaging Center of NTU (105R8916, 105R7732), and by the Ministry of Science and
Technology, Taiwan, under grant MOST-102-2112-M-002-018-MY3 and MOST-105-
2628-M-002-010-MY4. SWC acknowledge the generous support from the Foundation
for the Advancement of Outstanding Scholarship.
Author details
References
1.
Strasser, R.J., A. Srivastava, and Govindjee, Polyphasic chlorophyll a
fluorescence transient in plants and cyanobacteria. Photochemistry and
photobiology, 1995. 61(1): p. 32-42.
2.
Zarco-Tejada, P.J., et al., Chlorophyll fluorescence effects on vegetation
apparent reflectance: I. Leaf-level measurements and model simulation.
Remote Sensing of Environment, 2000. 74(3): p. 582-595.
3.
Krause, G. and E. Weis, Chlorophyll fluorescence and photosynthesis: the
basics. Annual review of plant biology, 1991. 42(1): p. 313-349.
4.
Horton, P., A.V. Ruban, and R.G. Walters, Regulation of Light Harvesting in
Green Plants (Indication by Nonphotochemical Quenching of Chlorophyll
Fluorescence). Plant Physiology, 1994. 106(2): p. 415.
5.
Dobrowski, S., et al., Simple reflectance indices track heat and water stress-
induced changes in steady-state chlorophyll fluorescence at the canopy scale.
Remote Sensing of Environment, 2005. 97(3): p. 403-414.
6.
Csintalan, Z., M.C. PROCTOR, and Z. TUBA, Chlorophyll fluorescence
during drying and rehydration in the mosses Rhytidiadelphus loreus (Hedw.)
Warnst., Anomodon viticulosus (Hedw.) Hook. & Tayl. and Grimmia pulvinata
(Hedw.) Sm. Annals of Botany, 1999. 84(2): p. 235-244.
7.
Bolhar-Nordenkampf, H., et al., Chlorophyll fluorescence as a probe of the
photosynthetic competence of leaves in the field: a review of current
instrumentation. Functional Ecology, 1989: p. 497-514.
8.
Trissl, H.-W., Y. Gao, and K. Wulf, Theoretical fluorescence induction curves
derived from coupled differential equations describing the primary
photochemistry of photosystem II by an exciton-radical pair equilibrium.
Biophysical journal, 1993. 64(4): p. 974.
9.
Schansker, G., et al., Chlorophyll a fluorescence: beyond the limits of the QA
model. Photosynthesis research, 2014. 120(1-2): p. 43-58.
10.
Papageorgiou, G.C., Photosystem II fluorescence: slow changes–scaling from
the past. Journal of Photochemistry and Photobiology B: Biology, 2011.
104(1): p. 258-270.
11.
DeEll, J.R. and P.M. Toivonen, Practical applications of chlorophyll
fluorescence in plant biology. 2012: Springer Science & Business Media.
12.
Lazár, D. and P. Ilik, High-temperature induced chlorophyll fluorescence
changes in barley leaves comparison of the critical temperatures determined
from fluorescence induction and from fluorescence temperature curve. Plant
Science, 1997. 124(2): p. 159-164.
13.
Schreiber, U., U. Schliwa, and W. Bilger, Continuous recording of
photochemical and non-photochemical chlorophyll fluorescence quenching
with a new type of modulation fluorometer. Photosynthesis research, 1986.
10(1-2): p. 51-62.
14. Mauzerall, D., Light-induced fluorescence changes in Chlorella, and the
primary photoreactions for the production of oxygen. Proceedings of the
National Academy of Sciences, 1972. 69(6): p. 1358-1362.
15.
Falkowski, P.G., et al., Relationship of steady-state photosynthesis to
fluorescence in eucaryotic algae. Biochimica et Biophysica Acta (BBA)-
Bioenergetics, 1986. 849(2): p. 183-192.
16.
Kolber, Z.S., O. Prášil, and P.G. Falkowski, Measurements of variable
chlorophyll fluorescence using fast repetition rate techniques: defining
methodology and experimental protocols. Biochimica et Biophysica Acta
(BBA)-Bioenergetics, 1998. 1367(1): p. 88-106.
17.
Barron-Gafford, G.A., et al., Herbivory of wild Manduca sexta causes fast
down-regulation of photosynthetic efficiency in Datura wrightii: an early
signaling cascade visualized by chlorophyll fluorescence. Photosynthesis
research, 2012. 113(1-3): p. 249-260.
18.
Papageorgiou, G., Light-Induced Changes in the Fluorescence Yield of
Chlorophyll a In Vivo: I. Anacystis nidulans. Biophysical journal, 1968. 8(11):
p. 1299-1315.
19.
Paddock, S., T. Fellers, and M. Davidson, Confocal microscopy. 2001:
Springer.
20.
Shotton, D.M., Confocal scanning optical microscopy and its applications for
biological specimens. Journal of Cell Science, 1989. 94(2): p. 175-206.
21. Matsumoto, B., Cell biological applications of confocal microscopy. Vol. 70.
2003: Academic Press.
22.
Hibbs, A.R., Confocal microscopy for biologists. 2004: Springer Science &
Business Media.
23.
Diaspro, A., et al. Functional imaging of living Paramecium by means of
confocal and two-photon excitation fluorescence microscopy. in International
Symposium on Biomedical Optics. 2002. International Society for Optics and
Photonics.
24.
Pawley, J. and B.R. Masters, Handbook of biological confocal microscopy.
Optical Engineering, 1996. 35(9): p. 2765-2766.
25.
Kautsky, H. and W. Appel, Chlorophyllfluorescenz und
Kohlensaureassimilation. Biochemistry, 1960. 322: p. 279-292.
26.
Bradbury, M. and N.R. Baker, Analysis of the slow phases of the in vivo
chlorophyll fluorescence induction curve. Changes in the redox state of
photosystem II electron acceptors and fluorescence emission from
photosystems I and II. Biochimica et Biophysica Acta (BBA)-Bioenergetics,
1981. 635(3): p. 542-551.
27. Maxwell, K. and G.N. Johnson, Chlorophyll fluorescence-a practical guide.
Journal of experimental botany, 2000. 51(345): p. 659-668.
28.
Stirbet, A., The slow phase of chlorophyll a fluorescence induction in silico:
Origin of the S–M fluorescence rise. Photosynthesis research, 2016: p. 1-21.
29.
Omasa, K., et al., Image analysis of chlorophyll fluorescence transients for
diagnosing the photosynthetic system of attached leaves. Plant Physiology,
1987. 84(3): p. 748-752.
30.
Kodru, S., et al., The slow S to M rise of chlorophyll a fluorescence reflects
transition from state 2 to state 1 in the green alga Chlamydomonas reinhardtii.
Photosynthesis research, 2015. 125(1-2): p. 219-231.
31.
Vermaas, W.F., et al., In vivo hyperspectral confocal fluorescence imaging to
determine pigment localization and distribution in cyanobacterial cells.
Proceedings of the National Academy of Sciences, 2008. 105(10): p. 4050-
4055.
32.
Chen, M.-Y., et al., Multiphoton imaging to identify grana, stroma thylakoid,
and starch inside an intact leaf. BMC Plant Biology, 2014. 14(1): p. 175.
33.
Stirbet, A., On the relation between the Kautsky effect (chlorophyll a
fluorescence induction) and photosystem II: basics and applications of the
OJIP fluorescence transient. Journal of Photochemistry and Photobiology B:
Biology, 2011. 104(1): p. 236-257.
34.
Reddy, G.D., et al., Three-dimensional random access multiphoton microscopy
for functional imaging of neuronal activity. Nature Neuroscience, 2008. 11(6):
p. 713-720.
35.
Omasa, K., et al., 3D confocal laser scanning microscopy for the analysis of
chlorophyll fluorescence parameters of chloroplasts in intact leaf tissues. Plant
and cell physiology, 2009. 50(1): p. 90-105.
36.
de Wijn, R. and H.J. van Gorkom, Kinetics of Electron Transfer from QA to
QB in Photosystem II. Biochemistry, 2001. 40(39): p. 11912-11922.
|
1101.0271 | 1 | 1101 | 2010-12-31T16:34:58 | Fast dynamics of odor rate coding in the insect antennal lobe | [
"physics.bio-ph",
"q-bio.NC"
] | Insects identify and evaluate behaviorally relevant odorants in complex natural scenes where odor concentrations and mixture composition can change rapidly. In the honeybee, a combinatorial code of activated and inactivated projection neurons (PNs) develops rapidly within tens of milliseconds at the first level of neural integration, the antennal lobe (AL). The phasic-tonic stimulus-response dynamics observed in the neural population code and in the firing rate profiles of single neurons is faithfully captured by two alternative models which rely either on short-term synaptic depression, or on spike frequency adaptation. Both mechanisms work independently and possibly in parallel to lateral inhibition. Short response latencies in local interneurons indicate that local processing within the AL network relies on fast lateral inhibition that can suppress effectively and specifically odor responses in single PNs. Reviewing recent findings obtained in different insect species, we conclude that the insect olfactory system implements a fast and reliable coding scheme optimized for time-varying input within the behaviorally relevant dynamic range. | physics.bio-ph | physics | Fast dynamics of odor rate coding in the insect antennal lobe
Martin P Nawrot1,2*, Sabine Krofczik2,3, Farzad Farkhooi1,2, and Randolf Menzel2,3
1 Neuroinformatics & Theoretical Neuroscience, Freie Universität Berlin, Germany
2 Bernstein Center for Computational Neuroscience Berlin, Germany
3 Institute of Biology – Neurobiology, Freie Universität Berlin
Insects identify and evaluate behaviorally relevant odorants in complex natural scenes where
odor concentrations and mixture composition can change rapidly. In the honeybee, a
combinatorial code of activated and inactivated projection neurons (PNs) develops rapidly
within tens of milliseconds at the first level of neural integration, the antennal lobe (AL). The
phasic-tonic stimulus-response dynamics observed in the neural population code and in the
firing rate profiles of single neurons is faithfully captured by two alternative models which
rely either on short-term synaptic depression, or on spike frequency adaptation. Both
mechanisms work independently and possibly in parallel to lateral inhibition. Short response
latencies in local interneurons indicate that local processing within the AL network relies on
fast lateral inhibition that can suppress effectively and specifically odor responses in single
PNs. Reviewing recent findings obtained in different insect species, we conclude that the
insect olfactory system implements a fast and reliable coding scheme optimized for time-
varying input within the behaviorally relevant dynamic range.
Keywords: combinatorial code, latency code, lateral inhibition, short term depression, spike
frequency adaptation, olfaction, honeybee
1. Introduction
In their natural environment, insects sense and evaluate blends of olfactory stimuli on a
complex background. In the honeybee, for instance, behaviorally relevant odorants range
from simple odors composed of single or few chemical compounds e.g. in pheromones used
as communication signals, to highly complex odor blends e.g. in food signals and flower
odors. Due to the turbulent nature of odor plumes, airflow, and animal movement the
concentration of odor blends fluctuate on variable time scales (e.g. Riffell et al., 2008). Thus,
under natural conditions, the olfactory system must deal with a complex and highly dynamic
input. Behavioral experiments showed that bees discriminate odor concentrations (Pelz et al.
1997, Wright et al. 2005) and that they distinguish between odor mixtures (Chandra and
Smith, 1998). More recently, it has been shown that bees can learn to discriminate odors that
are presented for only a very short period of 200ms (Wright et al., 2009; Fernandez et al.,
2009). Longer stimuli may be required to reach the same learning rate and performance in
odor discrimination under challenging conditions such as for low odor concentrations (Wright
et al., 2009) or when discriminating different mixture ratios (Fernandez et al., 2009). A
similar speed of discriminating dissimilar odors (Abraham et al., 2004) , different mixture
* Correspondance to: [email protected] Martin Paul Nawrot, Neuroinformatik, Freie Universität Berlin,
Königin-Luise-Strasse 1-3, 14195 Berlin
1
ratios of a binary mixture (Uchida and Mainen, 2003) or novel odors from learned odors
(Wesson et al., 2008) has been demonstrated in behavioral experiments with rodents.
Electrophysiological recordings from the mushroom body of honeybees showed that the
ensemble response of MB output neurons indicate the presence of a learned odor within
<200ms (Strube-Bloss, Nawrot, Menzel, submitted manuscript; Pamir et al., 2008). Odors are
thus encoded fast and reliably allowing the animal to track rapid changes in its environment.
The basic anatomy of the olfactory system is conserved across different insect species. In the
honeybee, it comprises 60,000 olfactory sensory neurons (OSNs; Esslen and Kaissling, 1976),
which are located predominantly in pore plate sensillae on the antennae. The antennal lobe
(AL) represents the first stage of olfactory processing. It is composed of ~160 glomeruli
which represent the synaptic sites between OSNs, local interneurons (LNs), and the projection
neurons (PNs), the latter forming the output of the antennal lobe (Fig. 1). Each OSN conveys
chemosensory information through one of the four antennal tracts (T1-4, Fig. 1A, Abel et al.
2001). Anatomically we distinguish two types of uniglomerular PNs. The lateral PNs (l-PNs)
receive input exclusively from T1 glomeruli (Figure 1A, green circles) and send their axons
along the lateral antennocerebralis tract to higher order brain centers, first to the lateral horn
(LH) and then to the mushroom body (MB). The median PNs (m-PNs) exclusively originate
in T2-4 glomeruli (Fig. 1A, magenta circles) and project along the median antennocerebralis
tract, first to the MB and then to the LH (Abel et al. 2001, Kirschner et al., 2006). While the
basic anatomical outline of the early olfactory system is similar in different insect species,
there are also marked inter-specific differences. For instance, the division into lateral and
median PN tracts exists in e.g. bees and ants (Hymenoptera), flies (Diptera), cockroaches
(Blattaria), and moths (Lepidoptera) but not in the locust (Orthoptera) (for a detailed review
see Galizia and Rössler, 2009). Additional differences relate to the size and structure of the
network of LNs in the AL. In the locust, the interneuron network comprises only about 300
inhibitory LNs which appear not to lead to sodium action potentials (Laurent and Davidowitz,
1996). In comparison, the honeybee AL network is composed of ~4000-5000 spiking
interneurons (Fonta et al., 1993, Abel et al. 2001). Excitatory interneurons were found in
Drosophila (Shang et al., 2007) but it is not known whether the same applies to the bee and
other insects. More commonality exists between species with respect to the number of PNs
which appear in the order of ~950 in the honeybee (Rybak & Eichmueller 1993; Kirschner et
al., 2006) and 830 in the locust (Leitch and Laurent, 1996).
We investigated neural processing in the AL network and odor encoding at the level of PNs
by means of in vivo intracellular recordings from individual PNs and LNs of the honeybee.
Single neurons were identified as LN, l-PN or m-PN by intracellular dye marking, the exact
recording position, and their spiking pattern (Krofczik et al., 2008). Successfully stained
neurons were subsequently reconstructed and registered into the honeybee standard brain
(HSB) atlas (Fig. 1B; Rybak et al., 2010; Brandt et al., 2005). Focusing on the early temporal
dynamics of odor processing in the AL, we review here our results in the context of recent
findings in other insect species. Specifically we address the following questions. Which
mechanisms could explain the experimentally observed stimulus-response dynamics of PN
firing? Can we determine an olfactory code carried by the PN population that evolves with
sufficient speed to explain the behavioral results of fast odor discrimination? How does local
inhibition interact with the excitatory input from OSNs to form PN output during this initial
response phase? And, finally, we asked whether the anatomically distinct types of
uniglomerular l-PNs and m-PNs could play a different role in odor processing.
2
2. Dynamic odor responses of projection neurons – experiment and model approaches
We investigated the in vivo response dynamics of single PN firing and of the PN ensemble
rate to constant odor stimuli. The prominent phasic-tonic rate profile can be captured by a
simple phenomenological input-output model. We review two alternative biophysical models,
which can provide a mechanistic explanation for our experimental observations.
2.1. Phasic-tonic response characteristic of PNs
First we tested the stimulus-response dynamics of single uniglomerular PNs in response to
odor pulses of 2s duration delivered through computer controlled valves. The intracellular
measurement of action potentials (Fig. 2A) during repeated stimulus presentations resulted in
series of single trial spike trains (Fig. 2B) from which we estimated the trial-averaged and
time-resolved firing rate as depicted in Fig. 2C. The excitatory PN response is composed of an
early phasic response (i.e. a transient firing rate) during the first 300ms - 600ms after the
stimulus onset, and a prolonged tonic response (i.e. a nearly constant firing rate) that outlasts
the stimulus duration. This phasic-tonic response type makes up for the largest part of the
observed PN responses and shapes the population response (Fig. 3D,E). The estimated mean
response latency, i.e. the temporal delay between stimulus onset and the earliest detection of
the excitatory PN response, was 130.8 ms. In few cases, we observed a delayed excitatory
response, which built up gradually with typical response latencies above 500ms (Müller et al.,
2002; cf. Fig. 3B). These cannot actively contribute to a fast encoding of the odor identity
during the early stimulus part. Finally, about 10% of all odor responses were of inhibitory
nature, expressed in a reduction of the spontaneous firing rate (Krofczik et al., 2008; Fig. 3B
and Fig. 4A-D). In addition, several neurons showed excitatory responses to the offset of a
stimulus. The excitatory phasic-tonic rate response profile is stereotypic in the honeybee
(Abel et al., 2001; Müller et al., 2002; Galizia and Kimmerle, 2004; Krofczik et al., 2008),
and one prominent feature of PN responses that has been commonly observed in various other
insect species (see Discussion), e.g. in the fruitfly (Drosophila melanogaster; Bhandawat et
al., 2007; Wilson et al., 2004, Olsen et al., 2007), in the silkmoth (Bombyx mori; Namiki and
Kanzaki, 2008), and in the locust (Schistocerca Americana; Stopfer et al., 2003; Mazor and
Laurent, 2005). The same stereotypic response dynamics has also been reported for mitral
cells in the olfactory bulb of the zebrafish (Danio rerio; Tabor et al., 2004; Friedrich and
Laurent, 2004).
2.2. Models of global PN response dynamics
Recently, Meister and colleagues (Geffen et al., 2009) presented a phenomenological rate
model for the response dynamics of individual PNs. Their model translates a stimulus S(t)
with time-depending amplitude into a time-varying firing rate R(t) in two steps (Filter model
in Fig. 2D-E). First, the time-varying stimulus S(t) passes through a linear filter f(t). In a
second step, the filtered signal g(t) is rectified according to a non-linear input-output function
G to produce the time-varying firing rate R(t). Crucial to this model is the adequate definition
of the filter function. To this end, Geffen et al. (2009) analyzed a large set of extracellular PN
recordings from the locust in response to 100ms short odor pulses. For each neuron they
obtained a non-parametric estimate of the filter shape f(t), approximating the linear impulse
response function. A subsequent principal component analysis showed that a combination of
the first two principal components could account for most of the single neuron response types.
Fig. 2B presents a simplification of this model accounting for excitatory responses. We
defined a parametric filter shape f(t) depicted in Fig. 2D with positive and negative
coefficients, which resembles the typical filter shape for excitatory PN responses in Geffen et
3
al. (2009). Convolution of a step stimulus S(t) with the filter shape f(t) and subsequent
rectification with the linear gain function G produces a phasic-tonic rate profile (red curve in
Fig. 2F) that qualitatively reproduces the experimental rate response profiles in Fig. 2C and
Fig. 3D,E.
The model by Geffen et al. (2009) provides a phenomenological description of the
experimentally observed rate response dynamics. However, it does not explain the underlying
physiological mechanisms. Here, we consider two candidate biophysical mechanisms that can
explain PN response dynamics. The first model is based on the recent finding by Kazama &
Wilson (2008). They discovered that excitatory synapses from the OSNs onto the PNs exhibit
synaptic short term depression (STD) as discovered by intracellular PN recordings combined
with the electrical stimulation of the antennal nerve. We captured this finding in our STD
model (Fig. 2G-I). For the depressive synapses between ORNs and PNs we used the model of
Tsodyks and Markram (Tsodyks and Markram,1997; Tsodyks et al., 1998) (Fig. 2G). The
recovery time constant determines the temporal dynamics of recovering the maximum
conductance amplitude. We chose rec=500ms. The postsynaptic PN is modeled as a
conductance based leaky integrate-and fire (LIF) neuron (Müller et al., 2007). We estimated
that each PN receives synaptic input from a population of 60 OSNs, which were modeled by
independent Poisson spike trains, each with a spontaneous firing rate of 4 Hz. During odor
stimulation, the presynaptic OSN firing rate increases by 7Hz. The trial-averaged PN firing
rate (Fig. 2I) reveals a phasic-tonic stimulus response, which matches the experimentally
observed phasic-tonic response dynamics in Fig. 2C reasonably well.
Elsewhere (Farkhooi et al., 2009b; Farkhooi et al., 2010), we used an alternative approach of
modeling odor induced PN responses which assumes a neuron-intrinsic mechanism of spike
frequency adaptation (SFA) in PNs, following the model by Sivan and Kopell (2006). In vitro
recordings from dissociated and cultured neurons from pupae of the honeybee (Grünewald,
2003) and the moth (Manduca sexta, Mercer and Hildebrand, 2002) have described calcium-
dependent K+ currents in PNs, which suggests a possible SFA mechanism. Our SFA model is
depicted in Fig. 2K-M. Again, we modeled the PNs by a conductance based LIF neuron, this
time incorporating a phenomenological model for SFA (Müller et all, 2007). This model
introduces a negative SFA conductance which adds to the synaptic conductance. With each
action potential, the model introduces a negative jump in the SFA conductance with
amplitude j=-1.5nS, which decays with a decay time constant SFA=400ms (Fig. 2K), that is
considerably longer than the membrane time constant (m=20ms). These two parameters
govern the response dynamics for transient inputs. We again provided Poissonian input from
the OSN population with a spontaneous rate of 300 Hz (equivalent to 60 neurons, each with a
rate of 5 Hz). During stimulation, the Poisson input was increased by 400Hz. In Fig 2L we
show repeated single trial responses of the model when stimulated during 2s. The phasic-tonic
rate response profiles in Fig. 2M match reasonably well the experimental observations in Fig.
2C.
3. Rapid odor encoding in the projection neuron population
The OSNs provide the olfactory input to the AL where it is processed within the local
network. The PNs divided into several different types represent the only AL output and thus
carry all sensory information about odor identity, odor intensity, and stimulus dynamics that
will enter the calyx of the mushroom body and the lateral horn. Here, we review two potential
coding schemes for odor identity: a combinatorial code of activated and inactivated PNs, and
a latency code where the stimulus is encoded in the spatio-temporal pattern of neuronal
response onsets.
4
3.1. Combinatorial code
A combinatorial odor code was first discovered in the honeybee in a series of Ca - imaging
studies (Joerges et al, 1997, Galizia and Menzel, 2000; Galan et al., 2004). Stimulation with
different odors revealed different spatial patterns of activated and inactivated glomeruli within
the AL. Similar findings were reported for other insect species, e.g. Drosphila (Wang et al.,
2003; Root et al., 2007), the ant Camponotus (Galizia, Menzel, Hölldobler, 1999), and the
moth Heliothis (Skiri et al. 2004). Each single glomerulus receives input only from OSNs
expressing the same chemoreceptor in Drosophila (Vosshall et al., 2000), and it is generally
believed that the same applies to other insect species. Thus, activation of a single glomerulus
reflects the chemoprofile of the respective receptor type, and the glomerular activation pattern
reflects the combination of activated receptor types.
We investigated the combinatorial code in our intracellular recordings from uniglomerular
PNs (Fig. 3). To this end, we pooled separate single neuron measurements from different
animals to construct a pseudo population of PNs, which we treated as if they had been
recorded simultaneously in the same animal. Fig. 3A shows the chemoprofiles of different
PNs and indicates that different odors evoke different patterns of neural activity. For instance,
the odor heptanon (2.7on) activates a subset of neurons, while other neurons remained silent
or were even suppressed in their spontaneous activity. Hexanol (6ol) activates a different but
overlapping subset of neurons. We may consider this type of neural ensemble activation as a
binary response pattern where each neuron will either be switched on or off in response to the
stimulus.
In our experiments, single neurons were activated by approx 45%-65% of the tested odors
(Krofczik et al., 2008). These numbers match well with those obtained in an independent
study that used Ca-imaging of PN presynaptic boutons in the honeybee mushroom body lip
region (Yamagata et al., 2009) and our results are in line with studies in different species
where a rather broad odor tuning was observed in single PN recordings (Wilson et al., 2004;
Schlief and Wilson, 2007; Perez-Orive et al., 2002). This means, in turn, that each odor
activates about half of the total PN population. If we assume a fixed number k of activated
PNs from a total population of n neurons, then, by simple combinatorics, it follows that we
can represent the largest number of different binary patterns (i.e. the largest number of
different odorants) if k=n/2. Thus, the empiric finding that, on average, about half of the
glomeruli are activated by a specific odor may indicate that the insect olfactory system is
optimized with respect to a maximized coding capacity. Noteworthy, the result of a broad
odor tuning in insect PNs is in contrast to the results in rodents where a sparse odor
representation was found in mitral cells of the olfactory bulb (Davison & Katz, 2007; Rinberg
et al., 2006).
How fast do the binary response patterns evolve in the PN population? Fig. 3B plots the firing
rate of 17 PNs as a function of time and in response to three different odors. The color
intensity represents the amplitude of the excitatory (red) or inhibitory (blue) rate change from
baseline. We find that the binary pattern (Fig. 3A) establishes rapidly. The amplitude of
individual activated neurons varies with odor and, thus, the amplitude yields additional
information above the binary activation pattern. In Fig. 3C we computed the average pair wise
distance d of the odor-specific population rate vectors in a time-resolved manner (Krofczik et
al., 2008). It shows that the response patterns evoked by different stimuli become significantly
different after only few tens of milliseconds. The maximum difference in firing rate of
average ~32 spikes/s per neuron is reached after ~150-200ms. A recent Ca-imaging study of
5
PNs in the honeybee AL (Fernandez et al., 2009) found a somewhat longer time to peak of
375ms which might be due to the slower Ca transients. Our observation is in accordance with
the results obtained from spike train analyses in other species. In the locust odor classification
as monitored by extracellular recordings of PN ensembles reached the near-maximum after
100–300 ms (Mazor and Laurent, 2005; Stopfer et al., 2003). In Drosophila odor tuning in
single PNs was strongest in the period of 130–230ms after stimulus onset (Wilson et al.,
2004), and in the moth a maximal separation of the PN population activity was reached 100–
300 ms after response onset (Bombyx mori, Namiki and Kanzaki, 2008; Namiki et al., 2009;
Manduca sexta, Daly et al., 2004; Staudacher et al., 2009). The activated neural ensemble
leads to a phasic-tonic response profile which is similar for all three odors (Fig. 3D,E). The
temporal evolution of the coding distance d is thus a result of the dominant phasic-tonic rate
dynamics of the activated neurons and the encoding is strongest during the initial phasic
response epoch.
3.2. Latency code
Visual inspection of single neuron spike responses to different odor stimuli give the
impression that different odors yield different response onset times in the same neuron (e.g.
Perez-Orive et al., 2002; Mueller et al., 2002; Wilson et al, 2004; Krofczik et al., 2008).
Likewise, the spike responses of a neuronal population to the same odor stimulus can show
neuron-specific response latencies (e.g. Wilson et al., 2004; Namiki & Kanzaki, 2008). This
suggests that the PN population may employ a latency code (Chase & Young, 2007; Gollisch
& Meister, 2008). We therefore statistically tested for the odor-specificity of single neuron
response latencies. To this end, we estimated the response onset in a spike train as the
maximum steepness of the initial rate increase (Krofczik et al., 2008). We found that the
variation of response latencies across odors (standard deviation: 47ms, pooled across all PNs)
is significantly larger than the variation of response latencies across repeated stimulations
with the same odor (standard deviation: 26ms, pooled across all PNs). Thus, on statistical
grounds, response latencies of single PNs are odor specific. In a population of PNs this
translates into an odor specific spatio-temporal pattern of response onset times across the
activated neural ensemble.
Where do these differences in response timing originate? Spors et al. (2006) performed
voltage sensitive dye imaging of single glomeruli in the mouse olfactory bulb and found that
odor-specific response latencies exist already at the level of OSN input to the olfactory bulb.
Using a statistical procedure similar to ours the authors estimated a standard deviation of ~40-
50ms across different glomeruli and a trial-to-trial standard deviation of ~20ms. Both results
quantitatively match ours (Krofczik et al., 2008). This may suggest that odor specific response
latencies evolve early on in the periphery, e.g. due to a receptor type specific activation
kinetics.
4. Fast inhibition might be essential for creating the early rate code
It is known from Ca imaging studies in the honeybee AL (Joerges et al., 1997; Deisig et al.,
2010) that single glomeruli can exhibit mixture suppression where the glomerulus is activated
for one or more of the individual components of a mixture, while the response to the mixture
itself is suppressed. Here we asked whether mixture suppression is seen in the subthreshold
graded potentials and spiking activity of uniglomerular l-PNs and m-PNs, and whether lateral
inhibition by LNs could underlie early response suppression.
6
4.1. Local inhibition leads excitatory projection
Fig. 4A presents an example of mixture suppression in a single l-PN which shows clear
responses to three compound odors when tested alone, but no response to the mixture of all
three compounds (Krofczik et al., 2008). Interestingly, in Fig. 4A and B we observe complete
suppression of the spontaneous activity and not even a single response spike indicating the
existence of a fast suppression mechanism. To explore the possibility that local inhibitory
interneurons cause this suppression we first analyzed the membrane potential of single PNs
during suppressed responses and found that it can show a strong hyperpolarization with a fast
onset in the range of ~60-90ms after stimulus onset (Fig. 4C,D; Krofczik et al., 2008).
Interestingly, our set of local interneurons exhibited similarly short spike response latencies
with an average of only 70.0ms (Fig. 4E,F). In contrast, PN spike responses showed a
considerably longer latency with average 130.8ms. Thus, LNs can respond fast enough to
suppress l-PN responses by means of lateral inhibition. Our finding of faster responses in LNs
have recently been supported in a Ca imaging study from labeled neurons in the moth AL
(Fujiwara & Kanzaki, 2009).
4.2. Do l-type and m-type projection neurons serve different roles?
Ca imaging of the honeybee AL has accessed so far only the most frontal glomeruli
innervated by the T1 tract of OSNs. These glomeruli are innervated exclusively by l -PNs that
form the lateral tract (Fig. 1). To investigate mixture interaction on the level of single
uniglomerular PNs we systematically tested in a set of neurons one tertiary mixture and its
individual components (hexanol, nonanol, heptanone). As expected from the Ca imaging
result, we could observe mixture suppression in our recordings from l-PNs. On average, l-PNs
showed inhibited spike responses (i.e. a suppression of the spontaneous firing rate) in 12% of
all stimulations with complex odors that were composed of a least three chemical compounds
(Krofczik et al., 2008). Surprisingly, we did not observe such suppression in the firing rates of
uniglomerular PNs from
the median antennocerebralis
tract,
indicating a possible
specialization of the l-PNs in mixture suppression. This hypothesis received recent support by
Yamagata et al. (2009) who performed Ca imaging at the level of synaptic terminals (boutons)
of uniglomerular PNs in the mushroom body lip region of the honeybee. Testing mixture
interaction for two binary mixtures, they observed mixture suppression for both tested
mixtures in about 5-20% of the l-PN boutons. In contrast, for m-PN boutons mixture
suppression was detected for only one of the mixtures in less than 4% of all boutons.
5. Conclusions and Discussion
The olfactory ensemble rate code evolves rapidly in the population of uniglomerular PNs of
the honeybee. This result is in line with recent findings in other insect species. Different odors
lead to significantly different ensemble rates within only few tens of milliseconds (Fig. 3).
Throughout the constant odor stimulus, the distance between two odor vectors is largely
determined by the combinatorial pattern of activated and inactivated neurons in the PN
population. The exact time course of this distance likely reflects a direct consequence of the
average phasic-tonic excitatory response profile of the single PNs, indicating an emphasis on
the initial stimulus epoch. We propose two biophysical mechanisms that can explain this
dominant phasic-tonic PN response dynamics either separately or in combination. Both
models work independently of lateral inhibition. The first mechanism is synaptic STD (Fig.
2G-I) which has been demonstrated for the synapses between OSNs and PNs in the fruit fly.
The second mechanism of SFA (Fig. K-M) relates to a ubiquitous phenomenon in neural
7
systems and different physiological mechanisms and computational models have been
described that can mediate SFA (Benda and Herz, 2003; Prescott and Sejnowski, 2008). With
respect to sensory computation in insects, SFA has been demonstrated to play an important
role in early auditory processing (Benda and Hennig, 2008; Hildebrandt et al., 2009), and it
has been suggested for neurons in the visual system of the locust (Peron an Gabbiani, 2009).
Sivan and Kopell (2006) have previously assumed an SFA mechanism in PNs in their
network model of the AL but experimental evidence for an SFA mechanism in AL neurons is
still weak (Mercer and Hildebrand 2002; Grünewald 2003). The neuron intrinsic property of
SFA can bear additional advantages, in particular it can reduce neuronal output variability
(e.g. Prescott and Sejnowski, 2008; Farkhooi et al, 2009a; Nawrot 2010) and could thus, at
least in parts, explain the observed lower response variability in PNs as compared to that of
OSNs (Bhandawat et al, 2007; Masse et al., 2009)
Local inhibition by LNs is fast and can effectively suppress single PN responses. This
suggests that one major role of the local inhibitory network is to rapidly process the peripheral
OSN input and to help constructing the combinatorial code in the PN output by silencing
single PNs, particularly in the case of mixture suppression of l-PNs. Such non-linear
processing supports previous work that underlined the importance of the local AL network in
shaping the olfactory code (e.g. Sun et al., 1993; Sachse and Galizia, 2002; Galizia and
Kimmerle, 2004; Wilson and Laurent, 2005; Stopfer, 2005; Tanaka et al., 2009), including
recent experimental evidence of learning-induced plasticity in the antennal lobe (Faber et al.,
1999; Galizia and Menzel 2000; Fernandez et al., 2009; Denker et al., 2010).
Adaptation in olfactory sensory neurons
Across species, OSNs show an adaptation of their response rate during constant odor
stimulation, which is typically weaker and slower than the adaptation observed in the PN
firing rate (e.g. de Bruyne et al., 1999; Bhandawat et al., 2007). The model of Sivan and
Kopell (2006) modeled OSN input by a nonhomogeneous Poisson process following phasic-
tonic intensity profile with a slow decay time constant. For simplicity we assumed a Poisson
input for both our models (Fig. 2G-M), which steps from an initial low intensity to a higher
intensity during odor stimulation. For this input model and appropriate parameters of strength
j and time constant τSFA we obtain a realistic phasic-tonic rate response profile. Including a
layer of slow adapting OSNs required weaker adaptation in the PNs (not shown; cf. Sivan and
Kopell, 2006) and did not qualitatively alter our results.
Slow temporal patterning of PN responses
Previous analyzes of PN responses often stressed the feature of a slow temporal patterning of
PN spiking in response to a constant stimulus observed in numerous species including the
honeybee (Müller et al., 2002), the locust (e.g. Stopfer et al., 2003; Brown et al., 2005) and
the moth (e.g. Daly et al., 2004; Ito et al., 2008). Response patterns can be composed of one
or several short response epochs of excitation and inhibition and often temporally extend
beyond the stimulus duration, and they may include off-responses. As suggested in the
modeling study of Sivan and Kopell (2006) the mechanism of SFA might play an active role
in slow patterning of PN responses because strong input can lead to an initial burst of output
spikes which can induce a strong adaptation current. This will act as a type of self-inhibition
that may suppress output spikes. After recovery from adaptation, spiking can reoccur.
PN response types were coarsely categorized in our data set. The most frequent type
resembled a stereotypic phasic-tonic excitatory response (Fig. 2A-C, Fig. 3B-D, Fig. 4). In
8
some cases it was very brief and was followed by inhibition during the later part of the
stimulus period. A second obvious stereotype was an inhibitory response where spontaneous
activity was suppressed (Fig. 3B,D, Fig. 4A-D). Often, this suppression was not sustained
throughout the complete stimulus duration, and spiking activity returned during the later part
of the stimulus (e.g. Fig. 4B). We here address the question how odor encoding in the
honeybee can be achieved so rapidly at the level of the AL output. Our analyses focused on
the very early response epoch. We find a combinatorial rate code that evolves within a few
tens of milliseconds, and which is carried by the presence or absence of an early phasic rate
response in a subset of the PN population. The phasic-tonic response type is most important
for this fast (<200ms) encoding scheme, while slow patterning during the later stimulus
epochs and beyond may serve different roles not considered here. In particular, such slow
response components may be relevant for refining evidence about odor identity under more
challenging conditions e.g. in the case of low odor concentration, a high olfactory
background, or for decoding additional stimulus features (Friedrich and Laurent, 2001). In
addition, the late response phase might be important for the acquisition of odor memories
where associations can be formed between short olfactory stimuli and a delayed reward.
Oscillations in the AL network
Oscillatory network activity in the AL, typically in the range of 10-40Hz, has been
demonstrated in various species, including the locust (e.g. MacLeaod and Laurent, 1996;
Perez-Orive et al., 2002), the honeybee (Stopfer et al., 1997), the fly (Tanaka et al., 2009), and
the moth (Ito et al., 2009). Blocking of GABAergic inhibition abolishes oscillations which
indicates that these are caused by on one or more local inhibitory networks. In the locust, the
oscillatory cycle has been implicated to restrict the integration time window in Kenyon cells,
imposing a mechanism for rhythmic coincidence detection at the level of the MB (Assisi et
al., 2007). The role of AL oscillations may be a different one in different species (Masse et
al., 2009; Tanaka et al., 2009). In the fly, oscillatory activity kicks in only in the late phase of
an odor stimulus and thus is likely irrelevant for a fast encoding of odor identity (Tanaka et
al., 2009). Also, oscillations transfer to the intracellular subthreshold activation of KCs in the
locust, but not in the fly (for review see Masse et al., 2009).
Little is known about the role of oscillations in the honeybee. Stopfer et al. (1997) showed
that abolishing oscillations may lead to a desynchronization among PN response spikes, while
the apparent slow response features evolving on a time scale longer than the typical
oscillation cycle remained unchanged. In our firing rate analyses, we thus ignored possible
fast oscillatory components in the spike trains. In our experiments, we devised intracellular
recordings from LNs and axons of uniglomerular PNs, which allowed us to determine cell
type (LN, l-PN, m-PN). However, we did not observe apparent strong oscillatory activity in
the neurons' membrane potential. Generally, single cell recordings are not well suited to
detect and analyze oscillatory activity and the role of oscillations in the honeybee AL should
be addressed by means of extracellular multiple single unit recordings, possible paired with
LFP recordings.
Encoding of stimulus dynamics and temporal sparseness in Kenyon cells
The uniglomerular PN ensemble activity provides the only olfactory input to the calyces of
the mushroom body (MB) where they connect to a large number of Kenyon cells (KCs; total
~160.000 in the bee). KCs show temporally sparse responses in the bee (Szyszka et al., 2006),
the locust (Perez-Orive et al., 2002; Brown et al., 2005; Broome et al., 2006), and the moth
(Ito et al., 2008), i.e. they respond with only few spikes to the onset or offset of a constant
9
odor stimulus. Different forms of inhibition dynamics have been suggested to be responsible
for the temporally sparse KC responses (Perez-Orive, 2002; Szyszka et al., 2005; Jortner et
al., 2007; Assisi et al., 2007). We recently suggested an alternative mechanism that produces
temporal sparseness in KCs independent of inhibition (Farkhooi et al., 2009b), which relies on
the assumption that KCs exhibit SFA. This assumption has recently gained experimental
support by Demmer and Kloppenburg (2009) who studied in detail the physiology of KCs of
the cockroach (Periplaneta Americana). In a network model of the olfactory pathway we
combined neural adaptation in two successive stages by implementing (1) SFA (or likewise
STD) dynamics in the PNs (Fig. 2), and (2) strong SFA dynamics in the KCs (Farkhooi et al.,
2010). A constant stimulus lead to phasic-tonic excitatory responses in PNs and temporally
sparse KC responses composed of very few spikes following the stimulus onset. Such a
network architecture with successive stages of fast neural adaptation approximates the
mathematical operation of temporal differentiation such that the output O(t) encodes the time
derivative of the input with O(t)=d/dt S(t) (Tripp and Eliasmith, 2009; Müller E, 2007;
Farkhooi et al., 2010). This might suggest that parts of the olfactory system of the insect could
be designed to focus on temporal changes and to largely neglect constancy in the olfactory
input. The experimental prediction is that, under naturally dynamic input conditions, the
activity of single KCs is not temporally sparse but rather driven by the naturally occurring
fluctuations of odor composition and odor concentration.
Acknowledgements
We thank Jürgen Rybak for his assistance with the visualization of registered neurons in the
Honeybee Standard Brain atlas and for his helpful comments on the manuscript. We are
grateful to Michael Schmuker for valuable discussions on data and models. We acknowledge
generous funding from the Deutsche Forschungsgemeinschaft (SFB 618), and from the
German Federal Ministry of Education and Research (BMBF) to the Bernstein Center for
Computational Neuroscience Berlin (01GQ1001D) and to the Bernstein Focus Learning and
Memory: Insect inspired robots – The role of learning and memory in decision-making
(01GQ0941).
10
References
Abel R, Rybak J, and Menzel R. (2001). Structure and Response Patterns of Olfactory Interneurons in the Honeybee, Apis
mellifera. J comp Neurol 437:363-383.
Assisi C, Stopfer M, Laurent G, Bazhenov M (2007) Adaptive regulation of sparseness by feedforward inhibition. Nat
Neurosci 10(9):1176-84
J. Benda and A. V. M. Herz (2003) A universal model for spike-frequency adaptation. Neural Comput. 15:2523-64
Benda J, Hennig RM (2008) Spike-frequency adaptation generates intensity invariance in a primary auditory interneuron, J
Comp Neurosci 24: 113-136.
Bhandawat, V., Olsen, S., Gouwens, N., Schlief, M., and Wilson, R. (2007). Sensory processing in the Drosophila antennal
lobe increases reliability and separability of ensemble odor representations. Nat. Neurosci. 10, 1474–1482.
Broome BM, Jayaraman V, Laurent G (2006) Encoding and decoding of overlapping odor sequences. Neuron 51(4):467 -82
Brown SL, Joseph J, Stopfer M. (2005) Encoding a temporally struc tured stimulus with a temporally structured neural
representation. Nat Neurosci. 8(11):1568-76
de Bruyne M, Clyne PJ, Carlson JR. (1999) Odor coding in a model olfactory organ: the Drosophila maxillary palp. J
Neurosci. 19:4520-4532.
Chandra S and Smith BH (1998) An Analysis of synthetic processing of odor mixtures in the honeybee. Journal of
Experimental Biology 201:3113-3121.
Chase SM, Young ED (2007) First-spike latency information in single neurons increases when referenced to population
onset. Proc Natl Acad Sci USA 104(12):5175-80
Daly KC, Wright GA, Smith BH (2004) Molecular features of odorants systematically influence slow temporal responses
across clusters of coordinated antennal lobe units in the moth Manduca sexta. J Neurophysiol 92:236-54
Davison, I., and Katz, L. (2007). Sparse and selective odor coding by mitral/tufted
neurons in the main olfactory bulb. J. Neurosci. 27, 2091–2101.
Deisig N, Giurfa M, Sandoz JC (2010) Antennal lobe processing increases separability of odor mixture representations in the
honeybee. J Neurophysiol. Feb 24. [Epub ahead of print]
Demmer H, Kloppenburg P (2009) Intrinsic membrane properties and inhibitory synaptic input of kenyon cells as
mechanisms for sparse coding? J Neurophysiol. 102(3):1538-50
Denker M, Finke R, Schaupp F, Grün S, Menzel R (2010) Neuronal correlates of odor learning in the honeybee antennal
lobe. Europ J Neurosci 31: 119-133
Esslen, J., and Kaissling, K. (1976) Zahl und Verteilung antennaler Sensillen bei der Honigbiene Apis mellifera L.
Zoomorphologie 83, 227–251.
Faber T, Joerges J, and Menzel R (1999) Associative learning modifies neural representations of odors in the insect brain.
nature neuroscience 2:74-78.
Farkhooi F, Strube M, Nawrot MP (2009a) Serial correlation in neural spike trains: experimental evidence, stochastic
modelling, and single neuron variability. Physical Review E 79: 021905
Farkhooi F, Müller E, Nawrot MP (2009b) Sequential sparsing by successive adapting neural populations. BMC
Neuroscience 10 (Suppl I):O10, doi:10.1186/1471-2202-10-S1-O10
Farkhooi F, Müller E, Nawrot MP (2010) Sequential sparsening by successive adaptation in neural populations.
arXiv:1007.2345v1
Fernandez PC, Locatelli FF, Person-Rennell N, Deleo G, Smith BH (2009) Associative conditioning tunes transient dynamics
of early olfactory processing. J Neurosci 29(33):10191-202.
Fonta, C., Sun, X., and Masson, C. (1993). Morphology and spatial distribution of bee antennal lobe interneurones responsive
to odours. Chem. Senses 18, 101–119.
Friedrich RW, Laurent G (2001) Dynamic Optimization of Odor Representations by Slow
Temporal Patterning of Mitral Cell Activity. Science 291:889-894
Friedrich RW, Laurent G (2004) Dynamics of olfactory bulb input and output activity during odor stimulation in zeb rafish. J
Neurophysiol 91(6):2658-69
Fujiwara T, Kazawa T, Haupt SS, Kanzaki R (2009) Ca2+ imaging of identifiable neurons labeled by electroporation in
insect brains. Neuroreport 20(12):1061-5
Galán R, Sachse S, Galizia CG, Herz AV (2004) Odor-driven attractor dynamics in the antennal lobe allow for simple and
rapid olfactory pattern classification. Neural Comput 16(5):999-1012
Galizia CG, Menzel R, Holldobler B (1999) Optical imaging of odor-evoked glomerular activity patterns in the antennal
lobes of the ant camponotus rufipes. Naturwissenschaften 86(11):533-7
Galizia, C., and Menzel, R. (2000). Odour perception in honeybees: coding information in glomerular patterns. Curr. Opin.
Neurobiol. 10, 504–510.
11
Galizia, C., and Kimmerle, B. (2004). Physiological and morphological characterization of honeybee olfactory neurons
combining electrophysiology, calcium imaging and confocal microscopy. J. Comp. Physiol. A 190, 21 –38. Galizia and
Rössler, 2009
Galizia CG and Sachse S. Representation of binary odor mixtures in the output neurons of the honeybee antennal lobe
reveals odor-specific interglomerular computation. Human Frontier Science Program Second Annual Awardees' Meeting,
Ottawa, Canada, June 9-12, 2002 , 51. 2002.
Geffen MN, Broome BM, Laurent G, Meister M (2009) Neural encoding of rapidly fluctuating odors. Neuron 61(4):570-86
Gollisch T, Meister M (2008) Rapid neural coding in the retina with relative spike latencies. Science 319(5866):1108 -11
Hildebrandt KJ, Benda J, Hennig RM (2009) The Origin of Adaptation in the Auditory Pathway of Locusts is Specific to Cell
Type and Function. J Neurosci 29:2626-2636
Ito I, Ong RC, Raman B, Stopfer M (2008) Sparse odor representation and olfactory learning. Nat Neurosci. 11(10):1177 -84
Ito I, Bazhenov M, Ong RC, Raman B, Stopfer M (2009) Frequency transitions in odor-evoked neural oscillations. Neuron
64:692-706
Joerges J, Küttner A, Galizia CG, and Menzel R (1997). Representation of odours and odour mixtures visualized in the
honeybee brain. Nature 387:285-288.
Jortner RA, Farivar SS, Laurent G (2007) A simple connectivity scheme for sparse coding in an olfactory system. J Neurosci
27(7):1659-69
Kazama H, Wilson RI (2008) Homeostatic matching and nonlinear amplification at identified central synapses. Neuron
58(3):401-13
Kirschner, S., Kleineidam, C., Zube, C., Rybak, J., Grünewald, B., and Roessler, W. (2006). Dual olfactory pathway in the
honeybee, Apis mellifera. J. Comp. Neurol. 499, 933–952.
Krofczik S, Menzel R and Nawrot MP (2008) Rapid odor processing in the honeybee antennal lobe network. Frontiers in
Computational Neuroscience 2:9
Leitch B, Laurent G (1996) GABAergic synapses in the antennal lobe and mushroom body of the locust olfactory system. J
Comp Neurol. 372(4):487-514.
Masse NY, Turner GC, Jefferis GS (2009) Olfactory information processing in Drosophila. Curr Biol 19:R700 -13
Mazor, O., and Laurent, G. (2005). Transient dynamics versus fixed points in odor representations by locust antennal lobe
projection neurons. Neuron 48, 661–673. Mueller02_359
Mercer AR, Hildebrand JG (2002) Developmental Changes in the Electrophysiological Properties and Response
Characteristics of Manduca Antennal-Lobe Neurons. J Neurophysiol 87:2650-2663
Müller, D., Abel, R., Brandt, R., Zoeckler, M., and Menzel, R. (2002). Differential parallel processing of olfactory
information in the honeybee, Apis mellifera L. J. Comp. Physiol. A 188, 359–370.
Müller E (2007) Markov Process Models for Neural Ensembles with Spike-Frequency Adaptation. PhD Thesis, Combined
Faculties for the Natural Sciences and for Mathematics,
Ruperto-Carola-University of Heidelberg, Germany
Müller E, Buesing L, Schemmel J, Meier K (2007) Spike-frequency adapting neural ensembles: beyond mean adaptation and
renewal theories. Neural Comput 19(11):2958-3010
Namiki, S., and Kanzaki, R. (2008). Reconstructing the population activity of olfactory output neurons that innervate identifi
able processing units. Front. Neural Circuits 2, 1
Namiki S, Haupt SS, Kazawa T, Takashima A, Ikeno H, Kanzaki R (2009) Reconstruction of virtual neural circuits in an
insect brain. Front Neurosci 3(2):206-13.
Nawrot MP (2010) Analysis and interpretation of interval and count variability in neural spike trains. In: Analysis of parall el
spike trains, ed. by. S Grün and S Rotter, Springer Series in Computational Neuroscience
Olsen, S. R., Bhandawat, V., and Wilson, R. I. (2007). Excitatory interactions between olfactory processing channels in the
Drosophila antennal lobe. Neuron 54, 89–103.
Pelz C, Gerber B, and Menzel R. (1997). Odorant intensity as a determinant for olfactory conditioning in honeybees: Roles in
discrimination, overshadowing and memory consolidation. Journal of Experimental Biology 200:837 -847.
Perez-Orive, J., Mazor, O., Turner, G., Cassenaer, S., Wilson, R., and Laurent, G. (2002). Oscillations and sparsening of odor
representations in the mushroom body. Science 297, 359–365.
Prescott SA, Sejnowski TJ (2008) Spike-rate coding and spike-time coding are affected oppositely by different adaptation
mechanisms. J Neurosci 28(50):13649-61
Riffell JA, Abrell L, Hildebrand JG (2008) Physical Processes and Real-Time Chemical Measurement of the Insect Olfactory
Environment. J Chem Ecol 34:837–853
Rinberg D, Koulakov A, Gelperin A (2006) Sparse odor coding in awake behaving mice.
J Neurosci 26(34):8857-65
Root, C. M., Semmelhack, J. L., Wong, A. M., Flores, J., and Wang, J. W. (2007). Propagation of olfactory information in
Drosophila. Proc. Natl. Acad. Sci. USA 104, 11826–11831.
12
Rybak J, Eichmüller S (1993) Structural plasticity of an immunochemically identified set of honeybee olfactory
interneurones. Acta Biologica Hungarica 44 (1): 61-65.
Sachse, S., and Galizia, C. (2002). Role of inhibition for temporal and spatial odor representation in the olfactory output
neurons: a calcium imaging study. J. Neurophysiol. 87, 1106–1117.
Schlief ML, Wilson RI (2007) Olfactory processing and behavior downstream from highly selective receptor neurons. Nat
Neurosci
10(5):623-30
Shang, Y., Claridge-Chang, A., Sjulson, L., Pypaert, M., and Miesenboeck, G. (2007). Excitatory local circuits and their
implications for olfactory processing in the fl y antennal lobe. Cell 128, 601–612.
Sivan E, Kopell N (2006) Oscillations and slow patterning in the antennal lobe. J Comput Neurosci 20:85-96
Skiri HT, Galizia CG, Mustaparta H (2004) Representation of primary plant odorants in the antennal lobe of the moth
Heliothis virescens using calcium imaging. Chem Senses 29(3):253-67
Spors H, Wachowiak M, Cohen LB, Friedrich RW (2006) Temporal dynamics and latency patterns of receptor neuron input
to the olfactory bulb. J Neurosci 26(4):1247-59
Stopfer M, Bhagavan S, Smith BH, Laurent G (1997) Impaired odour discrimination on desynchronization of odour -encoding
neural assemblies. Nature 390:70-4.
Stopfer M, Jayaraman V, Laurent G (2003) Intensity versus identity coding in an olfacto ry system. Neuron 39(6):991-100
Stopfer M. (2005) Olfactory coding: inhibition reshapes odor responses. Curr Biol. 20;15(24):R996 -8.
Sun XJ, Font C, Masson C (1993) Odour quality processing by bee antennal lobe interneurons. Chem Sens 18:355 -377
Staudacher EM, Huetteroth W, Schachtner J, Daly, KC (2009) A 4 -dimensional representation of antennal lobe output based
on an ensemble of characterized projection neurons. J Neurosci Meth 180: 208-223
Szyszka P, Ditzen M, Galkin A, Galizia G, and Menzel R (2005) Sparsening and Temporal Sharpening of Olfactory
Representations in the Honeybee Mushroom Bodies. J Neurophysiol 94: 3303 – 3313
Tabor, R., Yaksi, E., Weislogel, J., and Friedrich, R. (2004). Processing of odor mixtures in the zebrafi sh olfactory bulb. J.
Neurosci. 24, 6611–6620.
Tanaka NK, Ito K, Stopfer M (2009) Odor-evoked neural oscillations in Drosophila are mediated by widely branching
interneurons. J Neurosci 29: 8595-8603
Tripp B, Eliasmith C (2009) Population Models of Temporal Differentiation. Neural Comput, Epub ahead of print
Tsodyks MV, Markram H (1997) The neural code between neocortical pyramidal neurons depends on neurotransmitter
release probability. Proc Natl Acad Sci USA 94(2):719-23
Tsodyks M, Klaus Pawelzik, Henry Markram (1998) Neural networks with dynamic synapses. Neural computation 10, 821--
853
Vosshall LB, Wong AM, Axel R (2000) An olfactory sensory map in the fly brain.
Cell 102(2):147-59.
Wang JW, Wong AM, Flores J, Vosshall LB, Axel R (2003) Two-photon calcium imaging reveals an odor-evoked map of
activity in the fly brain. Cell 112(2):271-82
Wilson, R., Turner, G., and Laurent, G. (2004). Transformation of olfactory representations in the Drosophila antennal lobe.
Science 303, 366–370.
Wilson RI, Laurent G (2005) Role of GABAergic inhibition in shaping odor-evoked spatiotemporal patterns in the
Drosophila antennal lobe. J Neurosci 25(40):9069-79
Wright GA, Thomson MG, and Smith BH (2005). Odour concentration affects odour identity in honeybees. Proc Biol Sci
272:2417-2422.
Wright, G. A., Carlton, M., and Smith, B. H. (2009). A honeybee’s ability to learn, recognize, and discriminate odors
depends upon odor sampling time and concentration. Behav. Neurosci. 123, 36–43.
Yamagata N, Schmuker M, Szyszka P, Mizunami M and Menzel R (2009). Differential odor processing in two olfactory
pathways in the honeybee. Front. Syst. Neurosci. 3:16. doi:10.3389/neuro.06.016.2009
13
Figures
Figure 1: Parallel olfactory pathways in the honeybee. (A) Antennal lobe (AL) circuitry. OSN axons innervate
the AL through four antennal nerve (AN) tracts T1-T4. Uniglomerular l-PNs exclusively innervate glomeruli that
receive input through T1, uniglomerular m-PNs innervate glomeruli that receive input through the antennal tracts
T2-T4. (B) Morphological reconstructions of one l-PN innervating the glomerulus T1-22 (green) and one m-PN
innervating the glomerulus T2-03 (magenta) demonstrate their separate projections along the lateral and the
median antennocerebralis tracts to the mushroom body (MB) lip regions and to the lateral horn (LH). The local
interneuron LN (yellow) provides input to large parts of the AL homogenously.
14
Figure 2: Phasic-tonic response dynamics in projection neurons: experiment and model approaches. (A-C)
Experimental observation. (A) In vivo intracellular recording from an l-PN that innervated the glomerulus T1-22.
(B) Single trial spike trains measured during repeated stimulation with nonanal (9al). (C) Firing rate estimates
from the responses in B (red), and from repeated responses to the odor orange (gray) . (D-F) Phenomenological
filter model. (D) The step stimulus S(t) is convoluted with the impulse response function f(t) of a linear filter
which results in the input function g(t), depicted in F (gray). We modeled the filter kernel by the sum of two
alpha functions as f(t)= t*(1.3*exp(-t/1)-exp(-t/2)); 1=65ms, 2=140ms. (E) The input g(t) is rectified by the
transfer function G. The rectification onset in the input determines the spontaneous background rate bg=G(0).
The linear gain determines the amplitude of the output rate. (F) Time-varying response rate (red) in spikes per
seconds for the step stimulus S(t), and input function g(t) (gray) in arbitrary units. (G-I) Model with synaptic
STD. The postsynaptic PN receives input from 60 OSNs. Each connection is modeled by a depressive synapse.
A presynaptic AP will lead to the release of 40% of the available vesicle pool. The released vesicles recover with
a time constant rec=500ms. (G) Integrated EPSPs in response to 10 stimulations of a single OSN with 20Hz
(blue), 50 Hz (green), and 100Hz (brown). (H) PN spike trains in response to a 2s stimulus for 10 simulation
runs. Each single OSN input is modeled by a Poisson process with a baseline intensity of 4 Hz and a stimulus
intensity of 7 Hz. (I) Average PN response profile estimated from 50 repetitions expresses a strong phasic and a
moderate tonic component. (K-M) Model with postsynaptic SFA. (K) Sketch of negative SFA conductance.
Witch each PN action potential (black ticks) the SFA conductance is increased by a fixed value j=-1.5nS (red
flanks) and decays with a time constant of 400ms. (L) PN spike trains for 10 simulation runs. The total OSN
input is modeled as a Poisson process with a baseline intensity of 300 Hz and a stimulus intensity of 400 Hz. (M)
PN response averaged across 50 repetitions exhibit a phasic-tonic response.
15
Figure 3: Rapid odor encoding in the projection neuron population (A) Combinatorial code. Binary patterns of
inactivated and activated projection neurons in a pseudo population of 17 PNs differ for the three tested single
compound odors. The neuron ID refers to l-PNs and m-PNs and, where identified, indicates the antennal tract
(T1-T4) and the number of the AL glomerulus. (B) Ensemble rate code. For each neuron, the time-varying firing
rate is color-coded. Red signifies a rate increase from baseline (white), blue signifies a rate decrease. (C)
Average pair wise distance d (black) and Euclidian distance (gray, per neuron) between population rate vectors
for all three pairs of odors as a function of time. On average, d becomes significant (above baseline rate +5 times
SD) after 15ms (dotted line) and reaches 90% of its maximum after 145ms (dashed line). The firing rates were
estimated with an alpha-shaped kernel (τ=25ms; Krofczik et al., 2008) and strict causality adds 42ms to
estimated times in C. (D) Firing rate averaged across activated neurons describe a similar phasic-tonic response
profile for all three odors (dark colors). Average rate of inactivated neurons stays close to zero (light colors). (E)
Average peak-normalized firing rates of activated neurons (maroon). Standard deviation (black, shading) is
largest around the time of response onset mainly due to differences in onset latency.
16
Figure 4: Response suppression in PNs mediated by fast local inhibition. (A,B) Response suppression in two l-
PNs. Single trial spike trains (top) for three single compound odors and their tertiary mixture as indicated and
trial-averaged firing rate (bottom). (C,D) Membrane potential in the same two neurons in response t o the mixture
shows clear inhibition. Green triangles indicate estimated onset of inhibition. (E) Average response profiles in
LNs (green, N=6) and PNs (gray, N=24). (F) The average response latency is significantly shorter in LNs
(70.0±16.5ms) than in PNs (130.8±13.9ms).
17
|
1302.5007 | 1 | 1302 | 2013-02-20T15:58:32 | Spike train statistics and Gibbs distributions | [
"physics.bio-ph",
"q-bio.NC"
] | This paper is based on a lecture given in the LACONEU summer school, Valparaiso, January 2012. We introduce Gibbs distribution in a general setting, including non stationary dynamics, and present then three examples of such Gibbs distributions, in the context of neural networks spike train statistics: (i) Maximum entropy model with spatio-temporal constraints; (ii) Generalized Linear Models; (iii) Conductance based Inte- grate and Fire model with chemical synapses and gap junctions. | physics.bio-ph | physics |
Spike train statistics and Gibbs distributions
B. Cessac∗and R. Cofr´e†
NeuroMathComp team (INRIA, UNSA LJAD) 2004 Route des
Lucioles, 06902 Sophia-Antipolis, France.
Abstract
This paper is based on a lecture given in the LACONEU summer
school, Valparaiso, January 2012. We introduce Gibbs distribution in a
general setting, including non stationary dynamics, and present then three
examples of such Gibbs distributions, in the context of neural networks
spike train statistics: (i) Maximum entropy model with spatio-temporal
constraints; (ii) Generalized Linear Models; (iii) Conductance based Inte-
grate and Fire model with chemical synapses and gap junctions.
Keywords Neural networks dynamics; spike train statistics; Gibbs distribu-
tions.
1
Introduction
Neurons communicate among them by generating action potentials or "spikes"
which are pulses of electrical activity. When submitted to external stimuli,
sensory neurons produce sequences of spikes or "spike trains" constituting a
collective response and a dynamical way to encode information about those
stimuli. However, neural responses are typically not exactly reproducible, even
for repeated presentation of a fixed stimulus. Therefore, characterizing the re-
lationship between sensory stimuli and neural spike responses can be framed as
a problem of determining the most adequate probability distribution relating
a stimulus to its neural response. There exist several attempts to infer this
probability from data and / or general principles, based on Poisson or more
general point processes [1, 16, 51], Bayesian approaches [29, 23], maximum en-
tropy [47, 55] (for a review see [43]). In this paper we present several situations
where the notion of Gibbs distributions is appropriate to address this problem.
The concept of Gibbs distribution comes from statistical physics. We use it
here in a more general sense than the one usually taught in standard physics
courses, although it is part of mathematical statistical physics [22]. We argue
here that Gibbs distributions might be canonical models for spike train statistics
analysis. This statement is based on three prominent examples.
∗email:[email protected]
†Corresponding author, email:
rodrigo.cofre [email protected], Postal
2004 Route des Lucioles, 06902 Sophia-Antipolis, France.
Phone:
address:
+33-4-9238-2420, Fax:
+33-4-9238-7845
1
1. The so-called Maximum Entropy Principle allows one to propose spike
train statistics models considering restrictions based on empirical obser-
vations. Although this approach has been initially devoted to show the
role of weak instantaneous pairwise correlations in the retina [47], it has
been recently applied to investigate the role of more complex events such
as instantaneous triplets [19] or spatio-temporal events [55]. Probabil-
ity distributions arising from the Maximum Entropy Principle are Gibbs
distributions.
2. Other approaches such as the Linear-Non Linear (LN) or Generalized Lin-
ear Models (GLM) propose an ad hoc form for the conditional probability
that a neuron fires given the past network activity and given the stimu-
lus. Those models have been proven quite efficient for retina spike trains
analysis [41]. They are not limited by the constraint of stationarity, but
they are based on a questionable assumption of conditional independence
between neurons. As we show, the probability distributions coming out
from those models are also Gibbs distributions.
3. Recent investigations on neural networks models (conductance based integrate-
and-fire (IF) with chemical and electric synapses) show that statistics of
spike trains generated by these models are Gibbs distributions reducing to
1 when dynamics is stationary, and reducing to 2 in specific cases [7, 8, 14].
In the general case, the spike trains produced by these models have Gibbs
distributions which neither match 1 nor 2.
The paper is organized as follows. After some definitions regarding spike
train statistics and a presentation of Gibbs distributions we develop these three
examples, with a short discussion of their advantages and drawbacks in spike
trains analysis. Then, we discuss some relations between these models, mainly
based on the Hammersley-Clifford theorem [24, 2, 33, 39]. This paper is a
summary of several papers written by the authors and other collaborators
[7, 8, 34, 9, 14]. As such it does not contain original material (except the
presentation).
2 Definitions
2.1 Spike trains
We consider a network of N neurons. We assume that there is a minimal
time scale δ > 0 corresponding to the minimal resolution of the spike time,
constrained by biophysics and by measurements methods (typically δ ∼ 1 ms)
[11, 10]. Without loss of generality (change of time units) we set δ = 1, so
that spikes are recorded at integer times. One then associates to each neuron
k and each integer time n a variable ωk(n) = 1 if neuron k fires at time n
and ωk(n) = 0 otherwise. A spiking pattern is a vector ω(n)
k=1
which tells us which neurons are firing at time n. We note A = { 0, 1 }N the
set of spiking patterns. A spike block is a finite ordered list of spiking patterns,
written:
def
= [ ωk(n) ]N
ωn2
n1 = { ω(n) }{n1≤n≤n2} ,
2
where spike times have been prescribed between the times n1 to n2 (i.e.
,
n2 − n1 + 1 time steps). The range of a block is n2 − n1 + 1, the number of
time steps from n1 to n2. The set of such blocks is An2−n1+1. Thus, there
are 2N n possible blocks with N neurons and range n. We call a raster plot a
bi-infinite sequence ω
n=−∞, of spiking patterns. Obviously exper-
imental rasters are finite, but the consideration of infinite sequences is more
convenient mathematically. The set of raster plots is denoted Ω = AZ.
def= {ω(n)}+∞
2.2 Transition probabilities
The probability that a neuron emits a spike at some time n depends on the
history of the neural network. However, it is impossible to know explicitly its
form in the general case since it depends on the past evolution of all variables
determining the neural network state. A possible simplification is to consider
that this probability depends only on the spikes emitted in the past by the
network. In this way, we are seeking a family of transition probabilities of the
n−D ], the probability that the firing pattern ω(n) occurs at
time n, given a past spiking sequence ωn−1
n−D. Here, D is the memory depth of the
probability, i.e., how far in the past does the transition probability depend on the
n−D ] = Pn [ ω(n) ]
form Pn[ω(n) (cid:12)(cid:12) ωn−1
past spike sequence. We use the convention that Pn[ω(n) (cid:12)(cid:12) ωn−1
if D = 0 (memory-less case).
The index n of Pn[. . ] indicates that transition probabilities depend explic-
itly on the time n. We say that those transition probabilities are time-translation
] when-
(i.e the probability does not depend explicitely on time). In
invariant or stationary if for all n, Pn[ω(n) (cid:12)(cid:12) ωn−1
n−D ] = PD[ω(D) (cid:12)(cid:12) ωD−1
0
n−D = ωD−1
ever ωn−1
this case we drop the index n.
0
Transition probabilities depend on the neural network characteristics such
as neurons conductances, synaptic responses or external currents. They give
information about the dynamics that takes place in the observed neural net-
work. Especially, they have a causal structure where the probability of an event
depends on the past. This reflects underlying biophysical mechanisms in the
neural network, which are also causal.
2.3 Gibbs distribution
We define here Gibbs distributions (or Gibbs measures) in a more general setting
that the one usually taught in statistical physics courses, where Gibbs distribu-
tions are considered in the realm of stationary process and maximum entropy
principle. Here, we do not assume stationarity and the definition encompasses
the maximum entropy distributions. The Gibbs distributions considered here
are called chains with complete connections in the realm of stochastic processes
[18, 30] and g-measures in ergodic theory [27]. They are also studied in mathe-
matical statistical physics [22].
2.3.1 Continuity with respect to a raster
For n ∈ Z, we note An−1
−∞ . Assume that we are given
a set of transitions probabilities, like in the previous section, possibly depending
−∞ the set of sequences ωn−1
3
on an infinite past1, i.e. of the form Pn[ω(n) (cid:12)(cid:12) ωn−1
−∞ ]. We give in section 3.3 an
example of neural network model where such transition probabilities with an
infinite memory do occur.
Even if transition probabilities involve an infinite memory ωn−1
−∞ , it is rea-
sonable to consider situations where the effects of past spikes decreases expo-
nentially with their distance in the past. This corresponds to the mathematical
notion of continuity with respect to a raster. We note, for n ∈ Z, m ≥ 0, and r
integer:
m,n
= ω′
ω
if ω(r) = ω′(r), ∀r ∈ { n − m, . . . , n } .
Consider a function f depending both on discrete time n and on the raster part
of ω anterior to n. We write f (n, ω) instead of f (n, ωn−1
−∞ ). The function f is
continuous with respect to the raster ω if its m-variation:
varm [f (n, .)] := supn f (n, ω) − f (n, ω′) : ω
m,n
= ω′o
(1)
tends to 0 as m → +∞. This precisely means that the effect, on the value of f
at time n, as this change is more distant in the past.
2.3.2 Gibbs distribution
Definition 2.1 A Gibbs distribution is a probability measure µ : Ω → [0, 1]
such that:
(i) for all n ∈ Z and all F≤n-measurable functions f :
−∞ ω(n)(cid:1) Pn[ω(n) (cid:12)(cid:12) ωn−1
−∞ ]µ(dω).
Z f (cid:0)ωn
−∞(cid:1) µ(dω) = Z Xω(n)∈A
f (cid:0)ωn−1
−∞ , Pn[ω(n) (cid:12)(cid:12) ωn−1
−∞ ] > 0.
(ii) ∀n ∈ Z, ∀ωn−1
−∞ ∈ An−1
(iii) For each n ∈ Z, Pn[ω(n) (cid:12)(cid:12) ωn−1
−∞ ] is continuous with respect to ω.
The condition (i) is a natural extension of the condition defining the invariant
probability of an homogeneous Markov chain (see eq. (2) next section). In its
most general sense (i) does not require stationarity and affords the consideration
of an infinite memory. It defines so-called compatibility conditions. They state
that the average of a function f (n, ω) with respect to µ, at time n (left hand
side), is equal to the average computed from transition probabilities (right hand
side). This equality must hold for any time n.
There exist several theorems guaranteeing the existence and uniqueness of a
Gibbs distribution [22, 18]: this holds if the variation of transition probability
decays sufficiently fast with time (typically exponentially) as n − m → −∞.
1In this case, one has to assume that (i) for every ω(n) ∈ A , Pn[ω(n) . ] is measur-
−∞ ∈ An−1
n−1
−∞ ,
−∞ ; (ii) for every ω
able with respect to F≤n−1, the sigma-algebra on An−1
Pω(n)∈A
n−1
−∞ ] = 1.
ω
Pn[ω(n) (cid:12)
(cid:12)
(cid:12)
4
2.4 Markov chains
chain of length D is defined by a set of transition probabilities Pn[ω(n) (cid:12)(cid:12) ωn−1
Straightforward examples of Gibbs distributions defined that way are provided
by Markov chains with positive transition probabilities. Recall that a Markov
n−D ]
where the memory depth D > 0 is finite. These transition probabilities are
obviously continuous with respect to ω. If we assume moreover that they are
strictly positive ∀n ∈ Z, ∀ωn−1
−∞ then they match (ii) in the definition
above. Finally, in this case, (i) is equivalent to the following property. For any
time n1, n2, n2 − n1 ≥ D:
−∞ ∈ An−1
n2
µ(cid:2) ωn2
n1 (cid:3) =
Yl=n1+D
l−D ] µ(cid:2) ωn1+D−1
For any times n1, n2 as above, the Gibbs-probability µ(cid:2) ωn2
product of the Gibbs probability of the "initial block" µ(cid:2) ωn1+D−1
Pl[ω(l) (cid:12)(cid:12) ωl−1
products of transition probabilities from the initial time n1 + D to the last time
n2.
n1 (cid:3) is given by2 the
(cid:3) and the
n1
(2)
n1
(cid:3) .
Here we have considered transition probabilities depending explicitly on
time n. When they are time-translation invariant (homogeneous Markov chain)
the definition (2.1) is the definition of the unique invariant distribution of the
Markov chain (it is unique because we have assumed positive transition proba-
bilities).
Let us now state (2) in a different form. Define:
called a normalized Gibbs potential. Then, (2) can be stated using:
φn ( n, ω ) := log Pn[ω(n) (cid:12)(cid:12) ωn−1
n−D ],
n1 ωn1+D−1
n1
µ(cid:2) ωn2
n2
(cid:3) = exp
Xl=n1+D
φl ( l, ω ).
(3)
(4)
This form reminds the Gibbs distribution on spin lattices in statistical physics
where one looks for lattice translation-invariant probability distributions given
specific boundary conditions. Given a potential of range D the probability of a
spin block depends on the states of spins in a neighborhood of size D of that
block. Thus, the conditional probability of this block given a fixed neighbor-
hood is the exponential of the energy characterizing physical interactions within
the block as well as with the boundaries. Here, spins are replaced by spiking
patterns; space is replaced with time which is mono-dimensional and oriented:
there is no dependence in the future. Boundary conditions are replaced by the
dependence in the past.
The definition (3) of the normalized Gibbs potential extends to the case
D → +∞.
2One also says that µ is compatible with the set of transition probabilities.
5
3 Gibbs distributions and models of spike train
statistics
In this section we review several examples of models/concepts used to analyze
spike train statistics. All of them enter in the realm of Gibbs distributions
defined above.
3.1 Maximum entropy models
The definition (2.1) affords time-dependent transition probabilities. On the
opposite, in this section we assume that they do not depend explicitly on n, or,
equivalently, that they are time-translation invariant. This corresponds to the
physical concept of stationarity.
Assume that spike trains statistics is distributed according to an hidden
probability µ. How to approach µ from data ? Maximum entropy provides a
method that allows to approach µ. It selects among all the probability distribu-
tions consistent with empirical data constraints, the most random i.e. the one
with the highest entropy. But, why should we choose the maximum entropy dis-
tribution? The answer is that since entropy is a measure of information, then
one should choose the probability that includes the least amount of informa-
tion we have about the system and no more. The result probability is a Gibbs
distribution.
3.1.1 Entropy
We define the entropy rate (or Kolmogorov-Sinai entropy) of a probability µ ∈
Minv the set of time-translation invariant probability measures as:
h [ µ ] = − lim sup
n→∞
1
n + 1 Xωn
0
µ [ ωn
0 ] log µ [ ωn
0 ] ,
(5)
where the sum holds over all possible blocks ωn
Markov chain h [ µ ] also reads [15]:
0 . Note, that in the case of a
h [ µ ] = −XωD
0
µ(cid:2) ωD
0 (cid:3) P[ω(D) (cid:12)(cid:12) ωD−1
0
] log P[ω(D) (cid:12)(cid:12) ωD−1
0
],
Finally, when D = 0, h [ µ ] reduces to the usual definition:
h(µ) = −Xω(0)
µ [ ω(0) ] log µ [ ω(0) ] .
(6)
(7)
We used here the notation h(µ) instead of S or s, used in statistical physics.
This is the conventional notation in ergodic theory for the (Kolmogorov-Sinai)
entropy where the dependence on the measure µ is made explicit.
3.1.2 Observables
We call observable a function:
6
O : Ω → {0, 1},
ω 7→
r
Yu=1
ωku(nu)
(8)
i.e. a product of binary spike events where ku is a neuron index and nu a
time index, with u = 1, . . . , r, for some integer r > 0. Typical choices of
observables are ωk1(n1) which is 1 if neuron k1 fires at time n1 and is 0 otherwise;
ωk1 (n1) ωk2 (n2) which is 1 if neuron k1 fires at time n1 and neuron k2 fires at
time n2 and is 0 otherwise. Another example is ωk1 (n1) (1 − ωk2 (n2)) which is
1 is neuron k1 fires at time n1 and neuron k2 is silent at time n2. This example
emphasizes that observables are able to consider events where some neurons are
silent.
We say that an observable O has range R if it depends on R consecutive
spike patterns, e.g. O(ω) = O(ωR−1
). We consider here that observables do not
depend explicitly on time (time-translation invariance of observables). As a con-
sequence, for any time n, O(ωR−1
) whenever ωR−1
) = O(ωn+R−1
0
= ωn+R−1
.
n
0
0
n
3.1.3 Potential
A function of the form:
Hβ : Ω → R,
N
ω 7→
βkOk.
Xk=1
(9)
is called a potential, where the coefficients βk are finite3 real numbers. The
range of the potential is the maximum of the range of the observables Ok.
3.1.4 Variational principle
Fix a potential Hβ as in (9). Assume that it has finite range D. 4
In this case, a Gibbs distribution µ obeys the following variational principle:
P [ Hβ ] = sup
( h [ ν ] + ν [ Hβ ] ) = h [ µ ] + µ [ Hβ ] ,
(10)
ν∈Minv
where P [ Hβ ] is called the topological pressure, ν [ Hβ ] = PN
k=1 βkν [ Ok ] is the
average value of Hβ with respect to the probability ν and Minv is the set of
time-translation invariant probability measures on Ω. We use the notation ν(f )
for the average of a function f instead of < f > used in statistical physics or
Eν(f ) used in probability theory. Note that Observables and Gibbs potentials
are random functions that acts on the set of raster plots Ω.
Looking at the second equality, the variational principle (10) selects, among
all possible probabilities ν, a unique one, the Gibbs distribution, realizing the
3Thus, we do not consider here hard core potentials with forbidden configurations.
4The variational principle still holds if the range is infinite and its variation (1) decays
sufficiently fast with m, typically exponentially [46, 3, 12].
7
In this case ν [ Hβ ] becomes PN
supremum. A variant of this principle holds when the average value of observ-
ables Ok is constrained to a value Ck, fixed e.g. by experimental observations.
k=1 βkCk if the average value of all observables
Ok is constrained. In this case the variational principle (10) reduces to maximiz-
ing the entropy on the set of measures ν ∈ Minv such that ν [ Ok ] = Ck. Then,
one is lead to a classical Lagrange multipliers problem where the βks are the
Lagrange multipliers. This is the classical approach introduced by Jaynes [25].
In this setting (10) signifies: "maximizing the entropy given the information
that we have of the system" i.e. the observed average value of the observables
Ok is Ck.
3.1.5 Topological pressure
The topological pressure is the formal analogue of free energy density. It has
the following properties:
• P [ Hβ ] is a log generating function of cumulants. We have:
∂P [ Hβ ]
∂βk
= µ [ Ok ] .
∂2P [ Hβ ]
∂βk∂βl
=
∂µ [ Ok ]
∂βl
=
+∞
Xn=0
COkOl(n),
(11)
(12)
and
where COk Ol(n)
COk Ol(n) = µ [ Ok Ol ◦ σn ] − µ [ Ok ] µ [ Ol ] ,
is the correlation function between the two observables Ok and Ol at
time n and σ is the time shift operator. Note that correlation func-
tions decay exponentially fast whenever Hβ has finite range. So that
n=0 COk Ol(n) < +∞.
P+∞
Eq.
(12) characterizes the variation in the average value of Ok when
varying βl (linear response). The corresponding matrix is a susceptibility
matrix. It controls the Gaussian fluctuations of observables around their
mean (central limit theorem) [46, 38, 12].
• P(Hβ) is a convex function of β.
• Define:
Zn = Xωn
0
eHβ( ωn
0 ).
(13)
The topological pressure obeys:
P(Hβ) = lim
n→+∞
1
n
log Zn,
and is analogous to a thermodynamic potential density (free energy, free
enthalpy, pressure).
8
Remark 1 For D > 0 one cannot write the Gibbs distribution in the form:
µ [ ωn
0 ] =
1
Zn
eHβ [ ωn
0 ].
(14)
It only obeys: ∃A, B > 0 such that, for any block ωn
0
A ≤
µ [ ωn
0 ]
e−(n−D+1)P(Hβ)eHβ(ωn
0 ) ≤ B.
This is actually the definition of Gibbs distributions in ergodic theory [12].
3.1.6 Markov chain
The choice of the potential (9), i.e. the choice of a set of observables, fixes the
restrictions for the statistical model. A normalization procedure allows to find a
normalized potential φ equivalent5 to Hβ from which the transition probabilities
are constructed. This defines an homogeneous Markov chain whose invariant
measure is the Gibbs distribution associated with Hβ.
It is constructed as
follows.
Transition matrix
Consider two spike blocks w1, w2 of range D ≥ 1. The transition w1 → w2 is
legal if w1 has the form ω(0)ωD−1
ω(D). The vectors
ω(0), ω(D) are arbitrary but the block ωD−1
is common. Here is an example of
a legal transition :
and w2 has the form ωD−1
1
1
1
w1 = (cid:2) 0
0
1
1 (cid:3); w2 = (cid:2) 0
1
Here is an example of a forbidden transition
0
1
0
1
1
1
1
1
1
0 (cid:3).
1
0 (cid:3).
w1 = (cid:2) 0
0
1
1 (cid:3); w2 = (cid:2) 0
0
Any block ωD
the block w1 = ωD−1
0 of range R = D + 1 can be viewed as a legal transition from
0 ∼ w1w2.
0
1 and in this case we write ωD
to the block w2 = ωD
The transfer matrix L is defined as:
Lw1,w2 = (cid:26) eHβ(ωD
0,
0 )
if w1, w2
otherwise.
is legal with ωD
0 ∼ w1w2
.
(15)
Perron-Frobenius theorem
0 ) > −∞, eHβ (ωD
From the matrix L the transition matrix of a Markov chain can be constructed.
0 ) > 0 for each legal transition. As a consequence
Since Hβ(ωD
of the Perron-Frobenius theorem [21, 48], L has a unique real positive eigenvalue
sβ, strictly larger than the modulus of the other eigenvalues (with a positive
gap), and with associated right, R, and left, L, eigenvectors: LR = sβR, LL =
sβL.
The following holds:
5Two potentials are said "equivalent" or cohomologous if and only if they correspond to
the same Gibbs distribution [26].
9
• These eigenvectors have strictly positive entries R ( . ) > 0, L ( . ) > 0,
functions of blocks of range D. They can be chosen so that the scalar
product h L, R i = 1.
• We have:
P(Hβ) = log sβ.
• The following potential:
φ(ωD
0 ) = Hβ(ωD
0 ) − Gβ(ωD
0 )
(16)
(17)
(18)
(19)
(20)
with:
Gβ(ωD
0 ) = log R(cid:0) ωD−1
0
(cid:1) − log R(cid:0) ωD
1 (cid:1) + log sβ,
is equivalent to Hβ and normalized. It defines a family of transition prob-
abilities:
• These transition probabilities define a Markov chain which admits a unique
P[ω(D) (cid:12)(cid:12) ωD−1
0
def
= eφ(ωD
0 ) > 0.
]
invariant probability:
µ(ωD−1
0
) = R(cid:0) ωD−1
0
(cid:1) L(cid:0) ωD−1
0
(cid:1) .
which is the Gibbs distribution satisfying the variational principle (10).
• It follows that the probability of blocks of depth n ≥ D is:
µ [ ωn
0 ] =
eHβ ( ωn
0 )
sn−D+1
β
R(cid:0) ωn
n−D+1(cid:1) L(cid:0) ωD−1
0
(cid:1) .
(21)
• In the case D = 0 the Gibbs distribution reduces to (14). One can indeed
easily show that:
exp Gβ = sβ = Xω(0)
eHβ(ω(0)) = Zβ,
Additionally, since spike patterns occurring at distinct time are indepen-
dent in the D = 0 case, Zn in (13) can be written as Zn = Z n
β so that
P(Hβ) = log Zβ.
• In the general case of spatio-temporal constraints, the normalization re-
quires the consideration of normalizing function Gβ depending as well on
the blocks ωD
0 . Thus, in addition to function Hβ normalization intro-
duces a second function of spike blocks. This increases consequently the
complexity of Gibbs potentials and Gibbs distributions compared to the
spatial (D = 0) case where Gβ reduces to a constant.
10
3.1.7 Examples
We give here a few examples of Maximum Entropy Gibbs distributions, found
in the literature.
• Bernoulli model. Here only firing rates of neurons are constrained. The
potential has the form:
Hβ(ω(0)) =
N
Xi=1
βiωi(0).
This is a memory-less model, where transitions probabilities are given by
neuron firing rates λi = eβi
1+eβi . The Gibbs distribution has the form:
µ [ ωn
m ] =
n
Yl=m
N
Yk=1
λωk(l)
k
(1 − λk)1−ωk(l),
(22)
This is thus a product probability where neurons are independent.
• Ising model. This model was introduced by Schneidman et al [47] for
retina spike train analysis. Here, firing rates and instantaneous pairwise
synchronisation probabilities are constrained. The potential has the form:
Hβ(ω(0)) =
N
Xi=1
βiωi(0) +
N
Xi,j=1
βijωi(0) ωj(0).
This is a memory-less model where the Gibbs distribution has the classical
form (14).
• Extended spatial Ising model. A natural extension of Ising model has
been proposed by Ganmor et al [19], where triplets and more general
synchronous spike events are considered. The potential has the form:
Hβ(ω(0)) =
N
Xi=1
βiωi(0)+
N
Xi,j=1
βijωi(0) ωj(0)+
N
Xi,j,k=1
βijkωi(0) ωj(0) ωk(0)+. . .
This is a memory-less model where the Gibbs distribution has the classical
form (14).
• Spatio temporal Ising model.
In [31] Marre et al considered a spatio-
temporal extension of the Ising model where the potential has the form:
Hβ(ω1
0) =
N
Xi=1
βiωi(0) +
N
Xi,j=1
βij ωi(0) ωj(1).
Here spatio-temporal pairs with memory depth 1 are considered. Although
the Gibbs distribution has not the form (14), the authors use an approxi-
mation of the exact distribution by this form, based on a detailed balance
assumption. They applied this model for spike train analysis in the cat
parietal cortex.
• General Spatio temporal model. General models of the form (9) have been
considered in [55, 9, 34] for the analysis of retina spike trains. A C++
implementation of methods for fitting spatio-temporal models from data
is available at http://enas.gforge.inria.fr/v3/.
11
3.1.8 Applications
The maximum entropy principle has been used by several authors [47, 49, 52, 56,
35, 47, 19, 20] for Multi-electrode Arrays (MEA) spike train analysis. Efficient
methods have been designed to estimate the parameters of the potential, in the
spatial case [17] (Broderick et al., 2007) and in the spatio-temporal case [34].
This approach, grounded on statistical physics, attempts to find a generic
model for spike statistics based on a potential of the form (9), where the observ-
ables and their related β parameters summarize "effective interactions" between
spikes. Behind this approach exists, we believe, a physicists "dream": inferring,
from data analysis, the equivalent of the equation of states existing in thermo-
dynamics; that is, summarizing the behaviour of a big neuronal system by a few
canonical variables (analogous e.g. to temperature, pressure, volume in a gas).
To our opinion, recent remarkable investigations to exhibit critical phenomena
in retina spike train statistics are part of this project (Tkacic et al., 2006,2009)
The main advantage of this approach is the possibility of constructing dif-
ferent statistical models based on a priori hypotheses on the most statistically
significant events (single spikes, pairs, triplets, and so on). As such, it allows
to consider arbitrary forms of spatio-temporal correlations. But this strength is
also a weakness. Indeed, the possible forms of potentials are virtually infinite
and obviously, in the setting of neuronal dynamics, one does not have the equiv-
alent of mechanics or thermodynamics to construct the potential from general
principles.
Finally, this approach only holds for stationary data, a highly questionable
assumption as far as data from living systems are concerned.
3.2 Generalized Linear model
We now consider a second class of Gibbs distributions related to statistical mod-
els called Linear-Nonlinear (LN) model and Generalized Linear Model (GLM)
[4, 32, 50, 37, 54, 42, 40, 1, 41]. We focus here on the GLM and follow the
presentation of Ahmadian et al [1].
3.2.1 Conditional intensities
GLMs are commonly used statistical methods for modeling the relationship
between neural population activity and presented stimuli. Let x ≡ x(t) be
a time-dependent stimulus. In response to x the network emits a spike train
response r. This response does not only depend on x, but also on the network
history of spiking activity. The GLM (and LN) assimilate the spike response
r as an inhomogeneous point process: the probability that neuron k emits a
spike between t and t + dt is given by λk(t Ht) dt, where λk(t Ht) is called
"conditional intensity" and Ht is the history of spiking activity up to time t. In
the GLM this function is given by:
λk(t Ht) = f
bk + (Kk ∗ x)(t) +Xj
(Hkj ∗ rj)(t)
,
(23)
where:
12
• f is a non linear function (an exponential or a sigmoid);
• bk is some constant fixing the baseline firing rate of neuron k;
• Kk is a causal, time-translation invariant, linear convolution kernel that
mimics a linear receptive field of neuron k;
• ∗ is the convolution product;
• Hkj is the memory kernel that describes possible excitatory or inhibitory
post spike effects of the jth observed neuron on the kth. As such, it
depends on the past spikes, hence on ω. The diagonal components Hkk
describe the post spike feedback of the neuron to itself, and can account
for refractoriness, adaptation and burstiness depending on their shape;
• rj is the spike train of neuron j: rj(t) = Pr≥1 δ(t − t(r)
time of the rth spike of jth neuron.
j ), where t(r)
j
is the
The spike response has a history dependent structure that makes Poisson
models inappropriate. Point processes affords for history dependence and gen-
eralizes Poisson process. A point process can be completely characterized by its
conditional intensity function.
λk(t Ht) = lim
∆t→0
P(∆N[t+∆t)=1 Ht)
∆t
,
where N[t+∆t) is the counting process that gives the number of spikes occurring
in the interval [t + ∆t). Choosing ∆t to be a sufficiently small time interval
∼ 1ms, the probability of firing more than one spike is negligibly small compared
to the probability of firing one spike. This assumption is biophysically plausible
because neurons have refractory period. Therefore:
P(spike in [t + ∆t) Ht) ≈ λk(t Ht)∆t.
Here λk(t Ht) is defined in continuous time, and spikes are discrete events.
If we discretize the time to make the spikes emitted by the point process belong
to a single bin, we have:
P(ωk(n) = 1 Hn−1) ≈ λk(n Hn−1)∆t := pk(n)
3.2.2 Conditional independence
The GLM postulates that, given the history H and stimulus x, neurons are
independent (conditional independence upon the past and stimulus).
In the
context of transition probabilities defined on section 2.2, the response at time
n is a spiking pattern ω(n) while the history is the spike activity H. As a
consequence of the conditional independence assumption the probability of a
spike pattern follows a Bernoulli process:
Pn[ω(n) (cid:12)(cid:12) ωn−1
−∞ ] =
N
Yk=1
pk(n)ωk(n)(1 − pk(n))1−ωk(n).
(24)
13
3.2.3 Gibbs distribution
Transition probabilities are strictly positive whenever 0 < pk(n) < 1, for all k,n.
If f is e.g. a sigmoid this holds provided its argument bi + (Ki ∗ x)(t) +Pj(Hij ∗
rj)(t) remains bounded in absolute value. The continuity of λ with respect to
ω holds whenever f is continuous and the memory kernel H is continuous with
respect to ω. This second condition is fulfilled in two cases:
• H depends on a finite past;
• H depends on an infinite past, but the memory dependence decays suffi-
ciently fast to ensure continuity. Since H mimics synaptic influence it is
typically a sum of α-profiles that mimic PSPs (Post Synaptic Potentials).
α profiles decay exponentially fast with time, so they match this condition.
We come back to this point in section 3.3.
The Gibbs potential associated with (24) is:
φn(ω) =
N
Xk=1
( ωk(n) log pk(n) + (1 − ωk(n))(1 − pk(n)) ) ,
(25)
It is normalized by definition.
3.2.4 Applications
This model has been applied in a wide variety of experimental settings [5, 13, 53,
6, 36, 54, 40]. Efficient methods has been designed to estimate the parameters
[1].
To us, the main advantages of the GLM are:
• The transition probability is known (postulated) from the beginning and
does not require the heavy normalization (17) imposed by potentials of
the form (9);
• The model parameters have a neurophysiological interpretation, and their
number grows at most as a power law in the number of neurons.
• It has good decoding performances
• It holds for non stationary data.
Its main drawbacks are:
• It postulates an ad hoc form for the transition probability of the stochastic
process;
• It uses a quite questionable assumption of conditional independence: neu-
rons are assumed independent at time n when the past is given. On the
opposite, the maximal entropy principle does not require this assumption.
• To us, the biophysical interpretation of the parameters Hkj is unclear. Do
they correspond to "real" connectivity ? "functional" connectivity ?
14
3.3
Integrate and Fire neural networks
The previous examples were mainly developed for data analysis: one speculates
a form for transitions probabilities, performs parameters fitting, and then uses
the model to decode or to extrapolate the statistics of complex events. Here we
start from a different point of view asking the following questions: Can we have
a reasonable idea of what could be the spike train statistics studying a neural
network model? Do Gibbs distribution arise in these models ? What is the
shape of the potential ? We focus here on a model proposed in [7, 8, 14] where
these questions have been answered.
3.3.1 Model
The integrate-and-fire model remains one of the most ubiquitous model for sim-
ulating and analyzing the dynamics of neuronal circuits. Despite its simplified
nature, it captures some of the essential features of neuronal dynamics. Denote
V (t) the membrane potential vector with entries Vk(t). The continuous-time
dynamics of V (t) is defined as follows. Fix a real variable θ > 0 called "firing
threshold". For a fixed time t, we have two possibilities:
1. Either Vk(t) < θ, ∀k = 1, . . . , N . This corresponds to sub-threshold dy-
namics.
2. Or, ∃k, Vk(t) ≥ θ. Then, we speak of firing dynamics.
The model proposed here is an extension of the conductance based Integrate-
and-Fire neuron model introduced in [45]. The model-definition follows the pre-
sentation given in [11, 8]. Neurons are considered as points, with neither spatial
extension nor biophysical structure (axon, soma, dendrites). Dynamics is ruled
by a set of stochastic differential equations where parameters, corresponding to
chemical conductances, depend on the action potentials emitted in the past by
the neurons. In this way, the dynamical system defined here is ruled both by
continuous and discrete time dynamical variables.
Subthreshold dynamics
It is defined by:
Ck
dVk
dt
where:
= −gL,k(Vk − EL) −Xj
gkj (t, ω)(Vk − Ej) +Xj
gkj (Vj − Vk) + Ik(t),
(26)
• Ck is the membrane capacity of neuron k;
• Ik(t) = i(ext)
k
(t)+σB ξk(t) is a current when a time-dependent part i(ext)
(t)
(stimulus) and stochastic part σB ξk(t) where ξk(t) is a white noise and
σB controls the noise intensity;
k
• gL,k is the leak conductance and EL < 0 the leak Nernst potential;
• gkj mimics electric conductance (gap junctions) between neurons j and k;
these are passive and symmetric conductances;
15
• the term
gkj(t, ω) = Gkj Xr:t
(r)
j
(ω)<t
αkj (cid:16)t − t(r)
j (ω)(cid:17) ,
mimics the conductance of the chemical synapse j → k, where:
αkj (t) =
t
τkj
e
− t
τkj H(t),
(27)
(28)
mimics a PSP, H(t) is the Heaviside function (that mimics causality);
t(r)
j (ω) is the rth spike emitted by neuron j in the raster ω, therefore
gkj(t, ω) depends on the whole spike history; Ej is the reversal potential
of the chemical synapse j → k.
Firing dynamics and reset
If, at time t, some neuron k reaches its firing threshold θ, Vk(t) = θ, then this
neuron emits a spike. To conciliate the continuous time dynamics of membrane
potentials and the discrete time dynamics of spikes we define the spike and reset
as follows.
• The neuron membrane potential Vk is reset to 0 at the next integer time
after t.
• A spike is registered at time [t] + 1 where [ t ] is the integer part of t. This
allows us to represent spike trains as events on a discrete time grid. It
has the drawback of artificially synchronizing spikes coming from different
neurons, in the deterministic case [11, 28]. However, the presence of noise
in membrane potential dynamics eliminates this synchronization effect.
• Spikes are separated by a time scale τsep > 0 which is a multiple of δ (thus
an integer).
• Between [t] + 1 and [t] + τsep the membrane potential Vk is maintained
to 0 (refractory period). From time [t] + τsep on, Vk evolves according to
(26) until the next spike.
• When the spike occurs (at time [t] + 1), the raster ω as well conductances
gkj(t, ω) are updated.
3.3.2 Main results
This model has several variants: discrete time [7]; continuous time with chemical
synapses [8] and continuous time with chemical and electric synapses [14]. We
list here the main results concerning spike statistics and Gibbs distributions.
1. Whatever the values of the parameters the model admits a unique Gibbs
distribution in the general sense given in section 2.3.
2. When the noise is weak and without gap junctions, the normalized Gibbs
potential can be explicitly computed. It takes the form:
φn(ω) =
N
Xk=1
( ωk(n) log λk(n) + (1 − ωk(n) log(1 − λk(n)) ) ,
(29)
16
where
λk(n) = f (cid:16) bk(n − 1, ω) + Φ(ext)
k
(n − 1, ω) + Φ(syn)
k
(n − 1, ω)(cid:17) ,
(30)
where:
• f is a sigmoid function;
• bk(n − 1, ω) is a function depending on the threshold value, the leak
Nernst potential, and on the integrated noise, integrated from the
last time where has been reset (depending on ω) up to time n − 1;
• Φ(ext)
k
(n − 1, ω) corresponds to the integrated effects of the external
current i(ext)
k
on the membrane potential;
• the term Φ(syn)
k
(n − 1, ω) corresponds to the integrated effects of
chemical synapses on the membrane potential.
3. Eq. (29), expresses that in this case, neurons are conditionally independent
upon the past.
4. In this conductance-based model, conductances depend on the past via
(27). One can consider as well a current-based model where conductances
are fixed and current depend on the past spikes. In this case, the terms
Φ(ext)
(n − 1, ω) can be written as convolutions and
one recovers a potential with a form analogous to (25).
(n − 1, ω) and Φ(syn)
k
k
5. In the general case (gap junctions), neurons are not conditionally inde-
pendent. Gap junctions induce a coupling effect which does not allow any
more the factorization (29) of the potential.
6. Correlations (pairwise and higher order) are mainly due to chemical synapses
and gap junctions. Additional correlations can also be induced by the
stimulus using e.g. a current i(ext) where time fluctuations of i(ext)
are cor-
related with i(ext)
. But these are extra-correlations that disappear when
the stimulus is removed, whereas the dynamical correlations remains.
k
j
7. The potential has an infinite range (infinite memory). However, thanks to
the exponential decay of the alpha profile, one can show that the potential
is continuous. This allows to propose Markovian approximation of the
Gibbs distribution where the exact potential is replaced by a potential
with a finite range [7, 8].
3.3.3 Applications
What do we finally learn from the study of this model ?
• We have a positive answer to the existence of Gibbs distributions in neural
networks models.
• An explicit form for the potential is known in specific cases.
• The form (30) actually also fits with maximum entropy model, in the
stationary case, as shown in section 3.4.
17
• In this model the origin of correlations is essentially due to dynamics, not
to the stimulus.
• Gap junctions play here a central role in the structure of dynamical cor-
relations and dependence of dynamics upon history (see [14] for more
details).
• The analysis holds for non stationary data.
Considering potential uses of this study to fit real data the main criticism
is:
• This is a model. Is it sufficient to describe real neural networks ? For
example, its application to retina data is controversial since it considers
only spiking cells (that mimics ganglion cells), but retina has also non
firing cells like most amacrine and bipolar cells.
• In the general case there is no explicit form for the Gibbs potential.
• Even when there exists an explicit form for the potential, it has quite a
lot of parameters which can be difficult to fit from data.
3.4 Relations between these approaches
In this section, we establish a connection between the three examples of Gibbs
distributions considered in sections 3.1, 3.2, 3.3.
Consider a family of transition probabilities satisfying the positivity condi-
tion (ii) in section 2.3, where we furthermore assume that the memory depth is
finite and that transition probabilities are time-translation invariant. As stated
in section 2.4 this define an homogenous Markov chain. The transition prob-
abilities are thus functions of blocks ωD
0 (see section 2.2 and the definition of
time translation invariance). These functions can then take at most 2N (D+1)
values. The same holds for the normalized potential (3). Now, one can prove
that any such function can be written as:
φ(ω) ≡ φ(ωD
0 ) =
L
Xl=0
φl Ol(ω),
(31)
with L = 2N (D+1)−1 and Ol is an observable (8) where the time index ranges
from 0 to D. The index l parametrizes an enumeration of all possible observables
with N neurons and D + 1 time steps, where m0 is the constant observable
O0 = 1, and so on.
Now, using the positivity condition and the results in section 3.1 one can
show that any family of stationary transition probabilities with memory depth D
can be associated with a potential of the form (9). The correspondence is actu-
ally unique. This is a straightforward application of the celebrated Hammersley-
Clifford theorem [24, 2, 33, 39].
An immediate consequence of this result is that, in the stationary case with
finite memory, the GLM potential (23) and the Integrate and Fire (29) cor-
respond to a Maximum Entropy model with a potential of the form (9). The
18
parameters βk in (9) are then nonlinear functions of the parameters in (25) or in
(29) (see [7]). As a consequence, some of these parameters are redundant: there
are a priori 2N (D+1) non vanishing parameters βk while there are quite less pa-
rameters in the GLM or in the Integrate and Fire (of order N 2). However, the
GLM assumes conditional independence between neurons, while the maximum
entropy approach is precisely used to take care of (pairwise and higher order)
correlations between neurons. In this sense it is more general.
In the non stationary case one can no more apply the maximum entropy
principle (entropy is not defined). However, in the case where statistics depends
on time on a slow time scale (compared to spike characteristic time scale) one
can use a quasi-static approach where the parameters βk in (9) vary slowly in
time [44] (Tyrcha et al., 2012). On the opposite, the GLM allows to consider
non stationary data with efficient results [1, 41].
The IF model contains both maximum entropy models and GLM. It has a
maximum entropy Gibbs distribution in the stationary case, and it reduces to
GLM upon several simplifications. In its more general form it allows the consid-
eration of non stationarity and does not rely on the conditional independence
assumption. Unfortunately, its generality is a weakness since an explicit form
for the potential is not known yet in the general case.
4 Conclusion
In this paper we have argued that Gibbs distribution considered in a fairly
general sense could constitute generic statistical models to fit spike trains data.
The example of the Integrate and Fire model suggests that such distribution
could be also defined for more elaborated neural networks models (FitzHugh-
Nagumo or Hodgkin-Huxley).
In particular, the existence and uniqueness of
a Gibbs measure holds whenever there is continuity with respect to a raster,
with a sufficiently fast decay of the variation (1) [18]. As shown in [7, 8, 14]
this property is ensured when interactions between neurons decay exponentially
fast. This is typically the case for chemical synapses where the PSP (28) decays
exponentially fast with time.
The interest of proposing Gibbs distribution constructed from neural net-
work models is multiple. The model mimics a neurophysiological structure
where interactions between neurons, stimuli, and biophysical parameters are
well identified. As a consequence the model-parameters can be easily inter-
preted. Thus, the role of each specific biophysical parameter on spike statistics
can be easily analysed. Also, the potential obtained this way is already nor-
malized, while e.g. maximum entropy principle requires a complex procedure to
achieve normalisation. Finally, in this context, it is possible to study the effect
of a time-dependent stimulus on spike statistics.
However, this approach, to be efficient requires (i) to have an analytical
form for the potential; (ii) to be able to fit the many parameters of a non linear
problem. This is yet far from being achieved.
Acknowledgments This work was supported by the French ministry of Re-
search and University of Nice (EDSTIC), INRIA, ERC-NERVI number 227747,
KEOPS ANR-CONICYT and European Union Project # FP7-269921 (Brain-
Scales), Renvision # 600847 and Mathemacs # FP7-ICT-2011.9.7.
19
References
[1] Yashar Ahmadian, Jonathan W. Pillow, and Liam Paninski. Efficient
Markov Chain Monte Carlo Methods for Decoding Neural Spike Trains.
Neural Computation, 23(1):46 -- 96, January 2011.
[2] J. Besag. Spatial interaction and the statistical analysis of lattice sys-
tems. Journal of the Royal Statistical Society. Series B (Methodological),
36(2):192 -- 236, 1974.
[3] R. Bowen.
Equilibrium states and the ergodic theory of Anosov
diffeomorphisms, volume 470 of Lect. Notes.in Math. Springer-Verlag, New
York, 1975.
[4] D. R. Brillinger. Maximum likelihood analysis of spike trains of interacting
nerve cells. Biol Cybern, 59(3):189 -- 200, 1988.
[5] D. R. Brillinger. Nerve Cell Spike Train Data Analysis - a Progression of
Technique. J Amer Statist Assn, 87(418):260 -- 271, 1992.
[6] E. N. Brown, R. Barbieri, U. T. Eden, and L. M. Frank. Likelihood meth-
ods for neural spike train data analysis. Computational Neuroscience: A
Comprehensive Approach, 2003.
[7] Bruno Cessac. A discrete time neural network model with spiking neurons
ii. dynamics with noise. J. Math. Biol., 62:863 -- 900, 2011.
[8] Bruno Cessac. Statistics of spike trains in conductance-based neural net-
works: Rigorous results. Journal of Mathematical Neuroscience, 1(8), 2011.
[9] Bruno Cessac and Adrian Palacios. Spike train statistics from empirical
facts to theory: the case of the retina, volume in "Current Mathematical
Problems in Computational Biology and Biomedicine". Springer, 2012.
[10] Bruno Cessac, H´el`ene Paugam-Moisy, and Thierry Vi´eville. Overview of
facts and issues about neural coding by spikes. J. Physiol. Paris, 104(1-
2):5 -- 18, February 2010.
[11] Bruno Cessac and T. Vi´eville. On dynamics of integrate-and-fire neural
networks with adaptive conductances. Frontiers in neuroscience, 2(2), July
2008.
[12] J.R. Chazottes and G. Keller. Mathematics of Complexity and Dynamical
Systems, pages 1422 -- 1437. Springer, 2011.
[13] E. J. Chichilnisky. A simple white noise analysis of neuronal light responses.
Network: Comput. Neural Syst., 12:199 -- 213, 2001.
[14] Rodrigo Cofr´e and Bruno Cessac. Dynamics and spike trains statistics in
conductance-based integrate-and-fire neural networks with chemical and
electric synapses. Chaos, Solitons and Fractals, 2013. In press.
[15] I. P. Cornfeld, S. V. Fomin, and Ya. G. Sinai. Ergodic Theory. Springer,
Berlin, Heidelberg, New York, 1982.
20
[16] Vere-Jones D. Daley, D. J. An introduction to the theory of point processes
(2nd ed.). New York: Springer., 2003.
[17] M. Dud´ık, S. Phillips, and R. Schapire. Performance guarantees for regu-
larized maximum entropy density estimation. In Proceedings of the 17th
Annual Conference on Computational Learning Theory, 2004.
[18] Roberto Fernandez and Gr´egory Maillard. Chains with complete connec-
tions : General theory, uniqueness, loss of memory and mixing properties.
J. Stat. Phys., 118(3-4):555 -- 588, 2005.
[19] Elad Ganmor, Ronen Segev, and Elad Schneidman. The architecture of
functional interaction networks in the retina. The journal of neuroscience,
31(8):3044 -- 3054, 2011.
[20] Elad Ganmor, Ronen Segev, and Elad Schneidman. Sparse low-order in-
teraction network underlies a highly correlated and learnable neural popu-
lation code. PNAS, 108(23):9679 -- 9684, 2011.
[21] F. R. Gantmacher. The theory of matrices. AMS Chelsea Publishing,
Providence, RI, 1998.
[22] Hans-Otto Georgii. Gibbs measures and phase transitions. De Gruyter
Studies in Mathematics:9. Berlin; New York, 1988.
[23] Sebastian Gerwinn, Jakob Macke, and Matthias Bethge. Bayesian pop-
ulation decoding of spiking neurons. Front Comput Neurosci, page 3:21,
2009.
[24] J. M. Hammersley and P. Clifford. Markov fields on finite graphs and
lattices. unpublished, 1971.
[25] E.T. Jaynes.
Information theory and statistical mechanics. Phys. Rev.,
106:620, 1957.
[26] O. Jenkinson, R. D. Mauldin, and M. Urbanski. Zero temperature limits
of gibbs-equilibrium states for countable alphabet subshifts of finite type.
J. Stat. Phys, 119:765 -- 776, 2005.
[27] M. Keane. Strongly mixing g-measures. Invent. Math., 16:309 -- 324, 1972.
[28] C. Kirst and M. Timme. How precise is the timing of action potentials ?
Front. Neurosci., 3(1):2 -- 3, 2009.
[29] Shinsuke Koyama, Uri T. Eden, Emery N. Brown, and Robert E. Kass.
Bayesian decoding of neural spike trains. Annals of the Institute of
Statistical Mathematics, 62:pp 37 -- 59, 2010.
[30] G. Maillard.
Introduction to chains with complete connections. Ecole
Federale Polytechnique de Lausanne, winter 2007.
[31] O. Marre, S. El Boustani, Y. Fr´egnac, and A. Destexhe. Prediction of
spatiotemporal patterns of neural activity from pairwise correlations. Phys.
rev. Let., 102:138101, 2009.
21
[32] P. McCullagh and J. A. Nelder. Generalized linear models (Second edition).
London: Chapman & Hall, 1989.
[33] John Moussouris. Gibbs and markov random systems with constraints. J.
Stat. Phys., 10(1):11 -- 33, 1974.
[34] H. Nasser, O. Marre, and B. Cessac. Spatio-temporal spike trains analysis
for large scale networks using maximum entropy principle and monte-carlo
method. Journal Of Statistical Mechanics, 2013. in press.
[35] Ifije E. Ohiorhenuan, Ferenc Mechler, Keith P. Purpura, Anita M. Schmid,
Qin Hu, and Jonathan D. Victor. Sparse coding and high-order correlations
in fine-scale cortical networks. Nature, 466(7):617 -- 621, 2010.
[36] L. Paninski, M. Fellows, S. Shoham, N. Hatsopoulos, and J. Donoghue.
Superlinear population encoding of dynamic hand trajectory in primary
motor cortex. J. Neurosci., 24:8551 -- 8561, 2004.
[37] Liam Paninski. Maximum likelihood estimation of cascade point-process
neural encoding models. Network: Comput. Neural Syst., 15(04):243 -- 262,
November 2004.
[38] W. Parry and M. Pollicott. Zeta functions and the periodic orbit structure
of hyperbolic dynamics, volume 187 -- 188. Asterisque, 1990.
[39] P.Clifford. Disorder in Physical Systems: A Volume in Honour of John M.
Hammersley,, pages 19 -- 32. Oxford University Press, 1990.
[40] J W Pillow, J Shlens, L Paninski, A Sher, A M Litke, E J Chichilnisky,
and E P Simoncelli. Spatio-temporal correlations and visual signaling in a
complete neuronal population. Nature, 454(7206):995 -- 999, Aug 2008.
[41] Jonathan W. Pillow, Yashar Ahmadian, and Liam Paninski. Model-based
decoding, information estimation, and change-point detection techniques
for multineuron spike trains. Neural Comput., 23(1):1 -- 45, 2011.
[42] J.W. Pillow, L. Paninski, V.J. Uzzell, E.P. Simoncelli, and E.J.
Chichilnisky. Prediction and decoding of retinal ganglion cell responses
with a probabilistic spiking model. J. Neurosci, 25:11003 -- 11013, 2005.
[43] F. Rieke, D. Warland, Rob de Ruyter van Steveninck, and William Bialek.
Spikes, Exploring the Neural Code. The M.I.T. Press, 1996.
[44] Yasser Roudi and John Hertz. Mean field theory for non-equilibrium net-
work reconstruction. Phys. Rev. Lett., 106(048702), 2011.
[45] M. Rudolph and A. Destexhe. Analytical integrate and fire neuron models
with conductance-based dynamics for event driven simulation strategies.
Neural Computation, 18:2146 -- 2210, 2006.
[46] D. Ruelle. Thermodynamic formalism. Addison-Wesley,Reading, Mas-
sachusetts, 1978.
[47] E. Schneidman, M.J. Berry, R. Segev, and W. Bialek. Weak pairwise cor-
relations imply strongly correlated network states in a neural population.
Nature, 440(7087):1007 -- 1012, 2006.
22
[48] E. Seneta. Non-negative Matrices and Markov Chains. Springer, 2006.
[49] J. Shlens, G.D. Field, J.L. Gauthier, M.I. Grivich, D. Petrusca, A. Sher,
A.M. Litke, and E.J. Chichilnisky. The structure of multi-neuron firing
patterns in primate retina. Journal of Neuroscience, 26(32):8254, 2006.
[50] E. P. Simoncelli, J. P. Paninski, J. Pillow, and O. Schwartz. Charac-
terization of Neural Responses with Stochastic Stimuli. The cognitive
neurosciences, 2004.
[51] Miller M. I. Snyder, D. L. Random point processes in time and space. New
York: Springer., 1991.
[52] Aonan Tang, David Jackson, Jon Hobbs, Wei Chen, Jodi L. Smith, Hema
Patel, Anita Prieto, Dumitru Petrusca, Matthew I. Grivich, Alexander
Sher, Pawel Hottowy, Wladyslaw Dabrowski, Alan M. Litke, and John M.
Beggs. A maximum entropy model applied to spatial and temporal cor-
relations from cortical networks In Vitro. The Journal of Neuroscience,
28(2):505 -- 518, January 2008.
[53] F. E. Theunissen, S. V. David, N. C. Singh, A. Hsu, W. E. Vinje, and J. L.
Gallant. Estimating spatio-temporal receptive fields of auditory and visual
neurons from their responses to natural stimuli. Network, 12(3):289 -- 316,
January 2001.
[54] Wilson Truccolo, Uri T. Eden, Matthew R. Fellows, John P. Donoghue, and
Emery N. Brown. A point process framework for relating neural spiking
activity to spiking history, neural ensemble and extrinsic covariate effects.
J Neurophysiol, 93:1074 -- 1089, 2005.
[55] Juan Carlos Vasquez, Olivier Marre, Adrian G Palacios, Michael J Berry,
and Bruno Cessac. Gibbs distribution analysis of temporal correlation
structure on multicell spike trains from retina ganglion cells. J. Physiol.
Paris, 106:120 -- 127, 2012.
[56] Shan Yu, Debin Huang, Wolf Singer, and Danko Nikolic. A small world of
neuronal synchrony. Cereb. Cortex, 2008.
23
|
1005.0504 | 2 | 1005 | 2010-05-19T16:10:14 | Statistical Mechanics Model for Protein Folding | [
"physics.bio-ph",
"cond-mat.stat-mech",
"physics.chem-ph",
"q-bio.BM"
] | We present a novel statistical mechanics formalism for the theoretical description of the process of protein folding$\leftrightarrow$unfolding transition in water environment. The formalism is based on the construction of the partition function of a protein obeying two-stage-like folding kinetics. Using the statistical mechanics model of solvation of hydrophobic hydrocarbons we obtain the partition function of infinitely diluted solution of proteins in water environment. The calculated dependencies of the protein heat capacities upon temperature are compared with the corresponding results of experimental measurements for staphylococcal nuclease and metmyoglobin. | physics.bio-ph | physics | Statistical Mechanics Model for Protein Folding
Alexander V. Yakubovich∗, Andrey V. Solov'yov and Walter Greiner
Frankfurt Institute for Advanced Studies, Goethe University,
60438, Ruth-Moufang Str. 1, Frankfurt am Main, Germany
∗E-mail: [email protected]
We present a novel statistical mechanics formalism for the theoretical description
of the process of protein folding↔unfolding transition in water environment. The
formalism is based on the construction of the partition function of a protein obeying
two-stage-like folding kinetics. Using the statistical mechanics model of solvation
of hydrophobic hydrocarbons we obtain the partition function of infinitely diluted
solution of proteins in water environment. The calculated dependencies of the protein
heat capacities upon temperature are compared with the corresponding results of
experimental measurements for staphylococcal nuclease and metmyoglobin.
0
1
0
2
y
a
M
9
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
2
v
4
0
5
0
.
5
0
0
1
:
v
i
X
r
a
I.
INTRODUCTION
2
Proteins are biological polymers consisting of elementary structural units, amino acids.
Being synthesized at ribosome, proteins are exposed to the cell interior where they fold
into their unique three dimensional structure. The process of forming the protein's three
dimensional structure is called the process of protein folding. The correct folding of protein is
of crucial importance for the protein's proper functioning. Despite numerous works devoted
to investigation of protein folding this process is still not entirely understood. The current
state-of-the-art in experimental and theoretical studies of the protein folding process are
described in recent reviews and references therein [1 -- 5].
In this paper we develop a novel theoretical method for the description of the protein fold-
ing process which is based on the statistical mechanics principles. Considering the process
of protein folding as a first order phase transition in a finite system, we present a statistical
mechanics model for treating the folding↔unfolding phase transition in single-domain pro-
teins. The suggested method is based on the theory developed for the helix↔coil transition
in polypeptides discussed in [6 -- 17]. A way to construct a parameter-free partition function
for a system experiencing α-helix↔random coil phase transition in vacuo was studied in
[6]. In [8] we have calculated potential energy surfaces (PES) of polyalanines of different
lengths with respect to their twisting degrees of freedom. This was done within the frame-
work of classical molecular mechanics. The calculated PES were then used to construct a
parameter -- free partition function of a polypeptide and to derive various thermodynamical
characteristics of alanine polypeptides as a function of temperature and polypeptide length.
In this paper we construct the partition function of a protein in vacuo, which is the
further generalization of the formalism developed in [9], accounting for folded, unfolded and
prefolded states of the protein. This way of the construction of the partition function is
consistent with nucleation-condensation scenario of protein folding, which is a very common
scenario for globular proteins [18] and implies that at the early stage of protein folding the
native-like hydrophobic nucleus of protein is formed, while at the later stages of the protein
folding process all the rest of amino acids also attain the native-like conformation.
For the correct description of the protein folding in water environment it is of primary
importance to consider the interactions between the protein and the solvent molecules. The
hydrophobic interactions are known to be the most important driving forces of protein folding
3
[19]. In the present work we present a way how one can construct the partition function of
the protein accounting for the interactions with solvent, i.e. accounting for the hydrophobic
effect. The most prominent feature of our approach is that it is developed for concrete
systems in contrast to various generalized and toy-models of protein folding process.
We treat the hydrophobic interactions in the system using the statistical mechanics for-
malism developed in [20] for the description of the thermodynamical properties of the sol-
vation process of aliphatic and aromatic hydrocarbons in water. However, accounting solely
for hydrophobic interactions is not sufficient for the proper description of the energetics
of all conformational states of the protein and one has to take electrostatic interactions
into account. In the present work the electrostatic interactions are treated within a similar
framework as described in [21].
We have applied the developed statistical mechanics model of protein folding for two
globular proteins, namely staphylococcal nuclease and metmyoglobin. These proteins have
simple two-stage-like folding kinetics and demonstrate two folding↔unfolding transitions,
refereed as heat and cold denaturation [22, 23]. The comparison of the results of the the-
oretical model with that of the experimental measurements shows the applicability of the
suggested formalism for an accurate description of various thermodynamical characteristics
in the system, e.g. heat denaturation, cold denaturation, increase of the reminiscent heat
capacity of the unfolded protein, etc.
Our paper is organized as follows. In Sec. II A we present the formalism for the construc-
tion of the partition function of the protein in water environment and justify the assumptions
made on the system's properties. In Section III we discuss the results obtained with our
model for the description of folding↔unfolding transition in staphylococcal nuclease and
metmyoglobin. In Section IV we summarize the paper and suggest several ways for a fur-
ther development of the theoretical formalism.
II. THEORETICAL METHODS
A. Partition function of a protein
To study thermodynamic properties of the system one needs to investigate its potential
energy surface with respect to all the degrees of freedom. For the description of macromolec-
4
ular systems, such as proteins, efficient model approaches are necessary.
The most relevant degrees of freedom in the protein folding process are the twisting
degrees of freedom along its backbone chain [6, 7, 9 -- 11, 14, 15, 24, 25]. These degrees of
freedom are defined for each amino acid of the protein except for the boundary ones and
are described by two dihedral angles ϕi and ψi (for definition of ϕi and ψi see e.g. [6, 7, 9 --
11, 14, 15]).
The degrees of freedom of a protein can be classified as stiff and soft ones. We call the
degrees of freedom corresponding to the variation of bond lengths, angles and improper
dihedral angles as stiff, while degrees of freedom corresponding to the angles ϕi and ψi are
soft degrees of freedom [6]. The stiff degrees of freedom can be treated within the harmonic
approximation, because the energies needed for a noticeable structural rearrangement with
respect to these degrees of freedom are about several eV, which is significantly larger than
the characteristic thermal energy of the system (kT), being at room temperature equal to
0.026 eV [13 -- 15, 26 -- 28].
A Hamiltonian of a protein is constructed as a sum of the potential, kinetic and vibrational
energy terms. Assuming the harmonic approximation for the stiff degrees of freedom it is
possible to derive the following expression for the partition function of a protein in vacuo
being in a particular conformational state j [6]:
Zj = Aj(kT )3N −3− ls
2 Zϕ∈Γj
...Zψ∈Γj
e−ǫj({ϕ,ψ})/kT dϕ1...dϕndψ1...dψn,
(1)
where T is the temperature, k is the Boltzmann constant, N is the total number of atoms
in the protein, ls is the number of soft degrees of freedom, Aj is defined as follows:
Aj =
Vj · M 3/2 ·qI (1)
ls
(2π)
j I (3)
j I (2)
j Qls
2 π3NQ3N −6−ls
i=1
i=1qµs(j)
i
ω(j)
i
.
(2)
Aj is a factor which depends on the mass of the protein M, its three main momenta of inertia
I (1,2,3)
j
the generalized masses µs(j)
, specific volume Vj, the frequencies of the stiff normal vibrational modes ω(j)
corresponding to the soft degrees of freedom [6]. ǫi in Eq. (1)
and on
i
i
describes the potential energy of the system corresponding to the variation of soft degrees
of freedom. Integration in Eq. (1) is performed over a certain part of a phase space of the
system (a subspace Γj) corresponding to the soft degrees of freedom ϕ and ψ. The form of
the partition function in Eq. (1) allows one to avoid the multidimensional integration over
5
the whole coordinate space and to reduce the integration only to the relevant parts of the
phase space. ǫj in Eq. (1) denotes the potential energy surface of the protein as a function of
twisting degrees of freedom in the vicinity of protein's conformational state j. Note that in
general the proper choice of all the relevant conformations of protein and the corresponding
set of Γj is not a trivial task.
One can expect that the factors Aj in Eq. (1) depend on the chosen conformation of
the protein. However, due to the fact that the values of specific volumes, momenta of
inertia and frequencies of normal vibration modes of the system in different conformations
are expected to be close [9, 29], the values of Aj in all conformations become nearly equal,
at least in the zero order harmonic approximation, i.e. Aj ≡ A. Another simplification
of the integration in Eq. (1) comes from the statistical independence of amino acids. We
assume that within each conformational state j all amino acids can be treated statistically
independently, i.e. the particular conformational state of i-th amino acid characterized by
angles ϕi ∈ Γj and ψi ∈ Γj does not influence the potential energy surface of all other
amino acids, and vice versa. This assumption is well applicable for rigid conformational
states of the protein such as native state. For the native state of a protein all atoms of the
molecule move in harmonic potential in the vicinity of their equilibrium positions. However,
in unfolded states of the protein the flexibility of the backbone chain leads to significant
variations of the distances between atoms, and consequently to a significant variation of
interactions between atoms. Accurate accounting (both analytical and computational) for
the interactions between distant atoms in the unfolded state of a protein is extremely difficult
(see Ref. [30] for analytical treatment of interactions in unfolded states of a protein). In this
work we assume that all amino acids in unfolded state of a protein move in the identical
mean field created by all the amino acids and leave the corrections to this approximation
for further considerations.
With the above mentioned assumptions the partition function of a protein Zp (without
any solvent) reads as:
Zp = A · (kT )3N −3− ls
2
ξ
Xj=1
a
Yi=1Z π
−πZ π
−π
exp −
ǫ(j)
i (ϕi, ψi)
kT
! dϕidψi,
(3)
where the summation over j includes all ξ statistically relevant conformations of the protein,
a is the number of amino acids in the protein and ǫ(j)
is the potential energy surface as a
i
function of twisting degrees of freedom ϕi and ψi of the i-th amino acid in the j-th conforma-
6
tional state of the protein. The exact construction of ǫ(j)
a (ϕi, ψi) for various conformational
states of a particular protein will be discussed below. We consider the angles ϕ and ψ as
the only two soft degrees of freedom in each amino acid of the protein, and therefore the
total number of soft degrees of freedom of the protein ls = 2a.
Partition function in Eq. (3) can be further simplified if one assumes (i) that each amino
acid in the protein can exist only in two conformations: the native state conformation and
the random coil conformation; (ii) the potential energy surfaces for all the amino acids are
identical. This assumption is applicable for both the native and the random coil state. It is
not very accurate for the description of thermodynamical properties of single amino acids,
but is reasonable for the treatment of thermodynamical properties of the entire protein.
The judgment of the quality of this assumption could be made on the basis of comparison
of the results obtained with its use with experimental data. Such comparison is performed
in Sec. III of this work.
Amino acids in a protein being in its native state vibrate in a steep harmonic potential.
Here we assume that the potential energy profile of an amino acid in the native conformation
should not be very sensitive to the type of amino acid and thus can be taken as, e.g., the
potential energy surface for an alanine amino acid in the α-helix conformation [8]. Using
the same arguments the potential energy profile for an amino acid in unfolded protein
state can be approximated by e.g. the potential of alanine in the unfolded state of alanine
polypeptide (see Ref. [8] for discussion and analysis of alanine's potential energy surfaces).
Indeed, for an unfolded state of a protein it is reasonable to expect that once neglecting the
long-range interactions all the differences in the potential energy surfaces of various amino
acids arise from the steric overlap of the amino acids's radicals. This is clearly seen on
alanine's potential energy surface at values of ϕ > 0◦ presented in Ref. [8]. But the part of
the potential energy surface at ϕ > 0◦ gives a minor contribution to the entropy of amino
acid at room temperature. This fact allows one to neglect all the differences in potential
energy surfaces for different amino acids in an unfolded protein, at least in the zero order
approximation. This assumption should be especially justified for proteins with the rigid
helix-rich native structure. The staphylococcal nuclease, which we study here has definitely
high α-helix content. Another argument which allows to justify our assumption for a wider
family of proteins is the rigidity of the protein's native structure. Below, we validate the
assumptions made by performing the comparison of the results of our theoretical model with
7
the experimental data for α/β rich protein metmyoglobin obtained in [23].
For the description of the folding ↔ unfolding transition in small globular proteins obey-
ing simple two-state-like folding kinetics we assume that the protein can exist in one of
three states: completely folded state, completely unfolded state and partially folded state
where some amino acids from the flexible regions with no prominent secondary structure
are in the unfolded state, while other amino acids are in the folded conformation. With this
assumption the partition function of the protein reads as:
Zp= Z0 +
a
Xi=a−κ
κ!
(i − (a − κ))!(a − i)!
Zi,
(4)
where Zi is defined in Eq. (1), Z0 is the partition function of the protein in completely
unfolded state, a is the total number of amino acids in a protein and κ is the number of
amino acids in flexible regions. The factorial term in Eq. (4) accounts for the states in which
various amino acids from flexible regions independently attain the native conformation. The
summation in Eq. (4) is performed over all partially folded states of the protein, where a − κ
is the minimal possible number of amino acids being in the folded state. The factorial term
describes the number of ways to select i − (a − κ) amino acids from the flexible region of
the protein consisting of κ amino acids attaining native-like conformation.
Finally, the partition function of the protein in vacuo has the following form:
Zp = Zp · A(kT )3N −3−a,
where
a
Zp = Z a
u +
Xi=a−κ
−πZ π
exp(cid:18)−
−πZ π
exp(cid:18)−
−π
−π
Zb =Z π
Zu =Z π
κ!Z i
bZ a−i
u
exp (i · E0/kT )
(i − (a − κ))!(a − i)!
ǫb(ϕ, ψ)
kT (cid:19)dϕdψ
kT (cid:19)dϕdψ.
ǫu(ϕ, ψ)
(5)
(6)
(7)
(8)
Here we omitted the trivial factor describing the motion of the protein center of mass,
which is of no significance for the problem considered, ǫb(ϕ, ψ) (b stands for bound) is
the potential energy surface of an amino acid in the native conformation and ǫu(ϕ, ψ) (u
stands for unbound) is the potential energy surface of an amino acid in the random coil
conformation. The potential energy profile of an amino acid is calculated as a function of
its twisting degrees of freedom ϕ and ψ. Let us denote by ǫ0
b and ǫ0
u the global minima
on the potential energy surfaces of an amino acid in folded and in unfolded conformations
respectively. The potential energy of an amino acid then reads as ǫ0
u,b + ǫu,b(ϕ, ψ). E0 in
Eq. (6) is defined as the energy difference between the global energy minima of the amino
acid potential energy surfaces corresponding to the folded and unfolded conformations, i.e.
8
E0 = ǫ0
u − ǫ0
b. The potential energy surfaces for amino acids as functions of angles ϕ and ψ
were calculated and thoroughly analyzed in [8].
In nature proteins perform their function in the aqueous environment. Therefore the
correct theoretical description of the folding↔unfolding transition in water environment
should account for solvent effects.
B. Partition function of a protein in water environment
In this section we evaluate E0 and construct the partition function for the protein in
water environment.
The partition function of the infinitely diluted solution of proteins Z can be constructed
as follows:
Z =
ξ
Xj=1
p Z (j)
Z (j)
W ,
(9)
where Z (j)
a protein and Z (j)
p
W is the partition function of all water molecules in the j-th conformational state of
is the partition function of the protein in its j-th conformational state, in
which we further omit the factor describing the contribution of stiff degrees of freedom in the
system. This is done in order to simplify the expressions, because stiff degrees of freedom
provide a constant contribution to the heat capacity of the system since the heat capacity
of the ensemble of harmonic oscillators is constant. Below for the simplicity of notations we
put Zp ≡ Zp.
There are two types of water molecules in the system: (i) molecules in pure water and
(ii) molecules interacting with the protein. We assume that only the water molecules being
in the vicinity of the protein's surface are involved in the folding↔unfolding transition,
because they are affected by the variation of the hydrophobic surface of a protein. This
surface is equal to the protein's solvent accessible surface area (SASA) of the hydrophobic
amino acids. The number of interacting molecules is proportional to SASA and include only
the molecules from the first protein's solvation shell. This area depends on the conformation
9
of the protein. The main contribution to the energy of the system caused by the variation
of the protein's SASA associated with the side-chain radicals of amino acids because the
contribution to the free energy assosiated with solvation of protein's backbone is small [31].
Thus, in this work we pay the main attention to the accounting for the SASA change arising
due to the solvation of side chain radicals.
We treat all water molecules as statistically independent, i.e. the energy spectra of the
states of a given molecule and its vibrational frequencies do not depend on a particular
state of all other water molecules. Thus, the partition function of the whole system Z can
be factorized and reads as:
Z =
ξ
Xj=1
Z (j)
p Z Yc(j)
s
Z Nt−Yc(j)
w
,
(10)
where ξ is the total number of states of a protein, Zs is the partition function of a water
molecule affected by the interaction with the protein and Zw is the partition function of
a water molecule in pure water. Yc(j) is the number of water molecules interacting with
the protein in the j-th conformational state. Nt is the total number of water molecules in
the system. To simplify the expressions we do not account for water molecules that do not
interact with the protein in any of its conformational states, i.e. Nt = maxj{Yc(j)}.
To construct the partition function of water we follow the formalism developed in [20]
and refer only to the most essential details of that work. The partition function of a water
molecule in pure water reads as:
Zw =
4
Xl=0
[ξlfl exp(−El/kT )] ,
(11)
where the summation is performed over 5 possible states of a water molecule (the states in
which water molecule has 4,3,2,1 or 0 hydrogen bonds with the neighboring molecules). El
are the energies of these states and ξl are the combinatorial factors being equal to 1,4,6,4,1
for l = 0, 1, 2, 3, 4, respectively. They describe the number of choices to form a given number
of hydrogen bonds. fl in Eq. (11) describes the contribution due to the partition function
arising to to the translation and libration oscillations of the molecule.
In the harmonic
approximation fl are equal to:
fl =h1 − exp(−hν(T )
l /kT )i−3h1 − exp(−hν(L)
l /kT )i−3
,
(12)
Number of hydrogen bonds
0
1
2
3
Energy level, Ei (kcal/mol)
Energy level, Es
i (kcal/mol)
Translational frequencies, ν
(T )
i
, cm−1
Librational frequencies, ν
(L)
i
, cm−1
6.670 4.970 3.870 2.030
6.431 4.731 3.631 1.791 -0.564
26
86
61
57
210
197
374
500
750
750
10
4
0
Xl=0
TABLE I. Parameters of the partition function of water according to [20]
where ν(T )
l
and ν(L)
l
are translation and libration motions frequencies of a water molecule
in its l-th state, respectively. These frequencies are calculated in Ref. [20] and are given in
Table I. The contribution of the internal vibrations of water molecules is not included in
Eq. (11) because the frequencies of these vibrations are practically not influenced by the
interactions with surrounding water molecules.
The partition function of a water molecule from the protein's first solvation shell reads
as:
4
Zs =
[ξlfl exp(−Es
l /kT )] ,
(13)
where fl are defined in Eq. (12) and Es
interacting with aliphatic hydrocarbons of protein's amino acids. Values of energies Es
l denotes the energy levels of a water molecule
l are
given it Table I. For simplicity we treat all side-chain radicals of a protein as aliphatic
hydrocarbons because most of the protein's hydrophobic amino acids consist of aliphatic-
like hydrocarbons.
It is possible to account for various types of side chain radicals by
using the experimental results of the measurements of the solvation free energies of amino
acid radicals from Ref. [32] and associated works. However, this correction will imply the
reparametrization of the theory presented in [20] and will lead to the introduction of ∼ 20 · 5
additional parameters. Here we do not perform such a task since this kind of improvement
of the theory would smear out the understanding of the principal physical factors underlying
the protein folding↔unfolding transition.
In our theoretical model we also account for the electrostatic interaction of protein's
charged groups with water. The presence of electrostatic field around the protein leads to
the reorientation of H2O molecules in the vicinity of charged groups due to the interaction
of dipole moments of the molecules with the electrostatic field. The additional factor arising
in the partition function (11) of the molecules reads as:
ZE =(cid:18) 1
4πZ exp(cid:18)−
E · d cos θ
kT
(cid:19) sin θdθdϕ(cid:19)α
,
11
(14)
where E is the strength of the electrostatic field, d is the absolute value of the H2O molecule
dipole moment, α is the ratio of the number of water molecules that interact with the
electrostatic field of the protein (NE) to the number of water molecules interacting with
the surface of the amino acids from the inner part of the protein while they are exposed to
water when the protein is being unfolded (Nw), i.e. α = NE/Nw. Note that the effects of
electrostatic interaction turn out to be more pronounced in the folded state of the protein.
This happens because in the unfolded state of a protein opposite charges of amino acid's
radicals are in average closer in space due to the flexibility of the backbone chain, while in
the folded state the positions of the charges are fixed by the rigid structure of a protein.
Integrating Eq. (14) allows to write the factor ZE for the partition function of a single
H2O molecule in pure water in the form:
ZE = kT sinh(cid:2) Ed
kT(cid:3)
Ed
!α
.
(15)
This equation shows how the electrostatic field enters the partition function. In general,
E depends on the position in space with respect to the protein. However, here we neglect this
dependence and instead we treat the parameter E as an average, characteristic electrostatic
field created by the protein.
Let us denote by Ns the number of water molecules interacting with the proteins surface
in its folded state i.e. Nt = Ns + Nw; where Nt is defined in Eq (10). We assume that the
number of water molecules interacting with the protein (Yc) is linearly dependent on the
number of amino acids being in the unfolded conformation, i.e. Yc = Ns + iNw/a, where i
is the number of the amino acids in the unfolded conformation and a is the total number
of amino acids in the protein. Thus, the partition function (10) with the accounting for the
factor (15) reads as:
Z = Z Ns
s
·
ξ
Xj=1(cid:16)ZbZ Nw/a
w
Z Nw/a
E
exp (i · E0/kT )(cid:17)i(j)
s
(cid:0)ZuZ Nw/a
(cid:1)a−i(j)
,
(16)
where i(j) denotes the number of the amino acids being in the folded conformation when the
protein is in the j-th conformational state. Accounting for the statistical factors for amino
acids being in the folded and unfolded states, similarly to how it was done for the vacuum
case (see Eq. (6)), one derives from Eq. (16) the following final expression:
12
Z = (Zs)Ns ×
×"Z a
uZ Nw
s +
a
Xi=a−κ
κ! exp (i · E0/kT )
(i − (a − κ))!(a − i)!(cid:16)ZbZ Nw/a
w
(17)
Z Nw/a
E
(cid:17)i
(ZuZ Nw/a
s
)a−i# ,
where the term in the square brackets accounts for all statistically significant conformational
states of the protein.
Having constructed the partition function of the system we can evaluate with its use all
thermodynamic characteristics of the system, such as e.g. entropy, free energy, heat capacity,
etc. The free energy (F ) and heat the capacity (c) of the system can be calculated from the
partition function as follows:
F (T ) = −kT ln Z(T ),
∂2F (T )
c(T ) = −T
.
∂T 2
(18)
(19)
In this work we analyze the dependence of protein's heat capacity on temperature and
compare the predictions of our model with available experimental data.
III. RESULTS AND DISCUSSION
In this section we calculate the dependencies of the heat capacity on temperature for two
globular proteins metmyoglobin and staphylococcal nuclease and compare the results with
experimental data from [22, 23].
The structures of metmyoglobin and staphylococcal nuclease proteins are shown in Fig. 1.
These are relatively small globular proteins consisting of ∼150 amino acids. Under certain
experimental conditions (salt concentration and pH) the metmyoglobin and the staphylo-
coccal nuclease experience two folding↔unfolding transitions, which induce two peaks in
the dependency of heat capacity on temperature (see further discussion). The peaks at
lower temperature are due to the cold denaturation of the proteins. The peaks at higher
temperatures arise due to the ordinary folding↔unfolding transition. The availability of
experimental data for the heat capacity profiles of the mentioned proteins, the presence of
the cold denaturation and simple two-stage-like folding kinetics are the reasons for select-
ing these particular proteins as case studies for the verification of the developed theoretical
model.
13
FIG. 1. a) Structure of staphylococcal nuclease (PDB ID 1EYD [33]), and b) horse heart metmyo-
globin (PDB ID 1YMB [34]). Images have been rendered using VMD program [35].
A. Heat capacity of staphylococcal nuclease
Staphylococcal or micrococcal nuclease (S7 Nuclease) is a relatively nonspecific enzyme
that digests single-stranded and double-stranded nucleic acids, but is more active on single-
stranded substrates [36]. This protein consists of 149 amino acids. It's structure is shown
in Fig. 1a.
To calculate the SASA of staphylococcal nuclease in the folded state the 3D struc-
ture of the protein was taken from the Protein Data Bank [37] (PDB ID 1EYD). Using
CHARMM27 [28] forcefield and NAMD program [38] we performed the structural optimiza-
tion of the protein and calculated SASA with the solvent probe radius 1.4 A.
The value of SASA of the side-chain radicals in the folded protein conformation is equal
to Sf =6858 A2. In order to calculate SASA for an unfolded protein state, the value of all
angles ϕ and ψ were put equal to 180◦, corresponding to a fully stretched conformation.
14
Then, the optimization of the structure with the fixed angles ϕ and ψ was performed.
The optimized geometry of the stretched molecule has a minor dependence on the value
of dielectric susceptibility of the solvent, therefore the value of dielectric susceptibility was
chosen to be equal to 20, in order to mimic the screening of charges by the solvent. SASA
of the side-chain radicals in the stretched conformation of the protein is equal to Su =15813
A2.
The change of the number of water molecules those interacting with the protein due to
the unfolding process can be calculated as follows:
Nw = (Su − Sf )n2/3,
(20)
where Su = 15813 A2 and Sf = 6858 A2 are the SASA of the protein in unfolded and in
folded conformations, respectively and n is the density of the water molecules. The volume
of one mole of water is equal to 18 cm3, therefore n ≈ 30 A−3
To account for the effects caused by the electrostatic interaction of water molecules with
the charged groups of the protein it is necessary to evaluate the strength of the average
electrostatic field E in Eq. (15). The strength of the average field can be estimated as
E · d = kT , where d is the dipole moment of a water molecule, k is Bolzmann constant and
T=300 K is the room temperature. According to this estimate the energy of characteristic
electrostatic interaction of water molecules is equal to the thermal energy per degree of
freedom of a molecule.
The total number of water molecules that interact with the electrostatic field of the
protein can be estimated from the known Debye screening length of a charge in electrolyte
as follows:
NE = Nq
4πρ
3
λ3
d,
(21)
where Nq is the number of charged groups in the protein, ρ is the density of water and λ
is the Debye screening length. Debye screening length of the symmetric electrolyte can be
calculated as follows [39]:
λd =r ǫǫ0kT
2NAe2I
,
(22)
15
where ǫ0 is the permittivity of free space, ǫ is the dielectric constant, NA is the Avogadro
number, e is the elementary charge and I is is the ionic strength of the electrolyte.
The experiments on denaturation of staphylococcal nuclease and metmyoglobin were per-
formed in 100 mM ion buffer of sodium chloride and 10mM buffer of sodium acetate respec-
tively [22, 23]. The Debye screening length in water with 10 mM and 100 mM concentration
of ions is λd =30 A and λd =10 A at room temperature respectively.
The described method allows to estimate the number of water molecules (NE) interacting
with electric filed created by the charged groups of a protein. It should be considered as
qualitative estimate since we have assumed the average electric field as being constant within
a sphere of the radius λd, but in fact it experiences some variations. Thus, at the distances
∼15 A from the point charge the interaction energy of a H2O molecule with the electric field
becomes equal to ∼ 0.02 kT (for this estimate we have used the linear growing distance-
dependent dielectric susceptibility ǫ = 6R as derived in [40] for the atoms fully exposed to
the solvent). However, we expect that the more accurate analysis accounting for the spatial
variation of the electric field will not change significantly the results of the analysis reported
here, because it is based on the physically correct picture of the effect and the realistic
values of all the physical quantities. At physiological conditions staphylococcal nuclease has
8 charged residues [41]. The value of α for this protein varies within the interval from 1.29
to 31.27 for λd ∈[10..30] A. In our numerical analysis we have used the characteristic value
of α equal to 2.5.
Note that number of molecules interacting with the electrostatic field NE and the strength
of the electrostatic field E should be considered as the ef f ective parameters of our model.
In this work we do not perform accurate accounting for the spatial dependence of the elec-
trostatic field. Instead, we introduce the parameters α and E that can be interpreted as
effective values of the number of H2O molecules and the strength of the electrostatic field
correspondingly. Let us stress that the number of water molecules α and the strength of the
field E are not independent parameters of our model because by choosing the higher value
of E and smaller value of α or vice versa one can derive the same heat capacity profile.
In this work we do not investigate the dependencies of the heat capacity profiles on the
values of the parameters α and E. Below we focus on the investigation of the dependence
of the protein heat capacity on the energy E0 at the fixed value of α and E equal to 2.5 and
0.58 kcal/mol respectively.
16
pH value
7.0
5.0
4.5 3.88 3.23
E0 (kcal/mol) 0.789 0.795 0.803 0.819 0.890
TABLE II. Values of E0 for staphylococcal nuclease at different values of pH of the solvent
An important parameter of the model is the energy difference between the two states
of the protein normalized per one amino acid, E0 introduced in Eq. (6). This parameter
describes both the energy loss due to the separation of the hydrophobic groups of the protein
which attract in the native state of the protein due to Van-der-Waals interaction and the
energy gain due to the formation of Van-der-Waals interactions of hydrophobic groups of
the protein with H2O molecules in the protein's unfolded state. Also, the difference of
the electrostatic energy of the system in the folded and unfolded states is accounted for in
E0. The difference of the electrostatic energy may depend on various characteristics of the
system, such as concentration of ions in the solvent and its pH, on the exact location of the
charged sites in the native conformation of the protein and on the probability distribution
of distances between charged amino acids in the unfolded state. Thus, exact calculation of
E0 is rather difficult. It is a separate task which we do not intend to address in this work.
Instead, in the current study the energy difference between the two phases of the protein
is considered as a parameter of the model. We treat E0 as being dependent on external
properties of the system, in particular on the pH value of the solution.
Another characteristic of the protein folding↔unfolding transition is its cooperativity.
In the model it is described by the parameter κ in Eq. (4). κ describes the number of
amino acids in the flexible regions of the protein. The staphylococcal nuclease possesses a
prominent two-stage folding kinetics, therefore only 5-10% of amino acids is in the protein's
flexible regions. Thus, the value of κ for this protein is small. It can be estimated as being
equal to 149 · 7% ≈ 10 amino acids.
The values of E0 for staphylococcal nuclease at different values of pH are given in Table II.
For the analysis of the variation of the thermodynamic properties of the system during
the folding process one can omit all the contributions to the free energy of the system that
do not alter significantly in the temperature range between -50◦C and 150◦. Therefore, from
the expression for the total free energy of the system F we can subtract all slowly varying
contributions F0 as follows:
Z0(cid:19),
δF = F − F0 = −(kT ln Z − kT ln Z0) = −kT ln(cid:18) Z
17
(23)
(24)
From Eq. (23) follows that the subtraction of F0 corresponds to the division of the total
partition function Z by the partition function of the subsystem (Z0) with slowly varying
thermodynamical properties. Therefore, in order to simplify the expressions, one can divide
the partition function in Eq. (17) by the partition function of fully unfolded conformation
of a protein (by Z a
uZ Nw
s
) and by the partition function of Ns free water molecules (by Z Ns
w ).
Thus, Eq. (17) can be rewritten as follows:
Z =(cid:18) Zs
Zw(cid:19)Ns 1 +
a
Xi=a−κ
κ! exp (i · E0/kT )
(i − (a − κ))!(a − i)!(cid:18) Zb
Zu(cid:19)i(cid:18)ZwZE
Zs (cid:19)iNw/a! .
(25)
With the use of Eq. (19) on can calculate the heat capacity of the system as follows:
c(T ) = A + B(T − T0) − T
∂2F (T )
∂T 2
,
(26)
where the factors A and B are responsible for the absolute value and the inclination of the
heat capacity curve respectively. These factors account for the contribution of stiff harmonic
vibrational modes in the system (factor A) and for the unharmonic correction to these vi-
brations (factors B and T0). The contribution of protein's stiff vibrational modes and the
heat capacity of the fully unfolded conformation of protein is also included into these factors.
In our numerical analysis we have adjusted the values of A, B and T0 in order to match
experimental measurements. However, factors A, B and T0 should not be considered as
parameters of our model since their values are not related to the thermodynamic character-
istics of the folding↔unfolding transition and depend not entirely on the properties of the
protein but also on the properties of the solution, protein and ion concentrations, etc.
In our calculations for staphylococcal nuclease we have used the values of A = 1.25 JK−1g−1,
B = 6.25 · 10−3 JK−2g−1 and T0 =323 ◦K in Eq. (26).
The dependence of heat capacity on temperature calculated for staphylococcal nuclease
at different pH values are presented in Fig. III A by solid lines. The results of experimental
18
measurements form Ref. [22] are presented by symbols. From Fig. III A it is seen that
staphylococcal nuclease experience two folding↔transitions in the range of pH between 3.78
and 7.0. At the pH value 3.23 no peaks in the heat capacity is present.
It means that
the protein exists in the unfolded state over the whole range of experimentally accessible
temperatures.
FIG. 2. Dependencies of the heat capacity on temperature for staphylococcal nuclease (see Fig. 1a)
at different values of pH. Solid lines show results of the calculation, while symbols present experi-
mental data from Ref. [22].
Comparison of the theoretical results with experimental data shows that our theoretical
model reproduces experimental behavior better for the solvents with higher pH. The heat
capacity peak arising at higher temperatures due to the standard folding↔unfolding tran-
sition is reproduced very well for pH values being in the region 4.5-7.0. The deviations at
low temperatures can be attributed to the inaccuracy of the statistical mechanics model of
19
water in the vicinity of the freezing point.
The accuracy of the statistical mechanics model for low pH values around 3.88 is also
quite reasonable. The deviation of theoretical curves from experimental ones likely arise
due to the alteration of the solvent properties at high concentration of protons or due to
the change of partial charge of amino acids at pH values being far from the physiological
conditions.
Despite some difference between the predictions of the developed model and the experi-
mental results arising at certain temperatures and values of pH the overall performance of
the model can be considered as extremely good for such a complex process as structural
folding transition of a large biological molecule.
B. Heat capacity of metmyoglobin
Metmyoglobin is an oxidized form of a protein myoglobin. This is a monomeric protein
containing a single five-coordinate heme whose function is to reversibly form a dioxygen
adduct [42]. Metmyolobin consists of 153 amino acids and it's structure is shown in Fig. 1
on the right.
In order to calculate SASA of side chain radicals of metmyoglobin exactly the same
procedure as for staphylococcal nuclease was performed (see discussion in the previous sub-
section). SASA in the folded and unfolded states of the protein has been calculated and is
equal 6847 A2 and 16926 A2 respectively. Thus, there are 984 H2O molecules interacting
with protein's hydrophobic surface in its unfolded state.
The electrostatic interaction of water molecules with metmyoglobin was accounted for in
the same way as for staphylococcal nuclease. The parameter α in Eq. (15) was chosen to
be equal to 2.5. With this we derive that 10950 H2O molecules involve in the interaction
with the electrostatic field of metmyoglobin in its folded state. The strength of the field was
chosen the same as for staphylococcal nuclease.
The parameter κ for metmyoglobin in Eq. (4), describing the cooperativity of the
folding↔unfolding transition, differs significantly from that for staphylococcal nuclease.
The transition in metmyoglobin is less cooperative than the transition in staphylococcal
nuclease because metmyoglobin has intermediate partially folded states [43]. Thus, while
the rigid native-like core of the protein is formed, a significant fraction of amino acids in
20
pH value
4.10 3.70 3.84 3.5
E0 (kcal/mol) 1.128 1.150 1.165 1.2
TABLE III. Values of E0 for metmyoglobin at different values of solvent pH.
the flexible regions of the protein can exist in the unfolded state. We assume that 1/3 of
metmyoglobin's amino acids are in the flexible region, i.e. the parameter κ in Eq. (4) equal
to 50.
The values of E0 in Eq. (6) differ from that for staphylococcal nuclease and are compiled in
Table III. In our calculations for metmyoglobin we have used the values of A = 1.6 JK−1g−1,
B = 8.25 · 10−3 JK−2g−1 and T0=323 ◦K in Eq. (26).
FIG. 3. Dependencies of the heat capacity on temperature for horse heart metmyoglobin (see
Fig. 1b) at different values of pH. Solid lines show the results of the calculation. Symbols present
the experimental data from Ref. [23].
Solid lines in Fig. 3 show the dependence of the metmyoglobin's heat capacity on tem-
21
perature calculated using the developed theoretical model. The experimental data from
Ref. [23] are shown by symbols.
Metmyoglobin experiences two folding↔unfolding transitions at the pH values exceeding
3.5 which can be called as cold and heat denaturations of the protein. The dependence of
the heat capacity on temperature therefore has two characteristic peaks, as seen in Fig. 3.
Figure 3 shows that at pH lower than 3.84 metmyoglobin exists only in the unfolded state.
The comparison of predictions of the developed theoretical model with the experimental
data on heat capacity shows that the theoretical model is well applicable for metmyoglobin
case as well. The good agreement of the theoretical and experimental heat capacity profiles
over the whole range of temperatures and pH values shows that the model treats correctly
the thermodynamics of the protein folding process.
Our theory includes a number of parameters, namely the energy difference between two
phases E0, strength of the electrostatic field E, number of interacting H2O molecules α, the
parameter describing the cooperativity of the phase transition κ, as well as other parameters
introduced in Ref. [20] to treat the partition function of water. Three parameters, E, E0
and κ, are dependent on the properties of a particular protein and on the pH of the solvent.
We have adjusted the values of these parameters in order to reproduce the experimental
data. All other parameters of the model describing the structure of energy levels of water
molecules, their vibrational and librational frequencies, etc. are considered as fixed, being
universal for all proteins.
In spite of the model features of our approach, we want to stress that the complex behavior
and the peculiarities in dependencies of the heat capacity on temperature are all very well
reproduced by the developed model with only a few parameters. This was demonstrated
for two proteins and we consider this result as a significant achievement. This fact supports
our conclusion that the developed model can be used for the prediction of new features of
phase transitions in various biomolecular systems. Indeed, from Figs. III A and 3 one can
extract a lot of useful information on the heat capacity profiles: the concave bending of the
heat capacity profile for a completely unfolded protein, the temperature of the cold and heat
denaturation, the absolute values of the heat capacity at the phase transition temperature,
the broadening of heat capacity peaks. Another peculiarity which is well reproduced by
our statistical mechanics model is the decrease of the heat capacity of the folded state of
the protein in comparison with that for unfolded state and asymmetry of the heat capacity
peaks.
22
IV. CONCLUSIONS
We have developed a novel statistical mechanics model for the description of folding↔unfolding
processes in globular proteins obeying simple two-stage-like folding kinetics. The model is
based on the construction of the partition function of the system as a sum over all statisti-
cally significant conformational states of a protein. The partition function of each state is a
product of partition function of a protein in a given conformational state, partition function
of water molecules in pure water and a partition function of H2O molecules interacting with
the protein.
The introduced model includes a number of parameters responsible for certain physical
properties of the system. The parameters were obtained from available experimental data
and three of them (energy difference between two phases, cooperativity of the transition and
the average strength of the protein's electrostatic field) were considered as being variable
depending on a particular protein and pH of the solvent.
We have compared the predictions of the developed model with the results of experimental
measurements of the dependence of the heat capacity on temperature for staphylococcal
nuclease and metmyoglobin. The experimental results were obtained at various pH of solvent.
The suggested model is capable to reproduce well within a single framework a large number
of peculiarities of the heat capacity profile, such as the temperatures of cold and heat
denaturations, the corresponding maximum values of the heat capacities, the temperature
range of the cold and heat denaturation transitions, the difference between heat capacities
of the folded and unfolded states of the protein.
The good agreement of the results of calculations obtained using the developed formalism
with the results of experimental measurements demonstrates that it can be used for the
analysis of thermodynamical properties of many biomolecular systems. Further development
of the model can be focused on its advance and application for the description of the influence
of mutations on protein stability, analysis of assembly and stability of protein complexes,
protein crystallization process, etc.
V. ACKNOWLEDGMENTS
23
We acknowledge support of this work by the NoE EXCELL. We are grateful to Dr. Ilia
Solov'yov for the careful reading of the manuscript and helpful advice.
[1] V. Munoz, Annu. Rev. Biophys. Biomol. Struct. 36, 395 (2007).
[2] K. A. Dill, S. B. Ozkan, M. S. Shell, and T. R. Weikl, Annu. Rev. Biophys. 37, 289 (2008).
[3] J. N. Onuchic and P. G. Wolynes, Curr. Op. Struct. Biol. 14, 70 (2004).
[4] E. Shakhnovich, Chem. Rev. 106, 1559 (2006).
[5] N. V. Prabhu and K. A. Sharp, Chem. Rev 106, 1616 (2006).
[6] A. Yakubovich, I. Solov'yov, A. Solov'yov, and W. Greiner, Eur. Phys. J. D 46, 215 (2007),
arXiv:0704.3079v1 [physics.bio-ph], 23 Apr 2007.
[7] A. Yakubovich, I. Solov'yov, A. Solov'yov, and W. Greiner, Europhys. News 38, 10 (2007).
[8] I. Solov'yov, A. Yakubovich, A. Solov'yov, and W. Greiner, Eur. Phys. J. D 46, 227 (2008),
arXiv:0704.3085v1 [physics.bio-ph], 23 Apr 2007.
[9] A. Yakubovich, I. Solov'yov, A. Solov'yov, and W. Greiner, Eur. Phys. J. D 40, 363 (2006).
[10] A. Yakubovich, I. Solov'yov, A. Solov'yov, and W. Greiner, Eur. Phys. J. D 39, 23 (2006).
[11] A. Yakubovich, I. Solov'yov, A. Solov'yov, and W.Greiner, Khimicheskaya Fizika (Chemical
Physics) (in Russian) 25, 11 (2006).
[12] A. Yakubovich,
I. Solov'yov, A. Solov'yov, and W. Greiner, Eur. Phys. J. D DOI:
10.1140/epjd/e2008-00126-y (2008).
[13] I. Solov'yov, A. Yakubovich, A. Solov'yov, and W. Greiner, J. Exp. Theor. Phys. 103, 463
(2006).
[14] I. Solov'yov, A. Yakubovich, A. Solov'yov, and W. Greiner, Phys. Rev. E 73, 021916 (2006).
[15] I. Solov'yov, A. Yakubovich, A. Solov'yov, and W. Greiner, J. Exp. Theor. Phys. 102, 314
(2006).
[16] A. Yakubovich, A. Solov'yov, and W. Greiner, Int. J. Quant. Chem. 110, 257 (2010).
[17] A. Yakubovich, A. Solov'yov, and W. Greiner, AIP Conf. Proc. 1197, 186 (2009).
[18] B. Noetling and D. A. Agard, Proteins 73, 754 (2008).
[19] S. Kumar, C.-J. Tsai, and R. Nussinov, Biochemistry 41, 5359 (2002).
[20] J. H. Griffith and H. Scheraga, J. Mol. Struc. 682, 97 (2004).
[21] A. Bakk, J. S. Hye, and A. Hansen, BJ 82, 713719 (2002).
[22] Y. Griko, P. Privalov, J. Aturtevant, and S. Venyaminov, Proc. Natl. Acad. Sci. USA 85, 3343
24
(1988).
[23] P. Privalov, J. Chem. Thermodyn. 29, 447 (1997).
[24] S. He and H. A. Scheraga, J. Chem. Phys. 108, 271 (1998).
[25] S. He and H. A. Scheraga, J. Chem. Phys. 108, 287 (1998).
[26] W. Scott and W. van Gunsteren, in Methods and Techniques in Computational Chemistry:
METECC-95, edited by E. Clementi and G. Corongiu (STEF, Cagliari, Italy, 1995) pp. 397 --
434.
[27] W. Cornell, P. Cieplak, C. Bayly, and et al, J. Am. Chem. Soc. 117, 5179 (1995).
[28] A. MacKerell, D. Bashford, R. Bellott, and et al, J. Phys. Chem. B 102, 3586 (1998).
[29] S. Krimm and J. Bandekar, Biopolymers 19, 1 (1980).
[30] M. Cubrovic, O. Obolensky, and A. Solov'yov, Eur. Phys. J. D 51, 41 (2009).
[31] A. Finkelstein and O. Ptitsyn, Protein Physics. A Course of Lectures (Elsevier Books, Oxford,
2002).
[32] G. Makhatadze and P. Privalov, J. Mol. Biol. 232, 639 (1993).
[33] J. Chen, Z. Lu, J. Sakon, and W. Stites, J.Mol.Biol. 303, 125 (2000).
[34] S. Evans and G. Brayer, J.Mol.Biol. 213, 885 (1990).
[35] W. Humphrey, A. Dalke, and K. Schulten, J. Molec. Graphics 14, 33 (1996).
[36] F. A. Cotton, J. Edward E. Hazen, and M. J. Legg, Proc. Natl. Acad. Sci. USA 76, 2551
(1979).
[37] http://www.rcsb.org/(2009).
[38] J. C. Phillips, R. Braun, W. Wang, and et al, J. Comp. Chem. 26, 1781 (2005).
[39] W. Russel, D. Saville, and W. Schowalter, Colloidal Dispersions (Cambridge University Press,
1989).
[40] B. Mallik and T. L. A. Masunov, J. Comp. Chem. 23, 1090 (2002).
[41] H.-X. Zhou, BJ 83, 2981 2986 (2002).
[42] J. P. Collman, R. Boulatov, C. J. Sunderland, and L. Fu, Chem. Rev. 104, 561 (2004).
[43] D. Shortle and M. S. Ackerman, Science 293, 487 (2001).
|
1709.00793 | 1 | 1709 | 2017-09-04T02:51:00 | Cell contraction induces long-ranged stress stiffening in the extracellular matrix | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.QM",
"q-bio.TO"
] | Animal cells in tissues are supported by biopolymer matrices, which typically exhibit highly nonlinear mechanical properties. While the linear elasticity of the matrix can significantly impact cell mechanics and functionality, it remains largely unknown how cells, in turn, affect the nonlinear mechanics of their surrounding matrix. Here we show that living contractile cells are able to generate a massive stiffness gradient in three distinct 3D extracellular matrix model systems: collagen, fibrin, and Matrigel. We decipher this remarkable behavior by introducing Nonlinear Stress Inference Microscopy (NSIM), a novel technique to infer stress fields in a 3D matrix from nonlinear microrheology measurement with optical tweezers. Using NSIM and simulations, we reveal a long-ranged propagation of cell-generated stresses resulting from local filament buckling. This slow decay of stress gives rise to the large spatial extent of the observed cell-induced matrix stiffness gradient, which could form a mechanism for mechanical communication between cells. | physics.bio-ph | physics |
Cell contraction induces long-ranged stress stiffening in theextracellular matrixAuthor: Yu Long Han1†, Pierre Ronceray2†, Guoqiang Xu1, Andrea Malandrino1,3, RogerKamm1,4, Martin Lenz5, Chase P. Broedersz6* and Ming Guo1*Affiliations:1. Department op Mechanical Engineering, Massachusetts Institute op Technology, Cambridge,MA, 02139, USA;2. Princeton Center por Theoretical Science, Princeton University, Princeton, NJ 08544, USA;3. Institute por Bioengineering op Catalonia, Barcelona, 08028, Spain;4. Department op Biological Engineering, Massachusetts Institute op Technology, Cambridge,MA, 02139, USA; 5. Laboratoire de Physique Théorique et Modèles Statistiques (LPTMS), CNRS, UniversitéParis-Sud, Université Paris-Saclay, 91405 Orsay, France;6. Arnold-Sommerpeld-Center por Theoretical Physics and Center por NanoScience, Ludwig-Maximilians-Universität München, D-80333 München, Germany† These authors contributed equally to this work.*Correspondence:Ming GuoMassachusetts Institute op Technology, Cambridge, MA 02139, USA.Phone: +1 (617) 324-0136Email: [email protected] P. BroederszLudwig-Maximilians-Universität München, D-80333 München, GermanyPhone: +49 (0)89 2180-4514Email: [email protected] of Interest statement:The authors declare no conflict of interests.1 AbstractAnimal cells in tissues are supported by biopolymer matrices, which typically exhibit highlynonlinear mechanical properties. While the linear elasticity of the matrix can significantly impactcell mechanics and functionality, it remains largely unknown how cells, in turn, affect thenonlinear mechanics of their surrounding matrix. Here we show that living contractile cells areable to generate a massive stiffness gradient in three distinct 3D extracellular matrix modelsystems: collagen, fibrin, and Matrigel. We decipher this remarkable behavior by introducingNonlinear Stress Inference Microscopy (NSIM), a novel technique to infer stress fields in a 3Dmatrix from nonlinear microrheology measurement with optical tweezers. Using NSIM andsimulations, we reveal a long-ranged propagation of cell-generated stresses resulting from localfilament buckling. This slow decay of stress gives rise to the large spatial extent of the observedcell-induced matrix stiffness gradient, which could form a mechanism for mechanicalcommunication between cells.Introduction Living cells interact mechanically with their three-dimensional (3D) microenvironment. Manybasic cell functions, including migration, proliferation, gene expression, and differentiation,depend on how these forces deform and shape the surrounding soft extracellular matrix (ECM)(1-4). In addition, externally imposed forces on the matrix can impact cell behavior, for instancein beating cardiac cells on a two-dimensional (2D) substrate (5-7). Such external forces may begenerated by other cells and act as mechanical signals (8-10) leading to emergent collective celldynamics (11, 12). Nevertheless, it remains unclear how cell-generated forces propagate throughthe ECM and impact the mechanics of their 3D extracellular environment.The ECM is comprised of several biopolymers (13), such as collagen or fibrin, which arelargely responsible for its mechanical properties. Experiment and theory have shown thatbiopolymer networks exhibit a highly nonlinear mechanical response (14), involving the entropicelasticity of individual filaments, geometric effects due to fiber bending and buckling, and even2 collective network effects governed by critical phenomena (15-21). Recent works have indicatedthat this nonlinear response is highly relevant to cell-ECM interactions (22, 23). Due to a lack ofdirect characterization of local mechanics in a 3D matrix in the vicinity of a cell, however, therole of elastic nonlinearities in mechanical cell-ECM interactions has remained elusive.Ideally, cell-ECM interactions would be analyzed by determining the stress fieldgenerated by the cell. Unfortunately, standard microscopy techniques do not reveal thisinformation in a straightforward and unambiguous way. Some information about internalnetwork forces can be accessed by adding deformable particles (24) or by creating an interface,for example by laser ablation, and observing the resulting deformation of the system (25, 26).However, obtaining internal stresses with such invasive and destructive approaches requiresadditional assumptions about the network's local mechanical properties. The same is true ofapproaches that infer stresses from a combination of microscopy imaging and finite elementmodelling (23, 27, 28). The intrinsic heterogeneity (29-31) and a highly nonlinear mechanicalresponse (14, 32) of extracellular networks pose a daunting challenge to these techniques (33). To investigate how living cells mechanically modify their microenvironment, we usemicrorheology with optical tweezers to directly measure the local nonlinear elastic properties ina 3D ECM network. We observe that remarkably far-reaching stiffening gradients are generatedtowards the cell in a variety of biopolymer matrices. To investigate this, we introduce a novelmodel-independent measurement technique termed Nonlinear Stress Inference Microscopy(NSIM), enabling us to determine the stress in a region around the cell and study stresspropagation inside a 3D ECM. We use a combination of theory and simulations to demonstratethe ability of NSIM to accurately measure 3D local stress with high spatial resolution. UsingNSIM, we show that the observed long-range stiffness gradient around cells results fromremarkably long-ranged stresses, which are capable of exciting the matrix's nonlinear responseover distances exceeding the size of the cell. Our results demonstrate that cells strongly modifythe mechanics of the surrounding ECM, which could be crucial in shaping matrix-mediated cell-cell interactions. 3 Cells strongly stiffen their surrounding extracellular matrix by actively contractingTo study the mechanical interactions between cells and their surrounding ECM, we cultureMDA-MB-231 cells in a 1.5 mg/mL reconstituted 3D collagen network. The network is infusedwith 4.5 μm diameter latex beads, large enough to prevent slippage through the mesh. The cellsspread and start contracting the surrounding network within 4 hours (Fig. 1a). We probe the localmicromechanics of the matrix using optical tweezers to pull these beads away from the cell at aconstant speed of 1 μm/s (Figs. 1b-d and Supplementary Fig. 1). This low speed ensures that theviscous drag on the bead from the background fluid is negligible compared to the network'srestoring force, and at this speed the mechanical response of the matrix is rate-independent, fullyreversible and therefore predominately elastic (Supplementary Fig. 2). Thus, this protocolenables us to obtain the local force-displacement relationship F(x) that characterizes themicromechanics of the matrix.By probing a bead located far from the cell (>200 μm), we determine the intrinsicresponse of the collagen matrix. The resulting force-displacement relationship is shown by theblack line in Fig. 1e. The nonlinear differential stiffness knl(F)=dF/dx, defined as the slope of thisforce displacement curve, increases with applied force F, revealing a strong force-stiffeningbehavior. This is reminiscent of the well-characterized stress-stiffening behavior measured atlarge scales using macrorheology on collagen gels (14, 32). Interestingly, the matrix becomes substantially stiffer closer to the cell (Fig. 1e). Indeed,the local linear stiffness klin of the matrix, defined as the small force limit of knl(F), is two ordersof magnitude larger near the cell than at a remote location (Fig. 1f, red squares). This observationin the bulk of the network is consistent with prior 2D experiments showing cell-inducedstiffening of the surface of a collagen matrix with a cell migrating on top (22), as well as withsimulations (23).This dramatic stiffness gradient in the vicinity of the cell originates from the active forcesit exerts. Indeed, inhibiting cell contractility using 2 μM Cytochalasin D results in a strongattenuation of the cell-induced stiffening (Fig. 1f, blue circles). The weak residual gradient iswell explained by increased ECM density near the cell, under the assumption that the matrixrigidity scales as the square of the collagen concentration (c) (34) (Fig. 1f, gray diamonds).4 Indeed, by estimating c using confocal reflection microscopy, we determine that the enhancedmatrix density near the cell can account for a stiffening of up to a factor ~3 (Fig. 1f, graydiamonds). Finally, we rule out the effect of the passive rigidity of the cell on the matrixstiffness, which is very short-ranged in 3D (31). Thus, active forces exerted by the cell result inan extended stiffened region in the 3D matrix, reflecting the presence of a stress field decayingaway from the cell with stress values sufficiently large to excite the nonlinear response of thecollagen network, as illustrated in Fig. 1d.Nonlinear Stress Inference Microscopy To study the cell-induced stress fields, we use the network's nonlinear microrheological responseto our advantage and infer local stress values from our stiffness measurements. The nonlinearstiffening evidenced in Fig. 1e originates from two contributions: the force F exerted by theoptical tweezers acting on the bead, and the local stress σloc induced by the cell. This similarinfluence of force and stress suggests that we may be able to extract σloc at a specific distancefrom the cell by comparing the corresponding force-displacement relationship to the remotemeasurement at which σloc is negligible. This comparison is confounded, however, because offorce and stress being fundamentally different quantities: beyond having different dimensions,force transforms as an axial vector under spatial symmetry operations, while stress is a rank twotensor. This has an essential implication for the difference in the nonlinear response due to aforce as opposed to a stress: The local stiffness should be invariant under reversal of the forcevector, F to –F, while reversal of the stress tensor σ to – σ exchanges compression and tension,which can have a qualitatively different effect on the nonlinear mechanical response. Despite these differences, here we show that a correspondence between force and stresscontrolled stiffening can be established in the strongly nonlinear regime. First consider a simple1D system of nonlinear springs representing the network surrounding a bead in a geometry withfixed network stress σ (Fig. 2a) and one with fixed tweezer force F (Fig. 2b). For nonlinearsprings that stiffen under tension and soften under compression --- a generic characteristic ofbiopolymers (14, 18)--- we find that the functional form of the stiffness curves actually becomesimilar at large σloc and F, despite being qualitatively different in the weakly nonlinear regime.Indeed, the tensed spring in Fig. 2b dominates the differential stiffness experienced by the beadin the strongly nonlinear regime, rendering this case similar to the stress-controlled geometry,5 where the mechanical response is equally shared by two similarly tensed bonds (Fig. 2a). Thisquantitative similarity between the klin vs. σloc and knl vs. F curves in the strongly nonlinearregime enables us to use the latter, which we measure by nonlinear microrheology, as a"dictionary" to infer local stresses. This intuitive similarity between the force and stress controlled geometries in thenonlinear regime becomes mathematically exact when the springs' differential stiffness has apower-law dependence on tension, as widely observed for biopolymer networks (32, 35) (SIsection 4). Specifically, from a measurement of klin in a network with an unknown local stressσloc, we can obtain an effective force Feff defined such that knl(Feff) = klin(σloc), and this effectiveforce is directly proportional to the local stressσloc ≈ a Feff(1)provided large local stresses, such that klin ≫ k0, where k0 is the linear stiffness of the unstressednetwork. We determine the proportionality factor a by assuming that nonlinearity sets in at asimilar stress σ* at a macro and microscopic level. In practice, we adjust a to match the low- andhigh-stress asymptotes, in a log-log plot, of the macroscopic differential shear modulus K(σmacro)to those of the microrheology curve knl(F) (Figs. 2c-e, Supplementary Figs. 5 and 8). Togetherwith Eq. (1), this provides a procedure to infer stresses from nonlinear microrheology, which weterm Nonlinear Stress Inference Microscopy (NSIM). To demonstrate the validity and accuracy of NSIM, we perform simulations of theexperimental scenario in Fig. 1. We embed a contractile cell in a disordered 3D network of fiberswith power-law stiffening (SI Section 2). We model the cell as a rigid ellipsoidal body thatcontracts along its long axis, inducing strong stiffening in an extended conic region as depictedin Fig. 3a. We then simulate a microrheology experiment by applying a force on a selection ofnetwork nodes to obtain a local force-displacement curves at various distances r along thecontraction direction of the cell (Fig. 3b). From this, we determine the linear stiffness klin of thenetwork as a function of r (Fig. 3c), which exhibits a dependence similar to the experimentalmeasurements shown in Fig. 1f. We further confirm that this dependence vanishes in thedirection perpendicular to contraction and in the absence of an active contractile force, as inexperiments (Fig. 1f). We infer the local stress field from these linear stiffnesses using NSIM, as6 shown in Fig. 3d. We find excellent agreement with the "true" local stress in the strong stiffeningregime even when mechanical disorder gives rise to fluctuations in the stress field (Fig. 3e and SISection 3), thereby validating NSIM as a quantitative method to capture the spatial stressdistribution around a contractile cell in a disordered 3D fiber matrix (Figs. 3d-f). Long-ranged stress propagation leads to extended stiffness gradients around cellsTo unravel the mechanical origins of the far-reaching cell-induced stiffness gradient in collagen(Figs. 1,4a), we use NSIM to experimentally infer the local stresses σloc(r) around a cell inside thematrix. The inferred stress decays consistently with a power-law σloc ~ r2‒ over more than twodecades (Fig. 4b), in contrast with the power law σloc ~ r3‒ expected in a linear material (36). Thismeasured slow stress decay is consistent with both our simulations (Fig. 3f) and previoustheoretical predictions for nonlinear fiber networks (37-39). Deviations from linear elasticityhave previously been reported in experiments for the deformation fields (23, 27), but theimplications for the stress-field have remained unclear (40). Here, we establish a directexperimental measurement of the long-ranged transmission of cell generated stresses in fibermatrices. Conceptually, this increased range of stresses in fibrous materials found in simulationsresults from their asymmetric response to tension and compression: Fibers stiffen under tensionand soften due to buckling under compression (18, 41). Simply speaking, the matrix around astrong contractile cell effectively behaves as a network of ropes, where only tensile forces aretransmitted, unimpeded by orthoradial compressive resistance. Hence the total contractile forceexerted by the cell is conserved with distance, and the decay of radial stress simply reflects thisforce spreading over an increasing surface area (37). This buckling-based mechanism issupported by observations with confocal reflection microscopy of a larger amount of highlycurved collagen filaments in the vicinity of a contractile cell, as compared to the case wherecontraction is inhibited with Cytochalasin D (Fig. 4e-f). To explore the generality of our observations in collagen, we perform the same nonlinearmicrorheological experiments with MDA-MB-231 cells in a 2.5 mg/mL Matrigel (Fig. 4a, greencircles), a blend of biopolymers more complex than pure collagen (42), and for human umbilicalvein endothelial cells (HUVEC) in a 3.0 mg/mL fibrin gel (light blue triangles in Fig. 4a). In7 both cases, we find that cells are capable of generating large extended stiffness gradients alongthe cell's contraction direction (Fig. 4a). Using NSIM, we find that the slow stress decayconsistent with rope-like force transmission (σloc ~ r2‒) is observed in all three cases, despitesignificant variability in the absolute magnitude of the stresses (solid lines in Fig. 4b). Theseresults highlight the wide applicability of NSIM, and demonstrate the generality of long-rangedstress fields generated by the cell. This long-ranged stress transmission is intimately tied to theability of the cell to enhance the linear stiffness klin over an extended region of the ECM (Figs. 1f,4a): Slowly decaying stresses induced by the cell remain large enough to excite the nonlinearelastic response of the matrix over a distance exceeding the cell size. Cells not only actively modify the linear stiffness of their 3D matrix environment (Fig.4a), but also the nonlinear mechanical response. To reveal how cell stress and probe forcescombine to stiffen the surrounding network, we measure the full nonlinear microrheologicalresponse of the network in the vicinity of the cell in all three types of ECM model systems, asshown in Fig. 4c. Indeed, the nonlinear knl vs. F curves measured at different distances from thecell are clearly separated, indicating a significant contribution of cell generated stress on thenonlinear mechanical response. This contribution could be through nonlinear network elasticstiffening or network plastic deformation. We note that our stress inference is largelyindependent of the specifics of the ECM's nonlinear response, but does assume a predominantlyelastic response to the forces generated by the cell. Indeed, significant plastic deformations couldimply that the ECM's nonlinear response is systematically modified as a function of the distancefrom the cell. In the absence of plastic deformations, we expect that further stiffening aprestressed matrix by a large tweezer force would result in a nonlinear response that isfunctionally similar for all levels of cell stress. To test this, we plot all nonlinear stiffening curvesas a function of the combined force F+Feff, where the effective force Feff ∝ σloc is determined as inFigs 2a-b (Supplementary Table 1). Remarkably, we find that the data taken at different distancesto the cell collapse in any network composition onto a smooth master curve (Fig. 4d). The largecell-generated stress thus locally drives the ECM into an elastic nonlinear regime, which can befurther extended by the probe force we apply with optical tweezers. Cell contraction induces a local stiffening of the surrounding ECM by several orders ofmagnitude with a long-range decay over tens of micrometers. Here we infer the stresses8 responsible for this stiffening using NSIM, a conceptually new inference technique that does notrequire the materials' constitutive relationship of the stress field in terms of the strain field nor areference undeformed state. Due to its simplicity and insensitivity to the detailed material'sproperties, NSIM could be used in various conditions, including embryo or tumor development.The stresses inferred using this technique far from the cell are consistent with priormeasurements (27). Close to the cell, high stiffening renders the technique most accurate, andcorresponds to stresses of the order of 200 Pa, significantly larger than previously reported (27).These cell-induced stresses decay much more slowly than in a linear continuum material due tobuckling in fiber networks, resulting in far-reaching stiffness gradients as high as 50 kPa/mmover several cell diameters. Other cells in the surrounding matrix could sense and respond tosuch large gradients, suggesting that cell-induced ECM stiffening could mediate inter-cellmechanical communication and collective durotaxis. These observations highlight the criticalrole of nonlinear matrix mechanics not only in shaping cell-ECM interactions (8, 43), but alsofor matrix-mediated interactions between cells.MethodsCell culture and matrix preparation. Cells are maintained under 37 oC, 5% CO2 and 95%humidity. MDA-MB-231 cells were cultured in DMEM with 10% FBS, 1% penicillin andstreptomycin. Human Umbilical Vein Endothelial Cells (HUVEC) (Lonza) were cultured oncollagen I-coated flasks in EGM-2 growth medium (Lonza) and used between passages 6–8. Toprepare the collagen gel, 800 μL type I bovine collagen solution (3.0 mg/mL, PureCol, AdvancedBioMatrix) was mixed with 100 µL PBS (10X). We adjusted the solution to pH 7.2 with ~70 µL0.1 M NaOH. The collagen solution is then mixed with PBS (1X) to reach a final collagenconcentration of 1.5 mg/mL, and polymerized in the cell culture incubator for 30 min. To preparethe fibrin gel, fibrinogen from bovine plasma (F8630, Sigma) was dissolved in PBS at 6 mg/mL.Thrombin (T4648, Sigma) was dissolved at 2U/mL in PBS (for experiments without cells) or inEGM-2 (for experiments with cells). Then we mixed thrombin and fibrinogen at 1:1 volume ratioand polymerized it in the cell culture incubator for 15 min. For Matrigel preparation, thebasement membrane matrix (10 mg/mL, Corning) was diluted to 2.5 mg/mL with DMEM andpolymerized in the cell culture incubator for 30 min. For all cell-loaded gels, cell and beadsuspensions were added to the gel solution before polymerization, with a cell density around9 104/mL, and all measurements were conducted 12 hours after polymerization. To inhibitcontractility of MDA-MB-231 cells, we disrupted filamentous actin structures using 2-μMCytochalasin D (PHZ1063, Invitrogen) for 30 min. Optical tweezer measurements. We used a Thorlabs optical tweezers system to perform allmeasurements. Briefly, to optically trap a bead (4.5-μm carboxylate microspheres, Polyscience)that is embedded and confined in a 3D biopolymer network, the laser beam (5W, 1064 nm) istightly focused through a series of Keplerian beam expanders and a high numerical apertureobjective (100 x 1.4, oil, Leica). A high-resolution quadrant detector was used for positiondetection. The linear region of the detector and the trap stiffness (0.04 pN/nm) were calibratedwith the same bead in pure cell culture media by using an active power-spectrum method andequipartition theorem (44). To manipulate the trapped bead, a high-resolution piezo stage (P-545,PI nanoTM) was moved at a constant velocity of 1 μm/s, and the relative distance between laserand bead) was recorded, from which local force-displacement curves inside the matrix weredetermined (45) (see Supplementary Information for details). Bulk rheology. We performed bulk rheology measurements on a DHR-3 rheometer (TAInstruments) using a plate-plate geometry, with a 40-mm glass disk as the top plate and a 60-mmPetri dish as the bottom plate with a gap of 500 μm. All gels were formed in the gap at 37 oC andwere sealed by mineral oil to avoid evaporation. The polymerization process was monitored bystrain oscillations with a strain amplitude of 0.005 at a frequency of 1 rad/s. Afterpolymerization, a strain ramp was applied to the gel at a rate of 0.01/s, and the resulting stresseswere measured. Theoretical modelling and simulations. Numerical simulations presented in Fig. 3 areperformed using a model of nonlinear springs (force-extension relation p(x) = exp(µx) – 1), withregular removal of springs to introduce disorder in the network, while ensuring a fiber length Lf=10, in a spherical system of radius R=25.5. The contractile cell is a rigid ellipsoidal body of size14.2 × 2.8 × 2.8, with force and torque balance, contracted by 50% along its long axis. Thesurrounding network is flexibly clamped at the surface of the cell and at the boundary of thesystem. Mechanical equilibrium is attained by minimization of the energy using the BFGSalgorithm. Further details are provided in SI Sections 2 and 3.10 Imaging of collagen networks and image analysis. The 3D collagen networks near acontracting cell were imaged with confocal reflection microscopy using a 63x 1.2NA waterobjective (Leica SP8). To determine the boundary of the cell, the cytoplasm was stained with celltracker green (C7025, Thermofisher) and imaged at the same time under confocal microscope.To capture the fiber buckling process, we imaged the cell and its surrounding 3D fiber networksat a 5 min interval for 4 hr at 37 oC and with 5% CO2. To analyze the curvature of single collagenfiber, we manually selected 20 points on each individual collagen fiber; the fiber outline wasdetermined by cubic spline interpolation, from which the average curvature of the fiber wascalculated. Acknowledgements The authors thank Anna Posfai for useful comments. This work was supported by the NationalCancer Institute grant number 1U01CA202123 (to M.G.), the German Excellence Initiative viathe program 'NanoSystems Initiative Munich' (NIM) (to C.P.B.) and the DeutscheForschungsgemeinschaft (DFG) via project B12 within the SFB-1032 (to C.P.B.), a PCTSfellowship (to P.R.), and a MISTI-Germany seed fund (M.G. and C.P.B.). M. G. would also liketo acknowledge the support from the Department of Mechanical Engineering at MIT. A.M. issupported by EU's Seventh Framework Programme for Research (FP7, 625500). This work wasperformed in part at the Aspen Center for Physics, which is supported by National ScienceFoundation grant PHY-1607611. 11 FiguresFigure 1. Far-reaching stiffness gradient of extracellular matrix caused by a single contracting cellin a three-dimensional collagen network. a, Image of a MDA-MB-231 cell (blue) in a 3D collagennetwork (green). Scale bar, 10 µm. b-d, Schematics illustrating the force-displacement measurement withlaser tweezers and the relation between matrix stiffening (blue potential wells) and the cell-generatedstress field in the cell contraction direction. e, Local force-displacement curves, showing the localnonlinear stiffening response in the collagen network. Different colors represent measurements at variousdistances from the cell along the contraction direction. f, Quantification of the linear stiffness klin of thelocal 3D matrix as a function of distance to the cell r. Red and yellow symbols represent measurementsalong and perpendicular to the main contraction direction, respectively. Blue symbols are measured withcell contraction inhibited by Cytochalasin D treatment. Gray diamonds indicate the stiffness expectedsolely from the increased collagen concentration c. Here, "remote" stands for the locations that are faraway from the cell (>200 μm), where the matrix's response is not affected by cell contraction. Error barsrepresent standard deviation (n=15). 12 Figure 2. Nonlinear elastic responses can be used to infer cell-induced local stresses. a, 1D system ofnonlinear springs in a stress-controlled geometry with local stress σloc. b, force controlled geometry withforce F applied to the central bead, together with an expansion of stiffness dictated by symmetryproperties of the two scenarios and a schematic of the nonlinear response. The linear stiffness, klin, of thesystem in a can be measured by applying a small perturbation to the central bead, while the nonlinearstiffness, knl, is defined as the derivative of the force-displacement curve of the central bead in b. Thesprings represent the surrounding network. c, Schematics of linear microrheological stiffness as a functionof the local stress in the stress-controlled geometry on a logarithmic scale. d, Nonlinear microrheologicalstiffness for the force-controlled geometry. e, Differential shear modulus, K, as a function of applied shearstress σmacro as in a macroscopic rheology experiment. Our inference technique exploits a correspondencebetween the stress-controlled and force controlled geometries in the strongly nonlinear regime. 13 Figure 3. 3D Simulations of cell-generated stress fields inducing nonlinear network response andvalidations of NSIM. a, Simulated rigid ellipsoidal cell contracting inside a 3D nonlinear fiber network(in green). The linear stiffness klin is depicted by the spheres in a green-white logarithmic color gradient.b, Local force-displacement curves, showing the local nonlinear stiffening response in the simulatednetwork. Different colors represent measurements at various distances from the cell along the contractiondirection. c, Local linear stiffness klin of the 3D matrix as a function of distance to the cell r fromsimulations. Red and yellow symbols represent data parallel and perpendicular to the main contractiondirection, respectively. Blue symbols correspond to a non-contracting rigid cell. d, The inferred stressdepicted by spheres in a red-white logarithmic color gradient in the same simulation as in a. Absent pointsalong the direction perpendicular to the cell's contraction axis correspond to soft compressed regionswhere the local stiffness is smaller than k0, precluding the use of NSIM. e, Inferred stresses fromsimulated data in b using NSIM versus direct numerically determined stress, demonstrating that NSIMallows to correctly infer stresses within a factor of 2 in the nonlinear regime. f, Local stress obtained fromNSIM as a function of distance.14 Figure 4. Nonlinear matrix stiffening and long-ranged stress propagation in various 3D biopolymernetworks. a, Local linear stiffness klin is plotted against the distance to the cell r along its principalcontraction direction in collagen (red square), fibrin (blue triangle) and Matrigel (green circle). All threedifferent ECM model systems exhibit a strong cell induced stiffening gradient. b, The stress field σgenerated by the cell determined using NSIM is shown as a function of distance to the cell r, and thedashed line indicates a slope of -2. c, Local nonlinear differential stiffness knl is plotted against the appliedprobe force F for all three ECM model systems. d, Collapse of the date from panel c onto a master curvein each respective matrix obtained by plotting knl as a function of combined local force F+Feff, where theFeff is determined using NSIM. e, Time-lapse imaging shows the buckling process of a single fiber arounda contracting cell. The fiber undergoing buckling is highlighted in yellow. Scale bar, 2 µm. f, Fibercurvature distributions (bottom) and the cumulative probability (top) near the cell, within a 60 µmdistance along the principle cell contraction direction, before and after Cytochalasin D treatment. Errorbars in a and b represent standard deviation (n=15).(46-50)15 References1.Discher DE, Janmey P, & Wang Y-l (2005) Tissue cells feel and respond to the stiffnessof their substrate. Science 310(5751):1139-1143.2.Discher DE, Mooney DJ, & Zandstra PW (2009) Growth factors, matrices, and forcescombine and control stem cells. Science 324(5935):1673-1677.3.Mammoto A & Ingber DE (2009) Cytoskeletal control of growth and cell fate switching.Curr. Opin. Cell Biol. 21(6):864-870.4.Angelini TE, Hannezo E, Trepat X, Fredberg JJ, & Weitz DA (2010) Cell migrationdriven by cooperative substrate deformation patterns. Phys. Rev. Lett. 104(16):168104.5.De R, Zemel A, & Safran SA (2007) Dynamics of cell orientation. Nat. Phys. 3(9):655-659.6.Rehfeldt F & Discher DE (2007) Biophysics: Cell dipoles feel their way. Nat. Phys.3(9):592-593.7.Nitsan I, Drori S, Lewis YE, Cohen S, & Tzlil S (2016) Mechanical communication incardiac cell synchronized beating. Nat. Phys. 12(5):472-477.8.Ahmadzadeh H, Webster MR, Behera R, Jimenez Valencia AM, Wirtz D, Weeraratna AT,& Shenoy VB (2017) Modeling the two-way feedback between contractility and matrixrealignment reveals a nonlinear mode of cancer cell invasion. PNAS 114(9):E1617-E1626.9.Provenzano PP, Eliceiri KW, Campbell JM, Inman DR, White JG, & Keely PJ (2006)Collagen reorganization at the tumor-stromal interface facilitates local invasion. BMCmedicine 4(1):38.10.Bischofs IB & Schwarz US (2003) Cell organization in soft media due to activemechanosensing. PNAS 100(16):9274-9279.11.Trepat X, Wasserman MR, Angelini TE, Millet E, Weitz DA, Butler JP, & Fredberg JJ(2009) Physical forces during collective cell migration. Nat. Phys. 5(6):426-430.12.Tambe DT, Hardin CC, Angelini TE, Rajendran K, Park CY, Serra-Picamal X, Zhou EH,Zaman MH, Butler JP, & Weitz DA (2011) Collective cell guidance by cooperativeintercellular forces. Nat. Mater. 10(6):469-475.13.Griffith LG & Swartz MA (2006) Capturing complex 3D tissue physiology in vitro. Nat.Rev. Mol. Cell Biol. 7(3):211-224.14.Storm C, Pastore JJ, MacKintosh FC, Lubensky TC, & Janmey PA (2005) Nonlinearelasticity in biological gels. Nature 435(7039):191-194.15.Lieleg O, Claessens MM, Heussinger C, Frey E, & Bausch AR (2007) Mechanics ofbundled semiflexible polymer networks. Phys. Rev. Lett. 99(8):088102.16.Gardel M, Shin J, MacKintosh F, Mahadevan L, Matsudaira P, & Weitz D (2004) Elasticbehavior of cross-linked and bundled actin networks. Science 304(5675):1301-1305.17.Broedersz C, Sheinman M, & MacKintosh F (2012) Filament-length-controlled elasticityin 3D fiber networks. Phys. Rev. Lett. 108(7):078102.18.Broedersz CP & MacKintosh FC (2014) Modeling semiflexible polymer networks. Rev.Mod. Phys. 86(3):995.19.Onck P, Koeman T, Van Dillen T, & van der Giessen E (2005) Alternative explanation ofstiffening in cross-linked semiflexible networks. Phys. Rev. Lett. 95(17):178102.20.Wyart M, Liang H, Kabla A, & Mahadevan L (2008) Elasticity of floppy and stiff randomnetworks. Phys. Rev. Lett. 101(21):215501.16 21.Sharma A, Licup A, Jansen K, Rens R, Sheinman M, Koenderink G, & MacKintosh F(2016) Strain-controlled criticality governs the nonlinear mechanics of fibre networks.Nat. Phys. 12(6):584-587.22.van Helvert S & Friedl P (2016) Strain stiffening of fibrillar collagen during individualand collective cell migration identified by AFM nanoindentation. ACS Appl. Mat.Interpaces 8(34):21946-21955.23.Hall MS, Alisafaei F, Ban E, Feng X, Hui C-Y, Shenoy VB, & Wu M (2016) Fibrousnonlinear elasticity enables positive mechanical feedback between cells and ECMs.PNAS 113(49):14043-14048.24.Campas O, Mammoto T, Hasso S, Sperling RA, O'Connell D, Bischof AG, Maas R,Weitz DA, Mahadevan L, & Ingber DE (2014) Quantifying cell-generated mechanicalforces within living embryonic tissues. Nat. Meth. 11(2):183-189.25.Saha A, Nishikawa M, Behrndt M, Heisenberg C-P, Jülicher F, & Grill Stephan W (2016)Determining physical properties of the cell cortex. Biophys. J . 110(6):1421-1429.26.Nia HT, Liu H, Seano G, Datta M, Jones D, Rahbari N, Incio J, Chauhan VP, Jung K, &Martin JD (2016) Solid stress and elastic energy as measures of tumourmechanopathology. Nat. Biomed. Eng. 1:0004.27.Steinwachs J, Metzner C, Skodzek K, Lang N, Thievessen I, Mark C, Munster S, AifantisKE, & Fabry B (2016) Three-dimensional force microscopy of cells in biopolymernetworks. Nat. Meth. 13(2):171-176.28.Legant WR, Miller JS, Blakely BL, Cohen DM, Genin GM, & Chen CS (2010)Measurement of mechanical tractions exerted by cells in three-dimensional matrices. Nat.Meth. 7(12):969-971.29.Doyle AD, Carvajal N, Jin A, Matsumoto K, & Yamada KM (2015) Local 3D matrixmicroenvironment regulates cell migration through spatiotemporal dynamics ofcontractility-dependent adhesions. Nat. Commun. 6:8720.30.Jones CAR, Cibula M, Feng J, Krnacik EA, McIntyre DH, Levine H, & Sun B (2015)Micromechanics of cellularized biopolymer networks. PNAS 112(37):E5117-E5122.31.Beroz F, Jawerth LM, Münster S, Weitz DA, Broedersz CP, & Wingreen NS (2017)Physical limits to biomechanical sensing in disordered fibre networks. Nat. Commun.8:16096.32.Licup AJ, Münster S, Sharma A, Sheinman M, Jawerth LM, Fabry B, Weitz DA, &MacKintosh FC (2015) Stress controls the mechanics of collagen networks. PNAS112(31):9573-9578.33.Stout DA, Bar-Kochba E, Estrada JB, Toyjanova J, Kesari H, Reichner JS, & Franck C(2016) Mean deformation metrics for quantifying 3D cell–matrix interactions withoutrequiring information about matrix material properties. PNAS 113(11):2898-2903.34.MacKintosh F, Käs J, & Janmey P (1995) Elasticity of semiflexible biopolymer networks.Phys. Rev. Lett. 75(24):4425.35.Gardel M, Nakamura F, Hartwig J, Crocker J, Stossel T, & Weitz D (2006) Prestressed F-actin networks cross-linked by hinged filamins replicate mechanical properties of cells.PNAS 103(6):1762-1767.36.Landau LD & Lifshitz EM (1987) Course of Theoretical Physics. Course op TheoreticalPhysics, (Butterworth-Heinemann, Oxford).37.Ronceray P, Broedersz CP, & Lenz M (2016) Fiber networks amplify active stress. PNAS113(11):2827-2832.17 38.Wang H, Abhilash AS, Chen Christopher S, Wells Rebecca G, & Shenoy Vivek B (2014)Long-range force transmission in fibrous matrices enabled by tension-driven alignment offibers. Biophys. J . 107(11):2592-2603.39.Rosakis P, Notbohm J, & Ravichandran G (2015) A model for compression-weakeningmaterials and the elastic fields due to contractile cells. J. Mech. Phys. Solids 85:16-32.40.Polacheck WJ & Chen CS (2016) Measuring cell-generated forces: a guide to theavailable tools. Nat. Methods 13(5):415.41.Van Oosten AS, Vahabi M, Licup AJ, Sharma A, Galie PA, MacKintosh FC, & JanmeyPA (2016) Uncoupling shear and uniaxial elastic moduli of semiflexible biopolymernetworks: compression-softening and stretch-stiffening. Sci. Rep. 6:19270.42.Kleinman HK & Martin GR (2005) Matrigel: basement membrane matrix with biologicalactivity. Semin. Cancer Biol. 15(5):378-386.43.Winer JP, Oake S, & Janmey PA (2009) Non-linear elasticity of extracellular matricesenables contractile cells to communicate local position and orientation. PLOS ONE4(7):e6382.44.Jun Y, Tripathy Suvranta K, Narayanareddy Babu RJ, Mattson-Hoss Michelle K, & GrossSteven P (2014) Calibration of optical tweezers for in vivo force measurements: how dodifferent approaches compare? Biophys. J . 107(6):1474-1484.45.Hu J, Jafari S, Han Y, Grodzinsky AJ, Cai S, & Guo M (2017) Size- and speed-dependentmechanical behavior in living mammalian cytoplasm. PNAS.46.Ronceray P & Lenz M (2015) Connecting local active forces to macroscopic stress inelastic media. Sopt matter 11(8):1597-1605.47.Mark Galassi JD, James Theiler, Brian Gough, Gerard Jungman, Patrick Alken, MichaelBooth, Fabrice Rossi (2009) GNU Scientipic Library Reperence Manual (Network TheoryLtd).48.Jones E, Oliphant T, & Peterson P (2014) SciPy: open source scientific tools for Python.49.Hunter JD (2007) Matplotlib: A 2D graphics environment. Computing In Science &Engineering 9(3):90-95.50.Ramachandran P & Varoquaux G (2011) Mayavi: 3D visualization of scientific data.Computing in Science & Engineering 13(2):40-51.18 Supplementary Information for Cell contraction induces long-ranged stress stiffening in the extracellular matrixBy Yu Long Han, Pierre Ronceray, Guoqiang Xu, Andrea Malandrino, Roger Kamm, Martin Lenz, Chase P. Broedersz and Ming Guo 1 Experimental methods and supplementary experimental results. 1.1 Data analysis for active microrheology and measurements of collagen micromechanics To manipulate a trapped bead, a high-resolution piezo stage is displaced at a constant velocity, vstage = 1 µm/s. The distance between the laser and the bead is recorded using a quadrant photodiode detector (QPD) as a voltage signal, as shown by the black curve in Supplementary Figure 1a. This data is fitted with a quadratic function, V(t)= at2+bt (red curve), which is used in the following analysis to calculate the differential stiffness, knl. The voltage from the QPD is converted into the distance between the laser and the bead, Dlaser, through a proportionality factor β, i.e. Dlaser(t)= βV(t), as shown by the black curve in Supplementary Figure 1b. The bead displacement relative to the stage is then calculated using x(t)= vstage*t- Dlaser(t), as depicted by the red curve in Supplementary Figure 1b. The force on the bead is proportional to the distance between the laser and the bead, F(t) = k*Dlaser(t), where k is the trap stiffness. The force-displacement relationship for a bead inside a collagen gel is then obtained as shown in Supplementary Figure 1c. The slope of this curve, which represents the effective stiffness of the gel, increases significantly as the applied force increases, suggesting a strong nonlinear stiffening effect in the collagen gel at the microscale. Indeed, the stiffness of the gel increases by two orders of magnitude as the applied force is increased, as indicated by the relationship between differential stiffness, knl, and applied force shown in Supplementary Figure 1d. 1 Supplementary Figure 1: Data analysis for active microrheology. a, Voltage reading from the quadrant photodiode detector (QPD) during mechanical testing, which reflects the relative position between laser trap and particle. The black curve shows the raw data, and the red curve shows the quadratic fit. b, Position information of laser, bead, and stage. Stage displacement stands for the displacement that the stage moves. c, The force-displacement curve obtained in 1.5 mg/mL collagen I sample far away from the cell. d, Force-differential stiffness curve obtained in 1.5 mg/mL collagen I sample far away from the cell. 1.2 Rate dependence and reversibility of micromechanical response To investigate whether the mechanical response of the collagen matrix is rate dependent, we drag a 4.5-µm-diameter bead using optical tweezers to measure the force-displacement relationship at three different locations in a 1.5 mg/mL collagen I gel; at each location, we drag the bead three times with three different speeds, 0.5, 1, 1.5 µm/s, respectively, to measure the collagen micromechanics at different effective strain rates. We find that the resultant force-displacement relationships obtained at varying speeds are very close to each other (Supplementary Figure 2a), suggesting that the micromechanics of the collagen matrix is rate-independent within the range of loading speeds in our study. Furthermore, we have also confirmed that the nonlinear force-displacement relationship measured in the collagen sample is fully reversible with negligible plastic effects, as similar force-displacement curves are obtained from repeated cyclic loading using the same bead, as shown in Supplementary Figure 2b. 2 Supplementary Figure 2: Micromechanics of collagen matrix measured with optical tweezers. a, Force-displacement relationship measured using optical tweezers at different loading rates at 3 different locations in a 1.5 mg/mL collagen gel. At each location, 3 different speeds are tested: 0.5 µm/s (green), 1.0 µm/s (red), and 1.5 µm/s (blue). This result indicates that the mechanical response of the local matrix is largely rate-independent. b, Force-displacement relationship of five subsequent loading cycles at the same location in a 1.5 mg/mL collagen gel. This result indicates that the collagen matrix is predominately elastic in our study. To check the uncertainty and local variability of our measurements, 10 independent measurements in 1.5 mg/mL collagen gel far from a cell are performed, as shown in Supplementary Figure 3a. Interestingly, while the force-displacement curves can differ significantly from location to location, the resultant relationship between the differential stiffness and the applied force are similar, especially in the nonlinear regime, as shown in Supplementary Figure 3b. Supplementary Figure 3: Optical tweezers microrheology measurements in collagen samples. a, Ten independently measured force-displacement curves. b, Corresponding differential stiffness-force (knl vs. F) curves. The result indicates that although the collagen network is heterogeneous in linear stiffness, the nonlinear mechanics share a similar trend and collapse together. 3 Similar behavior is also observed in the cytochalasin D treated sample. We measure the nonlinear micromechanics at different distances to the cell in a 1.5 mg/mL collagen gel, with cell contractility inhibited by treatment of cytochalasin D, and plot the stiffness-force (knl vs. F) curves. We find that although the linear stiffness klin slightly increases closer to the cell, consistent with the observed increase in collagen density (Fig. 1f), the nonlinear mechanics at all distances approximately overlap for high forces, as shown in Supplementary Figure 4. This result indicates that the matrix density does not significantly affect the nonlinear mechanical properties of the collagen gel, consistent with previous work [32]. Supplementary Figure 4: Local nonlinear differential stiffness knl is plotted against the applied probe force F in a collagen matrix near a cytochalasin D treated cell. The difference in the linear stiffness klin is consistent with a change in fiber density, but all curves at different distances to the cell appear to asymptotically converge to the same response in the nonlinear regime, consistent with previous work [32], indicating that the matrix density does not affect the nonlinear mechanical properties of the collagen gel. 1.3 Conversion from local force on the bead to local matrix stress To use NSIM, we require a conversion between local force on the bead to local matrix stress (See Eq. 1 in the main text). This conversion is established by comparing nonlinear microrheology data with nonlinear macrorheology data. Specifically, we first shift the microrheology data (knl vs. F) in the vertical direction to align these data to the linear regime of the macrorheology data (K vs. s), as shown in Supplementary Figure 5a. Subsequently, we shift the microrheology data in the horizontal direction to align these data with the nonlinear regime of the macrorheology data (Supplementary Figure 5a); the shift factor in the horizontal direction involves a multiplication factor, which defines the conversion from the local force on the bead to matrix stress. This gives a factor of ~2, ~3 and ~10 for collagen, fibrin, and Matrigel respectively. Note, to empirically find the conversion factor between force and stress by shifting, we use a comparable stiffening range of the microrheology and macrorheology data where they share a similar power-law behavior, as indicated by the solid symbols in the Supplementary Figure 5.Interestingly, we observe that knl becomes increasingly stiffer at large force as indicated by the pinksymbols. However, this high stress regime is not accessible in macrorheology.4 Supplementary Figure 5: Determination of conversion factors for NSIM between effective force and local matrix stress using active microrheology (Optical Tweezers) and bulk rheology, resulting in a factor of ~2, ~3 and ~10 for collagen, fibrin and Matrigel respectively. Supplementary Table 1: The effective force Feff determined with NSIM using the data in Fig 4a Collagen Fibrin Matrigel r (μm) Feff (pN) r (μm) Feff (pN) r (μm) Feff (pN) 2.6 114.50 3.3 10.50 3.4 14.28 5.5 19.68 8.0 2.00 6.9 7.60 10.5 5.40 15.5 0.26 14.8 1.47 17.2 1.43 24.4 0.03 24.5 0.59 28.9 0.44 33.1 0.43 5 2 Simulation methods and parameter choicesThe simulation results presented in Figure 3 of the main text are obtained with a lattice model of nonlinear springs. In this model, nonlinear springs of unit rest length are arranged on the edges of a 3D face-centred cubic lattice, and the positions of the vertices thus represent the degrees of freedom of the system. To capture the stiffening behavior of the network, we choose a nonlinear force-extension relation for these springs as:f (x) = exp(µ x) − 1 (1)which results in an exponential stress-strain relationship at macroscopic scales, consistent with experimental observations in collagen [32]. At large forces, f , this choice yields a power-law de-pendence of local differential stiffness on tension: knl = d f /dx ∝ f .We simulate a large spherical region of the network of radius R with fixed boundary conditions. The contractile cell is modelled by a rigid ellipsoid replacing a portion of the network in the center. All lattice nodes that are initially inside the ellipsoid are constrained to translate and rotate as a single rigid body, which can be affinely deformed to simulate contractile forces. Force and torque balance over the rigid body are ensured by including its bulk rotational and translational degrees of freedom to the energy minimization.To capture network disorder, a fraction of the springs is randomly suppressed. This construction will result in substantial mechanical heterogeneities as well as non-affine network deformations, which are characteristic of fibrous networks [18]. However, a finite-size artefact occurs when a straight chain of springs remains that connects the cell at the center of the spherical network to the system's boundary. At large deformation, such straight lines tend to concentrate all the stress as they stiffen dramatically. Since 3D numerical simulations are limited in size, we choose to limit the length of such alignments to a given "fiber length" Lf < R. Specifically, for each line of aligned edges in the initial configuration, we randomly choose a starting point, and remove one spring every Lf + 1 edges, thereby ensuring that no straight line connecting the cell to the system's boundary remains.Local stiffnesses are obtained by measuring the response to point forces, exerted on selected lattice nodes, and directed towards the center of the ellipse. The "remote" force-displacement curve in Figure 3b of the main text, which serves as a reference for NSIM, is obtained in a similar system with no contractile cell, and with the probe placed at the center of the network.Local stress at a lattice node is defined as minus the dipole moment of the forces exerted by nearby nodes, divided by the unit cell volume. This construction is consistent with a macroscopic definition of stress.Macrorheological calibration, as discussed in Figure 2 of the main text, is performed on a large system with periodic boundary conditions, in the absence of cells or probes. The network is stressed by affinely deforming the periodic boundary conditions vectors, in a Lees-Edwards fashion. The stress is evaluated using the discrete mean-stress theorem [46].This system is simulated in C++14. Mechanical equilibrium is obtained by minimizing the total energy using the GNU Scientific Library [47] BFGS2 implementation of the Broyden-Fletcher-6 Goldfarb-Shanno algorithm. Data analysis and visualization are performed in Python2 using the SciPy [48], Matplotlib [49] and Mayavi2 [50] packages.In Figure 3 of the main text, we use the following parameters: system radius R = 25.5, spring stiffness parameter µ = 50, and fiber length Lf = 10. The cell has ellipse aspect ratio 5 : 1 : 1, with its long axis with length 14.2 along the (111) lattice direction. The cell is contracted by 50% along its long axis. We probe the plinear stiffness at each of the 352 points that are present in the plane perpendicular to the (1/√7, 6/7,0) vector (which includes the cell's long axis) and located at a distance > 5 from the system's boundary to avoid edge effects. These points are displayed in Figure 3a of the main text. Local linear stiffness is measured by exerting a force strong enough to displace the probe node by 10−3, small enough to be consistently in the linear response regime, yet large enough to prevent numerical imprecision. The "remote" reference curve is obtained by measuring the displacement in response to forces ranging from 10−2 to 105, in 22 logarithmic increments, and we used spline-fitting on the force-displacement curve in log-log space. This allows us to robustly evaluate the nonlinear differential stiffness knl(F), which we average over 20 realizations of the network. The macrorheological calibration system size is 403, with isotropic strain γ varying logarithmically from 10−5 to 0.46, and similar log-log spline-fitting is applied to macrorheological stress-strain curve. The consequences of this choice of geometry are discussed in Sec 3.4.7 3RobustnessofNSIMInthisSectionweprovidefurthersupportfortherobustnessofourstressinferencemethodinnumericalsimulations,byconsideringadditionalgeometries,whicharedifferentfromthecaseofstressesinducedbyanellipticcellpresentedinFigure3ofthemaintext.3.1HomogeneouslyprestressedsystemsWefirstconsiderthecaseofasystemwherethestressisspatiallyhomogeneous,andoriginatesfromthedeformationoftheboundaryofthesystem.Weconsiderasphericalsystemwithfixedboundaryconditions,withadeformationaccordingtoastrain:γ000−γ/2000−γ/2(2)wherethex-directionisalsotheforcedirection,whichwetaketobethe(111)latticedirection.Thisgeometryofrotationallysymmetricshearstraincorrespondstothestraininalinearelasticmaterialwithasphericallycontractilecell.WepresenttheresultsforthesesimulationsinSupplementaryFigure6,showingexcellentagreementbetweentheinferredstressanditsmeasuredvalue,uptoaproportionalityfactorofroughly∼0.3.ThisshowsthatNSIMaccuratelycapturesstressvariationsinadisorderednet-works,andthatourcalibrationallowsustocapturestresseswithinafactorof2-3dependingonthegeometryofthestrainfield.Thisaccuracyissimilartootherexperimentalerrorssuchasthetypicalexperiment-to-experimentvariationofmacrorheologicalprotocolsonbiopolymernetworks.8 0.00.20.40.60.81.01.21.4probedisplacementδx010002000300040005000probeforceFaremoteσ=2.3e−01σ=9.4e−01σ=3.0e+00σ=7.6e+00σ=1.8e+01σ=4.6e+01σ=1.2e+02σ=3.1e+0210−210−1100101102103104105probeforceF102103104105106localstiffnessknlb10−310−210−1100101102103104105stressσorforceF102103104105elasticmodulusKorlocalstiffnesskcK(σmacro)knl(F)shiftedknl(F)tofitK(σmacro)10−310−210−1100101102103localstressσloc(measured)10−310−210−1100101102103localstressσloc(inferred)dσNSIM=σmeasured±factor2SupplementaryFigure6:ResultsforNSIMinahomogeneouslyprestrainednetwork.a,Nonlin-earforce-displacementcurvesforvaryingpre-stress.b,Nonlinearstiffeningcurves(samecolorsasina).c,Calibrationusingmacrorheologydata:theprefactorforNSIMisobtainedbyaligningthemicro-andmacrorheologicalcurves,asillustratedinFigure2ofthemaintext.d,Inferredlo-calstressversusdirectmeasurement,showingproportionalityatlargestress.Parametersforthesesimulations:µ=100,Lf=12,R=15.5.3.2IsotropicallycontractilecellNext,weconsiderthecaseofasphericallysymmetriccontractilecell,modeledasarigidspherethatshrinksisotropically(SupplementaryFigure7a).WeperformNSIMona1000randompointsinthenetwork,whosedistancestothecellcenterareuniformlydistributedinlogscale.TheseresultsarepresentedinSupplementaryFigure7c,showingexcellentagreementbetweeninferredandmeasuredstress.Thestressislong-rangedanddecaysasr−2withdistancetothecell(Supple-mentaryFigure7e),consistentwithpreviousresultsandsimulationsofellipticcellspresentedinthemaintext.9 SupplementaryFigure7:NSIMinasphericallysymmetricstressfield.a,HerethecellmodelconsistsofarigidsphereofradiusRcell=5.1thatshrinksby55%inasphericalsystemofradiusR=25.5,withfixedboundaryconditions.Networkparameters:Lf=22,µ=50.b,The"remote"referencecurveisobtainedinanunstressednetwork.c,StressinferredwithNSIMvsmeasuredstress,fora1000randompointsinthenetwork.Theagreementbetweenthetwomeasuresisgoodoverthreedecadesofstress.d,Linearstiffnessvsstress.e,Stressdecayawayfromthecell.Whilethesignal-to-noiseratioishighduetothemechanicaldisorderintrinsictoourmodellednetworks,binningandaveragingyieldresultsconsistentwithrope-likeforcetransmission.f,Decayofthelocalstiffnessawayfromthecell.3.3NSIMcaptureslocalstressfluctuationsInthissectionweshowthatNSIMcapturesnotonlytheaveragestress,butalsolocalstressfluc-tuationsduetoheterogeneityinthelocalmechanicalproperties.Todemonstratethis,weconsideroursimulationsinthecaseofanisotropicallycontractilecell(Sec3.2)andanalyzethecorrelationsbetweenthreedatasets:distancestocellr,locallinearstiffnessklin,andlocalstressσloc(obtainedbydirectmeasurement).WefindthefollowingPearsonlinearcorrelationcoefficientsρ(forlog-logcorrelations):ρ(klin,r)=−0.63ρ(σloc,r)=0.56ρ(klin,σloc)=0.9710 Inotherwords,bothklinandσlocfluctuatestronglyandarenotdeterminedbythedistancetothecell,butstressfluctuationsarestronglycorrelatedwithlocalstiffnessfluctuations.Performingalog-loglinearregressionofthemeasuredlocalstressσlocwiththatobtainedusingNSIM,wefindthatthebestpower-lawfithasexponent1.04andprefactor0.567(theprefactorhereiswithisotropicprestrainforthecalibration;seediscussioninsec3.4).ThisisconsistentwiththeidealcaseforNSIMwheretheexponentwouldbe1,showingthatNSIMalsocapturesstressfluctuationsduetomechanicalheterogeneity.3.4MacrorheologicalgeometryandNSIMprefactorCalibrationisdonewithamacrorheologysimulation,measuringthestressresponsetoabulkstrainwithasystemsizeof403withperiodicboundaryconditions.Tensorialstressesaremeasuredusingthemeanstresstheorem.Ifγ=γeγisthestrainmatrix,wedefinethescalarstressasσ=Tr(γ·σ)/γandthediffer-entialmodulusas:K=∂σ∂γusingalogarithmicsplineinterpolationtodifferentiate.11 10−310−210−1100101102103104105macroscopicstressσ101102103104105106differentialmodulusisotropicσ∗=4.33uniaxialσ∗=1.76axialshearσ∗=2.66KSupplementaryFigure8:Macrorheologycurvesforcalibration.Differentstraingeometriesleadtodifferentvaluesofσ∗,whichimpactsthefinalvalueoftheforce-stressconversionfactor.Em-pirically,wefindthatanisotropicdilation(bluecurve)givesthemostaccurateresultforthecon-versionfactor.Thegreencurveshowsthecaseofauniaxialdilation,andtheredcurvecorrespondstouniaxialdilationwithcompensatingcompressionintheothertwodirectionstoensurevolumeconservationatlinearlevel.Asphericalcontractilecellinlinearelasticitycorrespondstothelattercase.12 4ProofofNSIMwithinaminimal1DmodelInthemaintextwementionedthattheNSIMcorrespondencebetweenlocalforceandlocalstressatequivalentstiffnessbecomesexactinthecaseofpower-lawstiffeningofthenetwork.Toestablishthisresult,weemployaminimalmodel,consistingoftwononlinearspringsdescribingthenetworkresponse.4.1ModelOurminimalmodel(SupplementaryFigure9)consistsoftwononlinearspringsinseries,withunitrestlengthandagenericforce-extensionrelation12f(δ)(the1/2beingforconvenienceofnotation).Thenetworkprestressismodeledbyadisplacementofthetwoboundarypointsbyγ(fixedstrain),resultinginastressσ(γ)=f(γ)(3)atthecentralpoint.Theforceexertedbyopticaltweezersresultsinadisplacementofthecentralpointofadistancex,correspondingtoaforceF(x,γ)=f(γ+x)−f(γ−x)2(4)Inparticular,intheabsenceofprestress,wedefinethereferenceforce-displacementcurvefortheprobe,Fref(x)=F(x,γ=0)=f(x)−f(−x)2=f(x)(5)wherefistheoddpartoff.Asdiscussedinthemaintext,theexperimentallymeasurablequantitiesareF(x,γ)foranun-knownprestress.Wedefineasbeforethelinearstiffnessinthepre-strainedstate:klin(γ)=∂F∂x(x=0,γ)=f′(γ)(6)andthenonlineardifferentialstiffnessintheabsenceofpre-stress:knl(x)=∂F∂x(x,γ=0)=f′(x)+f′(−x)2=f′(x)(7)4.2MathematicalformulationofNSIMcorrespondenceTheideaofNonlinearStressInferenceMicroscopyisthatthenetwork'sstiffnessdependsonitsstressstate(i.e.,fisnonlinear)suchthatwecanestablishacorrespondencebetweentheforce-inducedstiffeningcurveknlandthestress-inducedstiffeningcurveklin.13 SupplementaryFigure9:Aminimalmodelforourmicrorheologicalsystem.Twononlinearspringsdescribethenetworksurroundingabead.Startingfromanunstressedreferenceconfigura-tion(top),thecell-generatedprestress,σ,inducesasymmetricelongationofthetwospringsbyγ,andtheforceFexertedbytheopticaltweezerresultsinadisplacementxofthebead.Specifically,toeachpre-strainγweassociateaneffectivedisplacementxeff(γ)correspondingtothesamedifferentialstiffness,suchthatklin(γ)=knl(xeff(γ))(8)hencexeff(γ)=k−1nl(klin(γ))(9)Weareinterested,however,inmechanicalquantitiesFandσratherthanintheassociategeometri-calquantitiesxandγ.WethusdefinetheeffectivelocalforceFeff(σ)thatcorrespondstothesamedifferentialstiffnessasthelocalstressσ:Feff(σ)=Fref(cid:0)k−1nl(klin(γ(σ)))(cid:1)(10)OurstressmicroscopytechniquerequiresthatFeff(σ)∝σ;wenowstudytheconditionsunderwhichthisidentityholds.Mathematically,intermsofthegenericnonlinearforce-extensionrelationofthesprings,wehave:•σ(γ)=f(γ)(11)•klin(γ)=f′(γ)(12)•Fref(x)=f(x)−f(−x)2=f(x)(13)wherefistheoddpartoff.14 •knl(x)=f′(x)(14)therefore:Feff(σ)=hf◦(cid:0)f′(cid:1)−1◦f′◦f−1i(σ)(15)Schematically:σf−17−−→γf′7−→klin∼knl(f′)−17−−−−→xefff7−→Feff(16)4.3ConditionsforNSIMThequestionweneedtoaddressis:howdoesthecompositefunctionFeff=f◦(cid:0)f′(cid:1)−1◦f′◦f−1comparetotheidentityfunction?Whenthisisthecase,wewillbeensuredtohave:Feff(σ)∝σ.whichistherequirementforNSIM.NotefirstthatinordertobeabletoinferaneffectiveforceFeff,thevaluestakenbyknlandklinshouldbethesame.Inparticular,thisimpliesthatfshouldhaveanonlinearoddcomponent,suchasacubicorquinticterm.Otherwise,thereisnostiffeningunderdisplacementoftheprobe(astheweakeningofonesideexactlycompensatesthestiffeningoftheother),andthusnoinferenceispossibleasfisnotinvertible.Incontrast,iffisodd,Feff(σ)=σforallvaluesofσ;howeverthiscasecorrespondstoamaterialforwhichthenonlinearresponseundercompressionandtensionisthesame,whichdoesnotcorrespondtoanyphysicalsituation.Alsof′isevenbyconstruction,andthusinvertibleonlyonR+,i.e.,probeforcesFand−Fareindistinguishable.4.3.1WeaklynonlinearregimeNextweconsideragenericcasewherealltermsoftheTaylorexpansionoffarenonzero:f(δ)=δ+a22δ2+a33δ3+...(17)wherethefirstcoefficientissettounity(whichgivesthestiffnessscale,i.e.k0=1).Inthiscase,wehave,tofirstnonlinearorder:f′(δ)=1+a2δ+...(18)f(δ)=δ+a33δ3+...(19)f′(δ)=1+a3δ2+...(20)thusf−1(t)=t−a22t2(21)(cid:0)f′(cid:1)−1(k)=sk−1a3+...(22)15 andsoFeff(σ)=hf◦(cid:0)f′(cid:1)−1◦f′◦f−1i(σ)=√σ+(cid:16)a33−a24(cid:17)σ3/2+...(23)HenceintheweaklynonlinearregimeFeff(σ)∼√σ,precludinganaccurateinferenceofstressbythismethod.Thisisduetothedifferenceinsymmetrybetweenforceandstress,asdiscussedinFigure2ofthemaintext.4.3.2StronglynonlinearregimeWenowexaminetheconditionsforproportionalitybetweenstressandeffectiveforceinthestronglynonlinearregime.Wemakethefollowingassumptionsregardingthestiffeningfunctionfcharac-teristicofthematerial:•Divergingstiffness:limx→x∞f′(x)=∞(24)wherex∞standsforeitherafinitemaximumelongation(Worm-Like-Chain-likedivergence)orinfinity.Thisassumptionexcludeslinearandstrain-softeningmaterialsforinstance.•Compression-extensionratio:limx→x∞f(x)/f(−x)=−c(25)where0≤c<∞.Thenumbercisacharacteristicofthematerial:c=0foramaterialwhichbucklesandyieldsundercompression,andc=1foranoddforce-extensionrelation.Wealsodefineq=(1+c)/2,suchthat:f(x)∼x→x∞qf(x)(26)withq=1/2forabucklingmaterial.Undertheseassumptions,wenowderiveafunctionalrelationthattheforceextensioncurvefneedstoobeytoexactlysatisfyNSIMconditionsinthestronglynonlinearregime.FirstwedefineS(k)=f(cid:0)f′−1(k)(cid:1)andS(k)=f(cid:0)f′−1(k)(cid:1),suchthat:σ=S(klin)(27)andF=S(knl)(28)HencewecanrewriteEq.23asFeff(σ)=S(cid:0)S−1(σ)(cid:1)(29)TherequirementforourNSIMcorrespondenceisthatFeff(σ)∝σ,whichtranslatestoS(cid:0)S−1(σ)(cid:1)=bσforsomestress-forceproportionalityconstantb.Forthisproportionalitytoholdinthestrongly16 nonlinearasymptoticregimek→∞limit(undertheassumptionthatthenetworkstiffensundertension,i.e.Eq.(24)),weneedtohaveS(k)S(k)−−−→k→∞b;0<b<∞(30)Equation(26)impliesthatf′−1(k)∼k→∞f′−1(k/q)(31)CombiningthiswithEq.30,wegetS(k)∼k→∞qS(k/q)(32)TheconditionthatSandSbeproportional(Eq.(30))thusimplies:S(k)∼k→∞bqS(k/q)(33)Ofcourse,ifq=1(asymptoticallyoddfunction),thisequationonlyimpliesb=1,whichwealreadyknew(asdiscussedatthestartofSec.4.3).Foranyothervalueofq,includingthecaseofabucklingmaterial,Eq.(33)ishighlyconstraining:Assumingaphysicallyreasonablebehaviorforthematerial1,thisfunctionalrelationshipindeedimpliesapower-lawbehaviorprescribedbythevalueoftheproportionalityconstantb:S∼k→∞Akα(34)withα=1+logblogq(35)Notethatthevaluesofqandα,bothofwhicharephysicallymeasurableandmeaningful,setthevalueoftheconstantb.4.3.3ConsequencesonfRecallingthedefinitionofS(k)=f(cid:0)f′−1(k)(cid:1)(Eq.(27)),thepowerlawform(rewritingEq.(34)intermsoff−1)impliesthat:f′−1(k)∼k→∞f−1(Akα)(36)Weinvertthisequationtoarriveatf′(x)∼x→x∞(cid:20)1Af(x)(cid:21)1α(37)1MorepreciselyEq.(33)impliesthatS(k)=kαφ(logk),whereφisagenericlog(q)-periodicfunction.Itisthusnaturaltoassumethatφisaconstant,thealternativebeingthatthestiffnesscurveshowssomekindofnon-trivialself-similarstructure.17 Thisdifferentialequationonfhasthefollowingsolution,dependingonthevalueofα:f(x)∼xα1−α0<α<1exp(µx)α=1(x∞−x)−αα−1α>1(38)These solutions include standard models for force-extension relations of individual biopolymers, such as the Worm-Like-Chain model (α = 2) or exponential stiffening (α = 1 as in the numerical simulations presented in the main text).In summary, we have shown that the effective force Feff becomes exactly proportional to the local stress σloc in the strongly nonlinear regime if, and only if, the stiffness-tension relationship of the springs asymptotically behaves as a power-law.18 |
1808.10506 | 1 | 1808 | 2018-08-30T20:28:43 | Maximum Entropy Principle Analysis in Network Systems with Short-time Recordings | [
"physics.bio-ph",
"cs.IT",
"cs.IT",
"physics.data-an",
"stat.ME"
] | In many realistic systems, maximum entropy principle (MEP) analysis provides an effective characterization of the probability distribution of network states. However, to implement the MEP analysis, a sufficiently long-time data recording in general is often required, e.g., hours of spiking recordings of neurons in neuronal networks. The issue of whether the MEP analysis can be successfully applied to network systems with data from short recordings has yet to be fully addressed. In this work, we investigate relationships underlying the probability distributions, moments, and effective interactions in the MEP analysis and then show that, with short recordings of network dynamics, the MEP analysis can be applied to reconstructing probability distributions of network states under the condition of asynchronous activity of nodes in the network. Using spike trains obtained from both Hodgkin-Huxley neuronal networks and electrophysiological experiments, we verify our results and demonstrate that MEP analysis provides a tool to investigate the neuronal population coding properties, even for short recordings. | physics.bio-ph | physics |
Maximum Entropy Principle Analysis in Network Systems with Short-time
Recordings
Zhi-Qin John Xu1, Jennifer Crodelle2, Douglas Zhou3,∗ and David Cai1,2,3,4
1NYUAD Institute, New York University Abu Dhabi,
Abu Dhabi, United Arab Emirates,
2Courant Institute of Mathematical Sciences,
New York University, New York, New York, USA.
3School of Mathematical Sciences,
MOE-LSC and Institute of Natural Sciences,
Shanghai Jiao Tong University, Shanghai, P.R. China,
4Center for Neural Science, New York University,
New York, New York, USA.
(Dated: September 3, 2018)
In many realistic systems, maximum entropy principle (MEP) analysis provides an effective char-
acterization of the probability distribution of network states. However, to implement the MEP
analysis, a sufficiently long-time data recording in general is often required, e.g., hours of spiking
recordings of neurons in neuronal networks. The issue of whether the MEP analysis can be suc-
cessfully applied to network systems with data from short recordings has yet to be fully addressed.
In this work, we investigate relationships underlying the probability distributions, moments, and
effective interactions in the MEP analysis and then show that, with short recordings of network
dynamics, the MEP analysis can be applied to reconstructing probability distributions of network
states under the condition of asynchronous activity of nodes in the network. Using spike trains
obtained from both Hodgkin-Huxley neuronal networks and electrophysiological experiments, we
verify our results and demonstrate that MEP analysis provides a tool to investigate the neuronal
population coding properties, even for short recordings.
PACS numbers: 89.70.Cf, 87.19.lo, 87.19.ls, 87.19.ll
INTRODUCTION
ronal networks.
Binary-state models have been used to describe the
activity of nodes in many network systems, such as neu-
ronal networks in neuroscience [25, 27, 36]. Understand-
ing the distribution of network binary-state dynamics is
important in unveiling underlying network function, es-
pecially in neuroscience where both theoretical and ex-
perimental results indicate that populations of neurons
perform computations probabilistically through their fir-
ing patterns [11, 13]. For instance, statistical distribu-
tions of neuronal network firing patterns have been shown
to perform awake replays of remote experiences in rat
hippocampus [11]. Therefore, studying the characteris-
tics of neuronal firing pattern distributions will help to
understand how neuronal networks encode information
[17]. However, this is a difficult task, since the number
of all possible network states grows exponentially as the
network size increases, i.e., 2n for a network of n binary-
state nodes. This high dimensionality presents a chal-
lenge in directly measuring the distribution of network
states in electrophysiological experiments, especially for
the case of in vivo measurements on awake animals, thus,
the difficulty in understanding coding schemes in neu-
∗ [email protected]
The maximum entropy principle (MEP) analysis is a
statistical method used to infer the least biased prob-
ability distribution of network states by maximizing the
Shannon entropy with given constraints, i.e., moments up
to certain order [10]. The inferred probability distribu-
tion is called the MEP distribution. For example, under
constraints of low-order moments, the MEP distribution
gives rise to an accurate estimate of the statistical dis-
tribution of network states in many scientific fields, e.g.,
neuroscience [25, 27, 34, 36], biology [15, 22, 31], imaging
science [23], economics [7, 29], linguistics [30], anthropol-
ogy [8], and atmosphere-ocean science [12].
However, a practical and important issue remains un-
clear: how well the MEP analysis performs when the
recording time of dynamics of network nodes is short.
This is because a very long recording is often required to
carry out the MEP analysis, e.g., hours for a network of
10 neurons [25]. These long recordings are usually im-
practical to achieve due to cost or capability. For exam-
ple, physiological constraints such as fatigue of neurons
makes it very difficult to record the state of neuronal
networks over a long time, especially in vivo recordings
on awake animals. Meanwhile, data obtained by short
recordings poorly captures many activity states, leading
to incorrect descriptions for the probability distribution
of network states. The insufficient measurements due
to short recordings could lead to a misunderstanding of
information coding structure embedded in network activ-
ity states [20]. Therefore, it is important to estimate an
accurate probability distribution of network states from
short recordings where many network activities are un-
derrepresented.
In this work, we demonstrate that the MEP analy-
sis can give rise to an accurate estimate of the proba-
bility distribution of network states from a short-time
data recording if the activity of nodes in the network is
asynchronous (asynchronous network). To achieve this,
we first show the existence of a one-to-one mapping (de-
(n)
noted as the full-rank matrix U
PM for a network of size
n) between the probability distribution of network states
(denoted by a vector of size 2n, P(n)) and the corre-
sponding all moments (denoted by a vector of size 2n,
M(n)). Since the full-order MEP distribution (the distri-
bution obtained in the MEP analysis with constraints of
all moments M(n)) and the probability distribution P(n)
are the same (they have all the same moments M(n)),
we derive another one-to-one mapping (denoted as the
full-rank matrix L(n)
JP for a network of size n) between all
effective interactions (Lagrange multipliers correspond-
ing to constraints of all moments in the expression of
the full-order MEP distribution) and the probability dis-
tribution P(n). These mappings show that all moments
and all effective interactions can equivalently represent
a probability distribution of network states for a general
network of any size.
Next, we use the above equivalent representations to
show that, in an asynchronous network, low-order MEP
analysis gives rise to an accurate estimate of the prob-
ability distribution of network states from a short-time
recording. In an asynchronous network, the probability
of many nodes being active in one sampling time win-
dow (referred to as highly-active states) is very small.
Through the mapping from effective interactions to the
probability distribution, L(n)
JP , we observe that the prob-
ability of a highly-active state can be written as a sum-
mation of effective interactions corresponding to the con-
straints of moments of those active nodes. In an asyn-
chronous network, high-order effective interactions (La-
grange multipliers corresponding to the constraints of
high-order moments) are usually small [4, 25, 27, 34, 35],
as compared to the low-order effective interactions (La-
grange multipliers corresponding to the constraints of
low-order moments). Then, the probabilities of highly-
active states can be well estimated by the dominating
low-order effective interactions. The MEP analysis shows
that low-order effective interactions can be estimated by
low-order moments through a widely-used iterative scal-
ing algorithm (see Appendix A for details). Therefore,
the low-order moments may possibly be measured accu-
rately using short recording (as discussed below), and
may be used to perform the MEP analysis to obtain a
good estimate of the probability distribution of network
states.
To obtain an accurate estimation of the low-order mo-
2
ments, we use properties of the mapping between the
probability distribution of network states and the cor-
(n)
responding moments of the network, U
PM. This map-
ping shows that low-order moments can be written as a
summation of all probability states in which those nodes
corresponding to the constraints of moments are active,
e.g., the first-order moment of node i is a summation of
all network-state probabilities in which node i is active.
This summation includes both highly-active state prob-
abilities and low-active state probabilities in which few
nodes are active in one sampling time window. Since
the probability of observing a highly-active state is small
in an asynchronous network, the low-active state prob-
abilities dominate the summation. This results in good
estimation for the low-order moments because low-active
state probabilities can still be well-measured in a short
recording and the noise for the estimation of low-order
moments can be further reduced by the summation of
low-active state probabilities [24].
Although the procedure described in this work can be
applied to any asynchronous network with binary dynam-
ics, here we use spike trains obtained from both Hodgkin-
Huxley (HH) neuronal network dynamics simulations and
electrophysiological experiments to demonstrate that one
can perform the MEP analysis to accurately estimate the
probability distribution of network states from short-time
recordings. For the case of simulation data from HH neu-
ronal networks, we evolve the HH neuronal network for
a short run time of 120 s. With constraints of the first-
order and the second-order moments of the data from
this short recording, we obtain the distribution of net-
work states from the MEP analysis. To verify the accu-
racy of the MEP distribution, we evolve the HH neuronal
network for a long run time of 1.2 × 105 s and directly
measure the probability distribution of network states
for comparison. For the case of experimental data from
electrophysiological measurements, we use data recorded
from primary visual cortex (V1) in anesthetized macaque
monkeys (See Appendix E for details), in which we use
the data of 3824 s as a long recording data to obtain the
probability distribution of network states and perform
the second-order MEP analysis on the short-recording
data of 191.2 s (5% of the total recording). Our results
show that in both numerical simulation data and electro-
physiological experimental data, these two distributions,
i.e., the second-order MEP distribution and the distribu-
tion measured in the long recording data, indeed agree
very well with each other, whereas the probability distri-
bution measured from the short recording deviates sig-
nificantly from them and cannot capture many neuronal
activity states.
METHODS: THE MEP ANALYSIS
In this section, we will introduce the MEP analysis,
which has been applied to estimating the probability dis-
tribution of network states for many network systems
[3, 4, 25, 27, 34]. The process of performing the MEP
analysis is as follows. Let σ ∈ {0, 1} denotes the state
of a node in a sampling time bin, where 1 refers to an
active state and 0 an inactive state. Then, the state of
a network of n nodes in a sampling time bin is denoted
as Ω = (σ1, σ2, · · · , σn) ∈ {0, 1}n. In principle, to obtain
the probability distribution of Ω, one has to measure all
possible states of Ω, i.e., 2n states in total. For the case
of a network of n neurons, σ often represents whether
a single neuron fires in a sampling time bin in the net-
work, where 1 corresponds to a neuron that is firing and
0 corresponds to a neuron in the silent state. We choose
a typical bin size of 10 ms for the MEP analysis [27, 34].
The state, Ω, in the neuronal network would represent
the firing pattern of all neurons in the network.
The first-order moment of the state of node i, σi, is
given by
hσii =XΩ
P (Ω)σi(Ω),
(1)
where P (Ω) is the probability of observing the firing pat-
tern Ω in the recording, and σi(Ω) denotes the state of
the ith node in the firing pattern Ω. The second-order
moment of the state of node i, σi, and node j, σj, is given
by
hσiσji =XΩ
P (Ω)σi(Ω)σj(Ω),
(2)
with higher-order moments obtained similarly. Note
that, in Eqs. (1, 2), P (Ω) can be the true probability
distribution of Ω, in which case, one computes the true
moments; or P (Ω) can be an observed distribution of a
finite-time recording, in which case, one then estimates
the moments.
The Shannon entropy of a probability distribution
Pany(Ω) is defined as
S = −XΩ
Pany(Ω) log Pany(Ω).
(3)
By maximizing this entropy, S, subject to all moments
up to the mth-order (m ≤ n), one obtains the mth-order
MEP distribution for a network of n nodes [25, 27, 34].
Note that the kth-order moments consist of all the ex-
pectations of the product of any k nodes' states (e.g., the
second-order moments by Eq. (2) above for any pair of
i and j with i 6= j) and thus when considering the con-
straint of the kth-order moments, there are C k
n (the num-
ber of combinations of k nodes from n possible choices)
number of constraints of kth-order moments being con-
sidered. Finally, the mth-order probability distribution
is obtained from the following equation,
Pm(Ω) =
1
Z
exp m
Xk=1
n
Xi1<···<ik
Ji1···ik σi1 · · · σik! ,
(4)
where, following the terminology of statistical physics,
Ji1···ik is called the kth-order effective interaction (2 ≤
3
k ≤ m), i.e., the Lagrange multiplier corresponding to
the constraint of the kth-order moment hσi1 · · · σik i, and
the partition function, Z, is a normalization factor. Eq.
(4) is referred to as the mth-order MEP distribution. In
practice, one can utilize an iterative scaling algorithm
(see Appendix A for details) to numerically solve the
above MEP optimization problem to obtain the effective
interactions and thus the probability distribution in Eq.
(4). Here, for a network of size n, Pn(Ω) is referred to
as the full-order MEP distribution subject to moments
of all orders.
RESULTS
The results are organized as follows. We begin by
demonstrating that there is a wide range of dynamical
regimes where the neuronal network dynamics is asyn-
chronous. Then, we show that the probability distribu-
tion of network states measured from a short-time record-
ing (data recorded from the HH network dynamics with
short simulation time) cannot capture many network ac-
tivity states, and thus differs from the probability dis-
tribution of network states measured from a long-time
recording. Note that we have verified that the HH net-
work dynamics in a long simulation time of 1.2 × 105 s
reaches the steady state; therefore, the probability distri-
bution of network states measured from this long record-
ing can well represent the true probability distribution of
network states.
We next show that there exists a one-to-one mapping
between the probability distribution of network states
and the corresponding moments of the network. Then,
we combine this mapping with the full-order MEP dis-
tribution to show that there exists a one-to-one map-
ping between all effective interactions and the probabil-
ity distribution of network states. Using these mappings,
we further demonstrate that high-order effective interac-
tions are small in asynchronous networks; thus, to ac-
curately estimate the probability distribution of network
states, one may only require accurate estimation of the
low-order effective interactions. Finally, we make use of
low-order moments measured in short-time recordings to
estimate low-order effective interactions and show that
the obtained probability distribution from the low-order
MEP analysis agrees well with the probability distribu-
tion of network states. This is demonstrated by both
numerical simulations of asynchronous HH neuronal net-
work dynamics and also electrophysiological experiments.
Short-time recordings cannot represent all network
states
In this section, we first use numerical simulation data
from the HH neuronal network dynamics and show that
short-time recordings are often insufficient to accurately
estimate the probability distribution of network states.
Later, we will also demonstrate this issue using experi-
mental data from electrophysiological measurements.
We simulate a network of 80 excitatory and 20 in-
hibitory HH neurons, with a 20% connection density
among neurons in the network. As physiological exper-
iments can often only measure a subset of neurons, we
randomly select 10% of the neurons in the network (10
neurons) and demonstrate differences in the directly mea-
sured probability distribution between the short and the
long recording. Note that there are typically three dy-
namical regimes for neuronal networks [39]: (i) a highly
fluctuating regime where the input rate, µ, is low (Fig.
1a); (ii) an intermediate regime where µ is moderately
high (Fig. 1b); (iii) a low fluctuating or mean-driven
regime where µ is very high (Fig. 1c). We evolve the HH
neuronal network dynamics and record the spike trains
of all neurons for a duration of 1.2 × 105 s, which is suf-
ficiently long to obtain a stable probability distribution
of neuronal firing patterns. We then compare the proba-
bility distribution of network states directly measured in
the short recording of 120 s to that measured in the long
recording. As shown in Fig. 2, the measured probability
distribution of firing patterns in the short recording de-
viates substantially from that in the long recording, for
all three dynamical regimes.
The probability distribution of network states is im-
portant for a complete understanding of the underlying
function of networks [11, 13] despite the fact that a suffi-
cient long-time data recording is often impossible or im-
practical. Thus, it is essential to obtain an accurate es-
timate of the probability distribution of network states
from a short-time recording. To achieve this, we next
study relationships among the probability distribution,
the moments, and the effective interactions in the MEP
distribution and show that, for asynchronous networks,
the second-order MEP analysis gives rise to an accurate
estimate of the probability distribution of network states.
One-to-one mapping between the probability
distribution and the moments
1 , p(n)
2 , · · · , p(n)
To demonstrate the relationship between the true
probability distribution of network states, Ptrue(Ω), and
the corresponding moments, we introduce several nota-
tions for ease of discussion. First, denote the vector
P(n) = (p(n)
2n )T as the vector containing
the probability distribution of the network states for
a network of n nodes, and denote the vector M(n) =
(m(n)
2n )T as the vector containing all mo-
ments of the network. We arrange the entries in P(n) and
M(n) as follows. For an example network of size n = 2,
we assign values to the ith entry by expressing i − 1 using
the base-2 number system with total n = 2 digits, i.e.,
2 , · · · , m(n)
1 , m(n)
4
i
i Ei = e2e1 p(2)
i m(2)
1
P00
2
P10
3
P01
4
P11 hσ1σ2i
00
01
10
11
hσ1i
hσ2i
1
TABLE I: Table of entries for building the vectors P(n)
and M(n) for an example network of n = 2 nodes.
Ei = e2e1, where Ei represents the combined state of the
two nodes in the network, e1 and e2 (e.g., 00 corresponds
to both nodes being inactive), as shown in the second
column of Table I. Then, for each i, denote the proba-
bility of the network state (σ1 = e1, σ2 = e2) as p(2)
, as
shown in the third column of Table I. In neuroscience, the
vector P00 represents the probability of finding both two
neurons in the quiet state, P10 (P01) represents the prob-
ability of the first (second) neuron in the active state and
the second (first) neuron in the silent state, and P11 rep-
resents the probability of both neurons in the active state.
The entries in the vector M(2) are arranged similarly, i.e.,
m(2)
is the expectation of (σ1)e1 (σ2)e2 , as shown in the
fourth column of Table I.
i
i
For illustration, we show that, for a network of n = 2
nodes, there is a full-rank matrix, U(2)
PM, that transforms
from the probability distribution to moments. From Eqs.
(1) and (2), the expectation of σ1, σ2, and σ1σ2 can be
obtained by summing the probabilities in which those
nodes are active, leading to the following system
1 1 1 1
0 1 0 1
0 0 1 1
0 0 0 1
P00
P10
P01
P11
=
1
hσ1i
hσ2i
hσ1σ2i
,
(5)
i.e., U(2)
upper-triangular and of full rank.
PMP(2) = M(2). Clearly, from Eq. (5), U(2)
PM is
The above analysis can be extended to a network of any
size n. For each integer i, 1 ≤ i ≤ 2n, we can similarly ex-
press i−1 by the base-2 number system with n digits, de-
noted by Ei = enen−1 · · · e2e1. Then, write the probabil-
ity of the network state (σ1 = e1, σ2 = e2, · · · , σn = en),
denoted as p(n)
, and the moment i.e., the expectation
of (σ1)e1 (σ2)e2 · · · (σn−1)en−1 (σn)en , denoted as m(n)
, as
follows
i
i
P(n) =
P00···0
P10···0
P01···0
P11···0
...
P01···1
P11···1
, M(n) =
.
(6)
1
hσ1i
hσ2i
hσ1σ2i
...
j=2 σjE
DQn
j=2 σjE
Dσ1Qn
We prove that there is a full-rank matrix, U(n)
PM, that
5
(a)
(b)
(c)
FIG. 1: Raster plots for the HH neuronal network in three different dynamical regimes. Raster plots of
10 randomly selected neurons for each case are shown. A short bar indicates that the neuron with certain index fires
at certain time. The coupling strength is selected at random from the uniform distribution of the interval [0, s],
where s = 0.071 ms−1 (the corresponding physiological excitatory postsynaptic potential is ∼ 1 mV). The Poisson
input parameters for in (a), (b), and (c) are (µ = 0.6 ms−1, f = 0.05 ms−1), (µ = 1.1 ms−1, f = 0.04 ms−1) and
(µ = 2.5 ms−1, f = 0.03 ms−1), respectively.
10-2
10-3
10-4
t
n
e
m
e
r
u
s
a
e
m
t
r
o
h
S
10-2
10-3
10-4
t
n
e
m
e
r
u
s
a
e
m
t
r
o
h
S
10-2
10-3
10-4
t
n
e
m
e
r
u
s
a
e
m
t
r
o
h
S
10-4
Long measurement
10-3
10-2
10-4
Long measurement
10-3
10-2
10-4
Long measurement
10-3
10-2
(a)
(b)
(c)
FIG. 2: Measured probability distributions for short-time recordings compared with long-time
recordings. The frequency of each firing state measured from the short-time recording of network dynamics (120 s)
is plotted against the frequency measured from the long-time recording of network dynamics (1.2 × 105 s). Data for
these three cases are from three dynamical regimes shown (the same 10 selected neurons in each case) in Fig. 1,
respectively.
transforms from P(n) to M(n) as
U(n)
PMP(n) = M(n),
(7)
where U
pendix C for details).
(n)
PM is upper-triangular and of full rank (see Ap-
Therefore, all moments, which are shown in the en-
tries of M(n), can be used to describe the probability
distribution of network states for a network of n nodes
with binary dynamics. As the full-order MEP distri-
bution, Pn(Ω),
is subject to all moments, Pn(Ω) and
Ptrue(Ω) share the same moments for a sufficiently long-
time recording, i.e., M(n) in Eq. (7). Since U(n)
PM is of
full rank, Pn(Ω) is identical to Ptrue(Ω).
By directly substituting Ptrue(Ω) into the full-order
MEP analysis, we develop a relationship between effec-
tive interactions and the probability distribution, as dis-
cussed in the next section.
One-to-one mapping between effective interactions
and the probability distribution
the
To demonstrate
relationship between effec-
tive interactions and the true probability distribu-
tion of network states, we substitute all 2n states of
Ω = (σ1, σ2, · · · , σn) and the probability distribution,
Ptrue(Ω), into Eq. (4) with m = n, and then take the
logarithm of both sides. This results in a system of lin-
ear equations in terms of − log Z and all the effective
interactions,
n
n
− log Z +
Ji1···ik σi1 · · · σik = log Ptrue(Ω),
Xk=1
Xi1<···<ik
(8)
where − log Z can be regarded as the zeroth-order ef-
fective interaction, J0. By solving the system of linear
equations in Eq. (8), we can obtain all the 2n effective
interactions, J's, in terms of the true probability distri-
bution of network states, Ptrue(Ω).
We again turn to the small network case of n = 2 nodes
to demonstrate how to obtain a one-to-one mapping from
the linear system described by Eq.
(8). First, denote
the vector J(2) as the vector containing all the effective
interactions, with the index of each effective interaction,
i. We then express i − 1 by the base-2 number system
with total n = 2 digits, denoted by Ei = e2e1, and the
ith entry of J(2) is the coefficient of the term (σ1)e1 (σ2)e2
in Eq. (8), yielding J(2) = (J0, J1, J2, J12)T . Since the
ordering of the indices for J(2) is the same as that of
P(2), then the right hand side of Eq. (8) is simply the
logarithm of the vector P(2). Therefore, for a network of
n = 2 nodes, we have the following equation
6
network with size n = 2. Based on Eq. (9), we have
J0 = log P00,
J1 = log
P10
P00
,
J2 = log
P01
P00
,
and
J12 = log
P11
P01
− log
P10
P00
.
Next, we define a new term
J 1
1
, log
P11
P01
,
which describes the case in which the state of the second
node in the network is changed from inactive (state 0) to
active (state 1) (i.e., in J1, P10 → P11 and P00 → P01).
Note that with this notation, we can now express the
higher-order effective interaction, J12, as
(11)
(12)
1 0 0 0
1 1 0 0
1 0 1 0
1 1 1 1
J0
J1
J2
J12
=
log P00
log P10
log P01
log P11
.
(9)
J12 = J 1
1 − J1.
For a network of any size n, based on Eq. (10), we obtain
an expression for the first-order effective interaction
The above relation can be extended to a network of
any size n and the corresponding linear equations are as
follows
J1 = log
P10···0
P00···0
L(n)
JP J(n) = log P(n),
(10)
and for the second-order effective interaction
where L(n)
JP is a lower-triangular matrix with dimension
2n × 2n. For 1 ≤ i ≤ 2n, we can similarly express i − 1
by the base-2 number system with n digits, denoted by
Ei = enen−1 · · · e2e1. Then, the ith entry of J(n) is the
coefficient of the term (σ1)e1 (σ2)e2 · · · (σn−1)en−1 (σn)en
in Eq. (8). To show the linear transform between the
effective interactions in the full-order MEP distribution
J(n) and the probability distribution of network states
P(n) is a one-to-one mapping, one needs to demonstrate
that L(n)
JP is a matrix of full rank. Similarly, as the proof
of the one-to-one mapping between the probability dis-
tribution and the moments, we can show that L(n)
JP is the
transpose of U(n)
by mathemat-
ical induction. Therefore, L(n)
JP is lower-triangular and
full-rank, and one can use effective interactions to char-
acterize the probability distribution of network states for
a network of nodes with binary dynamics.
PM(cid:17)T
JP = (cid:16)U(n)
PM, i.e., L(n)
High-order effective interactions are small for
asynchronous networks
Using the mapping between the probability distribu-
tion and effective interactions, we show that there is a
recursive relationship among different orders of the effec-
tive interactions, where high-order effective interactions
are weak compared with low-order ones in asynchronous
networks. For illustration, we first discuss the case of a
J12 = log
P110···0
P010···0
− log
P10···0
P00···0
.
Then, the second-order effective interaction, J12, can be
equivalently obtained by the following procedure: First,
in J1 = log(P10···0/P00···0), we switch the state of the
second node from 0 to 1 to obtain a new term J 1
1 =
log(P110···0/P010···0). Then, note that if we subtract J1
from J 1
1 , we arrive at J12 as described in Eq. (12). The
above notation can be extended to the case of high-order
effective interactions (see Appendix D for a proof):
J12···(k+1) = J 1
12···k − J12···k,
(13)
where 1 ≤ k ≤ n − 1 and J 1
12···k is obtained by switch-
ing the state of the (k + 1)st node in J12···k from 0 to 1.
Next, we intuitively explain that the recursive structure
(Eq. (13)) leads to the hierarchy of effective interactions,
i.e., high-order effective interactions are much smaller
than low-order ones for asynchronous networks. J12···k
describes the effective interaction of the sub-network of
neurons {1, 2, 3, ..., k} when the state of the (k +1)st neu-
ron and states of neurons with indices {k + 2, k + 3, ..., n}
are inactive. The term, J 1
12···k, describes the case in which
the (k+1)st neuron becomes active and states of all other
neurons are not changed. For an asynchronous network,
the cortical interactions between neurons are relatively
weak, therefore, the effect on the change of network dy-
namics from the case when the (k + 1)st neuron is silent
to the case when the (k + 1)st neuron is active should
be small. In other words, the value difference between
the effective interaction of the sub-network where the
(k+1)st neuron is inactive, J12···k, and the effective inter-
action when the (k + 1)st neuron is active, J 1
12···k, should
be small, and thus the higher-order effective interaction,
J12···(k+1), is much smaller than the low-order effective
interaction, J12···k. Note that more strict mathematical
proof of the hierarchy of effective interactions in an asyn-
chronous network can be seen in our previous work [36].
Therefore, the recursive relation in Eq. (13) leads to a
hierarchy of effective interactions, i.e., low-order effective
interactions dominate higher-order effective interactions
in the MEP analysis.
Next, we investigate the HH neuronal network dynam-
ics to confirm that the first two-order effective interac-
tions in the full-order MEP distribution dominate higher-
order effective interactions. We evolve HH networks in a
long-time simulation of 1.2 × 105 s and measure the prob-
ability distribution of network states. We have verified
that the HH network dynamics in such a long simulation
time reaches the steady state, therefore, the measured
probability distribution of network states can well repre-
sent the true probability distribution of network states.
We then calculate effective interactions in the full-order
MEP distribution by Eq. (10) from the measured prob-
ability distribution of network states.
In Fig. 3, the
average strength of the kth-order effective interactions
is computed as the mean of the absolute value of the
kth-order effective interactions.
It can clearly be seen
that for three different dynamical regimes, the average
strength of effective interactions of high-orders (≥ 3) are
at least one order of magnitude smaller (in the absolute
value) than that of the first-order and the second-order
effective interactions.
Low-order MEP analysis with short-time recordings
Using the mappings described in the previous sections,
we show that the low-order MEP analysis can provide
an accurate estimate of the probability distribution of
network states with a short-time recording.
First, note that in a short recording of an asynchronous
network, the probability of observing a highly-active
state (one in which many nodes are active simultane-
ously) is usually too small to be measured accurately.
Thus, the probability distribution of network states mea-
sured from the short-time recording cannot capture many
network activity states as observed in the long-time
recordings (e.g., shown in Fig. 2). Note that the mapping
between the probability distribution and effective inter-
actions, i.e., the lower-triangular matrix L(n)
JP , suggests
that the probability of a highly-active state consists of the
summation of many effective interactions. For example,
in a small network of n = 2 nodes, the probability of the
highly-active state, P (σ1 = 1, σ2 = 1), can be obtained
from Eq. (9) by: log P11 = J0 + J1 + J2 + J12. Therefore,
to obtain an accurate estimation of such highly-active
states, it is important to obtain an accurate estimation
of the summation of the effective interactions.
7
For asynchronous networks, since high-order effective
interactions are small (e.g., shown in Fig. 3), the sum-
mation in the probability of a highly-active state will be
dominated by low-order effective interactions. Note that
low-order effective interactions can be derived from low-
order moments through an iterative scaling algorithm
(see Appendix A for details) in the MEP analysis, and
thus an accurate estimation of low-order moments is es-
sential. The mapping between the probability distribu-
tion of network states and corresponding moments of the
network, i.e., the upper triangular matrix U(n)
PM (Eq. (5)),
shows that low-order moments consist of the summation
of probabilities of many network activity states. For ex-
ample, in a network of n nodes, the first-order moment
of node 1 is the summation of network state probabil-
ities where node 1 is active (number of 2n−1 states in
total from low-active states to highly-active states), that
is hσ1i = P10···0 + P11···0 + · · · + P11···1. Note that low-
active states occurs frequently in an asynchronous net-
work; therefore, the summation in low-order moments
is dominated by probabilities of these low-active states
which can be accurately measured from a short-time
recording, leading to good estimation of low-order mo-
ments.
In addition, we point out that the estimation
error in low-order moments can be further reduced due
to linear summations. Therefore, one can accurately es-
timate the low-order moments and perform the low-order
MEP analysis in a short-time data recording of network
dynamics. It is expected that the low-order MEP distri-
bution provides a good estimate of the probability distri-
bution of network states.
In summary, the low-order MEP distribution can be
obtained by the following procedure: First, calculate low-
order moments from the experimental short-time record-
ing using Eqs.
(1) and (2) for the first-order and the
second-order moments, respectively. Second, the low-
order MEP analysis (with constraints of low-order mo-
ments) is carried out using the iterative scaling algorithm
(see Appendix A for details) to derive the low-order ef-
fective interactions with all higher-order effective inter-
actions set to zero. This determines J(n) in Eq. (10),
except for J0. Third, express the probability distribu-
tion, P(n), as a function of J0 using Eq. (10). Finally,
determine J0 by constraining the summation of all prob-
abilities to equal one. Once all the low-order effective
interactions are determined, Eq (4) is used to determine
the low-order MEP distribution, e.g., P1(Ω) and P2(Ω)
represents for the first-order and the second-order MEP
distribution, respectively.
8
100
10-2
n
o
i
t
c
a
r
e
t
n
I
100
10-2
n
o
i
t
c
a
r
e
t
n
I
100
10-2
n
o
i
t
c
a
r
e
t
n
I
First Second Third Fourth
First Second Third Fourth
First Second Third Fourth
order
(a)
order
(b)
order
(c)
FIG. 3: Hierarchy of effective interactions in the MEP analysis in three dynamical regimes. Mean
absolute values of effective interactions from the first-order to the fourth-order for three different dynamical regimes
(shown in Fig. 1) are plotted, with the standard deviation indicated by the error bar. Data for these three cases are
collected from the three experiments (the same 10 selected neurons in each case) in Fig. 1, respectively, with a long
recording time of 1.2 × 105 s.
Verification for the MEP analysis with short-time
recordings by HH neuronal network model
In this section, we use data from the HH neuronal net-
work model to verify that low-order MEP analysis can
accurately estimate the probability distribution of net-
work states from short-time recordings using the proce-
dure described in the previous section.
We first evolve the HH neuronal network model with
a short simulation time of 120 s and record the spike
trains of the network. Then we estimate the first-order
and the second-order moments from this short recording
and use these two-order moments to estimate both the
first-order and the second-order MEP distributions (Eq.
(4)). Finally, we compare the MEP distributions to the
probability distribution of network states measured from
the long recording of 1.2 × 105 s. As shown in Fig. 4,
for all three dynamical regimes, the probability distri-
bution of the first-order MEP analysis, P1 (green), de-
viates substantially from the measured probability dis-
tribution of network states in the long recording. The
probability distribution of the second-order MEP anal-
ysis, P2 (blue), however, is in excellent agreement with
the measured probability distribution of network states
in the long recording. These results indicate that the
low-order, i.e., second-order, MEP analysis provides an
efficient method to obtain the probability distribution of
network states with short recordings.
Verification for the MEP analysis with short-time
recordings by electrophysiological experiments
In general, it is difficult to obtain a stationary long-
time recording of spike trains from the brain of an awake
animal in vivo. To verify the validity of the MEP anal-
ysis on short recordings, we use the electrophysiologi-
cal experimental data recorded by multi-electrode array
from V1 in anesthetized macaque monkeys in multiple
trials. For each trial, an image stimulus was shown on
the screen for 100 ms, followed by a 200 ms uniform gray
screen [5, 14]. The number of different images is 956
in total and images are presented in pseudo-random or-
der with each presented 20 times. Experimental details
can be found in Appendix E. Here, we focus on the is-
sue of whether the second-order MEP analysis from a
short recording can accurately estimate the probability
distribution of neuronal firing patterns in a long record-
ing. Note that the presenting duration for each image is
only 2 s, which is too short for the recorded spike trains
to have a stable probability distribution. As an alterna-
tive, we put the spike trains recorded during the uniform
gray screen (200 ms in each trial) altogether to obtain a
long recording of 3824 s. We have verified that the spike
trains in such a long recording has a stable probability
distribution. For a short recording, we randomly select
5% length of the long recording, i.e., 191.2 s, and also
verified that the probability distribution in such a short
recording is quite different from that in the long record-
ing. To perform the MEP analysis, we randomly selected
eight neurons' spike train data from experimental mea-
surements.
As shown in Fig. 5a, these eight neurons are in an
asynchronous state. We then calculate effective interac-
tions in the full-order MEP distribution by Eq. (10) using
the measured probability distribution of network states
from the long recording of 3824 s. In Fig. 5b, the average
strength of the kth-order effective interactions is com-
puted as the mean of the absolute value of the kth-order
effective interactions. It can be seen clearly that the aver-
age strength of effective interactions of high-orders (≥ 3)
are almost one order of magnitude smaller (in the abso-
lute value) than that of the first-order and the second-
order effective interactions. Next, we consider a short
recording of 191.2 s and estimate the first-order and the
9
10-1
10-3
y=x
P1
P2
10-1
10-3
y=x
P1
P2
10-1
10-3
y=x
P1
P2
10-5
10-5
10-3
10-5
10-5
10-1
10-3
10-1
10-3
10-1
Long measurement
Long measurement
Long measurement
(a)
(b)
(c)
FIG. 4: First-order and second-order MEP distribution in comparison with the probability
distribution of network states. The frequency of each firing state from the distribution of the MEP analysis, P2
(blue) and P1 (green), is plotted against the frequency measured from the long recording (1.2 × 105 s). Data for three
different dynamical regimes (shown in Fig. 1) are collected from the three experiments (the same 10 selected neurons
in each case) in Fig. 1, respectively. Here, the MEP analysis is performed with a short recording (1.2 × 102 s).
second-order moments from this short recording and use
these moments to estimate the second-order MEP distri-
butions (Eq. (4)). As shown in Fig.5c, the probability
distribution estimated by the second-order MEP analysis
is in excellent agreement with the measured probability
distribution of network states in the long recording. How-
ever, the probability distribution measured in the short
recording clearly deviates from the probability distribu-
tion measured in the long recording.
DISCUSSION
The second-order MEP analysis has been used to in-
fer the probability distribution of network states under
various conditions, such as under spontaneous activity of
neuronal networks [27, 34] or visual input [25]. Since the
second-order MEP distribution can be obtained by max-
imizing the Shannon entropy with measured constraints
of only the first-order and the second-order moments,
the curse of dimensionality is circumvented and an accu-
rate estimation of the probability distribution of network
states is provided. A series of works have been initi-
ated to address various aspects of this approach, includ-
ing explorations of fast algorithms [2, 19], the inference
of spatial-temporal correlations [16, 18, 28, 34, 37], and
the characterization of network functional connectivity
[1, 6, 9, 21, 35, 38].
In addition to spiking neuronal networks [25], the MEP
analysis has been applied to analyzing various types of bi-
nary data. These applications include, for example, func-
tional magnetic resonance imaging data [35], in which
each brain region is considered to be either active or
silent, and stock market data [3], in which the price of
each stock is considered to be either increasing or not.
However, for all these binary data, long-time recordings
are often impractical to be obtained.
In this work, we begin by showing that there exist in-
vertible linear transforms among the probability distri-
bution of network states, the moments, and the effective
interactions in the MEP analysis for a general network of
nodes with binary dynamics. Based on these transforms,
we show that the second-order MEP analysis gives rise
to an accurate estimate of the probability distribution of
network states with a short-time recording. Finally, using
data from both simulated HH neuronal network model
and the electrophysiological experiment, we demonstrate
the good performance of the second-order MEP analysis
with short recordings. Therefore, the applicability of the
second-order MEP analysis in practical situations could
be significantly improved.
Finally, we point out that there are also some lim-
itations on low-order MEP analysis. First,
low-order
MEP analysis is often insufficient to accurately recon-
struct the probability distribution of network states in
synchronized networks since the high-order effective in-
teractions in such cases are often not small [36]. Second,
as both the order of moment constraints and the network
size, n, increase, the existing algorithms to estimate ef-
fective interactions for a large network become very slow
[19, 26]. Because the number all network states, i.e., 2n,
is too large when n is a large number, the existing algo-
rithms estimate moments of the MEP distribution using
Monte Carlo sampling or its variants from the MEP dis-
tribution, which are often very slow when the dimension
of the distribution is high [19, 26]. Therefore, how to effi-
ciently characterize the statistical properties of strongly-
correlated, e.g., nearly synchronized, network dynamics
and how to develop fast numerical algorithms for the ap-
plication of the MEP analysis in large-scale networks re-
main interesting and challenging issues for future studies.
8
6
4
2
x
e
d
n
i
n
o
r
u
e
N
0
n
o
i
t
c
a
r
e
t
n
I
100
10-1
10-2
100
50
time (ms)
(a)
10
y=x
Short
P2
100
10-2
10-4
First Second Third Fourth
10-4
10-2
100
Order
(b)
Long measurement
(c)
FIG. 5: Electrophysiological verification for the second-order MEP analysis with short-time
recordings. Spike trains of randomly selected eight neurons are recorded from V1 in anesthetized macaque
monkeys (see Appendix E). (a) A short bar indicates that the neuron with certain index fires at certain time. (b)
Mean absolute values of effective interactions from the first-order to the fourth-order are plotted, with the standard
deviation indicated by the error bar, computed by the long recording of 3824 s. (c) The frequency of each firing state
from the distribution measured in the short recording of 191.2 s (magenta) and the distribution of the second order
MEP analysis, P2 (blue), is plotted against the frequency measured from the long recording of 3824 s.
ACKNOWLEDGMENTS
hσiσj iP2(Ω) and hσiiP (Ω), hσiσjiP (Ω), the values of Ji and
Jij are adjusted by an iterative procedure:
This work was supported by NSFC-11671259, NSFC-
11722107, NSFC-91630208 and Shanghai Rising-Star
Program-15QA1402600 (D.Z.); by NSF DMS-1009575
and NSFC-31571071 (D.C.); by Shanghai 14JC1403800,
15JC1400104, and SJTU-UM Collaborative Research
Program (D.C. and D.Z.); and by the NYU Abu Dhabi
Institute G1301 (Z.X., D.Z., and D.C.).
Appendix A: The iterative scaling algorithm
We briefly describe the widely-used numerical algo-
rithm for the estimation of effective interactions from mo-
ments. More details about this algorithm can be found
in Ref.
[34]. For illustration, we present the procedure
to obtain the second-order MEP distribution, P2(Ω).
The interactions are initialized by: Ji = hσiiP (Ω) and
Jij = hσiσjiP (Ω), where h·iP (Ω) denotes the expectation
with respect to the measured probability distribution of
network states in data recording, P (Ω). The expected
values of the individual means hσiiP2(Ω) and pairwise
correlations hσiσj iP2(Ω) with respect to the second-order
MEP distribution P2(Ω) can be determined by
hσiiP2(Ω) ≡
2n
Xl=1
σi(Ωl)P2(Ωl),
J new
i
= J old
i + αsign(cid:16)hσiiP (Ω)(cid:17) log
ij + αsign(cid:16)hσiσjiP (Ω)(cid:17) log
hσiiP (Ω)
hσiiP2(Ω)
,
hσiσjiP (Ω)
hσiσj iP2(Ω)
,
J new
ij = J old
where the constant α is used to maintain the stability of
the iteration. We use α = 0.75 as in Ref. [34].
Appendix B: The Hodgkin-Huxley neuron model
The HH model is described as follows. The dynamics of
the membrane potential of the ith neuron, Vi, is governed
by [32, 33]
C
dXi
dt
with
,
i
= INa + IK + IL + I input
dVi
dt
INa = −(Vi − VNa)GNahim3
i ,
IK = −(Vi − VK)GKn4
i ,
IL = −(Vi − VL)GL,
= (1 − Xi)αX (Vi) − XiβX (Vi),
(B1)
where the gating variable X = m, n, h and
hσiσjiP2(Ω) ≡
2n
Xl=1
σi(Ωl)σj(Ωl)P2(Ωl),
αn(Vi) =
αm(Vi) =
0.1 − 0.01Vi
exp(1 − 0.1Vi) − 1
2.5 − 0.1Vi
exp(2.5 − 0.1Vi) − 1
,
βn(Vi) =
5
40
exp(−Vi/80),
, βm(Vi) = 4 exp (−Vi/18),
where σi(Ωl) is the state of the ith node in the network
state Ωl. To improve the agreement between hσiiP2(Ω),
αh(Vi) = 0.07 exp(−Vi/20),
βh(Vi) =
1
exp(3 − 0.1Vi) + 1
.
i
The current I input
describes inputs to the ith neuron
coming from the external drive of the network, as well
as interactions between neurons in the network, I input
=
G)GI
i + I I
I E
i,
where I E
i are excitatory and inhibitory input cur-
rents, respectively, and V E
G are their correspond-
ing reversal potentials. The dynamics of the conduc-
tance, GQ
i , for Q = E, I are described as follows,
i with I E
i and I I
i = −(Vi − V E
i = −(Vi − V I
G and V I
i and I I
G )GE
i
dGQ
i
dt
dH Q
i
dt
= −
= −
GQ
i
σQ
G
H Q
i
σQ
H
+ H Q
i ,
+Xk
F Q
i δ(t − T F
j
),
Cijg(V pre
i,k) +Xj6=i
j − 85)/2)(cid:1) , where F Q
j
i = f , F I
) = 1/(cid:0)1 + exp(−(V pre
with g(V pre
i
is the strength of the external Poisson input of rate µi
to neuron i with T F
i,k being the time of the kth input
event. We use F E
i = 0, µi = µ for all the
neurons in our simulation. The parameter Cij describes
the coupling strength from the jth presynaptic neuron
to the ith neuron, and V pre
is the membrane potential
of the jth presynaptic neuron. The adjacency matrix,
W = (wij ), describes the neuronal network connectivity
structure and Cij = wij C QiQj , where Qi, Qj is chosen as
E or I, indicating the neuron type of the ith neuron and
the jth neuron (C QiQj is one of CEE, CEI, CIE, CII).
The value wij = 1 if there is a directed coupling from the
jth presynaptic neuron to the ith postsynaptic neuron
and wij = 0 otherwise.
j
In this study, the values of parameters in the above
conductance equations are chosen as: VNa = 115 mV,
VK = −12 mV, VL = 10.6 mV (the resting potential
of a neuron is set to 0 mV), GNa = 120 mS · cm−2,
GK = 36 mS · cm−2, GL = 0.3 mS · cm−2, the membrane
capacity C = 1 µF · cm−2, V E
G = −15 mV,
σE
G = 0.5 ms, σE
H =
7.0 ms. We keep the Poisson input parameters fixed dur-
ing each single simulation.
G = 0.5 ms, and σI
H = 3.0 ms, σI
G = 65 mV, V I
In our numerical simulation, an explicit fourth order
Runge-Kutta method is used [32, 33] with time step
∼ 0.03 ms. The spike train data were obtained with a
sufficiently high sampling rate, e.g., 2 kHz.
Appendix C: Proof of one-to-one mapping between
the probability distribution and moments
We prove that there exists a full-rank matrix U(n)
PM that
transforms from P(n) to M(n), i.e.,
U(n)
PMP(n) = M(n),
(C1)
for a network of any size n. As an illustration with a
network of n = 2 nodes, the expectation of σ1, σ2, σ1σ2
can be obtained by
11
,
(C2)
1 1 1 1
0 1 0 1
0 0 1 1
0 0 0 1
P00
P10
P01
P11
=
1
hσ1i
hσ2i
hσ1σ2i
PMP(2) = M(2). Clearly, from Eq.
i.e., U(2)
(C2), U(2)
PM
is of full rank. We now prove the above result for any
n by mathematical induction. Suppose U(k)
PM is of full
rank. Then P(k+1) and M(k+1) can be decomposed into
two parts with equal length of 2k and U(k+1)
PM can be
decomposed into four sub-matrices with the dimension
of each sub-matrix being 2k × 2k as follows:
" U(k+1)
U(k+1)
11
21
12
U(k+1)
U(k+1)
22
#" P(k+1)
P(k+1)
1
2
# =" M(k+1)
M(k+1)
1
2
# ,
where
P(k+1)
1
=
P000···000
P100···000
P010···000
P110···000
...
P111···110
,
P(k+1)
2
=
P000···001
P100···001
P010···001
P110···001
...
P111···111
.
The expression of M(k+1)
of M(k) in Eq. (6), i.e., M(k+1)
be expressed as follows,
1
1
is the same as the expression
can
= M(k), and M(k+1)
2
M(k+1)
2
=
h1 · σk+1i
hσ1σk+1i
hσ2σk+1i
hσ1σ2σk+1i
...
j=3 σj σk+1E
j=3 σj σk+1E
Dσ2Qk
Dσ1σ2Qk
.
i
1
In the representation of the base-2 number system of k +
1 digits, the first digit of the representation of i − 1,
indicating the state of node k + 1, is 0 for 1 ≤ i ≤ 2k
and 1 for 2k + 1 ≤ i ≤ 2k+1. For any i with 1 ≤ i ≤ 2k,
suppose p(k)
is the probability of state (σ1, σ2, · · · , σk),
i.e., Pσ1σ2··· ,σk , then, the ith entry of P(k+1)
and P(k+1)
are Pσ1σ2··· ,σk ,0 and Pσ1σ2··· ,σk ,1, respectively. Thus, the
states of nodes from the first to the 2kth are all the same
for P(k), P(k+1)
is the same
as M(k), i.e., their expressions only consider nodes from
the first to the 2kth, the contribution from both P(k+1)
and P(k+1)
is the same as the contribution
from P(k) to M(k), i.e., U(k+1)
=
U(k)
PM. Since the state of node k + 1 is 0 in all entries of
P(k+1)
, i.e.,
U(k+1)
, there is no contribution of P(k+1)
= 0.
PM and U(k+1)
. Since M(k+1)
and P(k+1)
to M(k+1)
to M(k+1)
= U(k)
11
12
1
2
1
1
1
2
1
2
1
2
21
2
, the contribution from P(k+1)
Similarly, since the state of node k + 1 is 1 in all en-
to M(k+1)
tries of P(k+1)
only depends on the states of nodes from the first to the
2kth -- there is a one-to-one correspondence of each en-
try between P(k+1)
and
M(k) as aforementioned. Thus, U(k+1)
PM and we
can obtain a recursive relation, that is,
and P(k), and between M(k+1)
2
= U(k)
22
2
2
2
U(k+1)
PM =" U(k)
PM U(k)
0(k) U
PM # ,
PM
(k)
(C3)
where 0(k) is the zero matrix with dimension 2k × 2k.
Thus, U(k+1)
PM is also a matrix of full rank. By induction,
U(n)
PM is of full rank for any n.
Appendix D: Proof of the recursive relation among
effective interactions
For a network of n nodes, we prove the following recur-
sive relation in the full-order MEP analysis using math-
ematical induction:
J12···(k+1) = J 1
12···k − J12···k,
(D1)
where the term J 1
12...k is obtained from the kth order
effective interaction J12...k by changing the state of the
(k + 1)st node in it.
From the mapping between effective interactions and
(10)), it can be seen
the probability distribution (Eq.
k
12
that the effective interactions is simply a linear combina-
tion of logarithm of the probability distribution. To ex-
press this explicitly, we first introduce the notation H l
m
as
H l
m = XΩ∈Λl
m
log P (Ω),
(D2)
where Λl
j ≤ n}, 0 ≤ l ≤ m ≤ n.
m = {(σ1, σ2, · · · , σn)Pm
i=1 σi = l; σj = 0, m <
We then show that if the kth-order (1 ≤ k ≤ n) effec-
tive interaction, J12···k, can be expressed as
J12···k =
k
(−1)k−iH i
k,
Xi=0
(D3)
i
k+1 of Λ
k+1. For 0 < i < k + 1, we can split H i
Eq. (D1) is also valid as follows. For 1 ≤ i ≤ k, there are
i nodes out of k + 1 nodes active in the states described
by H i
into two terms, one is H i
inactive, the other term is W i−1
the (k + 1)st node is active. We also define W k
From Eq. (D3), we have
k+1
k, where the (k + 1)st node is
k, where
k ≡ H k+1
k+1 .
k ≡ H i
k+1 − H i
J12···(k+1) =
k+1
Xi=0
(−1)k+1−iH i
k+1.
By using H i
k+1 = W i−1
k + H i
k and H 0
k+1 = H 0
k , we have
J12···(k+1) = H k+1
k+1 +
(−1)k+1−i(W i−1
k + H i
k) − (−1)kH 0
k+1
(−1)k+1−iW i−1
k − (−1)kH 0
k+1 −
(−1)k−iH i
k
k
Xi=1
k
Xi=1
Xi=1
= W k
k +
k
=
Xi=0
(−1)k−iW i
k −
= J 1
12···k − J12···k,
(−1)k−iH i
k
k
Xi=0
where J 1
12···k is the quantity which switches the state of
the (k + 1)st node in the kth-order effective interaction
J12...k from inactive to active. Therefore, the recursive
relation (Eq.
(D3) is valid for
1 ≤ k ≤ n.
(D1)) is proved if Eq.
in the main text, we have
J1 = log
P10···0
P00···0
1 − H 0
1 .
= H 1
We next prove the validity of Eq. (D3) by mathemati-
cal induction as follows. For k = 1, as shown in Eq. (11)
We next want to prove that Eq. (D3) holds for k =
K + 1. We begin by show that an arbitrary gth-order
Therefore, Eq. (D3) is valid when k = 1. Now, we as-
sume Eq. (D3) is valid for k ≤ K.
(g ≤ K) effective interaction Ji1···ig can be expressed as
follows:
Ji1···ig =
g
Xi=0
(−1)g−iH i
Ag ,
g ≤ K
(D4)
where Ag = {i1, · · · , ig},
nodes but not the considered i nodes. Therefore, the
choice number of these g − i nodes is C g−i
K+1−i. Then, we
have
13
J12···(K+1) = H K+1
K+1 −H 0
K+1−
K
g
Xg=1
Xi=0
(−1)g−iC g−i
K+1−iH i
K+1.
(D7)
H l
Ag = XΩ∈Λl
Ag
log P (Ω),
For K + 1 > l > 0, the coefficient of H l
K+1 is
K
(−1)g−lC g−l
K+1−l = (−1)K+1−l.
(D8)
Ag = {(σi1 , σi2 , · · · , σin )Pij ∈Ag
and Λl
σij = l; σij =
0, g < j ≤ n}. To show the validity of Eq. (D4), we
can permute the neuronal indexes from {1, 2, · · · , n} to
{i1, i2, · · · , in} by the mapping j → ij for 1 ≤ j ≤ n.
Since Eq. (D3) is valid for g ≤ K by assumption, Eq.
(D4) is also valid.
Next, we study the relation between J12···(K+1) and
the effective interactions whose orders are smaller than
K + 1. By substituting Ω = (1, 1, · · · , 1, 0, · · · , 0) (nodes
from 1 to K + 1 are active and nodes from K + 2 to n
are inactive) into the full-order MEP analysis, we obtain
J12···(K+1) +
K
K+1
Xg=1
Xi1<···<ig
Ji1···ig = H K+1
K+1 − H 0
K+1. (D5)
For g ≤ K, from Eq (D4), by the induction assumption,
we have
K+1
Xi1<···<ig
Ji1···ig =
g
Xi=0
(−1)g−iC g−i
K+1−iH i
K+1,
(D6)
where C g−i
K+1−i is the number of the selection of g−i terms
from all the possible K + 1 − i choices. Since Ji1···ig is the
gth-order effective interaction, as in Eq. (D4), the sign
of the logarithm probability of a state in which there are
i nodes active is (−1)g−i. To consider the coefficient of
H i
K+1 in the right hand side of Eq. (D6), we can consider
that for each group of i nodes, how many groups of g
nodes in the left hand side of Eq. (D6) containing these
considered i nodes, where g ≥ i.
If there are g nodes
containing the considered i nodes, then, there are only
g − i nodes unknown. These g − i nodes can be chosen
from the group of nodes which belongs to the total K + 1
−
Xg=l
The coefficient of H 0
K+1 is
− 1 −
K
Xg=1
(−1)gC g
K+1 = (−1)K+1.
(D9)
Through Eqs. (D7), (D9) and (D8), we obtain
K+1
J12···(K+1) =
Xi=0
(−1)K+1−iH i
K+1.
(D10)
That is, Eq. (D3) is valid for k = K + 1. By induction,
we obtain that Eq. (D3) holds for any integer k with
1 ≤ k ≤ n. Therefore, we prove the validity of Eq. (D1).
Appendix E: Electrophysiological experiments
The data were collected in the Laboratory of Adam
Kohn at the Albert Einstein College of Medicine and
downloaded from crcns.org (pvc-8) [14]. These data con-
sist of multi-electrode recordings from V1 in anesthetized
macaque monkeys, while natural images and gratings
were flashed on the screen. Recordings were performed
using the "Utah" electrode array and spike-sorting algo-
rithm was used to determine spike trains corresponding
to each single neuron. Natural images were presented at
two sizes, 3 − 6.7 degrees and windowed to 1 degree, to
quantify surround modulation. All stimuli (956 in total)
were displayed in pseudo-random order for 100 ms each,
followed by a 200 ms uniform gray screen. Each stimulus
was presented 20 times. Experimental procedures and
stimuli are fully described in Ref. [5].
[1] J. Barton and S. Cocco, Ising models for neural ac-
tivity inferred via selective cluster expansion: structural
and coding properties, Journal of Statistical Mechanics:
Theory and Experiment, 2013 (2013), p. P03002.
[2] T. Broderick, M. Dudik, G. Tkacik, R. E.
Schapire, and W. Bialek, Faster
solutions of
the inverse pairwise ising problem, arXiv preprint
arXiv:0712.2437, (2007).
[3] T. Bury, Statistical pairwise interaction model of stock
market, The European Physical Journal B, 86 (2013),
p. 89.
[4] A. Cavagna, I. Giardina, F. Ginelli, T. Mora,
D. Piovani, R. Tavarone, and A. M. Walczak, Dy-
namical maximum entropy approach to flocking, Physical
Review E, 89 (2014), p. 042707.
[5] R. Coen-Cagli, A. Kohn, and O. Schwartz, Flexible
gating of contextual influences in natural vision, Nature
neuroscience, (2015).
[6] B. Dunn, M. Mørreaunet, and Y. Roudi, Correla-
tions and functional connections in a population of grid
cells, PLoS Comput Biol, 11 (2015), p. e1004052.
[7] A. Golan, G. G. Judge, and D. Miller, Maximum
entropy econometrics: Robust estimation with limited
data, Wiley New York, 1996.
[8] A. Hernando, R. Hernando, A. Plastino, and
A. Plastino, The workings of the maximum entropy
principle in collective human behaviour, Journal of The
Royal Society Interface, 10 (2013), p. 20120758.
[9] J. Hertz, Y. Roudi, and J. Tyrcha, Ising models
for inferring network structure from spike data, arXiv
preprint arXiv:1106.1752, (2011).
[10] E. T. Jaynes, Information theory and statistical me-
chanics, Physical review, 106 (1957), p. 620.
[11] M. P. Karlsson and L. M. Frank, Awake replay of
remote experiences in the hippocampus, Nature neuro-
science, 12 (2009), p. 913.
[12] S. Khatiwala, F. Primeau, and T. Hall, Reconstruc-
tion of the history of anthropogenic co2 concentrations in
the ocean, Nature, 462 (2009), pp. 346 -- 349.
[13] D. C. Knill and A. Pouget, The bayesian brain: the
role of uncertainty in neural coding and computation,
TRENDS in Neurosciences, 27 (2004), pp. 712 -- 719.
[14] A. Kohn and R. Coen-Cagli, Multi-electrode
to
gratings., CRCNS.org
recordings of anesthetized macaque v1 responses
static natural
http://dx.doi.org/10.6080/K0SB43P8, (2015).
images
and
[15] A. B. Mantsyzov, A. S. Maltsev, J. Ying, Y. Shen,
G. Hummer, and A. Bax, A maximum entropy ap-
proach to the study of residue-specific backbone angle dis-
tributions in α-synuclein, an intrinsically disordered pro-
tein, Protein Science, 23 (2014), pp. 1275 -- 1290.
[16] O. Marre, S. El Boustani, Y. Fr´egnac, and A. Des-
texhe, Prediction of spatiotemporal patterns of neural
activity from pairwise correlations, Physical review let-
ters, 102 (2009), p. 138101.
[17] A. S. Morcos and C. D. Harvey, History-dependent
variability in population dynamics during evidence accu-
mulation in cortex, Nature Neuroscience, (2016).
[18] H. Nasser and B. Cessac, Parameter estimation for
spatio-temporal maximum entropy distributions: Applica-
tion to neural spike trains, Entropy, 16 (2014), pp. 2244 --
2277.
[19] H. Nasser, O. Marre, and B. Cessac, Spatio-temporal
spike train analysis for large scale networks using the
maximum entropy principle and monte carlo method,
Journal of Statistical Mechanics: Theory and Experi-
ment, 2013 (2013), p. P03006.
[20] I. E. Ohiorhenuan, F. Mechler, K. P. Purpura,
A. M. Schmid, Q. Hu, and J. D. Victor, Sparse cod-
ing and high-order correlations in fine-scale cortical net-
works, Nature, 466 (2010), pp. 617 -- 621.
[21] Y. Roudi, J. Tyrcha, and J. Hertz, Ising model for
neural data: model quality and approximate methods for
extracting functional connectivity, Physical Review E, 79
(2009), p. 051915.
[22] D. Sakakibara, A. Sasaki, T. Ikeya, J. Hamatsu,
T. Hanashima, M. Mishima, M. Yoshimasu,
N. Hayashi, T. Mikawa, M. Walchli, et al., Pro-
tein structure determination in living cells by in-cell nmr
spectroscopy, Nature, 458 (2009), pp. 102 -- 105.
[23] S. Saremi and T. J. Sejnowski, Hierarchical model of
natural images and the origin of scale invariance, Pro-
14
ceedings of the National Academy of Sciences, 110 (2013),
pp. 3071 -- 3076.
[24] L. Saulis and V. Statulevicius, Limit theorems for
large deviations, vol. 73, Springer Science & Business Me-
dia, 2012.
[25] E. Schneidman, M. J. Berry, R. Segev, and
W. Bialek, Weak pairwise correlations imply strongly
correlated network states in a neural population, Nature,
440 (2006), pp. 1007 -- 1012.
[26] J. Shlens, G. D. Field,
J. L. Gauthier,
M. Greschner, A. Sher, A. M. Litke, and
E. Chichilnisky, The structure of
large-scale syn-
chronized firing in primate retina, The Journal of
Neuroscience, 29 (2009), pp. 5022 -- 5031.
[27] J. Shlens, G. D. Field, J. L. Gauthier, M. I.
Grivich, D. Petrusca, A. Sher, A. M. Litke, and
E. Chichilnisky, The structure of multi-neuron firing
patterns in primate retina, The Journal of neuroscience,
26 (2006), pp. 8254 -- 8266.
[28] J. Shlens, F. Rieke, and E. Chichilnisky, Synchro-
nized firing in the retina, Current opinion in neurobiol-
ogy, 18 (2008), pp. 396 -- 402.
[29] T. Squartini,
I. van Lelyveld, and D. Gar-
laschelli, Early-warning signals of topological collapse
in interbank networks, Scientific reports, 3 (2013).
[30] G. J. Stephens and W. Bialek, Statistical mechan-
ics of letters in words, Physical Review E, 81 (2010),
p. 066119.
[31] J. I. Su lkowska, F. Morcos, M. Weigt, T. Hwa,
and J. N. Onuchic, Genomics-aided structure predic-
tion, Proceedings of the National Academy of Sciences,
109 (2012), pp. 10340 -- 10345.
[32] Y. Sun, D. Zhou, A. Rangan, and D. Cai, Pseudo-
lyapunov exponents and predictability of hodgkin-huxley
neuronal network dynamics, Journal of computational
neuroscience, 28 (2010), pp. 247 -- 266.
[33] Y. Sun, D. Zhou, A. V. Rangan, and D. Cai, Library-
based numerical reduction of the hodgkin -- huxley neuron
for network simulation, Journal of computational neuro-
science, 27 (2009), p. 369.
[34] A. Tang, D. Jackson, J. Hobbs, W. Chen, J. L.
Smith, H. Patel, A. Prieto, D. Petrusca, M. I.
Grivich, A. Sher, et al., A maximum entropy model
applied to spatial and temporal correlations from corti-
cal networks in vitro, The Journal of Neuroscience, 28
(2008), pp. 505 -- 518.
[35] T. Watanabe, S. Hirose, H. Wada, Y.
Imai,
T. Machida, I. Shirouzu, S. Konishi, Y. Miyashita,
and N. Masuda, A pairwise maximum entropy model
accurately describes resting-state human brain networks,
Nature communications, 4 (2013), p. 1370.
[36] Z.-Q. J. Xu, G. Bi, D. Zhou, and D. Cai, A dynamical
state underlying the second order maximum entropy prin-
ciple in neuronal networks, Communications in Mathe-
matical Sciences, 15 (2017), pp. 665 -- 692.
[37] F.-C. Yeh, A. Tang, J. P. Hobbs, P. Hottowy,
W. Dabrowski, A. Sher, A. Litke, and J. M. Beggs,
Maximum entropy approaches to living neural networks,
Entropy, 12 (2010), pp. 89 -- 106.
[38] S. Yu, D. Huang, W. Singer, and D. Nikoli´c, A
small world of neuronal synchrony, Cerebral cortex, 18
(2008), pp. 2891 -- 2901.
[39] D. Zhou, Y. Xiao, Y. Zhang, Z. Xu, and D. Cai,
Granger causality network reconstruction of conductance-
based integrate-and-fire neuronal systems, PloS one, 9
(2014), p. e87636.
15
|
1510.03279 | 1 | 1510 | 2015-10-12T13:49:10 | Skewness and Kurtosis in Statistical Kinetics | [
"physics.bio-ph",
"cond-mat.stat-mech",
"physics.chem-ph"
] | We obtain lower and upper bounds on the skewness and kurtosis associated with the cycle completion time of unicyclic enzymatic reaction schemes. Analogous to a well known lower bound on the randomness parameter, the lower bounds on skewness and kurtosis are related to the number of intermediate states in the underlying chemical reaction network. Our results demonstrate that evaluating these higher order moments with single molecule data can lead to information about the enzymatic scheme that is not contained in the randomness parameter. | physics.bio-ph | physics |
Skewness and Kurtosis in Statistical Kinetics
1 II. Institut fur Theoretische Physik, Universitat Stuttgart, 70550 Stuttgart, Germany
2 Max Planck Institute for the Physics of Complex Systems, Nothnizer Strasse 38, 01187 Dresden,Germany
Andre C. Barato1,2 and Udo Seifert1
We obtain lower and upper bounds on the skewness and kurtosis associated with the cycle com-
pletion time of unicyclic enzymatic reaction schemes. Analogous to a well known lower bound on
the randomness parameter, the lower bounds on skewness and kurtosis are related to the number
of intermediate states in the underlying chemical reaction network. Our results demonstrate that
evaluating these higher order moments with single molecule data can lead to information about the
enzymatic scheme that is not contained in the randomness parameter.
PACS numbers: 87.14.ej, 05.40.-a
In enzyme kinetics one typically studies the relation
between the average rate of product formation and pa-
rameters like the substrate concentration [1]. If the num-
ber of enzymes is large, fluctuations of this rate of prod-
uct formation are small and can be neglected. Many
different enzymatic schemes can lead to the same phe-
nomenological expression for the rate of product forma-
tion as a function of the substrate concentration. There-
fore, in general, this quantity is not sufficient to pro-
vide essential information about the underlying enzy-
matic scheme, like, for example, the number of interme-
diate states.
Statistical kinetics [2 -- 6] is a novel field where one tries
to infer the underlying chemical reaction scheme from
measurements of the, typically, large fluctuations in sin-
gle molecule experiments [7 -- 12]. For example, experi-
mental data on the fluctuations related to the time to
complete an enzymatic cycle have been used to infer
properties of the enzymatic mechanism of the packing
motor in bactheriophage φ29 [13]. Related analysis for
kinesin and myosin can be found in [14 -- 18].
The time to complete an enzymatic cycle is a key quan-
tity in statistical kinetics. The size of its fluctuations is
quantified by the randomness parameter. The inverse of
this randomness parameter bounds the number of inter-
mediate states of a unicyclic enzymatic scheme [2]. This
main result of statistical kinetics implies that measure-
ments of the randomness parameter lead to a lower bound
on the number of intermediate states. Further important
information that can be inferred from this randomness
parameter is whether the chemical reaction network con-
stitutes of a cycle or is more complex. Specifically, a
randomness parameter larger than 1 implies that the re-
action scheme is not simply a cycle [19 -- 23]. Furthermore,
similar to the Michaelis-Menten expression for the rate of
product formation, the randomness parameter can also
be written as a function of the substrate concentration
[3] (see also [26 -- 28]).
The randomness parameter is related to the second
moment of the cycle completion time distribution, i.e., it
is the variance divided by the mean square of this distri-
bution. A natural and important question for statistical
kinetics is whether higher order moments can provide
further information about the enzymatic scheme. This
view is expressed in the recent review by Moffit and Bus-
tamante [2]: "... some of the most exciting advances in
statistical kinetics may come from generalizations of the
randomness parameter to higher moments ...".
In this letter, we analyze the third and fourth stan-
dardized moments associated with the probability distri-
bution of the cycle completion time, namely, the skewness
and kurtosis, respectively. We show that similar to the
randomness parameter these higher order moments have
upper and lower bounds in unicyclic enzymatic schemes,
with the lower bound depending on the number of in-
termediate states. Skewness and kurtosis can be used to
derive a stronger lower bound on the number of interme-
diate states and to identify the presence of extra cycles or
an extra state in the enzymatic scheme more effectively.
We consider a generic unicyclic enzymatic scheme of
the form
1
k+
1−−⇀↽−−k−
1
2
k+
2−−⇀↽−−k−
2
3 . . . N − 1
k+
N −1−−−⇀↽−−−k−
N −1
N
k+
N−−⇀ 1.
(1)
1
k+
k+
ES
E + S
1−−⇀↽−−k−
where k±
are transition rates and the numbers i =
i
1, 2, . . . , N represent different states of an enzyme E. For
example, the Michaelis-Menten scheme with N = 2 reads
2−−⇀ E + P. The last transition is taken to
be irreversible, i.e., k−
N = 0, and for the reminder of the
paper we fix the time-scale by setting k+
N = 1. The time
evolution of the probability of being in state i at time τ ,
Pi(τ ), is governed by the master equation dP/dτ = LP,
where P is a vector with dimension N and components
Pi(τ ). The stochastic matrix L is given by the z = 0 case
of the matrix
k+
i ezδi,N
k−
i
−k+
0
i − k−
i−1
if j = i + 1
if j = i − 1
if i = j
otherwise,
(2)
[L(z)]ji ≡
where z is a real variable, δi,N denotes the Kronecker
delta, and we assume periodic boundary conditions, i.e.,
i + 1 = 1 for i = N and i − 1 = N for i = 1. This
z-dependent matrix is needed for the calculations below.
The probability that a cycle is completed at time t is
The randomness parameter for the model in eq (1) is
denoted by ψ(t). Its moment of order n is defined as
bounded by [2]
2
tn ≡Z ∞
0
ψ(t)tn.
(3)
1/N ≤ R ≤ 1.
(11)
The main quantity in statistical kinetics is the random-
ness parameter
R ≡ σ2/t2
1,
(4)
where σ ≡ pt2 − t2
dardized moments are the skewness
1. The third and fourth order stan-
S ≡ (t3 − 3t1σ2 − t3
1)/σ3,
and the kurtosis
K ≡ (t4 − 4t3t1 + 6σ2t2
1 + 3t4
1)/σ4.
(5)
(6)
The moments of ψ(t) can be obtained from the gener-
ating function
as
ψ(s) ≡Z ∞
0
ψ(t)e−stdt
tn = (−1)n dn ψ
dsn .
(7)
(8)
As shown in [26] this generating function for the scheme
in eq (1) can be calculated explicitly. The expression for
ψ(s) involves the determinant of the matrix sI − L(z),
where I is the identity matrix. Specifically, in terms of
the coefficients {cj(z)}, defined through
det{sI − L(z)} ≡
N
Xj=0
cj(z)sj,
(9)
the generating function is given by [26]
N
1 k+
2 . . . k+
(10)
cjsj + Γ+
,
ψ(s) = Γ+,
Xj=1
where Γ+ ≡ k+
N and cj for j 6= 0 is independent
of z. This expression comes from a relation between ψ(t)
and the probability of completing M cycles up to time
t that is explained in [29]. The quite involved expres-
sion of these coefficients for arbitrary N is provided in
[26]. These coefficients can also be calculated from an
iterative procedure developed in [6]. Hence, we can ob-
tain an expression for the skewness and kurtosis in terms
of the transition rates with the following procedure: we
calculate the coefficients in eq (9), obtain the generating
function with eq (10) and calculate the first four moments
defined in eq (3), which finally lead to the skewness and
kurtosis with eqs (5) and (6), respectively.
These bounds are of central importance in statistical ki-
netics. First, measurements of R thus lead to the lower
bound N R
min ≡ 1/R on the number of intermediate states
in the enzymatic scheme. Second, a randomness param-
eter that is larger than 1 indicates that the underlying
enzymatic scheme is not just a simple cycle as in eq (1).
As our first main result, we can show that skewness
and kurtosis for the generic unicyclic network in eq (1)
are also bounded. These bounds are given by
2/√N ≤ S ≤ 2
for the skewness, and
6/N + 3 ≤ K ≤ 9
(12)
(13)
for the kurtosis. These results are based on strong nu-
merical evidence obtained in the following way. We calcu-
lated the skewness and kurtosis following the procedure
explained after eq (10) as functions of the 2N − 2 free
transition rates from N = 2 up to N = 8. With numer-
ical minimization (maximization), we obtain that they
are minimized (maximized) exactly at the value quoted
as lower (upper) bounds in eqs (12) and (13). Further-
more, we have also evaluated S and K at randomly cho-
sen transition rates and observe that they fulfill the above
bounds. This numerical evidence for the bounds in eqs
(12) and (13) is further discussed in [29], where we also
provide analytical evidence that includes a proof of the
bounds for the case N = 2.
For all three quantities, randomness parameter, skew-
ness, and kurtosis, the lower bound is reached when all
forward rates are k+
i = 1 and all backward rates are
k−
In this case ψ(t) = tN −1e−t/(N − 1)! is the
i = 0.
Erlang distribution [32]. The upper bounds are reached
when the rates are chosen such that ψ(t) becomes the ex-
ponential distribution, which is the distribution for the
case N = 1.
We now show that evaluating skewness S and kurtosis
K can lead to further information on the minimal number
of states that can not be obtained from the randomness
parameter R. A first presumption might be that the
bounds N S
min ≡ 6/(K − 3), as obtained
from eq (12) and eq (13), respectively, could be sharper
than N R
min. However, a thorough numerical investigation
up to N = 8 shows that this presumption is incorrect,
and the inequalities
min ≡ 4/S 2 or N K
N R
min ≥ N S
min ≥ N K
min,
(14)
hold. Hence, the randomness parameter provides a
tighter bound on the number of intermediate states.
Moreover, since a precise determination of higher or-
der moments requires better statistics, using the lower
TABLE I. Maximum of differences for N = 2, 3, . . . , 8.
3
3
4
5
6
2
N
8
max{D1} 0.325 0.692 1.090 1.510 1.950 2.403 2.869
max{D2} 0.073 0.161 0.261 0.370 0.487 0.610 0.739
max{D3} 0.382 0.816 1.286 1.782 2.299 2.833 3.382
7
bounds in eqs (12) and (13) to estimate the number of
states is pointless.
Nevertheless, skewness and kurtosis can still be used
to obtain further information on the minimal number of
states by evaluating the positive differences
D1 ≡ N R
min − N S
min ≥ 0,
D2 = N S
min − N K
min ≥ 0,
and
D3 = N R
min − N K
min ≥ 0.
(15)
(16)
(17)
Numerical investigation of these differences shows that
they have a maximum that depends on the number of
states N . The values of this maximum for different N ,
which are obtained with numerical maximization, are re-
ported in Table I, which constitutes our second main re-
sult. Note that these maxima increase with N . There-
fore, the upper bounds on the differences D can be used to
bound the number of states N from below. For example,
if D1 > 0.3246, then the number of states must be larger
than N = 2. As shown in Figure 1, for N = 3 there are
regions in the space of parameters where N R
min < 2, i.e.,
the randomness parameter indicates a number of states
that is not larger than 2, while the difference D1 > 0.3246
indicates N > 2. Hence, the difference D1 leads to
a stronger lower bound on the number of intermediate
states in these regions.
A posteriori, we understand why these upper bounds
on the differences D can lead to better lower bounds on
the number of intermediate states. The quantities N R
min,
N S
min, N K
min are equal to N when all forward transition
rates are equal and all backward transition rates are 0.
If the transition rates are such that ψ(t) is exponential,
then these quantities reach their minimal value 1. For the
differences D the situation is different. For both of these
extreme cases, irreversible transitions with uniform rates
and ψ(t) exponential, all differences D are 0. Therefore,
the differences are maximized somewhere in the space
of transition rates where N R
min are between
their extreme values. A precise analytical identification
of the regions in this space where the differences D lead to
a sharper lower bound on number of states for arbitrary
N , however, does not seem to be simple.
min, N K
min, N S
Skewness and kurtosis can also be used to identify more
efficiently than the randomness parameter whether the
1 = k−
(Color online) Illustration of the new information
FIG. 1.
provided by D1. The parameters are N = 3 and k−
2 = 0.
In the blue region marked by B, N R
min < 2 and D1 > 0.3246,
where 0.3246 is the maximum value of D1 for N = 2; in the
red region marked by C, N R
min < 2 and D1 < 0.3246; in the
green region marked by A, N R
min > 2 and D1 < 0.3246. The
region marked by B corresponds to the manifold in parameter
space where evaluating D1 leads to the information that N
must be larger than 2, while the randomness parameter gives
a lower bound below 2.
topology of the enzymatic network is more complex than
a single cycle. As a case study to demonstrate this fact
we consider the multicyclic enzymatic network from Fig-
ure 2(a). There are two intermediate states of the en-
zyme bound with substrate, which are denoted E1S and
E2S. The generation of product can then happen in two
different pathways:
and
E + S
k12−−⇀↽−−k21
E1S
k24−−⇀↽−−k42
EP
k41−−⇀ E + P,
E + S
k13−−⇀↽−−k31
E2S
k34−−⇀↽−−k43
EP
k41−−⇀ E + P,
(18)
(19)
where we identify the states E = 1, E1S = 2, E2S = 3,
and EP = 4. The method we use to calculate R, S and
K, which have quite long expressions, for this model is
explained in [29].
The maxima of randomness parameter R, skewness S
and kurtosis K for a unicyclic network are R = 1, S = 2,
and K = 9, see eqs (11), (12), and (13). If any of these
quantities exceeds these maximal values the enzymatic
scheme must be different from a single cycle. In order to
check which of these quantities is more effective in provid-
ing a signature of a non-unicyclic network we set the rates
as k12 = k24 = k41 = 1, k21 = k42 = 0, k13 = k43 = 10−2,
and k31 = k34 = k. For k ≫ 1, the cycle in eq (18)
dominates the network: in the rare occasions a jump to
the state E2S occurs the enzyme quickly returns to the
main cycle. In this case R, S, K are close to the values
they reach in the lower bounds with N = 3. If k ≪ 1,
EP
E+P
EP
E+P
E+S
E+S
EI
E S
1
E S
2
ES
(a)
(b)
FIG. 2. (Color online) Chemical reaction schemes. Scheme
(a) corresponds to the multicyclic model
in eqs (18) and
(19), and scheme (b) corresponds to the model with an
an off-pathway state that is obtained by setting the rates
k34 = k43 = 0. For both schemes, the red dotted link repre-
sents the transition that is irreversible.
Randomness Parameter
Skewness
Kurtosis
100
10
1
0.01
0.1
1
k
FIG. 3. (Color online) Randomness parameter, skewness, and
kurtosis for the model shown in Figure 2(a). The transi-
tion rates are k12 = k24 = k41 = 1, k21 = k42 = k14 = 0,
k13 = k43 = 10−2, and k31 = k34 = k. The horizontal dotted
lines indicate the respective maximum values for a unicyclic
network.
the system can get trapped in state E2S. Such a trap-
ping state makes very large cycle completion times more
probable, which leads to an increase in fluctuations. As
shown in Fig. 3, all three quantities cross their maximum
allowed values in a unicyclic network as k gets smaller.
The main result is that the kurtosis is the first to indicate
a more complex network followed by the skewness and
the randomness parameter. For example, at k = 0.376
the randomness parameter is R = 1.00 and the kurto-
sis is K = 148.7, way above the value K = 9. With a
numerical investigation in the whole parameter space we
find that higher order moments are always more effec-
4
tive in revealing whether the network contains a struc-
ture more complex than a cycle, which is our third main
result. This numerical investigation consists of, for ex-
ample, maximizing the skewness with the constraint that
R ≤ 1, which leads to a skewness exceeding 2. The op-
posite however is not possible, if the constraint S ≤ 2 is
satisfied, the randomness parameter never goes above 1.
A randomness parameter larger than one can also be
obtained in a case where there is only one cycle but with
an extra state that does not belong to the cycle [21, 22].
Such an off-pathway state may be related to an inhibitor.
This situation is already contained in the mathematical
model defined in eqs (18) and (19): by setting the rates
k34 = k43 = 0 and identifying the states E = 1, ES = 2,
EI = 3, and EP = 4 we obtain the enzymatic scheme
in Figure 2(b). Our numerics demonstrate that also for
this case with an off-pathway state due to an inhibitor
higher order moments are more effective in revealing that
the network is not just a single cycle. It remains as an
open question whether this property is true for arbitrary
networks that do not consist of a single cycle.
In conclusion, we have obtained lower and upper
bounds on the skewness and kurtosis associated with the
cycle completion time. These bounds can lead to infor-
mation about the enzymatic scheme that is not contained
in the randomness parameter. By evaluating skewness
and kurtosis for experimental data further information
on the number of intermediate states and on whether the
network of states is more complex than a cycle can now
be obtained. While for the former information excellent
statistics might be needed to evaluate higher order mo-
ments so that the upper bounds on the differences D can
be verified, for the latter even rough measures of higher
order moments may be enough: it is possible to find pa-
rameter regimes where the randomness parameter is close
to 1 but the kurtosis and skewness are way above their
maximum values for unicyclic schemes.
On the theoretical side, it is intriguing to ask whether
general expressions for the skewness and kurtosis in terms
of the substrate concentration, as the expression for the
randomness parameter derived in [3], exists and what
kind of information about the enzymatic scheme such ex-
pressions would reveal. Second, it would be worthwhile to
consider higher order moments associated with the num-
ber of generated product molecules for thermodynamic
consistent models, i.e., models without irreversible tran-
sitions. In this case stronger lower bounds on the number
of intermediate states depending on the chemical poten-
tial difference driving the reaction exist [24, 25].
[1] A. Cornish-Bowden, Fundamentals of Enzyme Kinetics
Natl. Acad. Sci. U.S.A. 107, 15739 (2010).
(Portland Press, London, 2013).
[4] M. Schnitzer and S. Block, Cold Spring Harbor Symp.
[2] J. R. Moffitt and C. Bustamante, FEBS J. 281, 498
Quant. Biol. 60, 793 (1995).
(2014).
[5] J. W. Shaevitz, S. M. Block, and M. J. Schnitzer, Bio-
[3] J. R. Moffitt, Y. R. Chemla, and C. Bustamante, Proc.
phys. J. 89, 2277 (2005).
5
[6] X. Li and A. B. Kolomeisky, J. Chem. Phys. 139, 144106
(2013).
[7] K. Svoboda, C. F. Schmidt, B. J. Schnapp, and S. M.
Block, Nature 365, 721 (1993).
[8] F. Ritort, J. Phys.: Condens. Matter 18, R531 (2006).
[9] W. J. Greenleaf, M. T. Woodside, and S. M. Block, Annu.
Rev. Biophys. Biomol. Struct. 36, 171 (2007).
[10] P. V. Cornish and T. Ha, ACS Chem. Biol. 2, 53 (2007).
[11] J. R. Moffitt, Y. R. Chemla, S. B. Smith, and C. Busta-
mante, Annu. Rev. Biochem. 77, 205 (2008).
[12] X.S. Xie, Science 342, 1457 (2013).
[13] J. R. Moffitt, Y. R. Chemla, K. Aathavan, S. Grimes,
P. J. Jardine, D. L. Anderson, and C. Bustamante, Na-
ture 457, 446 (2009).
[14] M. J. Schnitzer and S. M. Block, Nature 388, 386 (1997).
[15] K. Visscher, M. J. Schnitzer, and S. M. Block, Nature
400, 184 (1999).
[16] M.T. Valentine, P.M. Fordyce, T.C. Krzysiak, S.P.
Gilbert, and S.M. Block, Nat. Cell Biol. 8, 470 (2006).
[17] M.E. Fisher and A.B. Kolomeisky, Proc. Natl. Acad. Sci.
U.S.A. 98, 7748 (2001).
[18] A.B. Kolomeisky and M.E. Fisher, Biophys. J. 84, 1642
(2003).
[19] B. P. English, W. Min, A.M. van Oijen, K. T. Lee, G.
Luo, H. Sun, B. J Cherayil, S. C. Kou, and X. S. Xie,
Nat. Chem. Biol. 2, 87 (2006).
[20] S. C. Kou, J. B. Cherayil, W. Min, B. P. English, and
X. S Xie, J. Phys. Chem. B 109, 19068 (2005).
[21] J. R. Moffitt, Y. R. Chemla, and C. Bustamante, Meth-
ods Enzymol 475, 221 (2010).
[22] S. Saha, A. Sinha, and A. Dua, J. Chem. Phys. 137,
045102 (2012).
[23] The Fano factor, which is related to the randomness pa-
rameter [2], can be larger than one in reversible models
with a small chemical potential difference that drives the
reaction [17, 18, 24, 25].
[24] A. C. Barato and U. Seifert, Phys. Rev. Lett. 114, 158101
(2015).
[25] A. C. Barato and U. Seifert, J. Phys. Chem. B 119, 6555
(2015).
[26] Y. R. Chemla, J. R. Moffitt, and C. Bustamante, J. Phys.
Chem. B 112, 6025 (2008).
[27] S. Chaudhury, J. Cao, and N. A. Sinitsyn, J. Phys. Chem.
B 117, 503 (2013).
[28] W. Jung, S. Yang, and J. Sung, J. Phys. Chem. B 114,
9840 (2010).
[29] See supplemental material, which includes Refs. [30, 31],
for further calculations and numerics.
[30] Z. Koza, J. Phys. A: Math. and Gen. 32, 7637 (1999).
[31] J. L. Lebowitz and H. Spohn, J. Stat. Phys. 95, 333
(1999).
[32] D. Aldous and L. Sheep, Commun. Statist. Stochastic
Models 3, 467 (1987).
6
SUPPLEMENTAL MATERIAL: SKEWNESS AND KURTOSIS IN STATISTICAL KINETICS
I. RELATION BETWEEN CYCLE COMPLETION TIME DISTRIBUTION AND THE PROBABILITY
OF PRODUCT GENERATION
The relations derived in this section are valid for both the unicyclic scheme in eq (1) and for the multicyclic
scheme from eqs (18) and (19). The basic assumption that must hold is that the step leading to product formation is
irreversible and common to all cycles. We denote the probability of producing M products until time t by P (M, t).
The cycle completion time probability ψ(t) is related to the survival probability
Ps(t) ≡ 1 −Z t
0
ψ(t′)dt′,
(S1)
which is the probability of not completing a cycle until time t. These three probabilities are connected with the
expression
P (1, t) =Z t
0
ψ(t′)Ps(t − t′)dt′,
(S2)
which simply means that the probability of completing one cycle up to time t is the probability of completing a cycle
at time t′ multiplied by the probability of surviving for t − t′ integrated over t′ ≤ t. For M > 1 we have the recursion
relation
P (M, t) =Z t
0
ψ(t′)P (M − 1, t − t′)dt′.
(S3)
With the generating functions ψ(s), defined in eq (7) in the main text, and P (M, s) ≡R ∞
(S3) lead to
0 P (M, t)e−st, eqs (S2) and
From (S4), with the definition
we obtain the relation
P (M, s) = [ ψ(s)]M 1 − ψ(s)
s
.
P (z, s) ≡
∞
XM =0
P (M, s)ezM ,
1 − ψ(s)
s[1 − ez ψ(s)]
In the context of enzyme kinetics this relation has been derived in [4].
P (z, s) =
.
The master equation can be written in the form [30, 31]
d
dτ
P(z) = L(z)P(z),
P(z) is a vector with components Pi(z, t), with i = 1, 2, . . . , N . These components are defined as
Pi(z, t) ≡
∞
XM =0
Pi(M, t)ezM ,
(S4)
(S5)
(S6)
(S7)
(S8)
where Pi(M, t) is the probability that M cycles have been completed until time t and the enzyme is in the intermediate
i=1 Pi(M, t). With a Laplace
state i. This probability is related to P (M, t) through the expression P (M, t) = PN
transform in time, for which t → s, eq (S7) becomes
P(z, s) = [sI − L(z)]−1P(0),
(S9)
where the components of the initial condition vector P(0) are Pi(0) = δ1,i. Eqs (S6), where P (z, s) is the sum of the
components of the vector P(z, s), and (S9) are used in the derivation of expression (10) in the main text, which has
been obtained in [26].
For the multicyclic model in Eqs (18) and (19) in the main text, the modified generator in eq. (S7) reads
7
L(z) ≡
−r1 k21 k31 k41ez
0
k42
k12 −r2
0 −r3
k43
k13
k24 k34 −r4
0
,
(S10)
where r1 ≡ k12 + k13, r2 ≡ k21 + k24, r3 ≡ k31 + k34, and r4 ≡ k41 + k42 + k43. We can then obtain ψ(s) by solving eq
(S6) after calculating P (z, s) from eqs (S9) and (S10). With ψ(s) we obtain skewness and kurtosis from eqs (3), (5),
and (6) in the main text. The general expressions are quite long.
II. PROOF FOR N = 2 AND NUMERICAL EVIDENCE
For the case N = 2 the expressions for S and K are simple and the bounds (12) and (13) in the main text can be
proven explictly as follows. These expressions are obtained as explained in the main text after eq (10). The skewness
reads
(S11)
(S12)
(S13)
where A ≡ 1 + k+
1 + k−
1 and B ≡ A2 − 2k+
1 . Taking the derivative with respect to k−
1 we obtain
S = 2(A3 − 3k+
1 A)/B3/2
∂S
∂k−
1
= 12(k+
1 )2/B5/2 ≥ 0.
1 . The derivative with respect to k+
1 gives
Hence, S is a monotonically increasing function of k−
= −
1 − k−
B5/2
From these two equations it follows that S is minimized at k+
skewness is maximized for k+
bounds in eq (12) in the main text. The kurtosis reads
1 finite and k−
∂S
∂k+
6(k+
1
1 − 1)
.
1 = 0, where it becomes S = √2. The
1 → ∞, where it becomes S = 2. These results are in agreement with the
1 = 1 and k−
The derivatives are given by
K =
3[8(k+
1 )2 − 12k+
B2
1 A2 + 3A4]
∂K
∂k−
1
=
48(k+
1 )2A
C 3
≥ 0
and
∂K
∂k+
1
= −
24k+
1 (k+
1 − k−
C 3
1 − 1)2
,
.
(S14)
(S15)
(S16)
where C ≡ 1 + (k+
where it becomes K = 6, and maximized for k+
agreement with the bounds in eq (13) in the main text.
1 )2 + 2k−
1 )2 + (k−
1 (1 + k+
1 ). Similar to the skewness, the kurtosis is minimized at k+
1 = 0,
1 → ∞, where it becomes K = 9. These results are in
1 = 1 and k−
1 finite and k−
We cannot perform the same explicit proof for larger values of N as both functions become more complicated as
N increases. We did evaluate analytically the derivatives of S and K with respect to one of the forward rates at the
point where all forward rates are 1 and the backward rates are zero. In all cases this derivative is 0, in agreement with
S and K being minimized at this point, where they reach the lower bounds in eqs (12) and (13) in the main text.
The definite evidence for the bounds come from numerical minimization and maximization of S and K from L = 2
up to L = 8. As an independent check we have also evaluated these functions at randomly chosen transition rates.
Some of the results obtained with this second procedure are shown in Figure 4.
8
(b)
(d)
(f)
(a)
(c)
(e)
FIG. 4. Scatter plots illustrating inequalities (14) and (15) in the main text. The solid red line represents the upper bounds
and the dotted blue line the lower bounds in these equations. N = 4 for Figures 4(a) and 4(b), N = 5 for Figures 4(c) and
4(d), and N = 6 for Figures 4(e) and 4(f). For all figures k+
N = 1 and the other transition rates are taken as 10x, with x a
random number between −4 and 4. Each scatter plot has 106 points. The minima close to k+
1 = 1 comes from the fact that
skewness and kurtosis are minimized when all forward rates are 1 and the backward rates are 0.
|
1711.01160 | 4 | 1711 | 2019-12-17T00:21:29 | Functionality of disorder in muscle mechanics | [
"physics.bio-ph"
] | A salient feature of skeletal muscles is their ability to take up an applied slack in a microsecond timescale. Behind this remarkably fast adaptation is a collective folding in a bundle of elastically interacting bistable elements. Since this interaction has long-range character, the behavior of the system in force and length controlled ensembles is different; in particular, it can have two distinct order-disorder--type critical points. We show that the account of the disregistry between myosin and actin filaments places the elementary force-producing units of skeletal muscles close to both such critical points. The ensuing "double-criticality" contributes to the system's ability to perform robustly and suggests that the disregistry is functional. | physics.bio-ph | physics | Functionality of Disorder in Muscle Mechanics
Hudson Borja da Rocha1, 2, ∗ and Lev Truskinovsky2, †
1LMS, CNRS -- UMR 7649, Ecole Polytechnique, Université Paris-Saclay, 91128 Palaiseau, France
2PMMH, CNRS -- UMR 7636 PSL-ESPCI, 10 Rue Vauquelin, 75005 Paris, France
(Dated: December 18, 2019)
9
1
0
2
c
e
D
7
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
4
v
0
6
1
1
0
.
1
1
7
1
:
v
i
X
r
a
A salient feature of skeletal muscles is their ability to take up an applied slack in a microsecond
timescale. Behind this remarkably fast adaptation is a collective folding in a bundle of elastically
interacting bistable elements. Since this interaction has long-range character, the behavior of the
system in force and length controlled ensembles is different; in particular, it can have two distinct
order-disorder -- type critical points. We show that the account of the disregistry between myosin
and actin filaments places the elementary force-producing units of skeletal muscles close to both
such critical points. The ensuing "double-criticality" contributes to the system's ability to perform
robustly and suggests that the disregistry is functional.
If an isometrically activated muscle is suddenly short-
ened, the force first abruptly decreases but then partially
recovers over ∼ 1 ms timescale [1 -- 3]. Behind this remark-
ably swift contraction is a cooperative conformational
change in an assembly of actin-bound myosin heads (cross
bridges). Given that this "power stroke" takes place
at a timescale that is much shorter than the timescale
of the adenosine triphosphate (ATP)-driven attachment-
detachment (∼ 100 ms) [4 -- 6], such fast force recovery is
usually interpreted as a passive phenomenon [7, 8].
If it is an applied force, which is controlled, the mean-
field theory of fast force recovery, viewing filaments as
rigid and cross bridges as parallel [9], predicts metasta-
bility associated with a coherent response [10].
It also
predicts the existence of an order-disorder -- type critical
point, and it was argued that this critical point plays an
essential role in the functioning of the muscle machinery
[11, 12]. This is consistent with the fact that critical sys-
tems are ubiquitous in biology because of their adaptive
advantages, in particular, their robustness in the face of
random perturbations [13 -- 17]. Criticality is often linked
to marginal stability and, indeed, skeletal muscles are
known to exhibit near zero passive rigidity in physiolog-
ical (isometric contractions) conditions [2, 18 -- 20].
The mechanical functioning of this force generated sys-
tem system is complicated by the fact that muscle archi-
tecture involves both parallel and series connections (see
Fig. 1). Parallel elements respond to a common displace-
ment (hard device, Helmholtz ensemble), while series
structures sense a common force (soft device, Gibbs en-
semble). To fold coherently, individual contractile units
should be able to coordinate in both types of loading
conditions; however, the dominance of long-range inter-
actions [21, 22] induces different collective behavior in
force and length controlled ensembles [10].
In particu-
lar, the critical points corresponding to length and force
clamp loading conditions are strictly distinct [12].
In realistic conditions, however, they turn out to be
close to each other and, to ensure the robustness of the
response under a broad range of mechanical stimuli (flex-
ibility) [23], the system can still be poised in the vicinity
Figure 1. Schematic representation of a muscle myofibril, of
an elementary contractile unit (half-sarcomere) and of a par-
allel bundle of N cross bridges. In the model, the double-well
potentials are mimicked by spin variables.
of both critical points.
In this Letter, we argue that such "double criticality"
is actualized in the system of muscle cross bridges due
to quenched disorder. While skeletal muscles are often
compared to ideal crystals, the perfect ordering is com-
promised by the intrinsic disregistry between the period-
icities of myosin cross bridges and actin binding sites.
Binding of cross bridges is restricted to incompatibly
placed segments on actin filaments (target zones), and
experimental studies based on electron microscopy and
x-ray diffraction suggest that myosin heads are bound to
actin at seemingly random positions [24, 25]. To gain an
insight into the role of variable offsets, we assume that
the attachment sites are indeed chosen at random and
show that this gives us an analytically tractable model.
The idea that actomyosin disregistry brings the sys-
tem's stiffness to zero was pioneered in [26]. More re-
cently the utility of quenched disorder for the active as-
pects of muscle mechanics has been advocated in [27].
The beneficial role of random inhomogeneity has been
established in many other fields of physics from high-
temperature superconductivity in electronic materials
[28] to Griffiths phases in brain networks [29].
To explore the reachability of the "double criticality"
condition in realistic conditions, we reduce the descrip-
tion of the system of interacting cross bridges to a ran-
dom field Ising model (RFIM) and compute the equilib-
rium free energy applying techniques from the theory of
...x1v1xNvNκ0κ0κfymyosinactinzcontractile unit............myofibril N(cid:88)
i=1
glassy systems [30]. We then use the available experi-
mental data on skeletal muscles to justify the claim that
quenched disorder is the main factor ensuring the tar-
geted mechanical response.
We associate with each cross bridge a spin variable
x taking the value 0 in the pre-power-stroke state (un-
folded conformation) and −1 in the post-power-stroke
state (folded conformation). Each spin element is then
placed in series with a linear elastic spring of stiffness κ0.
If we nondimensionalize lengths by the power-stroke size
a and energy by κ0a2, the dimensionless energy of a cross
2 (y − x)2, where y is the dimen-
bridge reads (1 + x)v + 1
sionless displacement of myosin relative to actin and v
is the dimensionless energetic bias, see Fig.1. To model
disregistry, we assume that the parameter v is different
for different cross bridges [31].
Consider now a parallel bundle of N cross bridges
shown schematically in Fig. 1. Individual cross bridges
are attached to a backbone composed of myosin tails.
The elasticity of the backbone can be accounted through
a lump spring of stiffness κf in series with the bundle
[32 -- 34]. The system loaded in a hard device is then char-
acterized by the dimensionless energy
E =
[(1 + xi)vi +
1
2
(y − xi)2] + N
λf
2
(z − y)2,
(1)
where z is the applied displacement and λf = κf /(N κ0).
We assume that the parameters vi are independent iden-
tically distributed random variables with probability den-
sity p(v).
If we replace variables xi by si = 2xi + 1 = ±1 and
adiabatically eliminate y, assuming that ∂E/∂y = 0, the
energy (1) takes the form
(cid:88)
i,j
(cid:88)
i
E = −J/(2N )
sisj −
hisi + c,
where J = 1/4(1 + λf ), c is a z dependent constant,
and the coefficients hi are linear in vi (see Supplemental
Material [35]). We can then conclude that (1) is a version
of the mean-field RFIM, which is explicitly solvable [38,
39].
Using the self-averaging property of the free energy in
the thermodynamic limit, we write
F(β, z) = − lim
N→∞(N β)
−1(cid:104)log Z(β, z;{v})(cid:105)v,
where the averaging (cid:104)·(cid:105)v is over the disorder, β =
κ0a2/(kBT ), and
(cid:90)
(cid:88)
Z =
dy
x∈{0,−1}N
exp(−βE(x, y, z;{v}).
2
,
(2)
(cid:21)
.
In the thermodynamic limit, we obtain [35]
F(β, z) =
−
(cid:90)
(z − y0)2 +
dv p(v) log
λf
2
1
β
1
4
(cid:20)
(y0 + 1)2 +
1
2
(
y2
0
2
+ v0)
(cid:21)
2 cosh[
(1 + 2y0 − 2v)]
β
4
(cid:90)
(cid:20) β
4
where y0 must solve the self-consistency equation
p(v)
2λf z − 1
2(λf + 1)
+
dv
tanh
y0 =
2(λf + 1)
(1 − 2v + 2y0)
(3)
The multiplicity of solutions of Eq. (3) is a result of the
nonconvexity of the free energy with respect to y, which is
ultimately an effect of long-range interactions. The mul-
tiplicity leads to the possibility of discontinuous tension-
elongation curves t = ∂F/∂z = λf (z − y0).
If we assume that the disorder is Gaussian p(v) =
(2πσ2)−1/2 exp(− (v−v0)2
), the behavior of the system
will be fully defined by the temperature 1/β, the variance
of disorder σ2, and the parameter λf , characterizing the
degree of elastic coupling. The resulting phase diagram
is shown in Fig. 2. The disorder-free section σ = 0 of this
diagram was previously studied in [12]. At σ > 0 the sys-
tem responds as if it was subjected to a higher effective
temperature [40, 41]. The Helmholtz free energy F(β, z)
and the tension-elongation relations t(β, z) in the three
phases I, II, and III are illustrated in Fig. 3.
2σ2
Figure 2.
(a) Configuration of phases I, II, and III in the
parameter space (1/β, σ , λf ). (b) A section of this phase
diagram corresponding to λf = 0.54 ± 0.2; the shadowed re-
gion near the boundary of II and III reflects the uncertainty
in λf . The realistic dataset for skeletal muscles is presented
in (b) by a filled circle with the superimposed error bars indi-
cating uncertainty in temperature. Analytic approximations
in (b): dashed-dotted lines indicate low temperature; dashed
lines indicate low disorder.
In phase I, the cooperativity is absent and the cross
bridges fluctuate independently. In phase III, the cross
bridges can synchronously switch between two "pure
In the intermediate phase II, the tension-
states".
elongation relation exhibits negative stiffness.
The
boundary between phases II and III is defined by the
3(a)(b)0.30.4IIIIIIpqrsβ−1σλfIIIIII00.1ps00.1qrβ−1σ 3
Figure 3. (a) Representative Helmholtz free energies in each of
the phases I, II, III. (b) The corresponding tension-elongation
relations; z0 = (1 + λf )v0/λf − 1/2; t0 = v0.
Figure 4. The behavior of the parameter φ2 (solid lines) and
the Edwards-Anderson parameter qEA (dashed lines) near the
boundary between phases II and III at the realistic value of
disorder. (Inset) The case of weak disorder.
e
condition that ∂2 F(β, z, y)/∂y2 = 0, which is a condi-
tion that the three roots of (3) collapse into one.
In the limiting case σ → 0, the point p in Fig. 2(b) is
at β = 4(λf + 1). Around this point, the p − q curve is
described accurately by the low-disorder approximation
βe = 4(λf +1) where βe = (β−2 +σ2/2)−1/2 is the inverse
effective temperature (see Supplemental Material [35]).
In another limiting case β → ∞, the point q can be found
point the p − q curve is given by the small temperature
e = (σ2 +
2β−2) = 2β−2
is the variance of the effective disorder
[35].
from the equation σ = 1/(cid:112)2π(λf + 1) and around this
approximation σe = 1/(cid:112)2π(λf + 1), where σ2
N−1(cid:80)N
parameter qEA = N−1(cid:80)N
i=1 (cid:104)si(cid:105)β, where (cid:104)·(cid:105)β is the thermal average,
The boundary between phases II and III marks a
second-order phase transition: the order parameter φ =
is
double valued in phase III and single valued in phase II.
To distinguish between different microscopic configura-
tions, we also compute the Edwards-Anderson (overlap)
If qEA (cid:54)= 0
while φ = 0, the pre- and post-power-stroke symmetry
is broken and cross bridges may be locally frozen in ei-
ther of the two states, even though such local ordering
in time does not imply any spatial order. Figure 4 shows
that qEA is indeed different from zero in the phase II close
to the p − q boundary, which indicates weakly glassy be-
havior [38, 39, 42]. This is a hint that, in a more realistic
model, where the finite backbone stiffness is taken into
account, a real "strain glass" phase [43, 44] is likely to
appear.
i=1(cid:104)(cid:104)si(cid:105)2
β(cid:105)v [35].
To find the boundary between phases I and II, we need
to solve the equation ∂2F/∂z2 = 0 or ∂y0/∂z = 1, where
y0 is a solution of (3). When σ = 0, we obtain β = 4,
which defines the location of point s in Fig. 2(b) (see also
[10, 33]). The low-disorder approximation gives βe = 4.
In another limiting case β → ∞, the location of the point
r in Fig. 2(b) is given by σ = σe =
The boundary between the phases I and II can be also
interpreted as a line of second-order phase transitions,
but now in the soft device (force clamp) ensemble.
In
this case, the presence of a series spring is irrelevant and
(cid:112)
1/2π.
Figure 5. (a) Representative Gibbs free energies in each of
the phases I and II. (b) The corresponding tension-elongation
curves; z0 = (1 + λf )v0/λf − 1/2; t0 = v0.
we can assume that λf → 0, z → ∞, but λf z → t,
where tension t is the new control parameter. Follow-
ing the approach used in the case of a hard device, we
similarly obtain the Gibbs free energy G(β, t) and com-
pute the tension-elongation relation y = −∂G/∂t, (see
Supplemental Material [35]).
In Fig. 5, we show that the soft device tension-
elongation relation in phase II is monotone but discon-
tinuous. On the boundary of I and II [see Fig. 2(b)], the
stiffness becomes zero in stall conditions, which means
that it is a set of critical points in the soft device ensem-
ble. This line, targeted numerically in [26], represents
regimes that can be expected to deliver the optimal trade-
off between robustness and flexibility in the soft device
[45, 46].
So far, we have operated under an implicit assump-
tion that in the thermodynamic limit κf → ∞, while
λf remains finite. This assumption is based on the pic-
ture of a myosin filament as a parallel arrangement of N
myosin tails, all contributing to the lump stiffness of the
backbone. An alternative assumption may be that the
effective stiffness of the backbone κf does not depend on
the number of attached cross bridges N and, in this case,
we have a different scaling λf ∼ N−1. Then Fig. 2(a),
illustrating the size effect, suggests that the quasicriti-
(a)(b)−1−0.500.5100.20.4z−z0FIIIIII−1−0.500.51−10123z−z0tt0IIIIII00.5100.511/βqEAφ200.400.51σ=0.1σ=0.35(a)(b)0.511.500.20.40.6t/t0GIII−1−0.500.510.511.5y−v0+1/2tt0III cal behavior should be tightly linked to the particular
(optimal) number of cross bridges.
To apply our results to a realistic muscle system, we
use the data for Rana temporaria at T = 277.15 K [12].
From structural analysis, we obtain the value a ∼ 10 nm
[47 -- 49]. Measurements of the fiber stiffness in rigor mor-
tis, where all the 294 cross bridges per half-sarcomere
were attached, produced the estimate κ0 = 2.7 ± 0.9
pN/nm [18, 19]. The number of attached cross bridges in
physiological conditions is N = 106 ± 11 and experimen-
tal measurements at different N converge on the value
κf = 154 ± 8 pN.nm−1 for the lump filaments stiffness
[5, 50, 51]. This gives λf = 0.54± 0.2. Knowing κ0 and a
we can estimate the nondimensional inverse temperature
to be β = 71 ± 26.
Now, for y > y∗, where y∗ = v0− 1/2, the ground state
of a single cross bridge is in the pre-power-stroke state,
while for y < y∗ it is in the post-power-stroke state, so y∗
represents the characteristic offset for an individual cross
bridge. Knowing that y∗ ∼ 4nm [2, 26], we conclude
that v0 ∼ 24.3pN /(κ0a). It was experimentally shown
in [25] that at least 60% of the cross bridges are axi-
ally displaced within half of the spacing between actin
monomers, which corresponds to ∼ 2.76 nm shift from
the nearest actin binding site (see also [26]). Given the
linear relation between v0 and y∗ with the proportion-
ality coefficient equal to one, the variances of these two
quantities are the same. If the axial offsets are Gaussian
random numbers, we can estimate the standard deviation
of the energetic bias σ ∼ 3.3nm/a (see Supplemental Ma-
terial [35]).
Figure 6. The structure of the energy barriers in different
regimes for the case of the hard device. (a) -- (c) z dependence
i=1(cid:104)si(cid:105) in different regimes;
of the order parameter φ = N
(d) -- (f) matching free energies at fixed z = z0. λf = 0.35 and
v0 = 0.1
−1(cid:80)N
Based on these data we find that, rather remarkably,
the system appears to be operating in a narrow domain of
4
stability of phase II, close to both critical lines p − q and
r − s [see the point marked by a filled circle in Fig. 2(b)].
The gap between these boundaries corresponds to ∼ 1
nm difference in the cross bridge attachment positions,
which is rather small given that the size of a single actin
monomer is about 5.5 nm. The mechanical responses
in the adjacent critical regimes are structurally similar;
however, if in the hard device ensemble we can expect co-
herent fluctuations of stress (infinite rigidity), in the soft
device, criticality would manifest itself through system
size correlations of strain (zero rigidity).
The special nature of the critical regimes is illustrated
in Fig. 6 for the case of a hard device.
In phase I,
the response is uncorrelated, and the collective power
stroke is impossible [Figs. 6(a) and 6(d)]. In phase III,
the response is synchronous but at the cost of crossing
an energetic barrier that diverges in the thermodynamic
limit (F is the free energy per cross bridge), which facili-
tates freezing in the pure states, [see Figs. 6(b) and 6(e)].
The advantage of the critical regime is that the system
can perform the collective stroke without crossing a pro-
hibitively high macroscopic barrier, [see Figs. 6(c) and
6(f)]. The analysis is similar for the case of a soft device.
Our study then suggests that evolution might have
used quenched disorder to tune the muscle machinery to
perform near the conditions where both the Helmholtz
and the Gibbs free energies are singular. Such design is
highly functional when elementary force-producing units
are loaded in a mixed, soft-hard device. We recall that
the muscle architecture is characterized by hierarchical
structures with coupled modular elements loaded both in
parallel and in series. In such systems, the proximity to
only one of the two critical points will not be sufficient
to ensure high performance in a broad range of condi-
tions [23, 52]. Moreover, as we show in the Supplemental
Material [35], the very idea of ensemble independent local
constitutive relations for such systems becomes question-
able.
In conclusion, we established new links between mus-
cle physiology and the theory of spin glasses and revealed
a tight relation between actomyosin disregistry and the
optimal mechanical performance of the force-generating
machinery. At a price of neglecting many important fea-
tures of actual muscles, we were able to focus attention
on the role of quenched disorder in the functioning of
this biological system. The observed glassiness in the
regime of isometric contractions allows the system to ac-
cess the whole spectrum of rigidities from zero (adapt-
ability, fluidity) to infinite (control, solidity) and may
serve as the factor ensuring the largest dynamic reper-
toire of the "muscle material". Similar disorder-mediated
tuning towards criticality can be expected in other biolog-
ical systems relying on bistability and long-range inter-
actions [9], including hair cells, which employ elastically
coupled gating springs [53] and focal adhesions with their
cell adhesion molecules bound to a common substrate
(a)(b)(c)(d)(e)(f)phaseI(σ=0.2,β=4)phaseIII(σ=0.1,β=10)criticalstate(σ=0.1,β=6)−101−101z−z0φ−10100.1φF−101−101z−z0−1010.10.13φ−101−101z−z0−1010.10.14φ [54].
The authors thank M. Caruel and R. Garcia-Garcia
for helpful discussions. H.B.R. received support from an
Ecole Polytechnique Fellowship; L. T. was supported by
Grant No. ANR-10-IDEX-0001-02 PSL.
∗ [email protected]
† [email protected]
[1] R. J. Podolsky, Nature 188, 666 (1960).
[2] A. F. Huxley and R. M. Simmons, Nature 233, 533 (1971).
[3] M. Irving, V. Lombardi, G. Piazzesi, and M. A. Ferenczi,
Nature 357, 156 (1992).
[4] J. Howard, Mechanics of Motor Proteins and the Cy-
toskeleton (Sinauer Associates, Publishers, 2001).
[5] G. Piazzesi, M. Reconditi, M. Linari, L. Lucii, Y.-B. Sun,
T. Narayanan, P. Boesecke, V. Lombardi, and M. Irving,
Nature 415, 659 (2002).
[6] M. Kaya, Y. Tani, T. Washio, T. Hisada, and H. Higuchi,
Nature Communications 8, 16036 (2017).
(2003).
(2010).
(2016).
[8] L. Marcucci and L. Truskinovsky, Phys. Rev. E 81, 051915
[9] M. Caruel and L. Truskinovsky, Phys. Rev. E 93, 062407
[10] M. Caruel, J.-M. Allain, and L. Truskinovsky, Physical
Review Letters 110, 248103 (2013).
[11] M. Caruel and L. Truskinovsky, Journal of the Mechanics
and Physics of Solids 109, 117 (2017).
5
[25] R. T. Tregear, M. C. Reedy, Y. E. Goldman, K. A. Tay-
lor, H. Winkler, C. Franzini-Armstrong, H. Sasaki, C. Lu-
caveche, and M. K. Reedy, Biophysical Journal 86, 3009
(2004).
[26] A. F. Huxley and S. Tideswell, Journal of Muscle Re-
search & Cell Motility 17, 507 (1996).
[27] P. F. Egan, J. R. Moore, A. J. Ehrlicher, D. A. Weitz,
C. Schunn, J. Cagan, and P. LeDuc, Proceedings of the
National Academy of Sciences 114, E8147 (2017).
[28] J. Zaanen, Nature 466, 825 (2010).
[29] P. Moretti and M. A. Muñoz, Nature Communications 4,
2521 (2013).
[30] T. Castellani and A. Cavagna, Journal of Statistical Me-
chanics: Theory and Experiment 2005, P05012 (2005).
[31] This form of the quenched disorder is equivalent to the
explicit introduction of a pre-strain in each of the linear
springs and is also a signature of spatially inhomogeneous
ATP driving.
[32] M. Linari, G. Piazzesi, and V. Lombardi, Biophysical
Journal 96, 583.
[33] M. Caruel, J.-M. Allain, and L. Truskinovsky, Journal
of the Mechanics and Physics of Solids 76, 237 (2015).
[34] F. Jülicher and J. Prost, Phys. Rev. Lett. 75, 2618
[35] See Supplemental Material at [URL will be inserted by
publisher] for the details of the mapping on the RFIM, the
computation of Helmholtz and Gibbs free energies, the de-
scription of the boundaries between phases I and II and II
and III, the role played by the Edwards-Anderson param-
eter, of the random representation of the axial offset, and
the mechanical behavior of two half-sarcomeres in series,
which also includes Refs. [36, 37].
[36] I. Vilfan and R. A. Cowley, Journal of Physics C: Solid
[37] M. Suzuki and S. Ishiwata, Biophysical Journal 101, 2740
[38] T. Schneider and E. Pytte, Phys. Rev. B 15, 1519 (1977).
[39] F. Krzakala, F. Ricci-Tersenghi, and L. Zdeborová, Phys.
Rev. Lett. 104, 207208 (2010).
[40] S. Roux, Phys. Rev. E 62, 6164 (2000).
[41] A. Politi, S. Ciliberto, and R. Scorretti, Phys. Rev. E
[42] I. Vilfan, Physica Scripta 1987, 585 (1987).
[43] Y. Wang, X. Ren, and K. Otsuka, Phys. Rev. Lett. 97,
66, 026107 (2002).
225703 (2006).
[44] R. Vasseur, D. Xue, Y. Zhou, W. Ettoumi, X. Ding,
X. Ren, and T. Lookman, Phys. Rev. B 86, 184103 (2012).
[45] S. Kauffman, The Origins of Order: Self-Organization
and Selection in Evolution (Oxford University Press,
1993).
[46] C. Darabos, M. Giacobini, M. Tomassini, P. Provero,
and F. Di Cunto, in Advances in Artificial Life. Darwin
Meets von Neumann, edited by G. Kampis, I. Karsai, and
E. Szathmáry (2011) pp. 281 -- 288.
[47] R. Dominguez, Y. Freyzon, K. M. Trybus, and C. Cohen,
Cell 94, 559 (1998).
[48] I. Rayment, W. Rypniewski, K. Schmidt-Base, R. Smith,
D. Tomchick, M. Benning, D. Winkelmann, G. Wesenberg,
and H. Holden, Science 261, 50 (1993).
[49] I. Rayment, H. Holden, M. Whittaker, C. Yohn,
and R. Milligan, Science 261,
M. Lorenz, K. Holmes,
58 (1993).
[50] K. Wakabayashi, Y. Sugimoto, H. Tanaka, Y. Ueno,
Y. Takezawa, and Y. Amemiya, Biophysical Journal 67,
[7] A. Vilfan and T. Duke, Biophysical Journal 85, 818
(1995).
[12] M. Caruel and L. Truskinovsky, Reports on Progress in
State Physics 18, 5055 (1985).
Physics 81, 036602 (2018).
[13] E. Balleza, E. R. Alvarez-Buylla, A. Chaos, S. Kauffman,
I. Shmulevich, and M. Aldana, PLOS ONE 3, 1 (2008).
[14] J. Beggs and N. Timme, Frontiers in Physiology 3, 163
(2011).
(2012).
144, 268 (2011).
[15] T. Mora and W. Bialek, Journal of Statistical Physics
[16] D. Krotov, J. O. Dubuis, T. Gregor, and W. Bialek, Pro-
ceedings of the National Academy of Sciences 111, 3683
(2014).
[17] D. A. Kessler and H. Levine, ArXiv e-prints
(2015),
arXiv:1508.02414.
[18] E. Brunello, M. Caremani, L. Melli, M. Linari,
M. Fernandez-Martinez, T. Narayanan, M. Irving, G. Pi-
azzesi, V. Lombardi, and M. Reconditi, The Journal of
Physiology 592, 3881 (2014).
[19] G. Piazzesi, M. Reconditi, M. Linari, L. Lucii, P. Bianco,
E. Brunello, V. Decostre, A. Stewart, D. Gore, T. Irving,
M. Irving, and V. Lombardi, Cell 131, 784 (2007).
[20] M. Linari, I. Dobbie, M. Reconditi, N. Koubassova,
and V. Lombardi, Biophysical
M. Irving, G. Piazzesi,
Journal 74, 2459 (1998).
[21] A. Campa, T. Dauxois, and S. Ruffo, Physics Reports
480, 57 (2009).
87, 030601 (2001).
[22] J. Barré, D. Mukamel, and S. Ruffo, Phys. Rev. Lett.
[23] M. A. Muñoz, Rev. Mod. Phys. 90, 031001 (2018).
[24] R. T. Tregear, R. J. Edwards, T. C. Irving, K. J. Poole,
M. C. Reedy, H. Schmitz, E. Towns-Andrews, and M. K.
Reedy, Biophysical Journal 74, 1439 (1998).
[51] H. Huxley, A. Stewart, H. Sosa, and T. Irving, Biophys-
ical Journal 67, 2411 (1994).
[52] W. Bialek, Reports on Progress in Physics 81, 012601
2422 (1994).
(2018).
[53] V. Bormuth, J. Barral, J.-F. Joanny, F. Jülicher, and
P. Martin, Proceedings of the National Academy of Sci-
ences 111, 7185 (2014).
[54] U. S. Schwarz and S. A. Safran, Rev. Mod. Phys. 85,
1327 (2013).
Supplemental Material for the paper "Functionality
of Disorder in Muscle Mechanics"
Mapping to the Random-Field Ising Model
We start with the energy function (1) in the main text
and assume that the internal variable y is eliminated us-
ing the condition ∂E/∂y = 0. Then,
y =
λf z
1 + λf
+
1
N (1 + λf )
i
(cid:88)
xi.
i
1
+
xi
2N (1 + λf )
E(xi, z) = −
and the relaxed energy reads
2
(cid:88)
(cid:88)
Since xi is either 0 or -1, we may write(cid:80)
and (cid:0)(cid:80)
(cid:80)
i = −
i xi
In terms of
spin variables, 2xi = si − 1, with si = ±1 the relaxed
energy can be written as,
(cid:88)
(cid:88)
(cid:80)
j xixj = (cid:80)
(cid:88)
N λf z2
2(1 + λf )
= (cid:80)
(1 + xi)vi −
λf z
1 + λf
i,j xixj.
(cid:1)2
x2
i
2
xi +
i x2
i xi
+
1
i
i
i
i
E(si, z) = −
8N (1 + λf )
i,j
sisj
(cid:33)
si
2λf z − 1
4(1 + λf )
+
1
4 −
vi
2
vi
2 −
1
8(1 + λf )
hisi + f (z).
λf z(1 + z)
2(1 + λf )
(cid:88)
+
+
1
4
(cid:88)
J
2N
i
i,j
sisj −
= −
4(1+λf ), hi = 2λf z−1
8(1+λf ).
4 + vi
1
1
2 −
(cid:80)
where J =
λf z(1+z)
2(1+λf ) + 1
i
4(1+λf ) + 1
4 − vi
2 and f (z) =
(cid:32)
(cid:32)
(cid:88)
(cid:88)
i
i
−
+
Boundary between phases II and III
Using the expression for the partial free energy,
(cid:33)
(4)
F(β, z, y) =
−
(cid:90)
(z − y)2 +
dv p(v) log
1
4
λf
2
1
β
(cid:20)
(y + 1)2 +
2 cosh[
(cid:21)
(y2/2 + v0)
1
2
(1 + 2y − 2v)]
β
4
we can write the condition ∂2 F(β, z, y)/∂y2 = 0 in the
form
λf + 1 −
β
4
dv p(v) sech2 β
4
(1 − 2v + 2y0) = 0.
If we use the Gaussian distribution of disorder introduced
in the main text and use new variables η = β(1 + 2y0)/2
(cid:90)
6
Computation of the free energy
Using the self-averaging property of the free energy in
the thermodynamic limit, we write
F(β, z) = − lim
N→∞(N β)
−1(cid:104)log Z(β, z;{v})(cid:105)v,
where the averaging (cid:104)·(cid:105)v is over the disorder, β =
κ0a2/(kBT ), and
(cid:90)
(cid:88)
Z =
dy
x∈{0,−1}N
exp(−βE(x, y, z;{v}).
The mean field nature of the model allows one to rewrite
this expression in the form
Z =
dy exp(−βN [
λf
2
(z − y)2 −
1
βN
log Z]),
2 (y+1)2
where Z = e− β
+ e−β(y2/2+vi) is the partition
function of a single Huxley-Simmons element [1, 2]. In
the thermodynamic limit, we can use the saddle-point
approximation to obtain F(β, z) = F(y0, β, z), where
2 (z − y)2 − (cid:104)log Z(cid:105)v and y0(β, z) is the
F(y, β, z) = β λf
minimum of F. More explicitly,
(cid:21)
(cid:20)
(cid:90)
(z − y0)2 +
dv p(v) log
(y0 + 1)2 +
2 cosh[
+ v0)
y2
0
2
1
4
1
2
F(β, z) =
−
(1 + 2y0 − 2v)]
λf
2
1
β
β
4
(
,
(5)
where y0 solves the self-consistency equation,
(cid:20) β
(cid:21)
y0 =
2λf z − 1
2(λf + 1)
+
dv
p(v)
2(λf + 1)
tanh
(1 − 2v + 2y0)
.
4
N(cid:88)
i=1
(cid:90)
(cid:90)
and ¯v = βv we can rewrite this equation in the form
to require that the two functions are equally normalized
(cid:90)
λf + 1 −
β
4
(cid:112)
e
d¯v
− (¯v−βv0 )2
2σ2 β2
2πσ2β2
sech2 1
2
(η − ¯v) = 0.
1
4T
(cid:90)
Note that the variance of disorder appears in this formula
only in the combination σ2β2. This means that, modulo
some obvious adjustments, the small disorder σ → 0 and
large temperature β → 0 limits are complimentary. The
same can be said about the small temperature β → ∞
and the large disorder σ → ∞ limits.
Zero disorder limit. In the limit σ → 0 we have p(v) →
δ(v − v0) and the boundary between phase II and III is
defined by the equation
λf + 1 =
β
4
sech2 β
4
(1 − 2v0 + 2y0).
Since sech2 x ∈ [0, 1], this equation does not have solu-
tions y0 for β > 4(λf + 1) and therefore the point r is
defined by the condition β = 4(λf + 1).
To get the next term of the asymptotic expansion we
introduce the new variable ξ = (1 − 2v + 2y0)/4 and as-
sume that the temperature is large β → 0. Then we
can expand log sech2 βξ ≈ −β2ξ2 + O(β4), which im-
plies that sech2 βξ ≈ e−β2ξ2. Using this approximation
we can compute the integral and represent the boundary
between phase II and III in the form
e
2(2T 2+σ2)
−(y0−v0+1/2)2
(cid:112)
∈ (0, 1] the criticality condi-
2(2T 2 + σ2)
.
λf + 1 =
2
where T = 1/β. Since e−x2
tion is
(cid:112)
(λf + 1)2
2(2T 2 + σ2) = 1
The equivalent quenched disorder is then defined by the
condition σ2
eq = 2T 2 + σ2.
Zero temperature limit. In the zero temperature limit
β → ∞ we use the fact that limk→∞ k
2 sech2 kx → δ(x).
to rewrite the equation defining the boundary between
phase II and III in the form
(λf + 1)√2πσ2 = e
− (y0+1/2−v0 )2
2σ2
.
Here the r.h.s is defined in the interval (0, 1] and therefore
there are no solutions y0 if (λf + 1)√2πσ2 > 1 where we
used the fact that σ, λf > 0. The point q is then defined
by the condition (λf + 1)√2πσ2 = 1.
To obtain the next term of the asymptotic expansion
we assume that disorder is large σ → ∞.
In this case
we can still approximate the function sech2(x) by the
Gaussian distribution but now the approximation should
be good not at x = 0 but globally. To this end we need
7
dv sech2 1 − 2v − y0
(cid:90)
4T
1
√4πT 2
=
− (v−y0−1/2)2
4T 2
dve
= 1,
where again T = 1/β. With this normalization the inte-
gral can be again computed and we obtain the condition
2(2T 2+σ2)
−(y−v0+1/2)2
e
√2T 2 + σ2
.
The criticality criterion is then
(λf + 1)√2π =
(cid:112)
(λf + 1)
2π(2T 2 + σ2) = 1,
which allows us to introduce the effective disorder by the
condition σ2
e = 2T 2 + σ2.
Gibbs free energy
(cid:21)
(cid:20)
N(cid:88)
i=1
In the case of soft device the relevant potential is,
G =
(1 + xi)vi +
1
2
(y − xi)2
− ty.
(6)
Following the approach used in the case of hard device,
we obtain the expression for the Gibbs free energy
G(β, t) = −ty0 +
(cid:90)
1
4
(y0 + 1)2 +
(cid:20)
1
2
dv p(v) log
2 cosh[
(
+ v0)
y2
0
2
β
(1 + 2y0 − 2v)]
4
(cid:21) (7)
1
β
−
(cid:21)
where now y0 solves the equation
(cid:90)
(cid:20) β
t = y0 +
1
2 −
1
2
dv p(v) tanh
(1 − 2v + 2y0)
4
. (8)
The tension elongation relation is then a solution of y =
−∂G/∂t.
Edwards-Anderson order parameter
is
In the absence of disorder, a natural order parameter
N(cid:88)
i=1
φ =
1
N
(cid:104)si(cid:105)T ,
where si = 2xi+1. To find φ(z, β) we notice that since all
cross-bridges are the same we can write φ = 2(cid:104)xi(cid:105)T + 1
where
with
(cid:104)xi(cid:105)T = −Z(β, z)
−1e
−βE(xi=−1,y0,z)
−βN
Z(β, z) = e
(cid:20) λf
2 (z−y0)2− 1
β log(e
− β
2
(y0+1)2
−β(y2
0 /2+v))
+e
(cid:21)
.
main text and therefore
∂y0
∂z
=
+
λf
λf + 1
β
4(1 + λf )
(cid:90)
(cid:20) β
4
dv p(v) sech2
(1 − 2v + 2y0)
8
(cid:21) ∂y0
,
∂z
(9)
By combining these expressions we obtain
which is equivalent to
(cid:104)xi(cid:105)T = −
1
1 + eβ(y0−v+1/2) .
1 =
β
4
dv p(v) sech2
(cid:90)
(cid:20) β
4
(cid:21)
(1 − 2v + 2y0)
.
In the presence of disorder, the average values (cid:104)xi(cid:105)T are
different for different cross-bridges and the macroscopic
parameter φ(z, β) is no longer sufficient to differentiate
between microscopic configurations. To this end we can
introduce an analogue of the Edwards-Anderson param-
eter from the theory of spin glasses
The condition that this equation has a root y0 does not
contain λf and therefore the boundary between phases I
and II is λf independent.
Zero disorder limit. In the limit σ → 0 we can again
assume that the probability density p(v) is infinitely lo-
calized and compute the integral explicitly. We obtain
.
4
β
= sech2 β
4
(1 − 2v + 2y0).
Since sech2 x ∈ [0, 1], this equation does not have solu-
tions y0 if β < 4, hence βc = 4, which is the coordinate
of our point s. The higher order asymptotic expansion
can be obtained following the same procedure as in the
case of the boundary between phases II and III.
Zero temperature limit. In the limit β → ∞, we can
again use the fact that the function k
2 sech2 kx converges
to the delta function as k → ∞. Therefore, assuming
that the probability distribution p(v) is Gaussian we ob-
tain,
1 =
1
√2πσ2
e
− (y0+1/2−v0)2
2σ2
.
Using the same arguments as in the zero disorder limit
and noticing that e−x2
∈ (0, 1], we conclude that this
equation has solution only if σ ≥ 1/√2π. Therefore, the
critical value of the disorder in this limit is σc = 1/√2π,
which corresponds to our point r. The expansion around
this point can be obtained as in the case of the boundary
between phases II and III considered above.
N(cid:88)
(cid:69)
(cid:68)
(cid:104)si(cid:105)2
T
v
1
N
qEA =
i=1
and the ensemble average (cid:104)A(cid:105)v = (cid:82) dvp(v)A(v). If the
where we distinguish between the thermal average (cid:104)·(cid:105)T
parameter φ characterizes the average occupancy of the
pre-power stroke state, the nonzero value of qEA means
that individual cross bridges are 'frozen' either in pre- or
post-power-stroke states even if in average, both states
appear to be equally occupied. The knowledge of this
parameter is needed, for instance, if one is interested in
computing the effect of the random field on mechanical
susceptibility (stiffness) [3]
In terms of the variables xi the definition of qEA reads
(cid:69)
+ 4(cid:10)
(cid:21)
(cid:11)
qEA =
1
N
4
(cid:104)xi(cid:105)2
T
v
(cid:104)xi(cid:105)T
v + 1
,
(cid:20)
(cid:68)
(cid:90)
i=1
N(cid:88)
(cid:69)
(cid:11)
v
(cid:90)
(cid:68)
(cid:104)xi(cid:105)2
(cid:10)
T
(cid:104)xi(cid:105)T
where
and
=
dv
p(v)
(1 + eβ(y0−v+1/2))2 ,
v = −
dv
p(v)
1 + eβ(y0−v+1/2) .
Axial offset
Boundary between phases I and II
Note first that ∂t
∂z ), and therefore to get
zero stiffness we must have ∂y0/∂z = 1, Here y0 is found
from the self-consistency condition given by Eq. 5 in the
∂z = λf (1 − ∂y0
Experimental studies using electron microscopy (EM)
and x-ray diffraction have shown that the biding of cross-
bridges is restricted to limited segments of the actin fil-
ament known as target zones [4, 5]. These zones are
represented by two to three actin monomers, see Fig. 7.
Moreover, it was found [6] that the probability distribu-
tion of axial offsets from the target zone center is approx-
imately Gaussian and that at least 60% of the attached
cross-bridges are displaced within half of the spacing be-
tween actin monomers which corresponds to the offset of
2.76nm.
The offset can be represented by the reference elonga-
tion y0 = v0 − 1/2 which marks the boundary between
pre and post-power stroke states. Because the parame-
ters v0 and y0 differ by a constant, the variance of δv0
is equal to the variance δy0. Hence, placing disorder in
the energetic bias v0 is equivalent to introducing variable
axial offset.
Figure 7. Schematic representation of the attachment sites.
Each sphere represents an actin monomer; blue color delineate
target zones.
Gaussian distribution of offsets.
If we suppose that
the distribution of axial offsets between the myosin head
and the actin biding site is Gaussian we can estimate its
standard deviation by noting that the probability that
the variable deviation lies in the range ±kσ is given by,
(10)
Pr(µ − kσ ≤ X ≤ µ + kσ) = erf(
k
√2
),
we then use the fact that 60% is in the range ±2.76nm
to find k = 0.842 and σ = 3.3nm.
Critical response in soft and hard ensembles
In Fig. 8 we illustrate the mechanical responses in the
adjacent critical regimes marked as A and B in Fig. 2
of the main text. In the associated critical points, indi-
cated here by small circles and intended to represent the
physiological regime of isometric contractions, the sus-
ceptibilities diverge. The closeness of these two regimes
in the parameter space allows the system to exhibit the
whole repertoire of behaviors from zero to infinite rigid-
ity.
Two half-sarcomeres in series
Here we present an elementary illustration of the fact
that the equilibrium response of a bundle of contractile
units connected in series and placed in a hard device,
cannot be described by local equilibrium constitutive re-
lations obtained in either soft or hard device ensembles.
Instead, the system exhibits an intermediate behavior.
9
Figure 8. The response of the system in the critical regimes A
and B shown in Fig. 2 of the main text : (a) and (b) are the
Helmholtz free energy and the tension-elongation curve in the
hard device ensemble; (c) and (d) are the Gibbs free energy
and the associated tension-elongation curve in the soft device
ensemble. Critical points are marked by the small circles.
Consider two elementary contractile units in series, see
[7] for the analysis of M such elements. Each of the
two elements represents a parallel connection of N cross-
bridges. The total energy per cross bridge in dimension-
less form for a system placed in a hard device reads
1
N(cid:88)
N(cid:88)
N
i
1
N
i
E2 =
1
2
+
[(1 + xi1)vi1 +
1
2
(y1 − xi1)2 +
λf
2
(z1 − y1)2]
[(1 + xi2)vi2 +
1
2
(y2 − xi2)2 +
λf
2
(z2 − y2)2]
(11)
(cid:90)
The equilibrium response of the system is obtained by
computing the partition function
Z2(z, β) =
exp[−2βN E2]δ(z1 + z2 − 2z)dx
(cid:81)N
j dxj2dy2 and z is the (aver-
age) elongation imposed on the system. We can rewrite
(cid:26)
the expression for Z2 in the form
where dx =(cid:81)N
(cid:90)
i dxi1dy1
Z2(z, β) =
dy1dy2 exp
(cid:90)
λf
2
−βN [−
(cid:27)
(z − y1 − y2)2
dvp(v) log Z1(y1, v) Z2(y2, v)]
1
β
−
(12)
κκκκκMline14,5nm5,5nm(a)(b)(c)(d)A−1−0.500.510.20.4z−z0FA−1−0.500.510.911.1z−z0tt0B0.511.500.20.4t/t0GB−1−0.500.510.80.911.1z−z0tt0 2 (yi+1)2
+ e−β(y2
where Zi(yi, v) = e− β
i /2+v). The free en-
ergy per cross-bridge is then F2(z, β) = − 1
2N log Z2(z, β).
The equilibrium tension-elongation relation for this sys-
tem, obtained from the relation t(z, β) = ∂F2(z, β)/∂z,
is shown by the thick line in Fig. 9(a). Similar thick line
in Fig. 9(b) shows the equilibrium response of a single
contractile element placed in the hard device.
Figure 9. (a) Tension elongation relations for a system con-
taining two half-sarcomeres in series placed in a hard device.
Thick line: equilibrium response. Dotted (dashed) line: the
response of two contractile elements in series, each one en-
dowed with its own equilibrium the hard (soft) device con-
stitutive law. (b) Response of a single half-sarcomere. Thick
line: hard device; dashed line: soft device. β = 30, σ = 0,
v0 = 0, λf = 1.
We now compare this behavior with the one obtained
under the assumption that the two elements in series are
characterized by their equilibrium free energies computed
either in a hard or in a soft ensembles.
For instance, using the hard device ensemble we can
write the total (Helmholtz) free energy of the two element
system in the form Ehd
2 = F(z1, β) + F(z − z1, β), where
F is the free energy of a half-sarcomere given by Eq. 5.
The extra variable z1 can be eliminated using the equilib-
rium condition ∂F(z1, β)/∂z1 = ∂F(z − z1, β)/∂z1. The
resulting tension elongation curve is shown in Fig. 9 (a)
by a dotted line.
Similar analysis can be performed based on the re-
sponse functions for the elements loaded in a soft device.
10
Here we need to use equilibrium (Gibbs) free energies
of the elements (Eq. 5 in the main text) and since the
elements in series share the value of tension we obtain
2 = 2G(t, β). The ensuing response of the series bun-
GSD
dle is shown in Fig. 9(a) by a dashed line. In Fig. 9(b),
the dashed line show the equilibrium response of a single
contractile element loaded in a soft device.
Observe, first, that the equilibrium response predicted
by the two 'constitutive models' contains discontinuities,
while the response of the actually equilibrated system
(two half-sarcomeres in series) is smooth. Note also that
the actual response curves do not coincide with either
of the two 'constitutive models' and exhibit some inter-
mediate behavior with features mimicking both models
simultaneously. The observed discrepancy is due to the
fact that in a fully equilibrated system none of the con-
tractile elements is loaded in either soft or hard device
and that the overal response of the system is fundamen-
tally non-affine, see also [7, 8].
∗ [email protected]
† [email protected]
[1] A. F. Huxley and R. M. Simmons, Nature 233, 533 (1971).
[2] M. Caruel and L. Truskinovsky, Phys. Rev. E 93, 062407
[3] I. Vilfan and R. A. Cowley, Journal of Physics C: Solid
State Physics 18, 5055 (1985).
[4] R. T. Tregear, R. J. Edwards, T. C. Irving, K. J. Poole,
M. C. Reedy, H. Schmitz, E. Towns-Andrews, and M. K.
Reedy, Biophysical Journal 74, 1439 (1998).
[5] M. Suzuki and S. Ishiwata, Biophysical Journal 101, 2740
(2016).
(2011).
[6] R. T. Tregear, M. C. Reedy, Y. E. Goldman, K. A. Tay-
lor, H. Winkler, C. Franzini-Armstrong, H. Sasaki, C. Lu-
caveche, and M. K. Reedy, Biophysical Journal 86, 3009
(2004).
[7] M. Caruel and L. Truskinovsky, Reports on Progress in
Physics 81, 036602 (2018).
[8] A. Vilfan and T. Duke, Biophysical Journal 85, 191
(2003).
(a)(b)−2−1012−0.200.2z−z0t/t0−2−1012−0.4−0.200.20.4z−z0t/t0 |
1106.0674 | 2 | 1106 | 2011-08-18T16:37:17 | Stochastic theory of protein synthesis and polysome: ribosome profile on a single mRNA transcript | [
"physics.bio-ph",
"cond-mat.stat-mech",
"q-bio.SC"
] | The process of polymerizing a protein by a ribosome, using a messenger RNA (mRNA) as the corresponding template, is called {\it translation}. Ribosome may be regarded as a molecular motor for which the mRNA template serves also as the track. Often several ribosomes may translate the same (mRNA) simultaneously. The ribosomes bound simultaneously to a single mRNA transcript are the members of a polyribosome (or, simply, {\it polysome}). Experimentally measured {\it polysome profile} gives the distribution of polysome {\it sizes}. Recently a breakthrough in determining the instantaneous {\it positions} of the ribosomes on a given mRNA track has been achieved and the technique is called {\it ribosome profiling} \cite{ingolia10,guo10}. Motivated by the success of these techniques, we have studied the spatio-temporal organization of ribosomes by extending a theoretical model that we have reported elsewhere \cite{sharma11}. This extended version of our model incorporates not only (i) mechano-chemical cycle of individual ribomes, and (ii) their steric interactions, but also (iii) the effects of (a) kinetic proofreading, (b) translational infidelity, (c) ribosome recycling, and (d) sequence inhomogeneities. The theoretical framework developed here will serve in guiding further experiments and in analyzing the data to gain deep insight into various kinetic processes involved in translation. | physics.bio-ph | physics |
Stochastic theory of protein synthesis and polysome:
ribosome profile on a single mRNA transcript
Ajeet K. Sharma1 and Debashish Chowdhury∗1
1Department of Physics, Indian Institute of Technology, Kanpur 208016, India.
The process of polymerizing a protein by a ribosome, using a messenger RNA (mRNA) as the
corresponding template, is called translation. Ribosome may be regarded as a molecular motor for
which the mRNA template serves also as the track. Often several ribosomes may translate the same
(mRNA) simultaneously. The ribosomes bound simultaneously to a single mRNA transcript are
the members of a polyribosome (or, simply, polysome). Experimentally measured polysome profile
gives the distribution of polysome sizes. Recently a breakthrough in determining the instantaneous
positions of the ribosomes on a given mRNA track has been achieved and the technique is called
ribosome profiling [1, 2]. Motivated by the success of these techniques, we have studied the
spatio-temporal organization of ribosomes by extending a theoretical model that we have reported
elsewhere [3]. This extended version of our model incorporates not only (i) mechano-chemical
cycle of individual ribomes, and (ii) their steric interactions, but also (iii) the effects of (a) kinetic
proofreading, (b) translational infidelity, (c) ribosome recycling, and (d) sequence inhomogeneities.
The theoretical framework developed here will serve in guiding further experiments and in analyzing
the data to gain deep insight into various kinetic processes involved in translation.
Key words: ribosome traffic, master equation, extremum current hypothesis, distance-headway,
TASEP.
PACS numbers: 87.16.Ac 89.20.-a
I.
INTRODUCTION
Ribosome [4 -- 6] is a macromolecular complex and operates as one of the essential intracellular machines [7] that
participate in gene expression in all living cells [8, 9]. More specifically, it polymerizes a protein that is a linear hetero-
polymer consisting of amino-acid monomers each of which is linked to the next one by a peptide bond. Therefore,
growing protein is also called a polypeptide. For the synthesis of a protein, a messenger RNA (mRNA) serves as
the template; the sequence of the amino acid species in the protein is determined by that of the codons (triplets of
nucleotides) on the corresponding mRNA template. This process is called translation. Translation by every ribosome
goes through three main stages: (i) initiation, (ii) elongation, and (iii) termination. The start and stop codons mark
the positions on the template mRNA where initiation and termination of translation take place.
During the elongation stage, at every codon, the amino acid monomer required for elongating the protein is supplied
by an incoming tRNA molecule; the correct amino acid monomer is carried by those tRNA whose anti-codon is
complementary to the codon. The machinery of translation deploys a quality control mechanism which screens
the incoming tRNA through a multi-step selection process. However, in spite of this stringent selection process,
occasionally an incorrect amino acid may escape rejection by the quality control system; a translational error results
if the growing protein incorporates an incorrect amino acid monomer thereby lowering the fidelity of translation. In any
case, after the termination, a ribosome is partly disassembled. These parts can reach near the start codon by diffusion
in the surrounding aqueous medium. A ribosome can be assembled more quickly from these parts than from basic
constituents. Moreover, in case the start and the stop codons are close to each other because of the loop formation by
the mRNA, diffusive transfer of the parts of the ribosome from the stop codon to the start codon can be quite rapid
leading to a faster recycling of the ribosomes [10]. Rarely elongation process is aborted because of the premature
detachment of the ribosome from the mRNA track. Furthermore, often several ribosomes translate the same mRNA
transcript simultaneously, each polymerizing a distinct copy of the same protein. Because of the superficial similarities
with vehicular traffic on a given stretch of a highway [11 -- 13], the simultaneous collective translation of a mRNA by
several ribosomes is sometimes referred to as ribosome traffic. The ribosomes bound simultaneously to a single mRNA
transcript are the members of a polyribosome (or, simply, polysome) [14 -- 17]. Because of the mutual hindrance of
the ribosomes, the overall rate of protein synthesis is expected to attain a maximum at an optimum mean separation
between the ribosomes. Finally, the ongoing production and decay of mRNA transcripts and various feedback loops
in gene expression also control the rate of protein synthesis.
∗ Corresponding author(E-mail: [email protected])
2
It would be desirable to capture all the processes mentioned above within a single theoretical model of translation.
However, it is extremely unlikely that such a model can be analyzed analytically. Therefore, the aim of this paper
is more modest. Here we extend our earlier model [3, 18]. capturing (i) the mechano-chemical cycle of individual
ribomes, and (ii) their steric interactions, as well as (iii) the effects of (a) quality-control mechanisms, (b) translational
error, (c) ribosome recycling, and (d) sequence inhomogeity of the mRNA.
The overall rate of synthesis of proteins is a key quantity in any model of translation. However, the main focus
of our theoretical study here are the size of the polysome and the spatial distribution of ribosomes on a mRNA. We
identify the different parameter regimes of our theoretical model and characterize these in terms of the average density
of the polysome and the overall average rate of synthesis of proteins from a single mRNA transcript. Moreover, going
beyond the scope of all the previous theoretical works on this topic, we predict the nature of the fluctuations in the
spatial organizations of the ribosomes which throws light on the fluctuations in the size of ribosome clusters on a
given mRNA transcript.
In this paper we also suggest a new experiment for testing our theoretical predictions on the statistical properties of
polysomes. Traditional technique of polysome profiling [19, 20] provide the number of ribosomes bound to a mRNA,
but not their individual position at the instant when translation was stopped by the experimental protocol. An
improved version of this technique, called ribosome density mapping [21], provides more detailed information on the
numbers of ribosomes associated with specified segments of a particular mRNA by carrying out site-specific cleavage
of the mRNA transcript. The results obtained using these techniques are often adequate for getting a qualitative
indicator of the translational activity. However, the ribosomes are not expected to be uniformly distributed on a
mRNA because of the stochasticities in the steps of the mechano-chemical cycles of these cyclic machines. These
stochasticities arise from (i) intrinsic fluctuations in biochemical processes at low copy numbers of the molecules, and
(ii) extrinsic fluctuations arising from the sequence inhomogeneity of the mRNA. The most detailed picture of the
translational activity has been obtained by a recently developed technique, called ribosome profiling [1, 2]. For testing
some of our theoretical predictions, the older technique of polysome profiling is adequate whereas for testing the other
new results ribosome profiling would be necessary.
This paper is organized as follows: we introduce our model and write down the master equations for the stochastic
kinetics of this model in section II. In section III we solve the master equations in the steady state under periodic
boundary conditions to calculate the overall rate of protein synthesis. The results demonstrate the effects of steric
hindrance caused by congestion ribosome traffic. The spatio-temporal organization of the ribosomes in different
parameter regimes correspond to the different non-equilibrium phases on the "phase diagrams" which we plot in section
IV. The instantaneous spatial distribution of the ribosomes on a single mRNA is also characterized in terms of some
quantitative measures which we introduce in section V where we also explore the effects of sequence inhomogeneities
of mRNA. Finally, in section VI, the main results are summarized and important conclusions are drawn.
II. MODEL
The kinetic models of translation can be divided into three categories. Translation is just a single step in the
broader context of gene expression. However, in most of the kinetic models of gene expression [22], the details of
the mechano-chemistry of individual ribosomes as well as their mutual steric interactions are ignored. The rates
of synthesis and degradation of proteins are captured usually in these models by two rate constants without any
mechanistic details of these two processes. We are not concerned with a global picture of gene expression in this paper
and, therefore, such kinetic models will not be discussed further here.
There are models of translation which are intended to describe various key aspects of the stochastic mechano-
chemical kinetics of only a single ribosome. In contrast, another class of models of translation is motivated by the
polysome formation. Most of these models capture the effects of entire mechano-chemical cycle by a single parameter.
These models focus mainly on the effects of mutual steric interactions of the ribosomes on the overall rate of protein
synthesis. In this section we develop a model by capturing both these aspects of translation, namely, details of single-
ribosome mechano-chemistry and the effects of steric interactions among the ribosomes on the same mRNA transcript.
However, for the convenience of comparison of our work with earlier works, we summarize the main features of the
TASEP-type models in the next subsection.
A. TASEP-type models
Totally asymmetric simple exclusion process (TASEP) [23, 24] is one of the simplest models of interacting self-
propelled particles; it is used extensively for understanding the generic features of non-equilibrium steady-states of
interacting systems. TASEP and its various extensions exhibit interesting dynamical phase transitions [25]. For many
3
years, various biologically motivated extensions of TASEP [26 -- 36] have been used to model ribosome traffic. In the
TASEP-based models of ribosome traffic (see ref.[34] for a recent review) each lattice site represents a single codon.
Since a ribosome is much larger than a single codon, each ribosome is represented by a hard rod that covers ℓ (ℓ > 1)
sites simultaneously. But, the allowed step size of a rod is one lattice site (i.e., one codon). This extended version of
TASEP for hard rods will be referred to as ℓ-TASEP. As long as a site remains covered by a ribosome, it is inaccessible
to the other rods. The entry of a rod from one end (at a rate α) and its eventual exit from the other end (at a rate
β) model the initiation and termination stages of translation by a ribosome.
The steps of the mechano-chemical cycle of individual ribosomes during the elongation stage were not captured
explicitly in the simple TASEP-type models; instead, one single "hopping" parameter was used to describes the rate
of translation of one codon. Moreover, these TASEP-type models neither incorporate any mechanism for selecting
specifically the correct amino acid monomer, nor do these allow for the possibility of translational error. Therefore,
such TASEP-type models are too simple to account for the effects of various mechano-chemical processes on the
statistical properties of polysomes.
B. Our model: unification of single-ribosome mechano-chemistry and TASEP
In recent years, progressively more realistic models of translation have been developed [3, 18, 37 -- 39] and several
analytical results have been derived. Using the most recent version of this model [3], some statistical properties of
single ribosome have been derived analytically [3]. Here we extend this model even further to capture some features
of translation which were not included in its earlier version. Using this extended version of our model of translation,
we make experimentally testable predictions on the dependence of the statistical properties of the polysomes on the
various mechano-chemical processes involved in translation.
Each ribosome consists of two subunits which are designated as "large" subunit and "small" subunit, respectively.
The translation of the genetic message encoded in the codon is carried out by the small subunit while the elongation
of the polypeptide, by the formation of a peptide bond between it and the incoming amino acid, takes place in the
large subunit. The function of the two subunits is coordinated by the tRNA molecules. There are three binding sites
for a tRNA on each ribosome; these sites are designated as E,P and A. An incoming tRNA binds with an A site.
The amino acid carried by a tRNA is linked to the growing polypeptide by a peptide bond while the tRNA is bound
to the P site. Finally, the denuded tRNA exits from the ribosome from the E site. During the elongation stage,
in each complete mechano-chemical cycle, the ribosome steps forward on the template mRNA by one codon while,
simultaneously, the polypeptide gets elongated by one amino acid. These processes are captured explicitly in the our
kinetic model.
The distinct mechano-chemical states in our model and the allowed transitions among these states are shown
schematically in fig.1. At the beginning of each cycle, the system is in state 1 where the sites E and A are empty
while the site P is occupied by a tRNA that has just contributed its amino acid to the growing polypeptide. A tRNA
charged with an amino acid is called a aminoacyl tRNA (aa-tRNA). At this stage, an aminoacyl tRNA, bound to
an elongation factor Tu (EF-Tu) and a molecule of Guanosine triphosphate (GTP) enters and binds with the A site
on the Ribosome at the site A. This process takes place with rate ωa which causes transition of the system to the
chemical state 2. Thereafter non-cognate tRNAs are rejected, and the system reverts back to state 1, with rate ωr1,
through a quality control mechanisms based on the free energy of codon-anticodon matching.
However, the free-energy difference between the cognate and near-cognate tRNAs is too small to distinguish between
them. Therefore, usually, near-cognate tRNAs are not rejected at this stage. A second stage of quality control, called
kinetic proofreading [40, 41], is then activated. GTP, which is bound to the aa-tRNA, is hydrolyzed to GDP by EF-Tu
and this process is described by the transition from the state 2 to the state 3. At this stage, barring a few exceptional
cycles, the near cognate tRNAs are rejected from chemical state 3 which drives the system back to the chemical state
1; this happens with rate constant ωr2.
Although most often the noncongnate and near cognate tRNAs are rejected by the two-stage selection process, still
occasionally the quality control system fails to reject an incorrect (non-cognate or near-cognate) tRNA. Consequently,
there is a small, but non-vanishing probability, of a translational error when the growing polypeptide elongates by
the formation of a peptide bond with an incorrect amino acid. In our model, the incorporation of incorrect amino
acid leads to a branched pathway: in contrast to the transition 3 → 4 along the correct pathway, the wrong pathway
proceeds by the transition 3 → 4∗. Arrival of another elongation factor called EF-G, alongwith a molecule of GTP also
takes place at this stage. The transition 4 → 5 (or, 4∗ → 5∗) is reversible and essentially a Brownian rotation of the
two subunits relative to each other. This spontaneous Brownian rotation drives the two tRNA molecules back-and-
forth between the classical P/P, A/A state and the hybrid E/P, P/A state [42]. The rate constants for the forward
and backward Brownian rotations are denoted by ωbf and ωbr, respectively, along the correct pathway whereas the
same transitions along the wrong pathway take place with the rates Ωbf and Ωbr, respectively. Finally, hydrolysis of
4
GTP by EF-G drives the process of translocation at the end of which the two tRNA molecules are positioned at the
E and P sites while the ribosome finds itself poised to translate the next codon; the denuded tRNA molecule makes
an exit from the E site. The transition 5 → 1 and 5∗ → 1 take place with the rates ωh2 and Ωh2, respectively. The
completion of the full cycle elongates the protein by one amino acid (by correct amino acid along one pathway and by
an incorrect amino acid along another branch) and translocates the ribosome by one codon on the template mRNA
(For further details, see ref.[42]).
Since our model allows the possibility of translational error, we define [18] the fidelity φ of translation by the fraction
of the incorporated amino acids which are correct, i.e.,
φ = ωp/(ωp + Ωp)
(1)
In our model, the mRNA track is represented by a one-dimensional lattice where each of the total L sites corresponds
to a single codon. The length of a ribosome is denoted by ℓ in the units of the length of a single codon. A ribosome
can move forward by only one site (i.e., one codon) at a time. We use the convention that the leftmost site covered
by a ribosome is the one that is being translated by it; the leftmost site covered by a ribosome is also used in our
formulation to denote the position of a ribosome. Thus, throughout this paper, we follow the convention that, at any
instant of time, a ribosome "covers" ℓ sites but "occupies" only the leftmost of these ℓ sites.
According to our notation, the status of coverage of a site is denoted by 0 and 1; 0 represents an unoccupied lattice
site whereas 1 represents a covered site. Since many ribosomes move simultaneously on the same track they also
interact with each other. The simplest form of interaction would be mutual exclusion: if the ith site is "occupied"
by one ribosome, then all the ℓ sites from i to i + ℓ − 1 are "covered" by it and, therefore, none of the these ℓ sites
are accessible to any other ribosome at that instant of time. Moreover, a ribosome occupying the position i can
move forward if, and only if, the site i + ℓ is not simultaneously occupied by another ribosome. In our notation, the
symbol P (1..............1
0) represents the conditional probability that, given an uncovered site, there will be successive
ℓ
{z
}
ℓ
z
1..............1 0)
ℓ adjacent sites to its left all of which are covered simultaneously by a single ribosome. Similarly, P (
is the conditional probability of finding a empty site j, given that the successive ℓ sites on its left are covered by a
ribosome.
}
{
Using the same notation, we now define Q(ii + ℓ) as the conditional probability that the site i + ℓ is not occupied
by another ribosome, given that the site i is occupied by a ribosome, Similarly, given that the site i is occupied by a
ribosome, the probability that the site i− ℓ is not occupied by another ribosome is given by the conditional probability
Q(i − 1i). It is straightforward to show that [38, 39]
Q = (1 − ρℓ)/(1 + ρ − ρℓ)
(2)
Where ρ = N/L is the number density of the Ribosome on the mRN A track. Following the same prescription that
one of us (DC), and his collaborators, used in earlier simpler models of ribosome traffic [38, 39], we multiply the rate
constants ωh2 and Ωh2 by Q because a ribosome "feels" the mutual exclusion only when it tends to move forward to
the next codon.
By the symbol Pµ(i, t) we denote the probability of finding the ribosome in the µth chemical state while it occupies
the site i at time t. The master equations for the probabilities Pµ(i, t) are
dP1(i, t)/dt = −ωaP1(i, t) + ωr1P2(i, t) + ωr2P3(i, t) + ωh2QP5(i − 1, t) + Ωh2QP ∗
5 (i − 1, t)
dP2(i, t)/dt = ωaP1(i, t) − (ωr1 + ωh1)P2(i, t)
dP3(i, t)/dt = ωh1P2(i, t) − (ωp + Ωp + ωr2)P3(i, t)
dP4(i, t)/dt = ωpP3(i, t) − ωbf P4(i, t) + ωbrP5(i, t)
dP5(i, t)/dt = ωbf P4(i, t) − (ωh2Q + ωbr)P5(i, t)
dP ∗
4 (i, t)/dt = ΩpP3(i, t) − Ωbf P ∗
4 (i, t) + ΩbrP ∗
5 (i, t)
dP ∗
5 (i, t)/dt = Ωbf P ∗
4 (i, t) − (Ωh2Q + Ωbr)P ∗
5 (i, t)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Equations (3)-(7) correspond to the equations (52)-(56) of ref.[39]. However, there are some additional terms in (3)-(7)
because of (i) the kinetic proofreading, (ii) translational because of wrong amino acid selection, and (iii) reversible
nature of the transition between the states 4 and 5. Moreover, the precise interpretations of the respective states
were slightly different in ref.[39] (see ref.[39] for the detailed interpretations of the states and transitions between the
states).
Note that the normalization condition for the probabilities is
5
Pµ(i, t) + P ∗
4 (i, t) + P ∗
5 (i, t) = ρ
(10)
5
Xµ=1
Molecular mechanisms that lead to mistranslation have been under intense investigation for decades (see [43] for a
recent review). However, to our knowledge, none of the earlier models, including that developed in ref.[39], provides
a mathematical framework to treat the mechanisms of mistranslation analytically.
III. RATE OF PROTEIN SYNTHESIS: EFFECTS OF HINDRANCE IN RIBOSOME TRAFFIC
We solve the equations (3)-(9) in the steady state under normalization condition (10) and imposing periodic bound-
ary conditions (PBC). Using the steady-state solutions for Pµ we get the following expression for the steady-state
flux
JP BC = (P5ωh2 + P ∗
K −1
ef f = ω−1
a (cid:18)1 + (ωr1/ωh1)(cid:19)(cid:18)1 + (ωr2/ωp)(cid:19) + ω−1
+ (cid:18)Ωp/ωp(cid:19)(cid:20)ω−1
a (cid:18)1 + (ωr1/ωh1)(cid:19) + ω−1
h1 + Ω−1
5 Ωh2)Q = ρKef f(cid:18)1 + (Ωp/ωp)(cid:19)
h1(cid:18)1 + (ωr2/ωp)(cid:19) + ω−1
bf (cid:18)1 + (Ωbr/Ωh2Q)(cid:19) + (Ωh2Q)−1(cid:21)
p + ω−1
bf (cid:18)1 + (ωbr/ωh2Q)(cid:19) + (ωh2Q)−1
(11)
(12)
(13)
(14)
(15)
where
where
and
Separating out the Q-dependent and Q-independent parts, Kef f can be re-expressed as
K −1
ef f = k−1
1 + (k2Q)−1
k−1
1 = ω−1
a (cid:18)1 + (ωr1/ωh1)(cid:19)(cid:18)1 + (ωr2/ωp)(cid:19) + ω−1
h1(cid:18)1 + (ωr2/ωp)(cid:19) + ω−1
bf (cid:21)
+ (cid:18)Ωp/ωp(cid:19)(cid:20)ω−1
h1 + Ω−1
p + ω−1
bf
a (cid:18)1 + (ωr1/ωh1)(cid:19) + ω−1
h2(cid:18)1 + (ωbr/ωbf )(cid:19) + Ω−1
k−1
2 = ω−1
h2(cid:18)1 + (Ωbr/Ωbf )(cid:19)(Ωp/ωp)
Note that when simultaneously ωr2 → 0, Ωp → 0, ωbr → 0, the expressions for k1 and k2 reduce to Kef f and ωh2,
respectively.
So the JP BC is given by
JP BC =(cid:20)k2ρ(1 − ρℓ)(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)/(cid:20)(cid:18)1 + (k2/k1)(cid:19)(1 − ρℓ) + ρ(cid:21).
(16)
Since no premature detachment of ribosomes are allowed in our model, the flux of the ribosomes is also the total rate
of protein synthesis. In the special case ωr2 = ωbr = Ωp = Ωbf = Ωbr = Ωh2 = 0 the expression (16) reduces to
the corresponding expression for flux derived in ref.[39]. The expression (16) for JP BC (ρ) exhibits a single maximum
which occurs at the number density
ρ∗ =(cid:20)s(cid:18)1 + (k2/k1)(cid:19)ℓ−1(cid:21)/(cid:20)1 +sℓ(cid:18)1 + (k2/k1)(cid:19)(cid:21)
(17)
The equations (16) and (17) reduce, respectively, to the equations (58) and (63) of ref.[39].
6
IV. NATURE OF POLYSOMES: NON-EQUILIBRIUM PHASE DIAGRAM OF RIBOSOME TRAFFIC
In this section we impose open boundary conditions (OBC) which is more realistic for modeling translation. The
entry of the ribosomes at one open end captures the initiation of translation while the exit of the ribosomes from the
other open end mimics termination of translation.
A. Phase diagram
In a multi-dimensional abstract space spanned by some of the crucial model parameters, we identify the distinct
regions characterized by their distinctive properties which we describe below. The resulting diagram is referred to as
a "phase" diagram although the "phases" are not in thermodynamic equilibrium; these phases are non-equilibrium
steady states of the system. The theoretical prediction we make in this section can be tested by using the technique
of polysome profile [19, 20].
Our calculations here are based on the extremum current hypothesis (ECH) [44 -- 47] which relates the flux J , under
OBC, to the flux JP BC (ρ) of the same system, under PBC. We apply the ECH to our model in the same way in
which it was used earlier for the simpler versions of our model [38, 39]. We assume that the entrance and exit points
of the track (i.e., the start and stop codons) are connected to two infinite particle reservoirs where the respective
number densities are ρ− and ρ+, respectively. According to ECH, for systems with a single maximum in the JP BC (ρ)
function,
J = max JP BC (ρ) if ρ− > ρ > ρ+
(18)
These relations can be utilized to draw the surfaces separating the dynamical phases on the phase diagram. The
first step in this approach has already been completed by calculating the expression (17) for ρ∗. Next, we derive the
expressions for ρ− and ρ+ which would give rise to the same rates of initiation and termination as indicated by α and
β, respectively.
Suppose P jump
−
(∆t) is the probability that, given an empty site, a ribosome will hop onto it from left in the next
time interval ∆t. It is straightforward to see that
where P5 and P ∗
5 are given by the expressions
P jump
−
(∆t) = P (1..............1
0)(P5ωh2 + P ∗
5 Ωh2) × ∆t
ℓ
{z
}
P −1
5 = ωh2(k−1
1 + k−1
2 )
and
respectively, and as discussed in ref.[39],
P ∗
5 = (Ωp/ωp)P5,
Note that the solutions (20) and (21) have been obtained using the normalization condition
P (1..............1
0) = ρ/(1 + ρ − ρℓ)
ℓ
{z
}
Pµ + P ∗
4 + P ∗
5 = 1
5
Xµ=1
for the reservoir.
Now ρ− is the solution of the equation α = P jump
−
. Hence,
ρ− = α(cid:18)1 + (k2/k1)(cid:19)/(cid:20)(ℓ − 1)(cid:18)1 + (k2/k1)(cid:19)α + Pk2(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)
with
Pk2 = k2 (∆t)
(19)
(20)
(21)
(22)
(23)
(24)
(25)
relating the probability Pk2 with the rate constant k2 where k2 is given by (15). Similarly,
ℓ
P jump
+
(∆t) = P (
where
ℓ
z
1..............1 0)(P5ωh2 + P ∗
}
{
5 Ωh2)∆t
The unknown density ρ+ is the solution of the equation β = P jump
; hence,
+
P (
z
1..............1 0) = (1 − ρℓ)/(1 + ρ − ρℓ)
}
{
ρ+ =(cid:20)β(cid:18)1 + (k2/k1)(cid:19) − Pk2(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)/(cid:20)β(ℓ − 1)(cid:18)1 + (k2/k1)(cid:19) − ℓPk2(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)
7
(26)
(27)
(28)
Note that the equations (24) and (28) reduce, respectively, to the equations (67) and (70) of ref.[39] in the appropriate
limit.
1. Surface separating LD and MC phases
From MCH it follows that the surface separating the LD and MC phases on the phase diagram of the system is
obtained from the equation
by expressing ρ− and ρ∗ in terms of the rate constants for the elementary steps of the model kinetics. Hence, the
equation for this surface in the phase diagram of the model is found to be
ρ− = ρ∗
(29)
α =(cid:20)ρ∗Pk2(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)/(cid:20)(cid:18)1 + (k2/k1)(cid:19)[1 − (ℓ − 1)ρ∗](cid:21)
2. Surface separating HD and MC phases
Similarly, using the condition
ρ+ = ρ∗
the equation for the surface separating the HD and MC phases in the phase diagram of the system is given by
β =(cid:20)Pk2(1 − ρ∗ℓ)(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)/(cid:20)(cid:18)1 + (k2/k1)(cid:19)[1 − (ℓ − 1)ρ∗](cid:21)
3. Surface of coexistence of HD and LD phases
which gives
JP BC (ρ−) = JP BC (ρ+)
(30)
(31)
(32)
(33)
α =(cid:20)Pk2(cid:18)1 + (Ωp/ωp)(cid:19)β(cid:18)1 + (k2/k1)(cid:19)(cid:21)/(cid:20)Pk2(cid:18)1 + (Ωp/ωp)(cid:19)ℓ + β(cid:18)1 − ℓ + 2(k2/k1) − ℓ(k2/k1) + (k2
2/k2
1)(cid:19)(cid:21) (34)
Equivalently
β =(cid:20)αℓPk2(cid:18)1 + (Ωp/ωp)(cid:19)(cid:21)/(cid:20)(cid:18)1 + (k2/k1)(cid:19)(cid:26)Pk2(cid:18)1 + (Ωp/ωp)(cid:19) + α(ℓ − 1) − (k2/k1)α(cid:27)(cid:21)
(35)
8
Note the equations (30), (32), (34) and (35) reduce to the expressions (72), (74), (76) and (77), respectively, of ref.[39]
in the appropriate limit.
Figs.2 and 3 show the 3-dimensional phase diagrams plotted in the α− β − φ space from two different perspectives,
while figs.4 and 5 show the 3-dimensional phase diagrams plotted in the α − β − Pωr2 space also from two different
perspectives. Since none of the earlier models of ribosome traffic capture translational fidelity and kinetic proofreading
explicitly, these phase diagrams have not been reported ever before.
In order to compare the implications of these phase diagrams with that of the TASEP, we also project several
two-dimensional cross sections of these phase diagrams onto the α− β-plane (see figs.6 and 7). On the 2D projections,
transition from the LD phase to the MC phase takes place at α = α∗ whereas the transition from the HD phase to
the MC phase takes place at β = β∗.
As is evident from these 2D phase diagrams, the curvature of the lines of coexistence of HD and LD phases seems
to be the generic feature of all such models [39, 48]. The straight line α = β on which the LD and HD phases of
TASEP coexist is a manifestiation of the "particle-hole" symmetry in TASEP, a special property that is not shared
by our model of ribosome traffic.
Increasing φ shifts α∗ and β∗ to higher values. Increase of φ = ωp/(Ωp + ωp) can be viewed as a result of increasing
ωp which, in turn, increases the effective rate of hopping of a ribosome from one codon to the next. It is well known
[49] that higher values of effective hopping rate shifts α∗ and β∗ to higher values. Similarly, increasing ωr2 decreases
the effective hopping rate thereby shifting α∗ and β∗ to smaller values.
B. Effects of recycling on the phase diagram
Recycling of ribosomes can be captured by our model by replacing the constant initiation rate α by an effective
initiation rate αef f . Since the availability of ribosomes for initiation is proportional to the flux of ribosomes exiting
from the stop codon, we postulate that
αef f = α + qJ(αef f , β,{ωi})
(36)
where the coefficient q depends on the relative separation between the two ends of the mRN A transcript as well as
on the diffusion constant of the ribosome subunits in the solution.[32]. Note that the prescription (36) for recycling
is similar in spirit, but not identical quantitatively, to the prescription used by Gilchrist and Wagner [50] because the
latter model does not capture steric exclusion among the ribosomes.
On simple physical grounds, the effect of (36) on the phase diagram is expected to be non-trivial. Suppose, α is
varied keeping β fixed. At a very small value of α (α ≪ β), the flux would be determined by α. Starting from a very
small value, increasing α initially increases J which, in turn, increases the effective initiation rate αef f . However,
beyond a certain value of α, αef f is no longer rate limiting and the flux becomes independent of α.
Exploiting the relation between our model and ℓ-TASEP, we plot the phase diagram for the extended model that
captures recycling of ribosomes through αef f . We use the results reported in refs.[29, 34] for ℓ-T ASEP and get
J = [αef f (1 − αef f )]/[1 + αef f (ℓ − 1)]
αef f = α + q(cid:18)[αef f (1 − αef f )]/[1 + αef f (ℓ − 1)](cid:19)
which gives
So at LD and HD interface
αef f = [α(ℓ − 1) + q − 1 +p(α(ℓ − 1) + q − 1)2 + 4(ℓ − 1 + q)α]/[2(ℓ − 1 + q)]
(37)
(38)
(39)
(40)
(41)
at LD and MC boundary
and at HD and MC boundary
β = αef f
αef f = 1/(1 + √ℓ)
β = 1/(1 + √ℓ)
(42)
The resulting 2-d phase diagram in the α − β plane is plotted in fig.8. Increase of the recycling factor q shifts α∗ to
a smaller value while β∗ remains unchanged. This trend of variation is consistent with the fact that recycling affects
only the initiation rate without influencing the rate of termination. This is consistent with the trends of variation of
the coverage density and flux of ribosomes that we observed in our computer simulations which we descibe below.
9
1. Variations of flux and coverage density with the extent of recycling
In order to demonstrate the effects of recycling in a form that would be closer to one's physical intuition, we now
show the variation of the average flux and the average coverage density of the ribosomes with the parameter q which
is a measure of the extent of recycling. For this purpose, we carried out computer simulations of our model for several
different values of q. All the data reported here were obtained for L = 1000 and ℓ = 10. Since the linear size of a
ribosome is measured here in the units of codons, ℓ = 10 is a realistic choice because in recent experiments [1, 2] it
has been observed that each ribosome covers about 30 nucleotides on the mRNA track. Typical values of some of
the rate constants have been reported in the literature [51, 52]. For those rate constants whose numerical values are
not available in the literature, we have assumed some reasonable values based on physical intuition. However, our
conclusions are not sensitive to the precise numerical values of the rate constants.
The
rate
constants
which
we
used
for
the
simulations
are
ωa=25s−1,
ωh1=25s−1,ωh2=25s−1,ωp=25s−1,ωbf =25s−1,ωbr=25s−1,ωr1 =10s−1,ωr2 =10s−1,Ωp=5s−1,Ωbf =5s−1,ωbr=5s−1,ωh2=5s−1
All the rate constants were converted to dimensionless transition probabilities using the formula Pω =
1 − exp(−ω ∗ (∆t)) where time step ∆t = 0.005s is used for all the simulation runs.
At first sight, it may appear that there is some ambiguity in the defnition of αef f : what value of J should be
used in (36)? In order to avoid ambiguity, we average the spatially-averaged flux further over the time period elapsed
since the last entry of a ribosome (i.e., the initiation of translation by the ribosome closest to the start codon). This
"doubly-averaged" value of flux J is used in (36) to compute αef f .
In fig(9) we plot the flux and the coverage density of the ribosomes as functions of q. Starting from a vanishingly
small value, as q is increased, both the flux and coverage density increase. However, beyond a limiting value, both
become practically independent of q; this trend of variation is caused by a transition from the LD phase to either the
HD phase or to the MC phase, depending on the values of the set of other parameters.
'
C. Experimental tests of the phase diagram with polysome profile
The three different phases are characterized by three different densities; the expressions for these densities have
been derived above. Therefore, our theoretical predictions can be tested by measuring the average densities. For
this purpose, polysome profiling [19, 20] would be adequate. However, all the analytical calculations for the phase
diagram have been carried out for sequence-homogeneous mRNA strands. Therefore, a poly-U strand of mRNA, with
appropriate start and stop codons [3], should be used in the experiment.
'
V.
INSTANTANEOUS SPATIAL DISTRIBUTION OF RIBOSOMES: RIBOSOME PROFILE
In all the sections above we have calculated quantitative characteristics which do not require information on the
spatial distributions of the ribosomes on the mRNA transcript. In this section we explore some other quantitative
features of ribosome traffic which deal with the spatial distributions of the ribosomes. Our theoretical predictions on
the spatial distributions of the ribosomes can be tested with the ribosome profiling technique [1, 2].
'
A. Distance-headway distribution
The distance-headway (DH) is defined as the spatial separation between two successive ribosomes on the same
mRNA transcript. At any given instant of time, the magnitude of the DH fluctuates from one pair of ribosomes to
another, the instantaneous spatial distribution of the ribosmes is characterized by the corresponding distribution of the
DHs. The DH distribution is used extensively for quantitative characterization of macroscopic vehicular traffic [11, 12].
In this subsection we calculate the DH distribution for our kinetic model of ribosome traffic. Since ribosome profiling
[1, 2] provides the exact positions of the ribosomes at the instant when translation was stopped, DH distribution can
be extracted by repeatiting this profiling sufficiently large number of times.
Our system may be viewed as one that consists of M identical rods, each of length ℓ, distributed over a lattice of
L sites. First, assuming a ring-like mRNA track we get the DH distribution for the corresponding number density ρ
10
which, because of the PBC, does not fluctuate. In this case, the expression for the DH distribution is given by [29]
Pdh(m, ρ) = (ρ/ρs)(ρh/ρs)m
where ρh = 1 − ρℓ is the density of holes and ρs = ρ + ρh = 1 + ρ − ρℓ. Hence,
Pdh(m, ρ) = [ρ(1 − ρℓ)m]/[(1 + ρ − ρℓ)m+1]
(43)
(44)
The number density of the ribosomes for the real system under OBC is a fluctuating quantity. But, the mean density
deep inside the bulk (around the central region of the lattice) can be extracted numerically from computer simulations.
Substituting the numerically estimated density of the ribosomes under OBC into the expression (44) we get the DH
distribution under OBC. This distribution is plotted in fig.10. The straight lines on the semi-log plot reflects the
geometric nature of the distribution (discrete analog of the exponential distribution). In order to test the validity
of the approximate scheme used above to derive the DH distribution by a combination of analytical and numerical
arguments, we have also computed the DH distribution directly by computer simulation; the simulation data are also
plotted in fig.10. The theoretically derived lines are in reasonaly good agreemwnt with the DH distribution obtained
by computer simulation.
'
B.
Influence of slow codons on the density profile and flux
It is well known that translation of some codons take place at a very slow rate; these are often referred to as
"hungry" codon. However, we'll use the term "slow" to refer to the all those codons which get translated at a much
slower rate than other codons. In this subsection we explore the effects of bottlenecks created by such slow codons
against the forward movements of ribosomes. In particular, we investigate the effects of slow codons on the average
density profile and flux of ribosomes in ribosome traffic.
In the model that we simulated for this purpose, a mRN A transcript consists of only two different types of codons.
In the computer simulations of our model we assign ten times smaller numerical value to ωa for a slow codon compared
to that of a normal codon. Simultaneously, the numerical value of ωr1 assigned to a slow codon is ten times larger
than that of a normal codon. Since in this particular study we are interested mainly in the effects of bottlenecks, we
ignore the possibility of misincorporation by erasing the branched pathway putting Ωp = 0 = Ωh2.
In the first set of simulations, we put four slow codons at the center of the stretch of mRNA between the start
and the stop codon. Such a single extended bottleneck leads to a "phase-segregated" profile where on one side of the
bottleneck the average density is much higher than that on the other side (see fig11). Profiles are plotted for different
values of ωr2. Higher value of ωr2 reduce the effective hopping rate for both normal as well as rare codons. But, the
effective hopping rate for slow codon decrease more because of higher value of ωr1/ωh1. Thus, the higher is the value
of ωr2, the larger is the difference between the effective rates of hopping from normal and slow codons. This, in turn,
leads to the larger jump discontinuity of the density across the bottleneck at a larger value of ωr2.
In the second set of simulations, the slow codons were not clustered together. Instead, four equispaced slow codons
were placed at the sites 200, 400, 500, and 800 on a lattice of total length L = 1000.The average density profiles for
this case are plotted in fig.12 for two different values of ωr2. Both the profiles exhibit discontinuous jumps in the
coverage density; the position of each minimum in the coverage density coincides with the location of a slow codon.
Moreover, periodic oscillations are also observed in the vicinity of the the rare codon where periodicity is ℓ. Similar
results were obtained earlier in TASEP-type models of ribosome traffic [34]; however, unlike our data, shown in fig.12,
the effects of kinetic proofreading and futile cycles could not be addressed by the model of ref.[34].
In order to emphasize the effect of clustering of slow codons on the overall rate of protein synthesis we plot flux as
a function of ωr2 in fig13 for the two different conditions discussed above. The upper curve corresponds to the setup
where four slow codons are placed equidistant on lattice. The lower curve corresponds to the setup where all the four
slow codons are placed clustered together at the center of the system. The data clearly show that without increasing
the number of slow codons the rate of synthesis of proteins can be reduced drastically by clustering the slow codons
into a single bottleneck.
'
1. Probing spatial distribution of ribosomes
Ribosome profiling technique [1, 2] is ideally suited to probe the instantaneous spatial distribution of ribosomes
on the same mRNA transcript. But, our model does not take into account the variation of rate constants arising
11
from sequence inhomogeneity of the mRNA transcript. Therefore, at first sight, it may appear that a homogeneous
sequence (e.g., a poly-U) would be most appropriate transcript for testing our theoretical prediction. However, the
technique of ribosome profiling [1, 2] cannot locate the exact positions of the ribosomes on a sequence- homogeneous
mRNA transcript. Therefore, we suggest that the experiment should be performed with a special-type of sequence-
inhomogeneous mRNA transcript where, because of the intrinsic degeneracy of the genetic code, all the codons are
synonymous, i.e., correspond to the same amino acid. Only the cognate tRNA molecules carrying the correct amino
acid are to be supplied to the solution. We do not expect significant codon-to-codon variation of the rate constants in
this case. For studying the effects of translational fidelity and proofreading, tRNA molecules carrying a non-cognate
species of amino-acids should also be supplied.
In case it turns out to be difficult to extract the exact positions of the ribosomes from ribosome profiling of such
a mRNA strand where the degenerate codons are distributed randomly, we suggest an alternative strategy. Recall
that six synonymous codons correspond to the same amino acid Arg; similarly there is six-fold degeneracy also for
the amino acids Leu and ser (see fig.14(a)). In principle, one can synthesize an artificial mRNA transcript of the type
shown in fig.14(b) using six synonymous codons from any of the three possible sets shown in fig.14(a). A mRNA
transcript with such a non-random inhomogeneous codon sequence allows unambiguous identification of the positions
of the ribosomes on it while using the ribosome profiling technique [1].
VI. SUMMARY AND CONCLUSION
In this paper we have developed a theoretical framework that captures several key features of translation as well as
the spatio-temporal organization of polysomes. First, the selection of aa-tRNA in our model is a two-stage process; the
second stage captures kinetic proofreading. Second, our model allows occasional translational error and we calculate
several quantities as functions of the translational fidelity. We have also incorporated some of the other features of
the mechano-chemical cycle of ribosomes in the elongation stage which, to our knowledge, have not been incorporated
in any earlier model. On the basis of our hypothesis for capturing the effects of ribosome recycling, we have predicted
the effects of recycling on the spatial profile of the ribosomes as well as on the rate of protein synthesis.
Here we have also investigated the spatial organization of ribosomes in polysomes in terms of the distance-headway
distribution; it is a quantity that is used extensively to characterize crowding in vehicular traffic. We have also
identified the parameter regimes which display distinct characters of polysomes and the corresponding rates of protein
production in our model system. Finally, we have also demonstrated the effects of sequence inhomogeneity of the
mRNA transcript, particularly, that of the clustering of slow codons.
We hope this work will inspire experimental investigation for measuring new quantities. It is possible to test some
of our new predictions using polysome profile techniques. But, more interesting results on spatial distributions of the
ribosomes on the mRNA transcript would require ribosome profiling [1, 2].
Acknowledgements
This work is supported by IIT Kanpur through the Dr. Jag Mohan Chair professorship to one of the authors (DC).
DC also thanks the visitors program of the Max-Planck Institute for Physics of Complex Systems for hospitality in
Dresden during the preparation of this manuscript.
VII. REFERENCES
[1] N.T. Ingolia, S. Ghaemmaghami, J.R.S. Newman and J.S. Weissman, Genome-wide analysis in vivo of translation with
nucleotide resolution using ribosome profiling, Science 324 (2009) 218-223.
[2] H. Guo, N.T. Ingolia, J.S. Weissman and D.P. Bartel,Mammalian microRNAs predominantly act to decrease target mRNA
levels, Nature 466 (2010) 835-840.
[3] A.K. Sharma and D. Chowdhury, Distribution of dwell times of a ribosome: effects of infidelity, kinetic proofreading and
ribosome crowding, Phys. Biol. 8 (2011) 026005
[4] A. S. Spirin, Ribosomes, Springer (2000).
[5] A.S. Spirin, Ribosome as a molecular machine, FEBS Lett. 514 (2002) 2-10.
[6] J. Frank and C.M.T. Spahn, The ribosome and the mechanism of protein synthesis, Rep. Prog. Phys. 69 (2006) 1383-1417.
[7] J. Frank (ed.), Molecular machines (Cambridge University Press, 2011).
[8] B. Alberts et al., Essential Cell Biology, Garland Science (2003).
[9] H. F. Lodish, Model for the regulation of mRNA translation applied to haemoglobin synthesis, Nature 251 (1974) 385-388.
12
[10] G. Hirokawa, N. Demeshkina, N. Iwakura, H. Kaji and A. Kaji, The ribosome-recycling step: consensus or controversy? ,
Trends. Biochem. Sci. 31 (2006) 143-149.
[11] D. Chowdhury, L. Santen and A. Schadschneider, Statistical physics of vehicular traffic and some related systems Phys.
Rep. 329 (2000) 199-329.
[12] A. Schadschneider, D. Chowdhury and K. Nishinari, Stochastic transport in complex systems: from molecules to vehicles
ELSEVIER, Amsterdam, The Netherlands (2010).
[13] D. Chowdhury, A. Schadschneider and K. Nishinari, Physics of Transport and Traffic Phenomena in Biology: from molec-
ular motors and cells to organisms, Phys. of Life Rev. 2 (2005) 318-352.
[14] J.R. Warner, A. Rich and C.E. Hall, Electron microscope studies of ribosomal clusters synthesizing hemoglobin Science 138
(1962) 1399-1403.
[15] J.R. Warner, P.M. Knopf and A. Rich, A multiple ribosomal structure in protein synthesis, PNAS 49 (1963) 122-129.
[16] A. Rich, The excitement of discovery, Annu. Rev. Biochem. 73 (2004) 1-37.
[17] H. Noll, The discovery of polyribosomes, Bioessays 30 (2008) 1220-1234.
[18] A.K. Sharma and D. Chowdhury, Quality control by a mobile molecular workshop: Quality versus quantity, Phys. Rev. E
82 (2010) 031912.
[19] Y. Arava, Y. Wang, J.D. Storey, C.L. Liu, P.O. Brown and D. Herschlag, Genome-wide analysis of mRNA translation
profiles in saccharomyces cerevisiae, PNAS 100 (2003) 3889-3894.
[20] E. Mikamo, C. Tanaka, T. kanno, H. Akiyama, G. Jung, H. Tanaka and T. Kawai, Native polysomes of Saccharomyces
cerevisiae in liquid solution observed by atomic force microscopy, J. Struct. Biol. 151 (2005) 106-110.
[21] Y. Arava, F. Edward Boas,Patrick O. Brown and Daniel Herschlag, Dissecting eukaryotic translation and its control by
ribosome density mapping, Nucl. Acids Res. 33 (2005) 2421-2432.
[22] Zhdanov, Kinetic models of gene expression including non-coding RNAs, Phys. Rep. 500 (2011) 1-42.
[23] B. Derrida, Phys. Rep. 301 (1998) 65-83.
[24] G. M. Schutz, Phase Transitions and Critical Phenomena, vol. 19, Acad. Press (2001).
[25] D. Mukamel, Phase transitions in nonequilibrium systems, in: Soft and Fragile matter: nonequilibrium dynamics, metasta-
bility and flow, eds. M.R. Evans and M.E. Cates (Taylor and Francis, 2000).
[26] C. MacDonald, J. Gibbs and A. Pipkin, Kinetics of biopolymerization on nucleic acid templates, Biopolymers 6 (1968)
1-25.
[27] C. MacDonald and J. Gibbs,Concerning the kinetics of polypeptide synthesis on polyribosomes, Biopolymers, 7 (1969)
707-725.
[28] G. Lakatos and T. Chou, Totally asymmetric exclusion processes with particles of arbitrary size , J. Phys. A 36 (2003)
2027-2041.
[29] Leah B. Shaw, R.K.P. Zia and Kelvin H Lee, Totally asymmetric exclusion process with extended objects: A model for
protein synthesis, PRE 68 (2003) 021910
[30] L.B. Shaw, J.P. Sethna and K.H. Lee,Mean-field approaches to the totally asymmetric exclusion process with quenched
disorder and large particles, Phys. Rev. E 70 (2004) 021901.
[31] L.B. Shaw, A.B. Kolomeisky and K.H. Lee, Local Inhomogeneity in Asymmetric Simple Exclusion Processes with Extended
Objects J. Phys. A 37 (2004) 2105-2113.
[32] Tom Chou, Ribosome recycling, diffusion, and mRNA loop formation in translational regulation,Biophysical Journal 85
(2003) 755-773.
[33] T. Chou and G. Lakatos, Clustered Bottlenecks in mRNA Translation and Protein Synthesis, Phys. Rev. Lett. 93 (2004)
198101.
[34] R.K.P. Zia, J.J. Dong and B. Schmittmann, Modeling Translation in Protein Synthesis with TASEP: A Tutorial and Recent
Developments, J. Stat. Phys. (2011), and references therein.
[35] H. Zouridis and V. Hatzimanikatis, Effects of codon distributions and tRNA competition on protein translation, Biophys.
J. 95, (2008) 1018-1033.
[36] L.M.Y.T. Romero, M. Silber and V. Hatzimanikatis, The origins of time-delay in template biopolymerization processes,
PLoS Comp. Biol. 6, (2010) e1000726.
[37] L. Ciandrini, I. Stansfield and M.C. Romano, Role of the particle's stepping cycle in an asymmetric exclusion process: A
model of mRNA translation, Phys. Rev. E 81 (2010) 051904.
[38] A. Basu and D. Chowdhury, Traffic of interacting ribosomes: effects of single-machine mechano-chemistry on protein
synthesis , Phys. Rev. E 75 (2007) 021902.
[39] A. Garai, D. Chowdhury, D. Chowdhury and T.V. Ramakrishnan, Stochastic kinetics of ribosomes: Single motor properties
and collective behavior, Phys. Rev. E 80 (2009) 011908.
[40] J. J. Hopfield, Kinetic Proofreading: a new mechanism for reducing errors in biosynthetic process requiring high specificity,
Proc. Nat. Acad. Sci. 71 (1974) 4135-4139
[41] Jacques Ninio, Kinetic amplification of enzyme discrimination, Biochimie 57 (1975) 587-595
[42] X. Agirrezabala and J. Frank, Elongation in translation as a dynamic interaction among the ribosome, tRNA, and elon-
gation factors EF-G and EF-Tu, Quart. Rev. Biophys. 42 (2009) 159-200.
[43] N. M. Reynolds, B.A. Lazazzera and M. Ibba, Cellular mechanisms that control mistranslation, Nat. Rev. Microbiol. 8,
(2010) 849-856.
[44] J. Krug, Phys. Rev. Lett. Boundary-induced phase transitions in driven diffusive systems, 67 (1991) 1882-1885.
[45] V. Popkov and G. Schutz, Steady-state selection in driven diffusive systems with open boundaries, Europhys. Lett. 48
(1999) 257-263.
13
[46] J. Hager, J. Krug, V. Popkov and G. Schutz, Minimal current phase and universal boundary layers in driven diffusive
systems, Phys. Rev. E 63 (2001) 056110.
[47] J. Hager, Extremal principle for the steady-state selection in driven lattice gases with open boundaries, Phys. Rev. E 63
(2001) 067103.
[48] T. Antal and G.M. Schutz, Asymmetric exclusion process with next-nearest-neighbor interaction: Some comments on traffic
flow and a nonequilibrium reentrance transition, Phys. Rev. E 62 (2000) 83-93.
[49] A. B. Kolomeisky, Asymmetric simple exclusion model with local inhomogeneity J.Phys. A. 31 (1998) 1153-1164.
[50] M.A. Gilchrist and A. Wagner, A model of protein translation including codon bias, nonsense errors, and ribosome recycling,
J. Theor. Biol. 239 (2006) 417-434.
[51] R.C. Thompson, D.B. Dix and J. F. Eccleston, Single turnover kinetic studies of guanosine triphosphate hydrolysis and
peptide formation in the elongation factor Tu-dependent binding of aminoacyl-tRNA to Escherichia coli ribosomes, J biol.
chem. 255 (1980) 11088-11090.
[52] K. M. Harrington, I. A. Nazarenko, D.B.Dix, R.C. Thompson and O.C. Ulhenbeck, In vitro analysis of translational rate
and accuracy with an unmodified tRNA, Biochemistry 32 (1993) 7617-7622.
14
Fig.1: Detailed mechanochemical cycle of Ribosome on its track. The integer indices ..., i − 1, i, i + 1... label the
codons on the mRNA transcript. Although the same set of transitions are allowed from each codon, only those from
(and to) the codon i are shown explicitly.
Figure Captions
Fig.2: Phase diagram of ribosome traffic model in the 3-dimensional space spanned by α,β and φ.
Fig.3: Same data as in fig.2, except that plotted from a different perspective.
Fig.4: Phase diagram of ribosome traffic model in the 3-dimensional space spanned by α,β and Pωr2 .
Fig.5: Same data as in fig.4, except that plotted from a different perspective.
Fig.6: Projections of several two-dimensional cross sections, of the three-dimensional phase diagram, plotted in figs.2
and 3, onto the α − β plane. Each cross section corresponds to a fixed value of φ.
Fig.7: Projections of several two-dimensional cross sections, of the three-dimensional phase diagram, plotted in figs.4
and 5, onto the α − β plane. Each cross section corresponds to a fixed value of ωr2.
Fig.8: 2D Phase diagram of TASEP and ℓ-TASEP in the α − β plane in the presence of recycling of ribosomes.
Fig.9: Variation of the average flux and coverage density with the recycling factor q. The green and red curves have
the same α values whereas the red and blue curves have the same β values.
Fig.10: Distribution of distance-headways. The lines have been obtained by using the formula (44) whereas the
discrete data points have been obtained directly from computer simulations.
Fig11: Density profile of ribosomes for single bottleneck.
Fig.12: Density profile of ribosomes for slow sites at i = 200, 400, 600, 800.
Fig.13: Flux variation with ωr2 for both lattices (see text for detail).
Fig.14: Codon-sequence suggested for testing our theoretical predictions.
15
FIG. 1:
16
I: HD-LD phase boundary
II: HD-MC phase boundary
III: LD-MC phase boundary
II
0.002
III
0.003
0
0.001
0.002
α
0.003
I
1
0.9
0.8
0.7
0.6
φ
φ
0
β
0.001
FIG. 2:
I: HD-LD phase boundary
II: HD-MC phase boundary
III: LD-MC phase boundary
II
I
φ
φ
1
0.9
0.8
0.7
0.6
0.003
0.002
α
0
0.001
0.002
β
0.003
0
III
0.001
FIG. 3:
17
I: HD-LD phase boundary
II: HD-MC phase boundary
III: LD-MC phase boundary
II
I
0.02
0.016
0.012
Pω
Pω
r2
r2
0.008
0.004
0
0
0.001
β
0.002
III
0.003
0
0.001
0.002
α
0.003
FIG. 4:
I: HD-LD phase boundary
II: HD-MC phase boundary
III: LD-MC phase boundary
II
I
0.02
0.016
Pω
Pω
r2
r2
0.012
0.008
0.004
0
0.003
0.002
α
III
0.001
FIG. 5:
0
0.001
0.002
β
0.003
0
18
0.004
0.003
LD
β
0.002
0.001
0
0
0.0005
MC
φ=.1
φ=.3
φ=.5
φ=.7
φ=.9
HD
0.001
α
FIG. 6:
0.0015
0.002
LD
0.004
0.003
β
0.002
0.001
Pω
Pω
Pω
Pω
Pω
=0
r2
=.005
r2
=.010
r2
=.015
r2
=.020
r2
MC
HD
0
0
0.0005
α
0.001
0.0015
0.002
FIG. 7:
TASEP
MC
1
0.8
LD
β
0.6
0.4
0.2
0
0
HD
19
q=0.0
q=.25
q=.50
q=.75
q=1.0
β
0.5
0.4
0.3
0.2
0.1
l TASEP
LD
MC
q=0.0
q=.25
q=.50
q=.75
q=1.0
HD
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
α
0
0
0.1
0.2
α
0.3
0.4
FIG. 8:
y
t
i
s
n
e
D
e
g
a
r
e
v
o
C
1
0.8
0.6
0.4
0.2
0
0
0.0002
0.00015
0.0001
x
u
l
F
5e-05
0
α=.1, β=.7, LD→ HD
α=.05, β=.7, LD→ HD
α=.05, β=200, LD→ MC
1000
2000
q (recycling factor)
3000
4000
α=.1, β=.7, LD→ HD
α=.05, β=.7, LD→ HD
α=.05, β=200, LD→ MC
1000
2000
q(recycling factor)
3000
4000
FIG. 9:
20
Simulation α = 0.10
Formula α=0.10
Simulation α=0.20
Formula α=0.20
Simulation α=0.30
Formula α=0.30
50
60
70
ω
r2
ω
r2
-1
=0 s
-1
=40 s
Simulation α =0.10
Formula α=.010
Simulation α =0.20
Formula α =0.20
Simulation α =0.30
Formula α =.30
10 20 30 40 50 60
m
20
40
30
m
FIG. 10:
0.2
0.15
0.1
0.05
0
0
10
0
-4
-8
-12
h
d
P
)
h
d
P
(
g
o
L
-16
-20
0
1
e
g
a
r
e
v
o
c
f
o
y
t
i
l
i
b
a
b
o
r
P
0.8
0.6
0.4
0.2
0
0
200
400
codon
FIG. 11:
600
800
1000
21
1
e
g
a
r
e
v
o
c
f
o
y
t
i
l
i
b
a
b
o
r
P
0.8
0.6
0.4
0.2
ω
r2
ω
r2
-1
=0 s
-1
=40 s
0
0
200
400
600
800
1000
codon
FIG. 12:
Equidistant defects
Random defects
Central defect
0.00025
0.0002
x
u
l
F
0.00015
0.0001
0
20
40
ω
r2
-1
)
(s
60
80
100
FIG. 13:
(a)
22
(b)
FIG. 14:
|
1305.4715 | 1 | 1305 | 2013-05-21T05:24:11 | Reduced Dimensionality tailored HN(C)N Pulse Sequences for Efficient Backbone Resonance Assignment of Proteins through Rapid Identification of Sequential HSQC peaks | [
"physics.bio-ph",
"q-bio.BM"
] | Two novel reduced dimensionality (RD) experiments -(4,3)D-hNCOcaNH and (4,3)D-hNcoCANH- have been presented here to facilitate the backbone resonance assignment of proteins both in terms of its accuracy and speed. The experiments basically represent an improvisation of previously reported HN(C)N experiment [Panchal et. al., J. Biomol. NMR. (2002), 20 (2), 135-147] and exploit the simple reduced dimensionality NMR concept [Szyperski et. al. (2002), Proc. Natl. Acad. Sci. U.S.A. 99(12), 8009-8014] to achieve (a) higher dispersion and resolution along the co-evolved F1 dimension and (b) rapid identification of sequential HSQC peaks on its F2(15N)- F3(1H) planes. The current implementation is based on the fact that the linear combination of 15N and 13CO/13Ca chemical shifts offers relatively better dispersion and randomness compared to the individual chemical shifts; thus enables the assignment of crowded HSQC spectra by resolving the ambiguities generally encountered in HNCN based assignment protocol because of amide 15N shift degeneracy. Additionally, each of these experiments enables assignment of backbone 13CO/13Ca resonances as well. Overall, the reduced dimensionality tailored HNCN experiments presented here will be of immense value for various structural and functional proteomics studies by NMR; particularly of intrinsically or partially/unstructured proteins and medium sized (MW ~12-15 kDa) folded proteins. The experiments, like any other experiment that yields protein assignments, would be extremely valuable for protein folding and drug discovery programs as well. | physics.bio-ph | physics | Reduced Dimensionality tailored HN(C)N Pulse Sequences for Efficient Backbone
Resonance Assignment of Proteins through Rapid Identification of Sequential
Kumar D
HSQC peaks
Dinesh Kumar
Centre of Biomedical Magnetic Resonance, SGPGIMS Campus, Raibareli Road, Lucknow-226014, India
*Address for Correspondence:
Dr. Dinesh Kumar
(Assistant Professor)
Centre of Biomedical Magnetic Resonance (CBMR),
Sanjay Gandhi Post-Graduate Institute of Medical Sciences Campus,
Raibareli Road, Lucknow-226014
Uttar Pradesh-226014, India
Mobile: +91-9044951791,+91-8953261506
Fax: +91-522-2668215
Email: [email protected]
Webpage: http://www.cbmr.res.in/dinesh.html
KEYWORDS: Reduced Dimensionality NMR; Backbone Assignment; HN(C)N; (4,3)D-hNCOcaNH; (4,3)D-
hNcoCANH; Sequential Amide Correlations.
ABBREVIATIONS: NMR, Nuclear Magnetic Resonance; HSQC, Heteronuclear Single Quantum Correlation;
CARA, Computer Aided Resonance Assignment; RD: Reduced dimensionality.
1
Kumar D
Abstract:
Two novel reduced dimensionality (RD) experiments (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH have
been presented here to facilitate the backbone resonance assignment of proteins both in terms of its
accuracy and speed. The experiments basically represent an improvisation of previously reported HN(C)N
experiment [Panchal et. al., J. Biomol. NMR. (2002), 20 (2), 135-147] and exploit the simple reduced
dimensionality NMR concept [Szyperski et. al. (2002), Proc. Natl. Acad. Sci. U.S.A. 99(12), 8009-8014] to
achieve (i) higher dispersion and resolution along the co-evolved F1 dimension and (ii) rapid identification
of sequential HSQC peaks on its F2(15N)- F3(1H) planes. The current implementation is based on the fact that
the linear combination of 15N and 13C′/13Cchemical shifts offers relatively better dispersion and
randomness compared to the individual chemical shifts; thus enables the assignment of crowded HSQC
spectra by resolving the ambiguities generally encountered in HN(C)N based assignment protocol because
of amide 15N shift degeneracy. Additionally, each of these experiments enables assignment of backbone
13C′/13C resonances as well. Overall, the reduced dimensionality tailored HN(C)N experiments presented
here will be of immense value for various structural and functional proteomics studies by NMR;
particularly of intrinsically/partially unstructured proteins and medium sized (MW~12-15 kDa) folded
proteins. The experiments –like any other experiment that yields protein assignments- would be extremely
valuable for protein folding and drug discovery programs as well.
2
Kumar D
Introduction:
Backbone amide assignment forms the basis for variety of structural and functional proteomics
studies by NMR. Likewise in modern drug discovery research programs, the accurate assignment of 1H-15N
HSQC spectra of target proteins is extremely crucial where backbone amide (1H and 15N) shift
perturbations are used for screening the potential drug candidates [1]. However, the currently available
strategies [2-4] based on standard multidimensional NMR experiments [4] e.g. HNCA/ HN(CO)CA [5,6],
HN(CA)CO/HNCO [7], and CBCANH [8]/CBCACONH [9] are highly time consuming both in terms of data
acquisition and data analysis. Most of these strategies involve recording of several 3D spectra and
extensive analysis; therefore are neither ideally suited for high-throughput structural and functional
proteomics studies by NMR nor for modern drug discovery research programs. Further, the established
assignment strategies -where sequential connectivities between the neighboring residues are established
using correlated aliphatic 13Cand carbonyl 13C' chemical shifts [2,3]- require repeated scanning through
the 15N planes of the 3D spectra to search for the matching correlation peaks along the carbon dimension.
Therefore, backbone assignment based on these conventional NMR experiments has remained relatively a
slow and arduous process requiring an enormous investment of time and hard-work, particularly in cases
when the proteins are intrinsically/partially unstructured or are denatured for various folding/unfolding
studies by NMR. In all such cases, unambiguous and rapid identification of sequential HSQC peaks (using
these NMR spectra) remains the important rate-limiting step in the resonance assignment process.
In the above context, the pulse sequences proposed earlier for identifying direct sequential
correlations between amide (1H and 15N) chemical shifts of one residue with those of its adjacent neighbors
can be very useful [10-12]. However, the strategies based on some of these experiments involve the use
two 3D experiments which not only elaborate the analysis, but sometimes may lead to complications due to
inter-spectral variations of chemical shifts (a problem common in proteins which are unstable in solution).
Interestingly, the HN(C)N experiment proposed earlier [13] and based on novel assignment protocol [14]
can be extremely useful in this regard i.e. for rapid sequence specific assignment of backbone HN and 15N
resonances in (15N, 13C) labeled proteins. The protocol exploits the directly observable amide HN and 15N
sequential correlations and the distinctive peak patterns in the different planes of the HN(C)N spectrum ;
the self (intra-residue) and sequential (inter-residue) correlations appear opposite in peak sign (+ and -)
and thus can be easily identified without involving any additional complementary experiment . Glycines and
prolines, which are responsible for special patterns, provide many check/start points for mapping the
stretches of sequentially connected amide cross-peaks on to the primary sequence for assigning them
sequence specifically. Overall, the elegant spectral features of HN(C)N enable rapid data analysis and
render side chain assignments less crucial for the success of backbone amide assignment. Recently, this
3
Kumar D
novel protocol has also been automated by our group and the algorithm has been given the name AUTOBA
[15]. However for a given protein, the success of HN(C)N based assignment protocol depends critically on
the dispersion of peaks in its 1H–15N HSQC spectrum. Overlap of peaks in this spectrum can lead to
cancellation of + and - intensities in the HN(C)N spectrum [14]. This happens under two circumstances:
first, when the 15N chemical shifts of neighboring residues are too close and second, when the diagonals
having + and - intensities have accidentally close 15N chemical shifts [14]. In such a situation, the sequential
peaks can be absent (because of opposite sign and higher intensity of self peaks in the spectrum) and
assignment process may break or may be wrongly considered as proline break point. The situation arises
mainly because of the fact that both the indirect dimensions in the HN(C)N spectrum involve 15N chemical
shifts.
In this backdrop, two novel reduced dimensionality experiments (4,3)D-hNCOcaNH and (4,3)D-
hNcoCANH have been presented here to augment the novel HN(C)N based backbone assignment protocol
further, particularly, for protein systems which exhibit high-degree of amide 15N shift degeneracy. These
experiments are a result of a simple modification of the basic HN(C)N pulse sequence [13] and differ in the
way the t1 evolution is handled according to reduced dimensionality NMR approach [16,17]. The purpose of
current modification is to improve the dispersion and resolution of HN(C)N spectrum mainly to resolve the
problems arising because of amide 15N shift degeneracy. This has been achieved by joint sampling the
backbone 15Ni chemical shift with the backbone 13C
i-1/13C’i-1 chemical shift (where i is a residue number).
This implementation is based on the fact that the linear combination of 15Ni and 13C
i-1/13C’i-1 chemical shifts
provides dispersion better compared to the dispersion of the individual chemical shifts as elicited in Figure
S1 (Appendix I of Supplementary Material). As evident from Figure S1, the particular advantage with
these new experiments is the dispersion of peaks in the sum and difference frequency regions which
can be different; thus facilitates the analysis of the spectrum by simultaneously connecting
(15N+13C) and (15N+13C) correlations (here 13C represents backbone 13C’ or 13C chemical shift of
sequential i-1 residue). However, the only limitation of the method can be long experiment time
requirement in order to achieve sufficient resolution along co-evolved NC dimension where spectral width
may range from 2 to 3 times the original 15N spectral width. Thus for identical resolution, the total
experiment time in each case will be almost two-three times to that of the basic HN(C)N experiment.
However, for proteins with high shift degeneracy, the benefit incurred in terms of increased spectral
resolution and dispersion becomes more crucial compared to the increased total experiment time.
Nevertheless, the longer experiment time problem can be circumvented by using non-uniform
(sparse) sampling along indirectly detected dimensions [18]; thus making NMR data acquisition fast. Since
the experiments proposed here does not involve aliphatic protons, the overall experiment time can be
4
Kumar D
reduced further by modifying these experiments according to either BEST NMR [19] or L-optimization [20]
approaches which allow rapid data collection of NMR spectra via minimizing the inter-scan recycle delay
without loss of sensitivity. Rapid data collection generally becomes a crucial issue whenever the protein
under investigation is unstable in solution; the protein either precipitates or degrades inside the NMR
sample tube in matter of days. The other disadvantage associated with the experiments presented here is
their slightly reduced sensitivity compared to the basic HN(C)N experiment. The sensitivity will be lower
(a) by a factor of 2 with respect to the basic HN(C)N experiment and (b) by a factor of 2 compared to an
alternative 4D experiment with separate frequency labeling (e.g. like 4D-hNCOCANH). However with the
currently available higher magnetic fields and efficient cryo-probes offering higher signal to noise ratio, the
sensitivity of the experiment will not be a serious issue and both these experiments can be successfully
applied on higher (< 15 kDa) molecular weight proteins; particularly the intrinsically unstructured or
chemically denatured proteins where the inherent backbone flexibility renders increased sensitivity of the
experiment [21,22]. Further, both these experiments provide easy identification of self and sequential
correlation peaks because of their opposite signs; these are readily amenable to be exploited in automated
assignment algorithms like any other set of NMR experiments that yield protein assignments in an
automated manner [21]. Performance of these experiments has been tested and demonstrated here on two
folded proteins -bovine apo calbindin-d9k (Ca2+ free apo form) and human ubiquitin- and on a highly
disordered protein UNC60B denatured in 8M urea.
5
Kumar D
Materials and Methods:
The proposed reduced dimensionality experiments -(4,3)D-hNCOcaNH and (4,3)D-hNcoCANH- have
successfully been tested and demonstrated on three proteins including both folded (e.g. 75 amino acid long
bovine apo Calbindin-d9k and 76 amino acid long human ubiquitin) and unfolded ones (e.g. unfolded
UNC60B, a 152 amino acid protein denatured in 8M urea). The 13C/15N labeled bovine apo Calbindin
sample (1 mM in concentration, 50 mM Ammonium Acetate, pH 6.0, 10% D2O, in high quality NMR tube
sealed
under
inert
atmosphere)
has
been
purchased
from
Giotto
Biotech,
Itlay:
http://www.giottobiotech.com/); whereas 13C/15N labeled human ubiquitin (1.2 mM dissolved in phosphate
buffer pH 6.5 in 90% H2O and 10% D2O has been purchased from Cambridge Isotope Laboratories, Inc.,
USA (http://www.isotope.com). The 13C/15N labeled sample (final concentration ~ 0.8 mM and pH 6.0) of
8M urea-denatured state of UNC60B (a 152 amino acid ADF/cofilin family protein of Caenorhabditis
elegans) has been prepared as described in Appendix II (Supplementary Material). All the experiments
have been performed at 298 K on a Bruker Avance III 800 MHz NMR spectrometer equipped with a
cryoprobe. The various acquisition and processing parameters used in these experiments have been listed
in Table S1. The delays
,
and
were set to 28, 9, and 25 ms, respectively. Gaussian cascade Q3
pulses [23] were used for band selective excitation and inversion along the 13C channel. The 13C carrier
frequencies for pulses in 13C and 13C’ channel were set at 54.0 ppm and 174 ppm, respectively.
WATERGATE scheme [24] is used to suppress the water signal. Water saturation is minimized by keeping
water magnetization along the z-axis during acquisition with the use of water selective 90ᵒ pulses [25]. The
NMR data was processed using Topspin (BRUKER, http://www.bruker.com/) and analyzed using CARA
[26]. A point to be mentioned here is that though the experiments have been carried out at 800 MHz, the
high magnetic fields can be detrimental for the sensitivity of these experiments. This is because of the fact
that the transverse relaxation rate of the carbonyl carbons is dominated by the chemical shift anisotropy
mechanism [27] and both these experiments involve substantially long delays with transverse carbonyl
carbon magnetization required for the 13C’⟶13C transfers (Fig. 1A and 1B).
6
2NT2,2CN Kumar D
Results and Discussion:
Description of Pulse sequences
The schematic presentation of magnetization transfer pathway along with the respective frequency
labeling schemes and the pulse sequences of the proposed reduced dimensionality experiments (4,3)D-
hNCOcaNH and (4,3)D-hNcoCANH have been shown in Fig. 1. Figure 1A and 1B traces the magnetization
transfer pathways and Fig. 1C and 1D represent, respectively, the pulse sequences (4,3)D-hNCOcaNH and
(4,3)D-hNcoCANH. The pulse sequences have been derived by the simple modification of the basic 3D
HN(C)N experiment described earlier [13] and differ in the way the t1 evolution is handled according to
reduced dimensionality (RD) NMR [16,17]. In case of (4,3)D-hNCOcaNH experiment, the t1 evolution period
involves co-evolution of backbone 15Ni and 13C’i-1 chemical shifts (Fig. 1A and 1C); while in case of (4,3)D-
hNcoCANH experiment, the t1 evolution involves co-evolution of backbone 15Ni and 13C
i-1 chemical shifts
(Fig. 1B and 1D). Each experiment leads to the spectrum equivalent to basic HN(C)N spectrum, but with
higher chemical shift dispersion and randomness along the F1 axis, referred here as NC dimension.
Therefore, the transfer efficiency functions which dictate the intensities of self and sequential correlation
peaks in their respective spectra would be the same as those in the HN(C)N spectrum described previously
[13,28,29]. Therefore, only the distinguishing features of these spectra have been presented here. As shown
schematically in Figure 2, the peaks appear at the following coordinates in their respective spectra:
The letters “H” and “N” here refer to amide 1H and 15N chemical shifts, whereas the letters “
” and “
”
refer to the linear combination of 15Ni and 13Ci-1 chemical shifts (where i is residue number and 13C
represents the backbone 13C or 13C’ chemical shift). In Figure 2, the letters N and N, have been
represented as 15N and 15N, respectively, and depending upon the offset of RF pulses used along 13C
channel (i.e. 13Coffset), these can be evaluated as:
where “k” is the scaling factor and is equal to (13C)/ (15N)=2.48.
Thus, the F2(15N)-F3(1H) plane corresponding to
, shows two intra-residue amide correlation
peaks: a self (Hi, Ni) and a sequential (Hi-1, Ni-1) (Fig. 2, right panel) and the F1(NC)-F3(1H) plane
7
1231121311/, (,) (,),(,), (,)(,),(,),(,),(,)iiiiiiiiiiiiiiiFNNFFHNHNFNFFHNHNHNHNiNiN151513131151513131*() (1)*() (2)obsobsiiioffsetobsobsiiioffsetNNCCNNCC1/iiFNN Kumar D
corresponding to F2 = Ni, shows four correlation peaks: (
) (Fig. 2, left
panel). As shown in Figure 1, the self (notion used here for
,
and
) and sequential (notion
used here for
,
and
) correlation peaks have opposite peak signs in different planes of
these spectra except in special situations; in situations where a Glycine/Proline residue is involved, the
actual sign patterns of the two kinds of peaks in each set would vary as in HN(C)N [10]. For glycines,
different patterns arise depending on whether the ith or i-1th residue is a glycine or otherwise. For example,
in a stretch –XGYZ-, in the F1-F3 plane at the 15N chemical shift of G, all peaks appear with same sign
(positive) and likewise, at the 15N chemical shift of Y, all the four peaks will also have the same but
opposite sign (i.e. Negative). For the same reason, in a double glycine stretch like GG’Z the inter- and intra-
residue correlation peaks have distinct patterns. Similarly, in the case of proline at position i+1, absence of
amide proton results in the absence of sequential (
) correlation peaks. These special peak
patterns at glycines, proline and the residues present next to them provide important start or check points
during the course of sequential assignment process. Thus, for a given protein with a known amino acid
sequence, it is possible to identify several special triplet sequences simply by inspecting the various F1−F3
and F2-F3 planes of these spectra. Considering different triplets of residues, covering the general and all the
special situations, all the expected peak patterns in the F1-F3 and F2-F3 planes of these spectra are
schematically shown in Figure S2 (with details explained in Appendix-III, Supplementary material).
The above described features of the reduced dimensionality HN(C)N experiments have been tested
and demonstrated here on all the three proteins: bovine apo calbindin (75 aa), human ubiquitin (76 aa) and
unfolded UNC60B (152 aa). Figure 3 shows the experimental demonstration of these features for the
F1(NC)-F3(1H) and F2(15N)-F3(1H) planes of the (4,3)D-hNcoCANH spectrum of unfolded UNC60B. Figure
S3A and S3B shows the experimental demonstration of these features for the F1(NC)-F3(1H) planes of
(4,3)D-hNCOcaNH spectrum of human ubiquitin and (4,3)D-hNcoCANH spectrum of bovine apo-calbindin,
respectively. Figure S4 shows the experimental demonstration of these features for the F2(15N)-F3(1H)
planes of the (4,3)D-hNcoCANH spectrum of bovine apo-calbindin.
Facile Identification of Sequential Amide-Cross Peaks:
Like basic HN(C)N experiment, the beneficial feature of reduced dimensionality HN(C)N
experiments presented here is that they provide rapid identification of sequential HSQC peaks through
F2(15N)F3(1H) planes of their respective spectra along the co-evolved F1(NC) dimension. The process has
been illustrated experimentally in Fig. 4. As evident, the sequential (i+1) HSQC peaks can be identified on
8
11,, ,,, and ,iiiiiiiiHNHNHNHNiiHNiNiN11iiHN1iN1iN11, and ,iiiiHNHN Kumar D
the F2(15N)F3(1H) planes of these spectra at F1=
/
chemical shifts identified for residue, i.
Compared to the routine experiments (like HNCA, HN(CA)CO, CBCANH, etc), which also provide the
identical feature along the 13C dimension, the advantage here is the presence of opposite peak signs for self
and sequential correlations; therefore the sequential HSQC peaks (Hi+1, Ni+1) can be easily differentiated
from the self HSQC peaks (Hi,Ni). The process is facilitated further (both in terms of accuracy and speed) by
the random and well dispersed nature of 15N chemical shifts present along the jointly sampled F1(NC)
dimension. Further, a particular advantage is that there are two different sets of sequential correlations i.e.
15N+ (up-field domain) and 15N (down-field domain). Thus, any ambiguity in the identification of (i → i+1)
sequential correlation using up-field set of peaks can be resolved using down-field set of peaks and vice
versa.
A point to be mentioned here is that the separation between sum and difference frequencies (i.e.
15Nand 15N) in the final spectra of (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH experiments has been
achieved by keeping the offsets of 13C channel at 70 and 188 ppm, respectively. In order to achieve this
separation between the addition and subtraction frequencies (i.e. 15Nand 15N), one should be careful
about (i) the spectral width used along the co-evolved dimension and (ii) the offset frequencies used along
13C channel. The combined analysis of average 13C and 13C' chemical shifts of all the amino acids (Fig. 1;
data taken from BMRB) and the Eqns 1 and 2 revealed that for a given protein the two frequency regions
(i.e. 15Nand 15N) will be separated from each other (i) if the 13C carrier is kept 1–2 ppm away from the
most down fielded shifted 13C chemical shift; this is typically close to ~188 and ~70 ppm, respectively, for
13C’ channel [in case of (4,3)D-hNCOcaNH] and for 13C channel [in case of (4,3)D-hNcoCANH] and (ii) the
spectral width along the co-evolved F1(NC) dimension is increased accordingly; typically close to ~80 and
~150 ppm, respectively, for (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH spectra. Frequency selection along
all the indirect dimensions has been done by States-TPPI method [30] (in the coupled N-C dimension,
quadrature detection is performed only for the 15N) and the data is acquired in a way that it is processed
using normal Fourier Transformation (for detail see this reference [31]).
Overall, the true sequential HSQC peak (Hi+1,Ni+1) will be of opposite sign on both the 1H–15N planes
of spectrum at F1=
/
chemical shifts identified for residue, i. An illustrative example has been
shown in Figure 3 depicting the use of correlations observed in the F2(15N)F3(1H) planes of (4,3)D-
hNcoCANH spectrum of 8M urea-denatured UNC60B (for residues Thr36 to Val39) at the
chemical shifts identified for residue i for establishing a sequential (i ⟶ i+1) connectivity between the
HSQC peaks. As evident, the spectrum provides rapid and unambiguous identification of sequential HSQC
peaks which is particularly important while performing sequential assignment of intrinsically/partially
9
1iN1iN1iN1iN151511/iiNN Kumar D
unstructured proteins, proteins containing repetitive amino acids (e.g.…TSAAGTTTE…) or repetitive
stretches of amino acids (e.g. …QPLAGA… QPLAGA…) as well as for protein folding/unfolding studies by
NMR. Backbone assignment following conventional strategies has always been a great challenge in all such
cases where the poor dispersion of 13C chemical shifts (e.g. as in case of CBCANH) and their dependence
on the amino-acid type result in the crowding of self and sequential HSQC peaks on 1H–15N planes of these
spectra at the degenerate carbon chemical shifts. Likewise in HN(C)N spectrum, the 15N chemical shift
degeneracy may also lead to ambiguities in the identification of sequentially connected HSQC peaks (more
details are given below). Overall, the reduced dimensionality HN(C)N experiments presented here would
greatly facilitate the resonance assignment of proteins exhibiting crowded HSQC spectra (i) by resolving
the problems arising because of amide 15N shift degeneracy and (ii) by providing rapid and unambiguous
identification of sequential HSQC peaks.
Assignment Protocol:
The assignment protocol based on the reduced dimensionality HN(C)N experiment [i.e. (4,3)D-
hNCOcaNH or (4,3)D-hNcoCANH] has been illustrated schematically in Figure 4 using an example stretch
of amino acids -XYZ- where X represents residue i. As illustrated in Fig. 4, a sequential connectivity (i to
i+1) between the HSQC peaks can be established following the routine assignment procedure by matching
the self (here
) and sequential (here
) correlation peaks identified along the F1(NC)
dimension of F1(NC)–F3(1H) planes at 15Ni chemical shift. This is experimentally demonstrated in Fig. 3A
using the F1-F3 strips from (4,3)D-hNcoCANH spectrum of unfolded UNC60B. Like HN(C)N, the remarkable
feature of these RD HN(C)N spectra is that they provide easy discrimination between self and sequential
correlation peaks because of their opposite peak signs and thus no other complementary experiment is
required for discriminating them. The other advantage of the protocol is that it does not involve
cumbersome search through various F2 planes of the 3D spectrum for matching the frequencies along the
F1 dimension. Rather, a sequential i to i1 connectivity between two HSQC peaks can be established
unambiguously on the F2(15N)–F3(1H) planes of spectrum at
(Fig. 4B). The process has
been demonstrated experimentally in Fig. 3 using F2(15N)–F3(1H) planes of (4,3)D-hNcoCANH spectrum of
8M urea denatured UNC60B for residues Thr36-Val39. After establishing all the possible i → i+1
connectivities, the stretches of sequentially connected HSQC peaks are then mapped on to the primary
sequence for assigning them sequence specifically. For this, the strategy makes use of triplet specific peak
patterns present in these spectra as described in Appendix III and Figure S2 (Supplementary material). As
evident from the figure, these are the glycines and prolines which are responsible for special patterns of
10
and iiNN11 and iiNN1111 or iiFNFN Kumar D
self and sequential correlation peaks. These special patterns provide identification of certain specific triplet
sequences and thus serve as check points for mapping the stretches of sequentially connected HSQC peaks
on to the primary sequence for final assignment. The process has been illustrated schematically in Fig. 4C.
A point to be mentioned here is that like the basic HN(C)N, these experiments can also be modified to
generate additional amino-acid specific (i.e. alanines and serines/threonines specific) patterns of self and
sequential correlation peaks and hence enable identification of additional triplets of residues [28,29]. This
becomes important in those cases when the protein sequence has only a few glycines and prolines and they
are far removed along the sequence. As described previously [28,29], the additional check points can be
generated simply by changing the bandwidth of the 180 inversion pulse on 13C channel during the CN
evolution period (red encircled pulse, Fig. 2). Accordingly the experiment can have three variants: (a)
glycine variant (described here in the present paper, (b) alanine variant (which provides identification of
triplet stretches containing glycines/alanines and prolines), and (c) serine/threonine variant (which
provides identification of triplet stretches containing serines/threonines and prolines). However, to avoid
the confusion, the experimental results and the assignment protocol based on the normal experiment have
been presented here.
Following the above protocol, the complete sequence specific assignment of backbone (1H, 15N, 15N,
and 15N) resonances is established initially and then the backbone 13C or 13C' chemical shifts are
determined as:
where ‘k’ is the scaling factor and is equal to (13C)/ (15N)=2.48. Thus, the single reduced dimensionality
HN(C)N experiment provides direction specific sequential assignment walk without involving any
complementary experiment. Even the explicit side chain assignment would not be very necessary to decide
on the correctness of the sequential assignment because of the large number of various check points that
are generally available. The assignment protocol is thus very simple, swift and well suited for automation
as well. Indeed, we are in a process to automate the protocol as AUTOBA+ (the updated version of AUTOBA
[15]) which would also be able to handle the orthogonal projection planes of these RD HN(C)N
experiments (as shown in Fig S5) for the assignment of well-behaved medium sized (MW ~ 12-15 kDa)
folded proteins.
The described sequential assignment walk protocol relies mainly on the presence of sufficient
number of check points and moreover these should be well-dispersed all along the sequence. These check
points are generally obtained around either glycines or alanines/glycines, or serines/threonines
(depending upon the variant of the experiment used). However in case when these residues are far
removed along the sequence of polypeptide chain and/or when there are several ambiguous breaks in
11
1315151311()/2 (3)obsiiioffsetCNNC Kumar D
sequential connectivities due to missing amide cross-peaks (mainly due to conformational line broadening
effect), the above sequential assignment walk protocol may fail. Further, in case when a protein exhibits
degeneracy in both amide 1H and 15N shifts (i.e. when there is overlap of HSQC peaks), the strategy may
lead to ambiguities in backbone assignment. In such a case, residue type identifications will help to remove
the ambiguity. This is exactly the same deadlock situation encountered here for the assignment of unfolded
UNC60B protein (152 residues; MW~17 kDa). Figure S6A (Supporting Information) depicts the 1H-15N
HSQC spectrum of the unfolded UNC60B denatured in 8 M Urea at pH 6.0 and 298K. The limited chemical
shift dispersion of the backbone amide 1H resonances (within the range of ~0.75 ppm) indicated that the
polypeptide chain of UNC60B is highly denatured under the experimental conditions. However, due to
extensive conformational exchange and peak overlap at 298 K , only 88 amide-cross peaks were discerned
above the noise level out of 148 expected non-proline peaks. For these 88 peaks, the sequential
connectivities were established for 79 (~ 90 %) peaks unambiguously (like as depicted in Fig. S6B and
S6C). However, because of lack of sufficient number check points , the transformation of stretches of
sequentially connected HSQC peaks into the final sequence specific assignment was not possible. In such a
situation, the sequential 13C and 13C chemical shift information derived from 3D-CBCAcoNH experiment
has been used to transform these stretches into the final assignment. Therefore, the unambiguous
sequential walk approach based on these reduced dimensionality tailored HN(C)N experiments combined
with a 3D CBCAcoNH experiment (for amino acid type identification) provides an alternatively robust and
efficient approach for sequential backbone resonance assignment (however, the actual combination of
experiments will depend upon the situations at hand). Following this approach, we were able to assign
90% of backbone (1H, 15N, and 13C) chemical shifts of unfolded UNC60B sequence specifically (Fig. S6,
Supporting Information). A summary of all the observed sequential connectivities in the (4,3)D-
hNcoCANH spectrum of unfolded UNC60B protein has been shown by underlining along the sequence (Fig.
S6D). As evident from the assignment of unfolded UNC60B, only the N- and C-terminal residues of the
polypeptide chain give rise to observable peaks in the HSQC spectrum indicating extensive line broadening
of amide cross peaks signals from the core of the polypeptide chain.
12
Kumar D
Finally, the guidelines -regarding in what situation which of the two experiments should be used-
are also required to be mentioned here. As evident from Fig S1 and the experimental results depicted here,
the (4,3)D-hNcoCANH experiment provides relatively a better dispersion of chemical shifts compared to
(4,3)D-hNCOcaNH experiment, therefore the former should be used preferably. However for proteins
containing repetitive amino acids (like …TSAAGTTTE…) or repetitive stretches of amino acids (like
…GTSDE…GTSDE…), the (4,3)D-hNCOcaNH experiment will be a better option compared to (4,3)D-
hNcoCANH because of the random nature of backbone 13C’ chemical shifts.
Advantage over the Basic Experiment in its Application to Unfolded Proteins:
Unfolded proteins (including intrinsically unstructured, partially disordered and proteins
denatured for studying their folding mechanisms) have always posed a great challenge for NMR
investigations because of poor chemical shift dispersion of amide 1H and carbon resonances. Pulse
sequences relying heavily on well dispersed nitrogen chemical shifts –like HN(C)N [10,13]- have proven
extremely useful in this context. They provide direct sequential amide 1HN and 15N correlations along the
polypeptide chain and exploit the 15N chemical shift dispersion along two of the three dimensions, which is
generally very good for unfolded proteins. However, the HN(C)N based assignment protocols may fail when
the HSQC peaks of sequential residues have degenerate amide 15N chemical shifts. The fact has been
demonstrated here in Figure 5 (schematically) and Figure 6 (experimentally) using HN(C)N as an
example. The two situations may arise as discussed below:
13
Kumar D
(i)
When two HSQC peaks from self (i) and sequential (i+1) residues have almost the same 15N
chemical shift (Case-I, Figure 5): In this case, the sequential correlation peak (i.e. Hi,Ni+1) in the
F1(15N)-F3(1H) plane of the HN(C)N spectrum can disappear because of opposite sign and higher
intensity of self-correlation peaks (i.e. Hi,Ni) in this spectrum as shown in Fig. 5B, and sequential
assignment walk may end up with an ambiguity that means (a) residue i is present at the C-terminal
or (b) it is followed by a proline or (c) the sequential (i+1) residue has the degenerate 15N chemical
shift (i.e. 15Ni = 15Ni+1). However, the ambiguity is resolved here in the reduced dimensionality
HN(C)N experiments by breaking the 15N degeneracy condition by using linear combination of 15N
and 13C/13C’ chemical shifts. For example, the F1(NC)-F3(1H) strip of RD-HN(C)N spectrum in Fig.
5C clearly reveals that residue is not followed by proline; instead self ( i) and sequential (i+1) HSQC
peaks have the degenerate 15N chemical shifts.
(ii) When the residues (here i and j) having the degenerate 15N chemical shifts in the HSQC spectrum
(i.e. 15Ni = 15Nj) are sequentially connected to residues with almost same 15N chemical shift (i.e.
15Ni+1 = 15Nj+1) Case-II, Figure 5): In this case, the HN(C)N spectrum (or the F2(15N)-F3(1H) plane at
F1 = 15Ni+1 = 15Nj+1; as shown in Fig. 5E) may fail to identify the sequential (i+1 or j+1) HSQC peaks
corresponding to residues i and j, unambiguously. Under such situation, one has to consider both
the possibilities as shown in Fig. 5E which elaborates the analysis. However, in reduced
dimensionality HN(C)N experiments such an ambiguity (arising because of degenerate 15N chemical
shift) can be resolved along F1(NC) dimension (which is devoid of the 15N degeneracy condition),
where now
and
sequential amide correlations will appear on different F2(15N)-
F3(1H) planes i.e. at
, respectively, as clear from Figure 5.
Therefore in terms of data analysis particularly when two neighboring amino acid residues have
nearly identical backbone 15N chemical shifts, the presented reduced dimensionality tailored HN(C)N
experiments perform relatively better compared to the basic HN(C)N experiment (a situation generally
seen for unfolded or intrinsically unstructured polypeptide chains). To highlight the relative ease and
speed of sequential assignment of any protein based on these new experiments, the representative 2D F1-F3
strip plots and 2D F2-F3 planes of these spectra have been compared to those of basic HN(C)N spectrum in
Fig. 6. Depending upon the situation in hand, Fig. 6 shows comparison of the spectral features for
establishing a sequential i to i+1 connectivity between Leu17 and Leu18 of unfolded UNC60B. As evident
from Fig. 6B and 6C, the identification of sequential (i+1) correlation peak corresponding to HSQC peak
(HiNi; here it is Leu17) is ambiguous on F2-F3 planes of HN(C)N spectrum at F2=Ni, and one has to check for
14
1ii1jj151515151111/ and /iijjNNNN Kumar D
all the cross peaks corresponding to Ni+1 chemical shifts (identified initially from the F1-F3 plane of HN(C)N
spectrum at F2=Ni). In Fig. 6C, these possibilities are shown by the blue horizontal lines. However, on the
other hand, a sequential i to i+1 connectivity between the two HSQC peaks can be established in a very
simple, swift and accurate manner following the unambiguous RD HN(C)N based approach i.e. a true
sequential HSQC peaks (Hi+1,Ni+1) will be present on both the F2-F3 planes of these spectra at
(identified from the F1-F3 plane of RD HN(C)N spectrum at F2=Ni, Fig. 6D) These are shown by the grey
horizontal lines connecting the two red peaks in Fig. 6E.
Concluding Remarks:
In conclusion, two novel reduced dimensionality experiments -(4,3)D hNcoCANH and (4,3)D
hNCOcaNH- and an efficient assignment protocol has been demonstrated here for obtaining un-ambiguous
and accurate assignment of backbone amide (1H and 15N) and 13C or 13C’ resonances. The protocol is
basically an improvisation of the previously reported HN(C)N based protocol [14] and is based on the fact
that linear combination of backbone 15N and 13C (here either 13C or 13C') chemical shifts offers relatively
better dispersion and randomness compared to individual chemical shifts (Fig. S1); thus helps to resolve
the ambiguities arising because of amide 15N chemical shift degeneracy (as in case of HN(C)N based
assignment protocol). Overall, the experiments would greatly facilitate the assignment of crowded HSQC
spectra especially of intrinsically/partially unstructured proteins and medium sized helical proteins which
in general exhibit high degree of amide shift degeneracy. The performance of the experiments and the
assignment protocol has been demonstrated here two folded proteins (human ubiquitin and bovine apo
calbindin-d9k) and on a highly denatured UNC60B protein unfolded in 8M urea.
Compared to currently available strategies, the strength of the protocol lies in the fact that the
single RD HN(C)N experiment provides the complete sequence specific assignment of backbone amide
resonances. The method is very well suited for automated data analysis as well where keeping the total
number of spectra small is highly desirable in order to avoid the complications arising because of inter-
spectral variations of chemical shifts; the problem is common in proteins which are unstable in solution or
tend to precipitate in matter of days. The method will also serve as a valuable NMR assignment tool in drug
discovery research programs especially for SAR by NMR (i.e. studying Structure Activity Relations by NMR
and is the most commonly used method to verify binding, stoichiometry, and identification of the binding
site on the protein targets [32,33]). Likewise, the method can also be used to re-establish the lost resonance
assignment of proteins upon ligand binding or upon a mutation. Taken together, the experiments presented
here will serve as an efficient backbone amide assignment tool for various protein NMR studies.
15
111/iiFNN Kumar D
Acknowledgement:
This work is being financially supported by the Department of Science and Technology under SERC
Fast Track Scheme (Registration Number: SR/FT/LS-114/2011) for carrying out the research work. I
would also like to acknowledge the High Field NMR Facility at Centre of Biomedical Magnetic Resonance,
Lucknow, India. For demonstrating the method on an unfolded protein, the 13C/15N labeled sample of
protein UNC60-B was gifted by Dr. Ahish Arora of Molecular Structural Biology Division, CDRI, Lucknow. I
wish to dedicate this work to Prof. Ramakrishna V Hosur, TIFR, Mumbai, India who has always supported
and encouraged me to pursue research in NMR methodology development.
16
Kumar D
Reference:
[1] M. Pellecchia, D.S. Sem, K. Wuthrich, Nmr in drug discovery, Nat Rev Drug Discov 1 (2002) 211 -219.
[2] Cavanagh J, Fairbrother W J, Palmer A G, Skelton N J, Protein NMR Spectroscopy: Principles and Practice,
(2006)
[3] P. Permi, A. Annila, Coherence transfer in proteins, Progress in Nuclear Magnetic Resonance Spectroscopy 44
(2004) 97-137.
[4]
Sattler M, Schleucher J, Griesinger C, Heteronuclear multidimensional NMR experiments for the structure
determination of proteins in solution employing pulsed field gradients, Progress in Nuclear Magnetic
Resonance Spectroscopy 34 (1999) 93-158.
[5]
S. Grzesiek, A. Bax, Improved 3D triple-resonance NMR techniques applied to a 31 kDa protein, J. Magn Reson. A
96 (1992) 432-440.
[6] L.E. Kay, M. Ikura, R. Tschudin, A. Bax, Three-Dimensional Triple-Resonance NMR Spectroscopy of Isotopically
Enriched Proteins, J. Magn Reson. B 89 (1990) 496-514.
[7] R.T. Clubb, V. Thanabal, G. Wagner, A constant-time three-dimensional triple-resonance pulse scheme to
correlate intraresidue 1HN, 15N, and 13CO chemical shifts in 15N---13C-labelled proteins, J. Magn Reson. A 97
(1992) 213-217.
[8]
S. Grzesiek, A. Bax, An Efficient Experiment for Sequential Backbone Assignment of Medium -Sized Isotopically
Enriched Proteins, J. Magn Reson. B 99 (1992) 201-207.
[9]
S. Grzesiek, A. Bax, Correlating Backbone Amide and Side Chain Resonances in Larger Proteins by Multiple
Relayed Triple Resonance NMR, J. Am. Chem Soc 114 (1992) 6293.
[10] N.S. Bhavesh, S.C. Panchal, R.V. Hosur, An efficient high-throughput resonance assignment procedure for
structural genomics and protein folding research by NMR, Biochemistry 40 (2001) 14727 -14735.
[11] D. Kumar, R.V. Hosur, hNCOcanH pulse sequence and a robust protocol for rapid and unambiguous assignment
of backbone ((1) H(N) , (15) N and (13) C') resonances in (15) N/(13) C -labeled proteins, Magn Reson. Chem
(2011)
[12] J.G. Reddy, D. Kumar, Direct Sequential Hit Strategy for Unambiguous and Accurate Backbone Assignment of
13C/15N Labeled Proteins, Natl. Acad. Sci. Lett. 35 (2012) 389-399.
[13] S.C. Panchal, N.S. Bhavesh, R.V. Hosur, Improved 3D triple resonance experiments, HNN and HN(C)N, for HN and
15N sequential correlations in (13C, 15N) labeled proteins: application to unfolded proteins, J. Biomol. NMR 20
(2001) 135-147.
[14] A. Chatterjee, N.S. Bhavesh, S.C. Panchal, R.V. Hosur, A novel protocol based on HN(C)N for rapid resonance
assignment in ((15)N, (13)C) labeled proteins: implications to structural genomics, Biochem. Biophys. Res.
Commun. 293 (2002) 427-432.
[15] A. Borkar, D. Kumar, R.V. Hosur, AUTOBA: automation of backbone assignment from HN(C)N suite of
experiments, J. Biomol. NMR 50 (2011) 285-297.
[16] T. Szyperski, G. Wider, J.H. Bushweller, K. Wuthrich, Reduced dimensionality in triple-resonance NMR
experiments, J. Am. Chem Soc 115 (1993) 9307-9308.
17
Kumar D
[17] T. Szyperski, D.C. Yeh, D.K. Sukumaran, H.N. Moseley, G.T. Montelione, Reduced-dimensionality NMR
spectroscopy for high-throughput protein resonance assignment, Proc. Natl. Acad. Sci. U. S. A 99 (2002) 8009-
8014.
[18] D. Rovnyak, D.P. Frueh, M. Sastry, Z.Y. Sun, A.S. Stern, J.C. Hoch, G. Wagner, Accelerated acquisition of high
resolution triple-resonance spectra using non-uniform sampling and maximum entropy reconstruction, J Magn
Reson. 170 (2004) 15-21.
[19] P. Schanda, M.H. Van, B. Brutscher, Speeding up three-dimensional protein NMR experiments to a few minutes,
Journal of the American Chemical Society 128 (2006) 9042-9043.
[20] K. Pervushin, B. Vogeli, A. Eletsky, Longitudinal (1)H relaxation optimization in TROSY NMR spectroscopy, J Am.
Chem Soc 124 (2002) 12898-12902.
[21] R.L. Narayanan, U.H.N. D++rr, S. Bibow, J. Biernat, E. Mandelkow, M. Zweckstetter, Automatic Assignment of the
Intrinsically Disordered Protein Tau with 441-Residues, Journal of the American Chemical Society 132 (2010)
11906-11907.
[22] S. Mantylahti, O. Aitio, M. Hellman, P. Permi, HA-detected experiments for the backbone assignment of
intrinsically disordered proteins, J Biomol NMR 47 (2010) 171-181.
[23] L. Emsley, G. Bodenhausen, Gaussian Pulse Cascades: New Analytical Functions for Rectangular Selective
Inversion and In-Phase Excitation in NMR, Chemical Physics Letters 165 (1990) 469-476.
[24] M. Piotto, V. Saudek, V. Sklenar, Gradient-tailored excitation for single-quantum NMR spectroscopy of aqueous
solutions, J. Biomol. NMR 2 (1992) 661-665.
[25] O. Zhang, L.E. Kay, J.P. Olivier, J.D. Forman-Kay, Backbone 1H and 15N resonance assignments of the N -terminal
SH3 domain of drk in folded and unfolded states using enhanced-sensitivity pulsed field gradient NMR
techniques, J. Biomol. NMR 4 (1994) 845-858.
[26] R. Keller, The Computer Aided Resonance Assignment Tutorial CANTINA, Verlag, Goldau, Switzerland., (2004)
[27] A. Meissner, W. Sørensen, A Sequential HNCA NMR Pulse Sequence for Protein Backbone Assignment, Journal of
Magnetic Resonance A 150 (2001) 100-104.
[28] D. Kumar, J. Chugh, R.V. Hosur, Generation of Serine/Threonine Check Points in HN(C)N Spectra, J. Chem. Sci.
121 (2009) 955-964.
[29] A. Chatterjee, A. Kumar, R.V. Hosur, Alanine check points in HNN and HN(C)N spectra, J. Magn Reson. 181
(2006) 21-28.
[30] D. Marion, M. Ikura, R. Tschudin, A. Bax, Rapid Recording of 2D NMR Spectra without Phase Cycling. Application
to the Study of Hydrogen Exchange in Proteins, J. Magn Reson. 85 (1989) 393-399.
[31] D. Kumar, A. Arora, (15N +/- 13C') edited (4, 3)D-H(CC)CONH TOCSY and (4, 3)D-NOESY HNCO experiments for
unambiguous side chain and NOE assignments of proteins with high shift degeneracy, Magn Reson. Chem 49
(2011) 693-699.
[32] S.B. Shuker, P.J. Hajduk, R.P. Meadows, S.W. Fesik, Discovering high-affinity ligands for proteins: SAR by NMR,
Science 274 (1996) 1531-1534.
[33] R. Powers, Applications of NMR to structure-based drug design in structural genomics, J Struct Func Genom 2
(2002) 113-123.
18
Kumar D
[34] A.J. Shaka, C.J. Lee, A. Pines, Iterative schemes for bilinear operators; application to spin decoupling, Journal of
Magnetic Resonance (1969) 77 (1988) 274-293.
[35] A.J. Shaka, P.B. Barker, R. Freeman, Computer-Optimized Decoupling Scheme for Wideband Applications and
Low-Level Operation, Journal of Magnetic Resonance 64 (1985) 547-552.
[36] M.A. McCoy, L. Mueller, Bloch-Siegert phase shift, J. Magn Reson. 99 (1992) 18-36.
19
Kumar D
Figures:
Figure 1: Schematic illustrations of the selected coherence transfer pathways employed in reduced
dimensionality experiments (4,3)D- hNCOcaNH (A) and (4,3)D- hNcoCANH (B), respectively. The
magnetization flow from HN(i) is shown and the frequency labeling of the appropriate nuclei are indicated
where, 2TN,
are the delays during which the transfers indicated by the arrows take
place in the pulse sequence. (C) and (D) represent the pulse sequences for RD (4,3)D hNCOcaNH and
(4,3)D hNcoCANH experiments, respectively. Narrow (hollow) and wide (filled black) rectangular bars
represent non-selective 90° and 180° pulse, respectively. Unless indicated, the pulses are applied with
phase . Proton decoupling using the DIPSI-2 decoupling sequence [34] with field strength of 6.3 kHz is
applied during most of the t1 (15N/13C) and t2 (15N) evolution periods, and 15N decoupling using the GARP-1
sequence [35] with the field strength 0.9 kHz is applied during acquisition. Standard Gaussian cascade
pulses [23] shape Q3 for 180° inversion/refocusing (filled black and grey; width 200 s) and shape Q5 for
90° excitation (hollow, width 310 s) were used along the carbon channel. The bandwidth of the 13C pulses
(either for excitation, inversion or refocusing) is adjusted so that they cause minimal excitation of carbonyl
carbons and that of 180° 13C’ shaped pulse so that they cause minimal excitation of 13C. The blue encircled
red pulse on 13C channel is crucial for tuning the experiment for generation of different check points in the
final spectrum (like glycines/alanines and serines/threonines, for detail see these references [28,29]).
Yellow pulses were applied for compensation of off-resonance effects (Bloch-Siegert phase shift) [36]. The
values for the individual periods containing t1 evolution of 15Nnuclei are: A = t1/2, B = TN, and C =
. In (4,3)D hNCOcaNH pulse sequence, the 13C' nuclei are co-evolved with 15Nnuclei in a semi-constant time
manner, and the values for the individual periods containing t1 evolution of 13C’nuclei are:
,
and
. In (4,3)D hNcoCANH pulse sequence, the 13Cnuclei are co-evolved with
15Nnuclei in constant time manner and the values for the individual periods containing t1 evolution of
13Cnuclei are: G = t1/2, H =
, and J =
. The values for the individual periods containing t2
evolution of 15N nuclei are: D =
t2/2, E =
, and F = t2/2. The other delays are set to = 2.5 ms, = 5.4
ms, = 2.5 ms,
= 4.5 ms, TN = 14.0 ms,
= 12.5 ms and
= 13.5 ms. The
must be optimized and
is around 12-15 ms. The phase cycling for the experiment is 1 = 2(
), 2(-
); 2 = 3= , -
; 4 = 5 =
;
6 = 4(
), 4(-
); and receiver = 2(
), 4(-
), 2(
). The frequency discrimination in
and
has been
achieved using States-TPPI phase cycling [30] of 1 and 5, respectively, along with the receiver phase. The
gradient (sine-bell shaped; 1 ms) levels are as follows: G1=30%, G2=30%, G3=30%, G4=30%, G5=50%,
20
2, 2, and 2CCNNx1/2NTt1/2aCtbCC1/2cCCtCNC1/2CNtNNCCNNCNxxxxxxxxxx1t2t G6=80% and G7=20% of the maximum strength 53 G/cm in the z-direction. The recovery time after each
gradient pulse was 160 s. Before detection, WATERGATE sequence [24] has been employed for better
water suppression.
Kumar D
21
Kumar D
22
Kumar D
Figure 2: Schematic representation of the three-dimensional (4,3)D-hNCOcaNH/(4,3)D-hNcoCANH
spectrum. The correlations observed in the F1(NC)F3(1H) plane at the 15N chemical shift of residue i are
shown on
left
side
and
the
correlations observed
in
the F2(15N)F3(1H) planes
at
are shown on the right side. Squares and circles represent the self and
sequential peaks, respectively. Red and black represent positive and negative phase of peaks, respectively.
23
2211/ and /iiiiFNNFNN Kumar D
Figure 3: An illustrative example showing use of correlations observed in the F2(15N)F3(1H) planes (of
(4,3)D-hNcoCANH spectrum of 8M urea-denatured UNC60B) at the
chemical shifts identified
for residue i for establishing a sequential (i ⟶ i+1) connectivity between the backbone amide correlation
peaks of 1H-15N HSQC spectrum (like in case of HN(C)N based assignment protocol [14]). Red and black
contours represent positive and negative phase of the peaks, respectively. A sequential amide cross peak
(Hi+1, Ni+1) exists in both the F2(15N)F3(1H) planes of the spectrum at the
chemical shifts
identified for the residue i. The other advantage is the opposite peak signs for the self (H i, Ni) and
sequential (Hi+1, Ni+1) amide correlation peaks. These spectral features help to reduce the search for the
sequential amide correlations rather than making the search in various planes of the 3D spectrum (as is the
case with other presently used assignment strategies).
24
151511/iiNN151511 and iiNN Kumar D
Figure 4: Schematic showing sequential assignment walk protocol based on the reduced dimensionality
HN(C)N experiments. In (A) and (B), the strategy of establishing the sequential (i to i+1) connectivity has
been described using the polypeptide sequence –XYZ–. In (C), the sequential assignment walk procedure
through the F1(NC)F3(1H) planes of the reduced dimensionality HN(C)N spectra using the triplet
specific peak patterns has been shown. An arbitrary amino acid sequence is chosen to illustrate the start,
continue, check, and break points during the sequential assignment walk. Squares and circles represent the
self (intra-residue) and sequential (inter-residue) peaks, respectively. Red and black represent positive and
negative phase of peaks, respectively. The patterns of positive and negative peaks are drawn according to
the triplet specific peak patterns contained in this spectrum (as shown in Fig. 4). The residue identified on
the top of each strip identifies the central residue of the triplet.
25
Figure 5: Schematic showing the advantage of reduced dimensionality HN(C)N experiment for resolving
the ambiguities arising because of degenerate 15N chemical shifts as in case of HN(C)N based assignment
protocol.
Kumar D
26
Kumar D
Figure 6: Experimental demonstration on unfolded UNC60B protein, of the advantage of RD (4,3)D-
HN(C)N based assignment protocol for resolving the ambiguities arising due to degenerate 15N chemical
shifts. (A) A part of 1H-15N HSQC spectrum of unfolded UNC60B (in 8 M Urea at pH 6.0 and 298 K) showing
relative positions of sequentially connected HSQC peaks: Asp16 to His19. (B) F1(15N)-F3(1H) strips and (C)
F2(15N)-F3(1H) planes of 3D-HN(C)N spectra showing ambiguity in finding the sequential correlation to spin
system Leu17. (D) F1(NC)-F3(1H) strips and (E) F2(15N)-F3(1H) planes of of (4,3)D-hNcoCANH spectrum
showing unambiguous sequential connection between spin systems Leu17 and Leu18. The F2 (15N) values
are shown at the top for each F1-F3 strip. In (D), the top and bottom halves belong to the ‘minus’ and ‘plus’
domains of the spectrum, respectively.
27
Kumar D
Supplementary Material:
Reduced Dimensionality tailored HN(C)N Pulse Sequences for Efficient Backbone
Resonance Assignment of Proteins through Rapid Identification of Sequential
HSQC peaks
Dinesh Kumar
Centre of Biomedical Magnetic Resonance, SGPGIMS Campus, Raibareli Road, Lucknow-226014, India
*Address for Correspondence:
Dr. Dinesh Kumar
(Assistant Professor)
Centre of Biomedical Magnetic Resonance (CBMR),
Sanjay Gandhi Post-Graduate Institute of Medical Sciences Campus,
Raibareli Road, Lucknow-226014
Uttar Pradesh-226014, India
Mobile: +91-9044951791,+91-8953261506
Fax: +91-522-2668215
Email: [email protected]
Webpage: http://www.cbmr.res.in/dinesh.html
KEYWORDS: Reduced Dimensionality NMR; Backbone Assignment; HN(C)N; (4,3)D-hNCOcaNH; (4,3)D-
hNcoCANH; Sequential Amide Correlations.
ABBREVIATIONS: NMR, Nuclear Magnetic Resonance; HSQC, Heteronuclear Single Quantum Correlation;
CARA, Computer Aided Resonance Assignment; RD: Reduced dimensionality.
1
Kumar D
Appendix I
The particular advantage of new experiments is the dispersion of peaks in the sum and
difference frequency regions which can be different and this facilitates analysis of the spectrum by
simultaneously connecting (15N+13C) and (15N+13C) correlations (here 13C is either backbone 13C’
or 13C). The advantage is elicited from the fact that the linear combination of 15Ni and 13C
i-1/13C’i-1
chemical shifts provides dispersion better compared to the dispersion of the individual chemical shifts. The
fact has been demonstrated here in Fig. S1 using average 15N, 13C and 13C’ chemical shifts derived from the
BMRB database [1] for all the 20 amino acids. As shown in the figure, the average 13C’, 15N, and 13C
chemical shifts individually show dispersions of ~20.0 (=8.0*2.48), ~21.5 and ~53.7 (21.5 x 2.48) ppm,
respectively, (where 1 ppm = 80 Hz, along 15N dimension at 800 MHz spectrometer) (Fig. S1A, S1B and
S1C). However, the linear combination of these chemical shifts provides dispersion relatively higher than
the individual chemical shifts: (i) the subtraction and addition of 15Ni and 13C’j chemical shifts (i and j
represent the amino acid type) provide dispersion of 35.4 ppm and 35.5 ppm, respectively, (Fig. S1D) and
(ii) the subtraction and addition of 15Ni and 13C
j chemical shifts provide dispersion of 69.5 ppm and 68.0
ppm, respectively (Fig. S1E). The fact has been exploited here in the form of reduced dimensionality
experiments (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH to facilitate the assignment of backbone amide
resonances of complex protein systems .
2
Kumar D
Figure S1: Comparison between dispersion of individual 15N/13C/13C’ chemical shifts and that of linear combination
of these chemical shifts for 20 common amino acids. In (A), (B) and (C) the average chemical shifts of 13C’, 15N, and
13Care plotted against the residue types. The average chemical shifts have been taken from the BMRB statistical
table containing values calculated from the full BMRB database (http://www.bmrb.wisc.edu/ref_info/statful .htm#1)
[1]. This includes paramagnetic proteins, proteins with aromatic prosthetic groups, and entries where chemical shifts
are reported relative to uncommon chemical shift references. Standard deviations in these valu es are plotted as error
bars. In (D), the linear combinations of 15N and 13C’ average chemical shifts (as evaluated according to Eqns 1 and 2)
for all the amino acid types have been shown. In (E), the linear combinations of 15N and 13C average chemical shifts
(as evaluated according to Eqns 1 and 2) for all the amino acid types have been shown. The overall chemical shift
dispersion achievable in each case has been shown at the top of each plot. Comparison clearly shows that the linear
combination of chemical shifts lead to better dispersions compared to the individual chemical shifts.
3
Kumar D
Appendix-II
8M urea-denatured states of UNC60B from Caenorhabditis elegans:
The Caenorhabditis elegans UNC60B gene has been cloned into pETNH6 vector (Novagen) using
restriction site NcoI and BamHI (in Dr. Ashish Arora’s Lab, CDRI Lucknow). The cloning procedure added
extra residues at N-terminal including a hexa-histidine tag and TEV-protease site. The clone was over-
expressed in BL21 (DE3) strain of E. coli. Conditions for optimal over-expression and purification were
standardized. The yield of purified UNC-60B protein was 35 mg/L of culture medium. For isotopic labeling,
over-expression of UNC-60B was standardized in minimal media containing 15N-ammonium sulfate and
13C-glucose (CIL, MA, USA) as the sole nitrogen and carbon sources, respectively. The purity of the sample
was checked on SDS-PAGE. The protein sample was finally concentrated to ~800 µM using a centricon-
filter (molecular weight cut off 3K, Amicon). For 8M urea-denatured state of the UNC60B, the concentrated
protein (upto 0.8 mM) was exchanged with 50 mM phosphate buffer of pH 6.0 (1 mM EDTA, 50 mM NaCl,
0.1% NaN3, 1 mM DTT, and 8.0 M urea) containing 90% H2O and 10% D2O. The NMR experiments were
started after keeping the solutions for about ~3 h so as to reach equilibrium.
4
Kumar D
Table S1: Acquisition parameters used to record the various spectra on the uniformly 15N/13C labeled
Bovine apo-calbindin (75 aa), human ubiquitin (76 aa) protein and 8M urea denatured UNC60B.
Parameters
Bovine Apo Calbindin-d9k/Human
8M urea denatured UNC60-B
Ubiquitin
(4,3)D-hNCOcaNH
(4,3)D-
(4,3)D-hNCOcaNH
(4,3)D-hNcoCANH
hNcoCANH
Complex Data
1024 X 36 X 64
1024 X 36 X 96
1024 X 40 X 72
1024 X 40 X 108
Points (F3 X F2 X F1)
Spectral Width and
(offset) in ppm
F3(1H) = 10.6 (4.7)
F3(1H) = 10.6 (4.7)
F3(1H) = 10.6 (4.7)
F3(1H) = 10.6 (4.7)
F2(15N) = 27 (118
F2(15N) = 27 (118 )
F2(15N) = 26 (118 )
F2(15N) = 26 (118 )
)/33(118)
F1(NC) =160
F1(NC’) =100
F1(NC) =160
F1(NC’) =110/100 (185)
No. of scans per FID
8 (0.8 sec)
(118/70)
8 (0.8 sec)
(118/185)
8 (0.8 sec)
(118/70)
8 (0.8 sec)
(Recycle delay)
Zero filling
1024 X 128 X 256
1024 X 128 X 512
1024 X 128 X 256
1024 X 128 X 512
Total Expt. Time≠
~23 hr, 11 min
~1 day, 6 hr, 39 min
~1 day, 1 hr, 33 min
~1 day, 14 hr, 19 min
*Note: A point to be mentioned here is that the proposed RD experiments require longer time compared to the normal
3D HN(C)N experiment in order to reach the same resolution. This is because of the fact that one has to acquire more
data points along the co-evolved F1 dimension where the linear combination of chemical shifts renders increased
spectral width (generally two to three times of the primary 15N spectral width).
5
Kumar D
Appendix-III
Amino Acid Sequence-Dependent Peak Patterns
Like the basic HN(C)N experiment, the self (notion used here for
,
and
) and
sequential (notion used here for
,
and
) correlation peaks have opposite peak signs in
different planes of the (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH spectra except in special situations.
Considering different triplets of residues, covering the general and all the special situations, the expected
peak patterns in different planes of these spectra have been shown in Figure S3 where self and sequential
correlation peaks have been shown by squares and circles, respectively. Figure S3A shows the expected
peak patterns (for the example stretch -PZXGYP- covering all the various possibilities) in the F2(15N)-F3(1H)
planes at F1=
/
chemical shift identified for residue, i. Figure S3B shows the expected peak
patterns (for the triplet stretches shown above each panel) in the F1(NC)-F3(1H) planes of these spectra at
the F2(15N) chemical shift of the central residue, i. The filled and empty circles/squares represent positive
and negative peaks, respectively. The actual signs in the spectrum are dictated by whether the i−1 residue
is a glycine or otherwise, and of course by the phasing of the spectrum. Here, the self-correlation peaks (i.e.
,
) have been made positive (like the basic HN(C)N spectrum) for a triplet sequence –XYZ-,
where X, Y, and Z are any residues other than glycines and prolines which have been represented here by
letters “G” and “P”, respectively. Accordingly the peak patterns have been shown in Figure S3. From the
figure, it is clear that glycines make an important difference in the expected patterns of peaks. For glycines,
the evolutions of the magnetization components are slightly different because of the absence of the β-
carbon. This in turn generates some special patterns depending on whether the (i-1)th residue is a glycine
or otherwise (see Fig. S3). These special peak patterns –which help in the identification of residues
following glycines in the sequence– provide important start or check points during the course of sequential
assignment process. Prolines which do not have amide proton further give rise to special patterns in the
spectrum. Prolines, at position i-1 in the triplet stretches, results in the absence of sequential amide
correlation peaks (Hi-1, Ni-1) in F2(15N)-F3(1H) plane of the spectrum at F1=
/
(see Fig. S3A), whereas a
proline, at position
i+1
in the triplet stretches results
in the absence of correlation peaks
(
) in F1(NC)-F3(1H) plane of the spectrum at F2=15Ni (see Fig. S3B). These special peak
patterns –which help in the identification of residues following prolines in the sequence–, provide
important stop/break points during the course of the sequential assignment walk and thus facilitates the
assignment process further.
6
iiHNiNiN11iiHN1iN1iN1iN1iNiiHN and iiNNiNiN11, and ,iiiiHNHN
Kumar D
Figure S2: (A) Schematic peak patterns in the F2(15N)-F3(1H) planes of the (4,3)D-hNCOcaNH and (4,3)D-hNcoCANH
spectra at
. An arbitrary amino acid sequencePZXGYP (covering all the common and special peak
patterns) is chosen to illustrate the triplet specific peak patterns in these planes of the spectra. In the sequence, X, Y
and Z represent any residue other than glycine and proline. G and P, respectively, represent the glycine and proline
residues. Black and red colors represent positive and negative signs, respectively. Squares and circles represent the
self and sequential peaks, respectively. (B) Schematic peak patterns in the F1(NC)-F3(1H) planes and the
corresponding triplet of residues. The patterns involving G serve as start and/or check points during the sequential
assignment walk through the spectrum. GGP, which is likely to be less frequent than the others, may serve as a unique
start point. The patterns involving P identify break points during a sequential walk and thus they also serve as check
points. The pattern, which does not involve a G or a P, is the most common one occurring through the sequential walk.
7
211/iiFNN Kumar D
Figure S3: (A) The illustrative stretch of sequential walk through the strips along F1(NC) dimension and
centered about amide 1H chemical shift of residue i in F1(NC)-F3(1H) planes of the (4,3)D-hNCOcaNH
spectrum of human ubiquitin (for residues Lys6-Ile13) at 15N chemical shift of residue shown at the top of a
particular strip. (B) The illustrative stretch of sequential walk through the strips along F1(NC) dimension
and centered about amide 1H chemical shift of residue i in F1(NC)-F3(1H) planes of the (4,3)D-hNcoCANH
spectrum of bovine apo calbindin-d9k (for residues Leu6-Tyr13) at 15N chemical shift of residue shown at
the top of each individual strip. Each strip contains information about self (
) and sequential
(
) correlation peaks. The red and black contours indicate positive and negative peaks,
respectively. The labels and numbers at the top in each panel identify the residue and the respective F2(15N)
chemical shifts. A horizontal line connects a sequential peak (red here) in one plane to the self peak (black
here) in adjacent plane on the right. However, the special patterns appear for strips of glycines where both
self and sequential peaks appear positive/black in sign e.g. Gly10 and Gly8 strips in (A) and (B),
respectively. On the other hand, self and sequential peaks appear negative/red in sign on the strips
corresponding to residues following glycines in the sequence e.g. Lys11 and Ile9 strips in (A) and (B),
respectively.
8
and iiNN11 and iiNN Kumar D
Figure S4: An illustrative example showing use of correlations observed in the F2(15N)F3(1H) planes at the
chemical shifts identified for residue i for establishing a sequential (i ⟶ i+1) connectivity
between the backbone amide correlation peaks of 1H-15N HSQC spectrum (like in case of HN(C)N based
assignment protocol [2]). Red and black contours represent positive and negative phase of the peaks,
respectively. A sequential amide cross peak (H i+1, Ni+1) exists in both the F2(15N)F3(1H) planes of the
spectrum at the
chemical shifts identified for the residue i. The other advantage is the
opposite peak signs for the self (H i, Ni) and sequential (Hi+1, Ni+1) amide correlation peaks. These spectral
features help to reduce the search for the sequential amide correlations rather than making the search in
various planes of the 3D spectrum (as is the case with other presently used assignment strategies).
9
151511/iiNN151511 and iiNN Figure S5: F1-F3 projection planes, respectively, of (4,3)D-HNCOcaNH (A) and (4,3)D-hNcoCANH (B)
spectra of chicken SH3 domain projected along the F2(15N) axis. As described earlier in [3], the sequential
assignment walk can be performed directly on these planes for small to medium sized well folded proteins.
The details will be presented elsewhere.
Kumar D
10
Kumar D
Figure S6: (A) 1H-15N HSQC spectrum of UNC60B in 8 M Urea at pH 6.0 and 298 K. (B) Sequential walk
through the F1-F3 planes of the (4,3)D-hNCOcaNH spectrum. Sequential connectivities are shown for Thr36
to Val39 stretch. (C) Sequential walk through the F1-F3 planes of the (4,3)D-hNcoCANH spectrum of
UNC60B. Sequential connectivities are shown for Val5 to Asp8 stretch. The F2 (15N) values are shown at the
top for each strip. The black and red contours indicate positive and negative peaks , respectively. (D) The
summary of sequential assignment obtained along the amino acid sequence of UNC60B. The residues
underlined have been assigned here using the RD (4,3)D -hNcoCANH experiment (for establishing the
sequential connectivities) in combination with routine 3D-CBCAcoNH experiment (for amino acid type
identification). Gly shown in red served as the start/check point while the Pro shown in blue served as the
break point.
11
Kumar D
Reference:
[1] BMRB, Statistics Calculated for Selected Chemical Shifts from Atoms in the 20 Common Amino Acids
(http://www.bmrb.wisc.edu/ref_info/), (2011)
[2] A. Chatterjee, N.S. Bhavesh, S.C. Panchal, R.V. Hosur, A novel protocol based on HN(C)N for rapid resonance
assignment in ((15)N, (13)C) labeled proteins: implications to structural genomics, Biochem. Biophys. Res .
Commun. 293 (2002) 427-432.
[3] D. Kumar, A. Borkar, R.V. Hosur, Facile Backbone (1H, 15N, 13Ca and 13C') Assignment of 13C/15N labeled
proteins using orthogonal projection planes of HNN and HN(C)N experiments and its Automation, Magn Reson.
Chem. 50 (2012) 357-363.
13
|
1003.5358 | 2 | 1003 | 2010-05-07T08:53:20 | The geometrical origin of the strain-twist coupling in double helices | [
"physics.bio-ph",
"math-ph",
"math-ph",
"q-bio.BM"
] | The geometrical coupling between strain and twist in double helices is investigated. Overwinding, where strain leads to further winding, is shown to be a universal property for helices, which are stretched along their longitudinal axis when the initial pitch angle is below the zero-twist angle (39.4 deg). Unwinding occurs at larger pitch angles. The zero-twist angle is the unique pitch angle at the point between overwinding and unwinding, and it is independent of the mechanical properties of the double helix. This suggests the existence of zero-twist structures, i.e. structures that display neither overwinding, nor unwinding under strain. Estimates of the overwinding of DNA, chromatin, and RNA are given. | physics.bio-ph | physics | The geometrical origin of the strain-twist coupling in double
helices
Kasper Olsen∗ and Jakob Bohr†
Department of Physics,
Technical University of Denmark
Building 307 Fysikvej, DK-2800 Lyngby, Denmark
Abstract
The geometrical coupling between strain and twist in double helices is investigated. Overwinding,
where strain leads to further winding, is shown to be a universal property for helices, which are
stretched along their longitudinal axis when the initial pitch angle is below the zero-twist angle
(39.4◦). Unwinding occurs at larger pitch angles. The zero-twist angle is the unique pitch angle at
the point between overwinding and unwinding, and it is independent of the mechanical properties
of the double helix. This suggests the existence of zero-twist structures, i.e. structures that display
neither overwinding, nor unwinding under strain. Estimates of the overwinding of DNA, chromatin,
and RNA are given.
1
Introduction
If one pulls a double helix structure by the end, one might think that it would unwind by the applied
tension. In this paper we show why this is not always the case: A helix can unwind, overwind, or it
can stay at its current twist (which we denote a zero-twist (ZT) structure). Overwinding is contrary
to unwinding; unwinding is the de-twisting of the helices obtained by stretching the material. For the
zero-twist structure there is no coupling from strain to twist. The existence of a twist-stretch coupling
is a well-known phenomenon for helical steel wires [1] where it leads to unwinding, and design efforts go
into designing rotation resistant wire rope when desired [2, 3].
The geometrical investigation presented below is based on the study of packed double helices modeled
as two flexible tubes with hard walls. To be packed is defined by the constraint of the two tubes being
in contact. Does this mean that the helices are stretched? No, generally not, stretching is one way to
secure that a packed helix is obtained, however, for helices on the molecular size favorable molecular
interactions can also make it more preferable to be packed than not. A detailed analysis of packed helices
and their volume fractions showed that the helices with the highest volume fractions are noticeably
similar to the molecular structure of DNA [4]; this suggests that close-packing is at work as a structure
forming principle. For the description of compact strings and tube models, the importance of one kind of
optimum shape has been discussed by Gonzalez and Maddocks [5] and Maritan et al. [6], and one related
suggestion for the best packing of proteins and DNA has been considered by Stasiak and Maddocks [7].
A detailed analysis of the geometry of n-plies, and of their self-contacts, has been given by Neukirch and
van der Heijden [8].
2 Model
The close-packed (CP) structure with an optimized volume fraction has a pitch angle of 32.5◦ [4]: This
structure that has a central channel is shown in Figure 1a. Under a pull, the pitch angle is increased
∗[email protected]
†[email protected]
0
1
0
2
y
a
M
7
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
2
v
8
5
3
5
.
3
0
0
1
:
v
i
X
r
a
and the diameter of the central channel gets smaller, and eventually, the inner channel disappears at a
pitch angle of 45◦. Whether a helix overwinds or unwinds is then determined from the balance between
the gain in length from the reduction in the helical radius versus untwisting. The crossing point -- which
we denote as the zero-twist angle -- is at 39.4◦ (Figure 1b) and is smaller than the 45◦, where the helical
radius becomes equal to the diameter of the tubes, and is maintained for all pitch angles above 45◦. The
45◦ motif, here denoted the tightly packed (TP) double helix, is shown in Figure 1c.
Figure 1: Different geometries of a double helix of tubes of fixed diameter D. a) Close-packed (CP)
structure of pitch angle 32.5◦ measured from horizontal. b) Zero-twist (ZT) structure with a pitch angle
of 39.4◦. It is at the point between overwinding/unwinding. c) Tightly-packed (TP) structure of pitch
angle 45◦. Overwinding from stretching takes place from the a) to the b) confirmations; unwinding from
b) to c).
Geometrically, the double helix is given by two tubes of diameter D, whose centerline defines two
helices with simple parametric equations. A helix is a curve of constant curvature, κ, and torsion, τ , and
it can be specified by two parameters, for example a and H, where a is the helix radius (the radius of
the cylinder hosting the helical lines) and H the helical pitch (the raise of the helix for each 2π rotation).
The tangent to each of the helical curves is at an angle v⊥ (the pitch angle) with the horizontal axis,
and it is determined by tan v⊥ = h/a, where h = H/2π is the reduced pitch. We say that the double
helix is packed when the shortest distance between the centerline of one helical tube to the next one
equals the diameter D of the tubes, i.e. the double helix is packed when the tubes are in contact. The
volume fraction can be calculated using, as a reference volume, an enclosing cylinder of height H = 2πh
and volume VE = 2π2h(a + D/2)2, and comparing it to the combined volume occupied by the two
circumscribed helical tubes, VH = π2hD2/ sin v⊥. The volume fraction is the ratio of the two volumes,
i.e. fV = VH /VE, which reads
)2)1/2 · (
+ 1)−2
2a
D
fV = 2(1 + (
a
h
(2.1)
With this choice of reference volume the packing fraction depends only on the shape of the double helix
structure, which can be described by one parameter, e.g. the pitch angle, v⊥. The maximum of fV defines
the close-packed (CP) helix. For the double helix this maximum is at v∗
V = 0.796
[4]. For the CP structure, the channel radius is about 17 % of a [4]. Generally, the radius of the central
channel, which is given by Ri = a − D/2, is a decreasing function of v⊥; this can be seen from Figure 2
⊥ = 32.5◦, where f∗
2
a) b) c)Figure 1. which shows 2a/D depending on the pitch angle. For v⊥ ≥ 45◦ there is no central channel as 2a/D = 1,
see Figure 2.
Figure 2: Graph showing the ratio 2a/D as a function of pitch angle, v⊥ [deg.], where a is the helix radius
and D the diameter of the helical tubes. The tightly packed double helix has a pitch angle of vT P = 45◦;
it is the helix with the smallest pitch angle obeying the criterion that 2a = D.
3 Results
Consider a long straight segment of a double helix consisting of two long molecular strands each of length
LM . The length of the double helix is HM = LM sin v⊥ and the total twist is ΘM = LM cos v⊥/a. In
Figure 3 the dimensionless ratio DθM /2LM is shown as a function of the pitch angle. One can see that
for v⊥ < vZT there is overwinding while for v⊥ > vZT there will be unwinding. We find numerically that
vZT = 39.4◦.
We can determine the amount of overwinding and unwinding in the following way. If a long double
helical segment is stretched a bit, the pitch angle, v⊥, will change by a small amount dv⊥, and hence HM
changes by
and ΘM by
so that
dHM = LM cos v⊥dv⊥
dΘM = −LM
sin v⊥
a
dv⊥ − LM
a2 cos v⊥
da
dv⊥
dv⊥
dΘM
dHM
= − 1
a
tan v⊥ − 1
a2
da
dv⊥
(3.2)
(3.3)
(3.4)
If this derivative is positive, then the helix will overwind, and if it is negative, it will unwind. The
derivative in Eq. (3.4) has dimension of inverse length. From a geometrical viewpoint it is more natural
to look at the dimensionless function of v⊥, obtained by multiplying with the common radius of the
tubes, (D/2), namely:
(3.5)
(cid:18) D
(cid:19)
D
2
dΘM
dHM
= − D
2a
tan v⊥ +
d
dv⊥
3
2a
TP0204060800123456v!2aDFigure 2. Figure 3: The total twist, θM , for a long segment of the double helix; the dimensionless quantity DθM /2LM
is shown as a function of the pitch angle, v⊥ [deg.]. The maximum value is obtained for the pitch angle
vZT = 39.4◦ and mark the transition from overwinding to unwinding. At the ZT structure there is zero
coupling between twist and strain.
This equation can be given a simple interpretation. The first term is negative and determines the amount
of unwind, while the second term is positive and determines the amount of overwind. The graph of this
derivative, that dictates the coupling between strain and twist, is depicted in Figure 4. Notice that the CP
double helix will always overwind since dΘM /dHM > 0. At the close-packed structure, the overwind is
(D/2)dΘM /dHM = 0.665. The extension is therefore universally determined just by giving the diameter,
D, of the tubes making up any close-packed double helix. At the zero-twist structure, vZT = 39.4◦,
there is neither overwinding, nor unwinding. For larger pitch angles the overwind, (D/2)dΘM /dHM ,
is negative and the double helix will unwind under strain. It is therefore crucial, that the pitch angle
is below that of the zero-twist (39.4◦) for overwinding to be observed, but it also indicates that elastic
properties of the material are not essential to understanding the phenomenon.
4 Discussion
In the following we discuss some molecular examples. The phenomenon of overwinding in DNA was
first observed in 2006, see Lionnet et al. [9] and Gore et al. [10] using magnetic tweezers to control the
wringing [9] and optical tweezers to control the pulling [10]: For small deformations, DNA overwinds
when stretched, i.e. it rotates counter to unwinding. During overwinding the extension of a long chain of
DNA-B has been reported to be 0.42 ± 0.2 nm per 2π rotation [9] and 0.5 nm per 2π rotation [10]. Very
recently, it has been suggested that in the absence of tension DNA is an order of magnitude softer [11].
Using the above mathematical solution for the double helical structure of DNA we find the change of
length ∆H to be determined by
∆H =
dHM
dΘM
∆Θ
(4.6)
The diameter of the molecular tubes that make up the DNA helix is D = 1.15 nm, which is given from
our previous analysis of the close-packed structures [4]. We then estimate ∆H per full 2π turn to be
π(0.665)−1 × 1.15 nm = 5.4 nm, see Figure 4. Our result seems to support the findings of ref.
[11].
4
ZT0204060800.00.20.40.60.81.0v!DΘM2LMFigure 3. Figure 4: Graph showing the calculated overwind of double helices (solid line), i.e. Eq. (3.5) as a function
of v⊥ [deg.]. A positive overwind means that the double helix will exhibit overwinding, while a negative
overwind means that the double helix will exhibit unwinding. The zero-twist structure (ZT) is indicated
with an arrow at vZT = 39.4◦, the close-packed structure (CP) is indicated by an arrow at vCP = 32.5◦.
The first derivative is discontinuous at vT P = 45◦ where the helix radius can not get smaller. The dashed
line is the overwind for a triple helix, which has a zero-twist angle of 42.8◦.
The geometrical restriction imposed by base pairing and its influence on dΘM /dHM has not been taken
into account. The numerical analysis has been performed for the symmetrical double helix where the
close-packed structure has a pitch angle of 32.5◦. The asymmetrical DNA-B has a close-packed pitch
angle of 38.3◦ and, as one can show, a zero-twist angle of 41.8◦. Theoretical work on understanding
the overwinding of DNA has focused on constructing elastic models which show a negative twist-stretch
coupling [12] and on incorporating stochastic effects [13]. One elastic model was considered by Gore et al.
[10], and consists of a rod with a stiff helical wire (analogous to the sugar-phosphate backbone) attached
to its surface. As this system is stretched, the inner rod decreases in diameter and the helix will overwind.
Smith and Healey has argued that a linear material law is inadequate for the description and suggest a
non-linear elastic rod [14].
For chromatin, the above results can be related to recent experiments in twisting chromatin fibers,
see e.g.
[15, 16]. For a close-packed 30 nm chromatin fiber, in the so-called two-start geometry, we
estimate a tube diameter of 30/(2a/D + 1) nm= 30/(1.2 + 1) = 13.6 nm, where 2a/D is determined
from Figure 2. For the close-packed 30 nm chromatin structure we then estimate ∆H per full 2π turn
to be π(0.665)−1 × 13.6 nm = 64 nm. It is interesting to note that the numbers reported in ref. [16] are
measurements of ∆H for Xenopus chromatin per turn at a pulling force of 0.3 pN. Using the depicted
[16] we have estimated an average extension of ∼ 60 ± 40 nm per turn. Here, we have
data in ref.
assumed the two-start helix to behave like a tubular packed double helix -- that is a view which ignores
the intricate details of the structure, details which are discussed for example by Barbi et al. [25], where
elaborate mechanical models are described, including one which maintain its twist while being stretched.
We have presented a simple geometrical explanation for overwinding of helices -- an effect which has
been observed before for the double helix of DNA and for chromatin, and which is contrary to usual
unwinding. Our model of unwinding and overwinding can be applied to any symmetric double helix
which is packed in the sense that the two helices touch each other, i.e. remain at the distance D from
5
CPZT020406080!4!2024v!OverwindFigure 4. each other. Packed double helical structures will show an overwinding behavior similar to those already
observed, as long as their initial pitch angle is sufficiently small. Perhaps, the analysis will be relevant for
other helical structures such as nanofabricated quartz cylinders [17], fabricated twisted polymer nanofibers
[18], and for the beautiful double helical structures formed from helical carbon nanotubes [19]. Further,
the phenomenon may be important for some aspects of the working of molecular motors during gene
expression and regulation [20]. The analysis presented in this paper is straightforwardly applicable to
RNA double helices [21], which we therefore predict will show overwinding. Using a value of 26 A[22] for
the molecular diameter of the double helix, we estimate an overwinding of 5.6 nm. Necturus chromatin
fibers [23] are known to pack as a double helix with a pitch angle of v⊥ = 32±3◦ a value suggestive of being
close-packed. Thus it follows that these chromatin double helices will overwind as well (other chromatin
fibers with a different linker length would not necessarily overwind). Such predictions for overwinding
and unwinding can nowadays be studied on single biomolecules using magnetic traps [24]. Furthermore,
the derived geometrical expressions for overwinding are straightforwardly extended to helices with more
than two strands. In Figure 4 we have shown the solution for a triple helix (dashed line) which has a
zero-twist angle of 42.8◦.
Maybe one will even find examples, where Nature has build zero-twist structures, i.e. structures that
display neither overwinding, nor unwinding. Chromatin with an appropriate linker length, and collagen
are possible candidates for structures with such properties.
Acknowledgements
We would like to thank Jean-Marc Victor for helpful comments on a first version of the manuscript.
6
References
[1] W.S. Utting, N. Jones, The response of wire rope strands to axial tensile loads - Part I. Experimental
results and theoretical predictions, International Journal of Mechanical Sciences 29, 605 (1987).
[2] D.L. Pellow, Rotation resistant wire rope, United States Patent 4365467 (1982).
[3] R.B. Waterhouse, Fretting in steel ropes and cables - a review in "Fretting fatigue: Advances in basic
understanding and Applications", Eds. Y. Mutol, S.E. Kinyon, D.W. Hoeppner, AST International,
West Conshohochen, PA (2003).
[4] K. Olsen, J. Bohr, The generic geometry of helices and their close-packed structures, Theor. Chem.
Acc. 125, 207 (2010).
[5] O. Gonzalez, J.H. Maddocks, Global curvature, thickness and the ideal shapes of knots, PNAS 96,
4769 (1999).
[6] A. Maritan, C. Micheletti, A. Trovato, R. Banavar, Optimal shapes of compact strings, Nature 406,
287 (2000).
[7] A. Stasiak, J.H. Maddocks, Mathematics: Best packing in proteins and DNA, Nature 406, 251 (2000).
[8] S. Neukirch, G.H.M. van der Heijden, Geometry and mechanics of uniform n-plies: from engineering
ropes to biological filaments, Journal of Elasticity 69, 41 (2002).
[9] T. Lionnet, S. Joubaud, R. Lavery, D. Bensimon, V. Croquette, Wringing out DNA, Phys. Rev. Lett.
96, 178102 (2006).
[10] J. Gore, Z. Bryant, M. Nollmann, M.U. Le, N.R. Cozzarelli, C. Bustamante, DNA overwinds when
stretched, Nature 442, 836 (2006).
[11] R.S. Mathew-Fenn, R. Das, P.A.B. Harbury, Remeasuring the Double Helix, Science 322, 446 (2008).
[12] M.Y. Sheinin, M.D. Wang, Twist-stretch coupling and phase transition during DNA supercoiling,
Phys. Chem. Chem. Phys. 11, 4800 (2009).
[13] C.C. Bernido, M.V. Carpio-Bernido, Overwinding in a stochastic model of an extended polymer,
Physics Letters A 369, 1 (2007).
[14] M.L. Smith, T.J. Healey, Predicting the onset of DNA supercoiling using a non-linear hemitropic
elastic rod, Int. Jour. of Non-Linear mechanics 43, 1020 (2008).
[15] A. Bancaud, N. Conde e Silva, M. Barbi, G. Wagner, J.-F. Allemand, J. Mozziconacci, C. Lavelle,
V. Croquette, J.-M. Victor, A. Prunell and J.-L. Viovy, Structural plasticity of single chromatin fibers
revealed by torsional manipulation, Nature structural & molecular biology 13, 444 (2006).
[16] A. Celedon, I. M. Nodelman, B. Wildt, R. Dewan, P. Searson, D. Wirtz, G. D. Bowman and S. X.
Sun, Magnetic tweezers measurement of single molecule torque, Nano Lett. 9, 1720 (2009).
[17] C. Deufel, S. Forth, C.R. Simmons, S. Dejgosha, M.D. Wang, Nanofabricated quartz cylinders for
angular trapping: DNA supercoiling torque detection, Nature Methods 4, 223 (2007).
[18] B.K. Gu, M.K. Shin, K.W. Shon, S.I. Kim, S.J. Kim, S.K. Kim, H. Lee, J.S. Park, Direct fabrication
of twisted nanofibers by electrospinning, Applied Physics Letters 90, 263902 (2007).
[19] J. Liu, X. Zhang, Y. Zhang, X. Chen, J. Zhu, Nano-sized double helices and braids:
interesting
carbon nanostructures, Materials Research Bulletin 38, 261 (2003).
7
[20] J. Michaelis, A. Muschielok, J. Andrecka, W. Kugel, J.R. Moffitt, DNA based molecular motors,
Physics of Life Reviews, 6 250-266 (2009).
[21] K.J. Baeyens, H.L. De Bondt, S.R. Holbrook, Structure of an RNA double helix including uracil-
uracil base pairs in an internal loop, Nature Structural Biology 2, 56 (1995).
[22] A. Varshavsky, Discovering the RNA double helix and hybridization, Cell 127, 1295 (2006).
[23] S.P. Williams, B.D. Athey, L.J. Muglia, R.S. Schappe, A.H. Gough, J.P. Langmore, Chromatin fibers
are left-handed double helices with diameter and mass per unit length that depend on linker length,
Biophysical Journal 49, 233 (1986).
[24] A. Meglio, E. Praly, F. Ding, J.F. Allemand, D. Bensimon, V. Croquette, Single DNA/protein studies
with magnetic traps, Current Opinion in Structural Biology 19, 615 (2009).
[25] M. Barbi, J. Mozziconacci, J.-M.Victor, How the chromatin fiber deals with topological constraints,
Phys. Rev. E 71, 031910 (2005).
8
|
1511.01845 | 3 | 1511 | 2016-03-01T17:02:10 | Shape dynamics of growing cell walls | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.CB"
] | We introduce a general theoretical framework to study the shape dynamics of actively growing and remodeling surfaces. Using this framework we develop a physical model for growing bacterial cell walls and study the interplay of cell shape with the dynamics of growth and constriction. The model allows us to derive constraints on cell wall mechanical energy based on the observed dynamics of cell shape. We predict that exponential growth in cell size requires a constant amount of cell wall energy to be dissipated per unit volume. We use the model to understand and contrast growth in bacteria with different shapes such as spherical, ellipsoidal, cylindrical and toroidal morphologies. Coupling growth to cell wall constriction, we predict a discontinuous shape transformation, from partial constriction to cell division, as a function of the chemical potential driving cell-wall synthesis. Our model for cell wall energy and shape dynamics relates growth kinetics with cell geometry, and provides a unified framework to describe the interplay between shape, growth and division in bacterial cells. | physics.bio-ph | physics |
Shape dynamics of growing cell walls
Shiladitya Banerjee,1 Norbert F. Scherer,1, 2, 3, † and Aaron R. Dinner1, 2, 3, †
1James Franck Institute, The University of Chicago, Chicago IL 60637
2Institute for Biophysical Dynamics, The University of Chicago, Chicago IL 60637
3Department of Chemistry, The University of Chicago, Chicago IL 60637
We introduce a general theoretical framework to study the shape dynamics of actively growing
and remodeling surfaces. Using this framework we develop a physical model for growing bacterial
cell walls and study the interplay of cell shape with the dynamics of growth and constriction. The
model allows us to derive constraints on cell wall mechanical energy based on the observed dynamics
of cell shape. We predict that exponential growth in cell size requires a constant amount of cell wall
energy to be dissipated per unit volume. We use the model to understand and contrast growth in
bacteria with different shapes such as spherical, ellipsoidal, cylindrical and toroidal morphologies.
Coupling growth to cell wall constriction, we predict a discontinuous shape transformation, from
partial constriction to cell division, as a function of the chemical potential driving cell wall synthesis.
Our model for cell wall energy and shape dynamics relates growth kinetics with cell geometry,
and provides a unified framework to describe the interplay between shape, growth and division in
bacterial cells.
INTRODUCTION
Understanding the growth of structures in living sys-
tems presents new challenges for soft matter theory
owing to the interplay of
irreversible dynamics and
mechanochemical forces.
In turn, elucidating how in-
trinsic molecular factors and extrinsic environmental fac-
tors combine quantitatively to determine morphologies is
important for understanding many biological processes,
including wound healing, tissue morphogenesis, tumor
metastasis, and plant cell wall formation. The purpose of
this paper is to report a theoretical framework for mod-
eling the dynamics of actively growing and remodeling
shapes. In particular, we investigate the growth of bac-
terial cell walls, which epitomize growing active matter.
Active growth arises from cell wall enzymes that catalyze
the assembly reactions driving the synthesis of cell wall
material.
Bacteria exhibit a remarkable diversity in cell shapes
and sizes [1]. The shapes of most bacteria are defined by
the peptidoglycan cell wall, in association with cytoskele-
tal proteins and internal turgor pressure. Cell walls are
thicker and stiffer than most polymeric membranes and
are capable of maintaining cell shapes while sustaining
large amounts of turgor pressure. The significant variety
of shapes ranging from spherical cocci to rod-shaped E.
coli to crescent-shaped C. crescentus implies that the
maintenance of each specific shape requires a distinct
physical mechanism[2, 3].
Cell shape has a direct relation to the observed quan-
titative laws governing cell size growth. For instance,
exponential longitudinal growth is observed in rod-like
bacterial cells, such as E. coli [4], and B. subtilis [5], or
crescent-shaped C. crescentus [6], and even in eukaryotic
cells such as the ellipsoidal shaped budding yeast [7]. In
addition, recent experiments on S. aureus, a model sys-
tem for round bacteria, reveal that the cell volume grows
exponentially throughout the cell cycle with increasing
aspect ratio [8]. Thus, an anisotropic cell geometry can
be linked to their exponential growth via lateral pepti-
doglycan insertion.
Starting with Koch's hypothesis that surface stresses
determine bacterial cell shape [9], a number of theoretical
models have been proposed in recent years to account for
the shape and growth of bacterial cell walls. These in-
clude growth by plastic deformations as surface stresses
exceed a critical value [10], elastic growth of peptido-
glycan networks driven by assembly reactions [11], and
dislocation driven growth of partially ordered peptido-
glycan structures [12]. These models however do not
make clear the relationship between cell shape, kinet-
ics of growth and constriction, and the mechanochemical
energies driving growth. For instance, how does expo-
nential longitudinal growth arise from isotropic pressure
in rod-like cells? How does cell shape influence growth
and division kinetics?
Motivated by our recent experimental work that pro-
vides detailed growth and contour data of single C. cres-
centus cells across a large number of generations [6, 13],
we introduce a general theoretical model for the shape
dynamics of growing cell walls based on a principle of
minimal energy dissipation. For a bacterial cell wall,
the dissipative forces arise from the insertion of pepti-
doglycan strands, whereas the driving forces arise from
changes in the mechanochemical energy, E, associated
with maintaining the shape of the cell wall. The depen-
dence of E on cell geometry directly determines which
shape parameters grow and which are size limited. We
discuss how the condition of growth, more specifically
exponential growth, imposes constraints on the form of
the energy function. We show that our model for cell
shape dynamics encompasses previous phenomenological
models of cell wall growth for specific geometries such as
cylinders or spheres [14, 15].
We use the model to study the interplay of cell shape,
growth and division control in bacteria. With new shape
analysis of our experimental data, we demonstrate how
cell shape features such as width and curvature can in-
fluence the rate of cell size growth. In addition to elu-
cidating the relationship between cell shape and growth
dynamics, our model is capable of describing cell wall
constriction. We predict that a threshold chemical po-
tential for septal synthesis is required for completing cell
division such that on reaching the threshold, cell shape
discontinuously switches from partial to full constriction.
We compare the predictions of our model with available
experimental data on cell shape and growth of single bac-
terial cells.
THEORETICAL FRAMEWORK
Equations of Shape Dynamics
We parametrize the geometry of the cell wall by N
shape degrees of freedom specified by the generalized co-
ordinates qi (i = 1, . . . , N ) and the generalized velocities
qi. For example, a sphere is parametrized by its radius,
whereas a cylinder has two degrees of freedom, its radius
and length (Fig. 1A-C). For a general surface described
by a mesh of triangles, the generalized coordinates are
given by the individual vertex positions. The mechanical
energy of the cell wall is a function of the generalized
coordinates, Em({qi}). The generalized forces driving
changes in cell wall geometry are given by the deriva-
i = −∂Em/∂qi, ∀i. The
tives of the energy function, F m
equilibrium shape of the cell wall is simply given by min-
imizing the mechanical energy, ∂Em/∂qi = 0, ∀i.
In the absence of external forces and strong thermal
noise, the mechanical forces Fm counterbalance the ac-
tive forces (Fa) and the dissipative forces (Fd) associated
with irreversible cell wall growth, Fa + Fd + Fm = 0.
The active forces are of non-equilibrium origin and they
arise from distributed molecular components in the cell
that convert chemical energy into mechanical work. For
example,
in the case of a growing bacterial cell wall,
the active forces arise from cell wall enzymes that cat-
alyze the assembly reactions driving peptidoglycan syn-
thesis. We define the energy due to active processes as
Ea =(cid:80)
i
(cid:82) dqiF a
i .
The dissipative forces are not symmetric under time-
reversal and they are derived from minimizing the rate of
i = −∂D/∂ qi. The dissi-
energy dissipation, D, using F d
pation function, D, represents the amount of work done
to the medium when the shape deforms at a rate qi. The
force-balance relation follows from minimizing the total
rate of energy change in the system and the medium,
i.e., ∂( Em + Ea + D)/∂ qi = 0, which is a statement of
Rayleigh's principle of least energy dissipation [16]. This
variational principle can be shown to hold for general ir-
2
FIG. 1. Examples of shape parameters for (A) spherical,
(B) cylindrical, and (C) curved cells. (D) Illustration of the
growth law. Rate of growth of the shape parameter qi is pro-
portional to the energy dissipated (qiFi) per unit volume (Vi).
Inset: Physical picture of growth for a rod-like cell where the
shaded region represents the dissipated volume and Fi is the
driving force.
(cid:80)
reversible processes and is equivalent to maximizing the
rate of the entropy production in the system [17].
The rate of dissipated energy is given by D =
1
i Viσi( qi/qi), where Vi is the volume over which dissi-
2
pation of qi occurs, σi is the dissipative stress, and qi/qi is
the strain rate. Assuming the deformation is small com-
pared to current cell size, the dissipative stress is given
by σi = ηi( qi/qi), where ηi is the associated viscosity
constant. The resultant equations of motion are
ηiVi
qi
q2
i
= − ∂Em
∂qi
+ F a
i = Fi ,∀i ,
(1)
such that the rate of growth is proportional to the to-
tal energy dissipated per unit volume (Fiqi/Vi), as illus-
trated in Fig. 1D. In the limit of a rod-like cell, when the
dissipated volume scales linearly with cell length, Vi ∝ qi
(Fig. 1D, inset), our model reduces to the phenomeno-
logical growth model proposed by Jiang and Sun [15].
Eqn (1) can be written in a simple and instructive form
by choosing logarithmic strain, Φi(t) ≡ log [qi(t)/qi(0)],
as our new dynamic variable,
ηi
dΦi
dt
= − 1
Vi
∂E
∂Φi
,∀i ,
(2)
where we defined E = Em + Ea, as the total internal
energy of the system. Eqn (2) represents the familiar
constitutive law of Newtonian flow such that the internal
stress (right hand side), is proportional to the rate of
strain. Eqn (2) also illustrates that the laws of shape
dynamics are isomorphic to the overdamped motion of a
particle with coordinate Φi(t) in a potential E.
Energy requirements for exponential growth
We now discuss how the specific form of the growth
law puts constraints on the scaling of the energy func-
tion with the shape parameters. A necessary condition
for growth in the shape parameter qi is ∂E/∂qi < 0.
Growth is arrested when ∂E/∂qi = 0.
It follows from
eqn (1) that qi(t) grows exponentially if E scales as the
dissipated volume Vi, such that a constant stress E/Vi
drives material growth. For a growing bacterial cell wall,
we assume that dissipation is dominated by peptidogly-
can insertion over the thin shell defining the cell wall,
such that Vi = hAi, where Ai is the surface area over
which dissipation occurs and h is the thickness of the cell
wall assumed to be constant and uniform. A minimal
model for exponential growth thus requires E ∝ −Ai.
For a thin spherical shell of radius r the dissipative
volume scales as V ∝ r2. It then follows from the dy-
namics of cell radius, r ∝ −∂E/∂r, that for cells to grow
exponentially i.e., r ∝ r, the energy would need to scale
as E ∝ −r2. Using eqn (1), we can thus conclude that a
minimal energy model for exponential growth of isotropic
cells is given by E = −εA, where A is the surface area
and ε is a positive constant representing the chemical
potential for adding unit surface area.
As an example of growth dynamics in anisotropic cells,
we consider a thin cylindrical shell of length L and radius
r. From eqn (1) it follows that the radius and the length
L/L ∝ −r−1∂E/∂L, and
grow exponentially with rates,
r/r ∝ −L−1∂E/∂r. The shape-dependence of growth
rates implies that a minimal energy model E = −εA,
with ε > 0, can describe exponential growth in both
cell radius and length. However most rod-like bacteria
elongate in length while maintaining a fixed radius, sug-
gesting a more complex shape dependence of the growth
energy.
Mechanical Energy model
The mechanical energy of a growing cell wall is given
by the sum of contributions from an internal turgor pres-
sure, Π, acting to expand the cell volume, V , surface
tension, γ, resisting increase in the cell surface area, A,
and the mechanical energy of interaction with cytoskele-
tal bundles, Ecyto, which controls cell shape. That is,
Em = −ΠV +
dA γ + Ecyto .
(3)
(cid:90)
The surface tension is determined by the stored elastic
energy per unit area of the cell wall, possibly offset by
favorable peptidoglycan interactions at the surface [11].
Contributions to Ecyto arise from MreB bundles that con-
trol cell width [18, 19], FtsZ filaments that drive cell wall
constriction [20], and crescentin bundles that control cur-
vature in C. crescentus cells [21].
3
In the following Results section, we use eqn (1) to
study shape dynamics in spherical, ellipsoidal, rod-like,
and curved bacteria, by considering specific forms for the
energy function Em. We compare our predictions and
results against available experimental observations and
data.
RESULTS
Growth in round cells
We first consider the simplest case of a spherical cell
as a model for round bacteria like S. pneumoniae or
S. aureus, where cytoskeletal bundles such as MreB are
known to be absent (Ecyto = 0). We model the active
growth energy as Ea = −ΠaV , where Πa is the energy
released per unit volume of cell wall synthesis. Neglect-
ing cell division, the internal energy is simply given by,
Eround = −P V + γA, where P = Π + Πa is the effec-
tive growth pressure. The dynamics of the cell radius, r,
follow from eqn (1):
(cid:18)
(cid:19)
,
(4)
dr
dt
= 4πµrr
P r − 2γ − r
dγ
dr
where µr = 1/4πhηr is the mobility coefficient. We
consider two distinct models for cell wall mechanics.
If the cell wall deforms like a plastic material, γ is
a constant [10] such that there exists a critical radius
rc = 2γ/P (given by Laplace's law [22]), at which the cell
size is stationary. By minimizing the energy one finds
that the cell grows for r > rc and shrinks for r < rc.
Thus, a newborn cell must at least attain a critical size
rc for growth and survival. If, however, the surface ten-
sion originates from the elastic strain energy stored in
the pressurized spherical shell, we get γ = γ0(r/rc)2/2,
where γ0 = Y h/2(1−ν), Y is the Young's modulus and ν
is the Poisson ratio of the cell wall [23]. In this case, the
cell radius attains the steady-state value, rc = 2γ0/P ,
which corresponds to an absolute minimum in the in-
ternal energy. The latter case is relevant for spherical
bacteria that maintain a stable cell size before the onset
of cell division [24]. Thus, a plastic cell wall can support
indefinite growth if nutrient availability is optimal and
division is inhibited, whereas an elastic cell wall cannot
support growth beyond a threshold size rc.
However, in reality bacteria such as S. aureus are not
perfectly spherical but have an ellipsoidal shape. Recent
experiments show that S. aureus cells grow in volume
throughout their cell cycle while their aspect ratio ini-
tially decreases followed by a period of increase [8]. For
a more realistic description of S. aureus geometry and to
facilitate closer comparisons with experiments we model
the shape of a S. aureus bacterium as a spheroid, with
semi-axes a and b (b > a) defining the shape parameters
4
where Ecyto = Ewidth is the mechanical energy for main-
taining the cell width.
(cid:19)2
(cid:18) 1
r
− 1
R0
(cid:19)2
,
(cid:90)
(cid:18) 1
r
Ewidth =
k
2
dA
− 1
R0
= kπrL
FIG. 2. Growth modes in ellipsoidal bacteria.
(A) Oblate
growth in spheroidal cells, in the absence of septum tension
(f = 0). Cell volume (red curve) and aspect ratio (blue
curve), respectively, increase and decrease with time. τ de-
fines the generation time and V0 is the initial cell volume. In-
set: Cell shape evolution during growth. (B) Prolate growth
in spheroidal cells in the presence of tension in the septal ring,
f = 0.5. Both volume and aspect ratio grow during the cell
cycle. Inset: (Top) cell shape evolution during growth. (Bot-
tom) Schematic of a spheroidal cell with semi axes a and b,
and f is the tension in the septal ring in units of P a(t = 0)2.
Parameters: γ = 0.4, P = 1, µa = µb = 1.25.
(Fig. 2B, inset). Their dynamics are given by
1
a
1
b
da
dt
db
dt
= − aµa
A
= − bµb
A
∂Eround
∂a
∂Eround
∂b
,
,
(5a)
(5b)
where A and Eround are the surface area and the energy
of the spheroidal bacterium with µa and µb defining the
growth mobility along the semi-axes a and b respectively.
We model the active energy as Ea = −ΠaV +2πaf , where
f is the tension due to the division septum at the midcell
(Fig. 2B, inset). Net energy is thus given by Eround =
−P V + γA + 2πaf . Prior to the formation of the division
septum, f = 0, and the cell exhibits oblate growth such
that it increases in volume but decreases in aspect ratio.
As shown in Fig. 2A, an initial spheroidal cell with b > a
will assume a spherical shape with b = a. In contrast,
for non-zero f , the cell exhibits prolate growth such that
volumetric growth is accompanied by increasing aspect
ratio (Fig. 2B), in agreement with recent experiments on
S. aureus [8].
Growth and shape control in rod-like cells
Rod-like cells such as E. coli assume the shape of a
sphero-cylinder parametrized by the radius (r) and the
length (L) (Fig. 3A, inset). E. coli cells grow by lateral
insertion of peptidoglycan material [25]. We neglect the
hemispherical poles that are mechanically rigid and in-
ert [26]. The internal energy for the cylindrical cell is
given by
Erod = −P (πr2L) + γ(2πrL) + Ecyto ,
(6)
where R0 is the preferred radius of cross section of the
cell wall and k is the circumferential bending rigidity.
Contributions to k can come from the elasticity of gly-
can strands in the peptidoglycan cell wall as well as from
membrane bound cytoskeletal proteins such as MreB,
MreC and RodZ that are known to be responsible for
maintaining rod-like cell shape [14, 27, 28].
In this section, we neglect constriction to establish the
basic growth dynamics. Such a situation can be realized
experimentally by suppressing division, which gives rise
to filamentous cells [29]. The internal energy assumes the
scaling form Erod(r, L) = U (r)L, where the energy den-
sity, U , is solely a function of the cell radius. According
to eqn (1), the length and the radius evolve as
1
L
1
r
dL
dt
dr
dt
= −µL
U
r
= −µr
dU
dr
(7a)
(7b)
,
where µL = 1/2πhηL and µr = 1/2πhηr are the longi-
tudinal and radial mobility coefficients, respectively, and
ηL and ηr are the associated viscosities. From eqn (7a)
the cell length grows exponentially if U has a minimum
at r = rs such that U (rs) < 0 and is a constant. Fig. 3A
shows the dynamics of length and radius in the regime of
parameters that allow exponential elongation at constant
radius.
Assuming that the cell maintains a constant effec-
tive growth pressure, we can rescale the energy density
by U0 = πP R2
0. The shape dynamics are then con-
trolled by two dimensionless parameters, γ = γ/P R0
In the limit k (cid:29) 1, the cell radius
and k = k/P R3
0.
approaches R0, and the cell assumes a stationary shape
defined by the value γ = 1. As such, the numerical
values for γ and κ are cell type dependent (Table 1).
For Gram-negative E. coli cells with Young's modulus
Y (cid:39) 25 − 50 MPa, h (cid:39) 3 nm and P (cid:39) 0.3 MPa [32],
the estimated values for γ lie in the range 0.25 − 0.5.
Whereas for Gram-positive B. subtilis cells with smaller
values for Young's modulus Y (cid:39) 15− 30 MPa, and larger
values for thickness and pressure, h (cid:39) 30 − 40 nm and
P (cid:39) 1.5 MPa [33], the estimated values for γ lie in the
range 0.3 − 0.8. In Fig. 3B we show the dependence of
the cell width (2r) on the rate of exponential growth,
κ = −µLU (r)/r, for parameter values corresponding to
E. coli and B. subtilis. The parameters are determined
by fitting our model prediction to the available data on
E. coli [30] and B. subtilis [31]. In agreement with ex-
perimental data [30, 31, 34, 35], our model quantitatively
captures the positive correlation between κ and r for both
based on mechanical measurements [13, 32, 36] and sug-
gest that rod-like bacteria operate close to the triple point
in Fig. 3C.
5
Response to shape perturbations
Having discussed growth and shape dynamics under
steady environmental conditions, we now consider how
rod-like cells modulate their growth dynamics in response
to morphological perturbations. It has been experimen-
tally observed that upon addition of A22, which causes
disassembly of MreB, wave-like bulges form on the cell
wall [14]. Thus, the loss of MreB, which corresponds
to lower values of k, can induce morphological instabil-
ities in the cell wall. Motivated by this observation, we
now investigate the robustness of the rod-like geometry
to external perturbations as a function of k. We examine
the stability of a cylindrical shape under a small peri-
odic perturbation of the steady-state cell radius rs along
the axial direction z: r(z, t) = rs(t) + δr(t) cos (2πz/λ),
where δr (cid:28) rs is the amplitude and λ is the wavelength
of the perturbation (Fig. 4A). To leading order in δr, the
internal energy of the cell integrated over one cycle of the
perturbation is given by Erod/λ = U + α(λ)δr2 + O(δr4),
where the decay rate of the perturbation, α(λ), is an in-
creasing function of the wavenumber 2π/λ for all values
of k (Fig. 4B). The stability of the cylindrical shape is de-
termined by the sign of α. For α < 0 the cylindrical shape
is unstable to perturbations of wavelength greater than
λ, such that wave-like bulges nucleate on the cell-wall
with growing amplitude. For k = 0 the cylindrical shape
is unstable for λ > λmin = 2π(cid:112)γrs/P . However, as k
increases beyond a critical value, the cylindrical shape is
stable to perturbations of any wavelength.
Another experiment that allows examining predictions
of our model
involves studying bacterial growth and
movement in sub-micron microfluidic channels that ge-
ometrically confine growth [37, 39] (Fig. 4C). Rod-like
bacteria such as E. coli or B. subtilis are able to grow
in very narrow microfluidic channels of width compara-
ble to or even smaller than their unperturbed diameters.
Furthermore, bacterial cell walls can deform (the ceiling
of) the microchannels, which are made of elastic material
such as PDMS. To verify if our growth model can cap-
ture the experimental results [37], we include the elas-
tic interaction between the channel and the cell wall as
Eint(rc < rs) = 1
nel stiffness, r is the radius of the bacterial cell wall, rc is
the radius of the cylindrical channel and rs is the steady
state radius of a freely growing bacterium. If the chan-
nel is wider than rs then there is no interaction between
the channel and the cell wall, Eint(rc > rs) = 0. The
total energy is then given by Erod + Eint. By solving the
coupled equations for cell length and radius (eqn (1)) we
derive the growth rate dependence on channel width. For
(cid:82) dl(r − rc)2, where kch is the chan-
2 kch
FIG. 3. Growth dynamics of rod-like cells. (A) Dynamics of
length (L) and radius (r) normalized by their initial values
(L0 and r0) in semi-log scale with time normalized by the
timescale for growth, κ−1 (see text). The surface tension and
bending rigidity are γ = 0.3 and k = 6, respectively. Inset:
Schematic of a longitudinally growing cylindrical cell. (B) De-
pendence of cell width (2r) on growth rate (κ). Open circles
represent experimental data for E. coli [30] (blue) and B. sub-
tilis [31] (red) and solid curves represent model fits. Fitting
parameters: (E. coli) γ=0.56, k=3.2; (B. subtilis) γ=0.53,
k=3.6. (C) Phase diagram in γ-k plane showing different re-
gions of steady-state behavior. Insets: Representative plots
of energy density U as a function of cell radius r in the three
regions of parameter space.
cell types. The predicted cell width for E. coli is more
sensitive to changes in growth rate, presumably due to
the fact that the cell wall is softer and thinner in E. coli
because it is gram negative.
The steady-state behavior at different values of γ and
k is shown in Fig. 3C. The corresponding plots of the en-
ergy densities are shown in the insets to Fig. 3C. While
radial growth occurs for smaller values of k, exponen-
tial elongation with constant radius occurs for γ (cid:46) 0.5
and k (cid:38) 4. Using the typical range of estimates for the
internal pressure in gram-negative bacteria P (cid:39) 0.1-0.5
MPa [26, 32] and the preferred radius of cross-section
R0 (cid:39) 0.1-0.5 µm [14], we predict the upper bound on
surface tension to be γmax (cid:39) 50 nN/µm and a lower
bound on the circumferential rigidity to be kmin (cid:39) 0.4
nNµm. These values are in agreement with estimates
TABLE I. List of parameters used in the energy model
6
Parameter Description
Π
Turgor Pressure
Function
Cell wall expansion
Associated Molecules Numerical Estimate
Peptidoglycan
Πa
Growth pressure
Cell wall synthesis
PBPs, MreB
Surface tension
Cell shape maintenance
Peptidoglycan
0.3 MPa (E. coli)
1.5 MPa (B. subtilis) [32, 33]
0.4 MPa (E. coli)
1.5 MPa (B. subtilis)
19 nN/µm (E. coli)
113 nN/µm (B. subtilis) [32, 33]
γ
k
R0
kc
Rc
ε
f
Circumferential bending rigidity Cell width control
MreB, MreC, RodZ 0.03 MPaµm3 (E. coli)
0.3 MPaµm3 (B. subtilis) [33]
Preferred radius of cross-section Cell width maintenance
MreB, MreC, RodZ 0.38 µm (E. coli)
Longitudinal bending rigidity
Preferred radius of curvature
Chemical potential for growth
Line tension
Cell curvature control
Crescentin
Cell curvature maintenance Crescentin
Septum synthesis
Constriction force
PBPs, divisomes
FtsZ
0.43 µm (B. subtilis) [14, 37]
1.5 nNµm2 [13, 15]
2-6 µm
>12 nN/µm (prediction)
8-80 pN [33, 38]
channels wider than unperturbed cell radius, rc > rs, the
cells grow at a constant rate κ = κs. For rc < rs, the
dependence of κ on rc is controlled by the dimension-
less parameter kchR0/γ, describing the stiffness of the
channel relative to the cell wall (Fig. 4D). For channels
softer than the cell wall, we find that the growth rate is
insensitive to channel radius. However if the channel is
stiffer than the cell wall, we predict that the growth rate
increases monotonically with rc, and no growth occurs
below a critical channel radius. We quantitatively com-
pare our model predictions with the experimental data
on doubling times of E. coli cells vs channel radius [37].
As shown in Fig. 4E, our model is in good quantitative
agreement with the data and predicts that the longitudi-
nal growth rate is insensitive to changes in channel width
beyond 0.5 µm.
Curved cells
As an example of a curved cell, we explore the shape
dynamics of a C. crescentus bacterium. Note that the
results are not specific to crescent-shaped bacteria, and
apply equally well to helical bacteria, for example. We
model the geometry of a C. crescentus cell by a toroidal
segment parametrized by the radius of cross section r,
centerline radius of curvature R, and the spanning angle θ
(Fig. 5A, inset). Experiments have shown that the curva-
ture of C. crescentus cells is maintained by intermediate
filament-like bundles of crescentin proteins that adhere to
the concave face of the cell wall [40]. Although the molec-
ular mechanism by which crescentin maintains cell cur-
vature is not precisely known, proposed models include
modulation of elongation rates across the cell wall [40],
which can originate from bundling with a preferred cur-
vature [15]. We thus model the curvature energy in the
FIG. 4. Response of growth dynamics to shape perturba-
tions. (A) An initial cylindrical cell (solid line) undergoing
morphological perturbation of harmonic form (dashed curve).
(B) Decay rate α for the amplitude of the harmonic pertur-
bation of the cell radius, as a function of the dimensionless
wavenumber 2πR0/λ at various values of k. The system is
stable to perturbations for α > 0 and unstable for α < 0. (C)
Schematic of rod-like bacteria squeezed into narrow channels
of width 2rc and rigidity kch [37]. When not constrained
to grow in the microchannels, the cells grow while maintain-
ing a constant diameter 2rs. (D) Dependence of longitudinal
growth rate (normalized by the steady-state value κs) on the
channel radius (normalized by steady-state cell radius rs) for
different values of the dimensionless parameter kchR0/γ, de-
scribing the relative stiffness of the channel to the cell wall.
Parameters: γ = 0.45, k = 6. (E) Doubling time vs channel
width for E. coli cell parameters (Table 1). Solid blue curve is
the model prediction. The data (red) are taken from Ref. [37].
7
Ecurv(R, r, θ) = θUc(r, R). The dynamics of the shape
parameters R, r, and θ follow from eqn (1), character-
ized by the viscosity parameters ηR, ηr and ηθ, respec-
tively. The cell exhibits hoop-like growth [41] with R and
r remaining constant and θ growing exponentially as
1
θ
dθ
dt
= −µθ
Uc(r, R)
rR
,
(9)
with a rate κ = −µθUc/rR, where µθ = 1/(2πhηθ) is
the hoop growth mobility. The shape variables R and
r attain constant steady-state values determined by the
global minimum of Uc(r, R) (Fig. 5B). Our model pre-
dicts that the angular growth rate, κR, is a decreasing
function of the curvature, 1/R (Fig. 5A, red curve). This
coupling between angular dynamics and curvature arises
through the curvature dependence of the bending energy
(Ecres) that increases with cell curvature. The growth
rate is proportional to −Ecres through the dependence of
Uc on R, and κ is consequently larger for straight cells
(larger R).
We test this prediction of our model with our experi-
mental shape data on single C. crescentus cells [13]. The
scatter plot showing the dependence of angular growth
speed on cell curvature demonstrates that angular growth
is slower for curved cells (Fig. 5A). The fitted model (red)
is in excellent agreement with the binned data (black
points) at the value of the fitting parameter kc = 1.75,
which further constrains the physical values of γ in the
range 0.2-0.5, as discussed below (see Fig. 5D).
In the absence of crescentin (kc = 0), the internal en-
ergy assumes the scaling form Ecurv(R, r, θ) = θRU (r),
such that both R and θ grow exponentially as expected
during self-similar growth. This leads to cell straight-
ening (Fig. 5C), as observed for cells lacking creS [40].
The cell curvature, C = 1/R, decays as dC/dt = −κ(cid:48)C,
whereas the angle grows according to dθ/dt = κ(cid:48)(cid:48)θ,
with κ(cid:48)/κ(cid:48)(cid:48) = ηθ/ηR. The cell length (L = Rθ) conse-
quently grows exponentially with a rate κ(cid:48) + κ(cid:48)(cid:48). The
ratio κ(cid:48)/(κ(cid:48) + κ(cid:48)(cid:48)), which quantifies the propensity of cell
straightening, has been experimentally determined to be
(cid:39) 0.57 [42]. We thus estimate the ratio of viscosities
characterizing the angular and curvature dynamics to be
ηθ/ηR (cid:39) 1.3, implying that angular growth is slower than
the decay of cell curvature. The steady-state behavior
at different values of γ and kc, which control cell size
and shape respectively, is shown in Fig. 5D for fixed val-
ues of pressure and width.
In particular, we find that
the cell elongates exponentially while maintaining a con-
stant curvature in the range 0.2 < γ < 0.5. At smaller
values of surface tension, γ, the cell wall cannot support
curvature-induced stresses and relaxes to a straight mor-
phology. For γ > 0.5, the cell maintains a stationary size
and shape.
FIG. 5. Growth and shape dynamics of curved cells. (A) In-
set: Schematic of a C. crescentus cell contour and the shape
parameters [13]. Dependence of angular growth rate (κR) on
the curvature of the cell 1/R. We determine the cell cur-
vature and the spanning angle for each generation from the
splined contour of the cell boundary. We then obtain the an-
gular growth rate, κ, by fitting an exponential to the data
for θ(t). Gray points indicate single-generation data. Ex-
perimental binned data [13] are shown by solid black circles
and the model fit is given by the red dashed curve. Error bars
represent ±1 standard deviation. (B) Curvature maintenance
in the presence of crescentin at a value of the dimensionless
bending rigidity kc = kc/P R4
0 = 4. Dynamics of the radius of
curvature (R), radius of cross-section (r), and spanning angle
(θ) normalized by their initial values in a semi-log plot. (C)
Cell straightening in the absence of crescentin (kc = 0). Cell
length (L), radius of curvature (R) and spanning angle (θ)
grow exponentially (shown in a semi-log plot, normalized by
the respective initial values). Parameters: k = 5, γ = 0.3,
Rc = R0. (D) Phase diagram in γ-kc plane illustrating the
steady-state growth behaviors.
crescentin bundle as
Ecres =
(cid:90) (cid:96)c
0
kc
2
(cid:18)
d(cid:96)
C − 1
Rc
(cid:19)2
,
where (cid:96)c = (R−r)θ is the contour length of the crescentin
bundle, C is the longitudinal cell wall curvature, kc is the
bending rigidity, and Rc is the intrinsic radius of curva-
ture of the bundle. The energy term, Ecres, accounts
for the compressive stresses generated by the crescentin
bundle on one side of the cell wall, thereby leading to dif-
ferential growth across the sidewall. The total internal
energy is given by
Ecurv = −P V + γA + Ecyto ,
(8)
where Ecyto = Ewidth + Ecres. In the presence of cres-
centin, kc (cid:54)= 0, the internal energy has the scaling form
Cell wall constriction
We now study how cell wall growth couples with con-
striction in bacteria. The onset of cell wall constriction
influences the overall shape dynamics of the cell. For sim-
plicity, we first consider the case of a rod-like bacterium.
Bacterial cell division is driven by a large complex of pro-
teins, known as the divisome, that assembles the Z-ring
near the mid-plane of the cell [43]. The Z-ring comprises
FtsZ filaments that form a patchy band structure [44].
It is believed that these filaments generate constrictive
and bending forces [20]. In addition the divisome triggers
peptidoglycan synthesis and directs formation of the sep-
tum [45]. We assume that the shape of the constriction
zone is defined by two intersecting and partially formed
hemispheres with radii r, equal to the radii of the new
poles (Fig. 6A). The shape parameter defining the mid-
cell radius, rmin(t), equals r at the onset of constriction
and reaches 0 at the completion of division. We assume
that the Z-ring proteins exert a mechanical tension f
on the cell wall and trigger septal growth by releasing
an energy ε per unit surface area. The chemical poten-
tial ε is related to the activity of MreB and penicillin-
binding proteins (PBPs) that synthesize peptidoglycan
by localizing to the division site near the mid-plane of
the cell [46, 47]. These active mechanisms contribute an
energy Ea = f (2πrmin)−εS, where S is the septal surface
min. With Ecyto = Ewidth,
the energy of the constricting cell thus takes the scaling
form Erod(r, L, rmin) = U (r)L + E (rmin, r), where E de-
fines the effective energy of constriction. Therefore, the
steady-state values for rmin are controlled by the tension
f and the chemical potential ε.
area given by S = 4πr(cid:112)r2 − r2
To determine the minimum values of ε and f that are
required to achieve full constriction, we first examine the
dependence of E on rmin at different values of the di-
mensionless chemical potential ε = ε/P R0 while keeping
f fixed (Fig. 6B). At ε = 0 the energy is minimized for
rmin (cid:39) r, and no constriction occurs. As ε is increased,
the local minimum of the energy at rmin/r (cid:39) 1 shifts
towards a more constricted state, but division is still un-
successful. For ε (cid:38) εc, the local minimum is lost in favor
of a global minimum at rmin = 0, corresponding to a fully
constricted state.
This energy minimization approach reveals the funda-
mental mechanism behind constriction: cell shape main-
tenance enforces a competition between ε (and f ) that
minimizes the mid-plane perimeter and the surface ten-
sion γ that resists the associated increase in surface
area [48]. The steady-state ratio rmin/r obtained by min-
imizing the energy functional, gives us an order parame-
ter for cell division, such that division is unsuccessful for
ε < εc and successful for ε > εc (rmin = 0). The bifur-
cation diagram in Fig. 6C shows the dependence of the
order parameter rmin/r as a function of ε. For smaller
8
FIG. 6. Mechanics of cell wall constriction. (A) Schematic of
the constricting cell. The arrows indicate the forces driving
constriction arising from tension (f ) and growth (ε). (B) Con-
striction energy E as a function of rmin/r at a fixed tension
f = f /P R2
0 = 0.2 and different values of the dimensionless
chemical potential ε = ε/P R0: 0 (red), 0.15 (purple), 0.3
(brown) and 0.4 (blue). The corresponding minima are indi-
cated by solid circles. (C) Bifurcation diagram showing the
dependence of the division order parameter rmin/r (green)
on the chemical potential ε. Cell division is successful for
ε > εc (cid:39) 0.2, where a discontinuous transition occurs be-
tween partial and full constriction ( f = 0.2).
(D) Critical
tension fc required for full constriction as a function of the
chemical potential ε. For ε > 0.45, no mechanical force is
required for cell division, with fc < 0.015.
values of ε, rmin/r decreases, whereas for ε > εc (cid:39) 0.2
(when f = 0.2), there is a discontinuous transition to
a fully constricted state. The prediction of a threshold
force for completion of constriction could be tested ex-
perimentally by treating cells with controlled amounts
of Divin, a small molecule inhibitor of bacterial divisome
assembly that reduces peptidoglycan remodeling and pre-
vents cytoplasmic compartmentalization [49]. Consistent
with this suggestion, it was found experimentally that a
threshold amount of Divin is required to inhibit bacterial
cell division.
Having discussed the mechanisms for cell constriction,
it is pertinent to consider the relative contributions of
the mechanical tension f and the chemical potential ε in
executing cell wall constriction. Fig. 6D shows the depen-
dence of the critical force fc required for full constriction
on the magnitude of the chemical potential ε. While a
large mechanical force is required for low values of ε, we
predict that for ε (cid:38) 0.45 little (or no) mechanical force is
required to complete division. Using P = 0.03 MPa [32]
and R0 = 0.4 µm, we predict a numerical value for the
(cid:39) 72
upper bound of the Z-ring mechanical force f max
pN which translates to fc = 0.015 in dimensionless units.
This estimate is consistent with the mechanical proper-
c
9
eters controlling longitudinal and circumferential curva-
tures of the cell wall. Our model can reproduce different
families of known bacterial shapes (cocci, bacilli, vibrio)
by varying the mechanical rigidities controlling the cur-
vatures of the cell wall.
In this paper we obtained the following key conclu-
sions:
• Exponential growth in cell size requires a constant
amount of energy dissipation per unit volume.
• Cell shape, as opposed to simply size, controls the
rate of exponential growth in cell size.
• Cell division can be explained as a discontinuous
(first-order) shape transformation controlled by the
interplay between cell wall surface tension and the
chemical potential required for the addition of new
cell wall material.
• Cell growth and constriction are both driven by the
addition of new cell wall material, and thus their
kinetics are same. This insight provides a physi-
cal explanation for the recent experimental obser-
vation that a single time scale governs growth and
division [6].
The microscopic formulation of the equations of mo-
tion makes it convenient for their adoption in compu-
tational modeling of cell wall growth and morphology.
It is, however, important to recognize that the underly-
ing structure of the bacterial cell wall is highly dynamic,
and cellular mechanical properties may fluctuate due to
molecular scale defects and stochastic forces. Our model
is thus valid on timescales comparable to measurable cell
wall growth (∼minutes) that are much larger than the
timescales of molecular processes (∼seconds) involving
peptidoglycan bond rupture and subsequent insertions
of new cell wall material. In future work we aim to in-
corporate the effects of stochasticity and spatiotemporal
variations in cellular material parameters to better un-
derstand the statistical mechanics of shape fluctuations
in living cells.
ACKNOWLEDGEMENTS
We thank Charles Wright and Srividya Iyer-Biswas for
help with reproducing the experimental data shown in
Fig. 5A from ref. [13]. We gratefully acknowledge fund-
ing from the NSF Physics of Living Systems program
(NSF PHY-1305542), NSF Materials Research Science
and Engineering Center (MRSEC) at the University of
Chicago (NSF DMR-1420709). NFS acknowledges par-
tial support from the ONR NSSEFF program. We thank
an anonymous reviewer for bringing Refs. 29, 30, 37, 38
to our attention.
FIG. 7. Bacterial polymorphism. Shape stability diagram
as functions of longitudinal rigidity, kc (normalized by P R4
0),
and circumferential rigidity, k (normalized by P R3
0).
ties of FtsZ filament bundles [50]. Previous models have
also suggested that a force in the range 8-80 pN (0.0017-
0.017 in our dimensionless units) is sufficient for pinch-
off during division of rod-like bacteria [33, 38]. We thus
claim that typical rod-like bacterial cells operate in the
regime ε > 0.45 (ε >12 nN/µm) such that constriction is
entirely driven by cell wall synthesis at the septum. The
predicted minimum value for the chemical potential is
roughly one-fourth of the surface tension measured for
Gram-negative bacteria [32].
CONCLUSIONS
How cells regulate their shapes and sizes through the
processes of growth and division poses a fundamental
question at the interface of physics and biology. To ad-
dress this fundamental question, we have developed a
broadly applicable model for the shape dynamics of grow-
ing cell walls that are driven by mechanical and active
forces (eqn (1)). Our model takes advantage of recent
technological advances in single cell imaging [4, 6, 13, 51]
that have yielded unprecedented amounts of quantitative
information about the shapes of single bacteria as they
grow and divide. The active forces arise from proteins
driving cell wall growth and constriction, whereas the
mechanical forces arise from tensions in the peptidogly-
can cell wall and associated cytoskeletal bundles. The
equations for the shape dynamics in combination with
the appropriate energy models (see Table I for a sum-
mary of the model parameters), describe a wide range of
phenomena that occur in bacterial cells, including expo-
nential growth, steady-state sizes, shape robustness and
constriction. Using the energy model, we demonstrate
how width and curvature control can be achieved in bac-
terial cells and discuss the mechanical instabilities that
can lead to morphological transformations (Figs. 4 A,B
and 5C). In Fig. 7 we show the shape stability diagram for
our energy model as functions of the mechanical param-
† To whom correspondence may be addressed. Email: din-
[email protected] or [email protected]
[1] M. G. Pinho, M. Kjos, and J.-W. Veening, Nature Re-
views Microbiology 11, 601 (2013).
[2] M. T. Cabeen and C. Jacobs-Wagner, Annual Review of
Genetics 44, 365 (2010).
[3] A. Gahlmann and W. Moerner, Nature Reviews Micro-
biology 12, 9 (2014).
[4] P. Wang, L. Robert, J. Pelletier, W. L. Dang, F. Taddei,
A. Wright, and S. Jun, Current Biology 20, 1099 (2010).
[5] A. L. Koch, Antonie van Leeuwenhoek 63, 45 (1993).
[6] S. Iyer-Biswas, C. S. Wright, J. T. Henry, K. Lo,
S. Burov, Y. Lin, G. E. Crooks, S. Crosson, A. R. Dinner,
and N. F. Scherer, Proceedings of the National Academy
of Sciences 111, 15912 (2014).
[7] S. Di Talia, J. M. Skotheim, J. M. Bean, E. D. Siggia,
and F. R. Cross, Nature 448, 947 (2007).
[8] X. Zhou, D. K. Halladin, E. R. Rojas, E. F. Koslover,
T. K. Lee, K. C. Huang, and J. A. Theriot, Science 348,
574 (2015).
[9] A. L. Koch, Bacterial Growth and Form (Springer, 2001).
[10] A. Boudaoud, Physical Review Letters 91, 018104
(2003).
[11] H. Jiang and S. X. Sun, Physical Review Letters 105,
028101 (2010).
[12] A. Amir and D. R. Nelson, Proceedings of the National
Academy of Sciences 109, 9833 (2012).
[13] C. S. Wright, S. Banerjee, S. Iyer-Biswas, S. Crosson,
A. R. Dinner, and N. F. Scherer, Scientific Reports 5,
9155 (2015).
[14] H. Jiang, F. Si, W. Margolin, and S. X. Sun, Biophysical
Journal 101, 327 (2011).
[15] H. Jiang and S. X. Sun, Soft Matter 8, 7446 (2012).
[16] L. Rayleigh, Proc. Math. Soc.(London) 4, 363 (1873).
[17] M. Doi, Journal of Physics: Condensed Matter 23,
284118 (2011).
[18] L. J. Jones, R. Carballido-L´opez, and J. Errington, Cell
104, 913 (2001).
[19] R. M. Figge, A. V. Divakaruni, and J. W. Gober, Molec-
ular Microbiology 51, 1321 (2004).
[20] H. P. Erickson, D. W. Taylor, K. A. Taylor,
and
D. Bramhill, Proceedings of the National Academy of
Sciences 93, 519 (1996).
[21] N. Ausmees, J. Kuhn, and C. Jacobs-Wagner, Cell 115,
705 (2003).
[22] P.-G. De Gennes, F. Brochard-Wyart, and D. Qu´er´e,
drops, bubbles,
Capillarity and wetting phenomena:
pearls, waves (Springer, 2004).
[23] A. F. Bower, Applied mechanics of solids (CRC press,
2011).
[24] E. Kuru, H. Hughes, P. J. Brown, E. Hall, S. Tekkam,
F. Cava, M. A. de Pedro, Y. V. Brun, and M. S. Van-
Nieuwenhze, Angewandte Chemie International Edition
51, 12519 (2012).
[25] J.-V. Holtje, Microbiology and Molecular Biology Re-
views 62, 181 (1998).
[26] J. J. Thwaites and N. H. Mendelson, Advances in Micro-
bial Physiology 32, 173 (1991).
10
[27] M. Wachi, M. Doi, S. Tamaki, W. Park, S. Nakajima-
Iijima, and M. Matsuhashi, Journal of Bacteriology 169,
4935 (1987).
[28] N. Iwai, K. Nagai, and M. Wachi, Bioscience, Biotech-
nology, and Biochemistry 66, 2658 (2002).
[29] A. Amir, F. Babaeipour, D. B. McIntosh, D. R. Nelson,
and S. Jun, Proceedings of the National Academy of Sci-
ences 111, 5778 (2014).
[30] B. Volkmer and M. Heinemann, PloS one 6, e23126
(2011).
[31] M. E. Sharpe, P. M. Hauser, R. G. Sharpe, and J. Erring-
ton, Journal of bacteriology 180, 547 (1998).
[32] Y. Deng, M. Sun, and J. W. Shaevitz, Physical Review
Letters 107, 158101 (2011).
[33] G. Lan, C. W. Wolgemuth, and S. X. Sun, Proceedings
of the National Academy of Sciences 104, 16110 (2007).
[34] N. Nanninga, Canadian Journal of Microbiology 34, 381
(1988).
[35] S. Taheri-Araghi, S. Bradde, J. T. Sauls, N. S. Hill, P. A.
Levin, J. Paulsson, M. Vergassola, and S. Jun, Current
Biology 25, 385 (2015).
[36] S. Wang, H. Arellano-Santoyo, P. A. Combs, and J. W.
Shaevitz, Proceedings of the National Academy of Sci-
ences 107, 9182 (2010).
[37] J. Mannik, R. Driessen, P. Galajda, J. E. Keymer, and
C. Dekker, Proceedings of the National Academy of Sci-
ences 106, 14861 (2009).
[38] J. F. Allard and E. N. Cytrynbaum, Proceedings of the
National Academy of Sciences 106, 145 (2009).
[39] J. Mannik, F. Wu, F. J. Hol, P. Bisicchia, D. J. Sher-
ratt, J. E. Keymer, and C. Dekker, Proceedings of the
National Academy of Sciences 109, 6957 (2012).
[40] M. T. Cabeen, G. Charbon, W. Vollmer, P. Born, N. Aus-
mees, D. B. Weibel, and C. Jacobs-Wagner, The EMBO
Journal 28, 1208 (2009).
[41] R. Mukhopadhyay and N. S. Wingreen, Physical Review
E 80, 062901 (2009).
[42] O. Sliusarenko, M. T. Cabeen, C. W. Wolgemuth,
C. Jacobs-Wagner, and T. Emonet, Proceedings of the
National Academy of Sciences 107, 10086 (2010).
[43] H. P. Erickson, D. E. Anderson, and M. Osawa, Micro-
biology and Molecular Biology Reviews 74, 504 (2010).
[44] S. J. Holden, T. Pengo, K. L. Meibom, C. F. Fernandez,
J. Collier, and S. Manley, Proceedings of the National
Academy of Sciences , 201313368 (2014).
[45] A. Typas, M. Banzhaf, C. A. Gross, and W. Vollmer,
Nature Reviews Microbiology 10, 123 (2011).
[46] R. A. Daniel, E. J. Harry, and J. Errington, Molecular
Microbiology 35, 299 (2000).
[47] A. V. Divakaruni, C. Baida, C. L. White, and J. W.
Gober, Molecular Microbiology 66, 174 (2007).
[48] H. Turlier, B. Audoly, J. Prost, and J.-F. Joanny, Bio-
physical Journal 106, 114 (2014).
[49] Y.-J. Eun, M. Zhou, D. Kiekebusch, S. Schlimpert, R. R.
Trivedi, S. Bakshi, Z. Zhong, T. A. Wahlig, M. Thanbich-
ler, and D. B. Weibel, Journal of the American Chemical
Society 135, 9768 (2013).
[50] G. Lan, B. R. Daniels, T. M. Dobrowsky, D. Wirtz, and
S. X. Sun, Proceedings of the National Academy of Sci-
ences 106, 121 (2009).
[51] M. Campos, I. V. Surovtsev, S. Kato, A. Paintdakhi,
B. Beltran, S. E. Ebmeier, and C. Jacobs-Wagner, Cell
159, 1433 (2014).
|
1012.4624 | 1 | 1012 | 2010-12-21T12:09:53 | Enhanced diffusion due to active swimmers at a solid surface | [
"physics.bio-ph",
"cond-mat.soft"
] | We consider two systems of active swimmers moving close to a solid surface, one being a living population of wild-type \textit{E. coli} and the other being an assembly of self-propelled Au-Pt rods. In both situations, we have identified two different types of motion at the surface and evaluated the fraction of the population that displayed ballistic trajectories (active swimmers) with respect to those showing random-like behavior. We studied the effect of this complex swimming activity on the diffusivity of passive tracers also present at the surface. We found that the tracer diffusivity is enhanced with respect to standard Brownian motion and increases linearly with the activity of the fluid, defined as the product of the fraction of active swimmers and their mean velocity. This result can be understood in terms of series of elementary encounters between the active swimmers and the tracers. | physics.bio-ph | physics |
Enhanced diffusion due to active swimmers at a solid surface
Gast´on Mino1, Thomas E. Mallouk2, Thierry Darnige1, Mauricio Hoyos1, Jeremy Dauchet1,
Jocelyn Dunstan3, Rodrigo Soto3, Yang Wang2 , Annie Rousselet1, and Eric Clement1
1 PMMH-ESPCI, UMR 7636 CNRS-ESPCI-Univ.Paris 6 and Paris 7,
2 Department of Chemistry,
10 rue Vauquelin, 75005 Paris, France.
The Pennsylvania State University, USA. 3 Departmento de F´ısica, FCFM, Univ. de Chile, Chile.
(Dated: May 28, 2018)
We consider two systems of active swimmers moving close to a solid surface, one being a living
population of wild-type E. coli and the other being an assembly of self-propelled Au-Pt rods. In
both situations, we have identified two different types of motion at the surface and evaluated the
fraction of the population that displayed ballistic trajectories (active swimmers) with respect to
those showing random-like behavior. We studied the effect of this complex swimming activity on
the diffusivity of passive tracers also present at the surface. We found that the tracer diffusivity is
enhanced with respect to standard Brownian motion and increases linearly with the activity of the
fluid, defined as the product of the fraction of active swimmers and their mean velocity. This result
can be understood in terms of series of elementary encounters between the active swimmers and the
tracers.
PACS numbers: 87.16.-b, 05.65.+b
Since the pioneering work of Wu and Libchaber [1]
considerable efforts have been made to understand hy-
drodynamic properties of active suspensions. Generally
speaking, this is the name borne by fluids laden with self-
swimming entities such as bacteria [2–5], algae [6, 7] or
collections of active artificial swimmers [8]. Assemblies of
microscopic motors dispersed in a fluid display emergent
properties that differ strongly from passive suspensions.
The momentum and energy transfer balances as well the
constitutive transport properties are deeply modified by
the momentum sources distributed in the bulk [2, 9].
Some of these anomalous properties have already been
identified such as active diffusivity [1, 2], anomalous vis-
cous response [7, 10], active transport and mixing [11]
as well as the possibility to use fluctuations to extract
work [12]. The presence of living and apparently gregari-
ous entities also offers the possibility to move collectively
and organize at the mesoscopic or macroscopic level in
the form of flocks and herds [9, 14]. Similar collective
effects were also identified in suspensions of self-propeled
inorganic particles [15]. In the bulk, swimming bacteria
with flagella such as E. coli create in the far field-limit,
a force-dipole velocity field and consequently, experience
a hydrodynamic attraction toward surfaces [16]. Then,
it has been observed that E. coli smooth out their run-
and-tumble movement and spend long times parallel to
the surface undergoing circular motion as a consequence
of the torque-free condition [17, 18]. When the concen-
tration becomes large, the E. coli population eventually
associates collectively to form a bio-film. Even in the
low concentration limit, the quantitative analysis of the
near surface motion increase tremendously in complex-
ity, a reason being the close field hydrodynamic forces
that become prevalent and require a complex treatment
of the lubrication hydrodynamic fields. However, even
in this frame of description, it remains unclear whether
the motion close to the surface is hydrodynamically sta-
ble and if the presence of thermal noise is essential to
account for the bacterium dwelling time at a surface[19].
Beyond the hydrodynamic interactions, more complex in-
gredients may come into play such has the surface inter-
action potentials (electrotastic or van der Waals) [20] or
more refined details of the bacterium physiology such as
swimming speed variations and desynchronization dur-
ing bacteria cell cycle [21, 22]. From the perspective of
providing a fully consistent treatment of active hydrody-
namics, with important applications for understanding
bacterial transfer in biological micro-vessels, microfluidic
devices, or the formation of bio-films, a reliable descrip-
tion of fluid activity in the vicinity of a solid surface is
strongly needed. In this letter, to tackle this open and
timely question, we compared the behavior of two kinds
of active micrometric swimmers: wild type E. coli K12
and artificial self-propelling rods [8], with completely dif-
ferent propulsion mechanisms. In both cases, we moni-
tor the swimmers' motions and their ability to activate,
beyond Brownian motion, passive tracers, hence charac-
terizing the active momentum transfer to the fluid.
Following the experimental procedure described in Ref.
[16], wild type E. coli K12 were grown overnight in rich
medium (LB). After washing, they were transferred into
MMAP, a motility medium supplemented with K-acetate
(0.34 mM) and polyvinyl pyrolidone (PVP: 0.005%).
They were incubated for at least an hour in that medium
and, in some cases, so-called "baby cells" were selected by
centrifugation and resuspended in MMAP. To avoid bac-
terial sedimentation (isodense conditions), Percoll was
mixed with MMAP (1vol/1vol). We checked that under
these conditions, the suspending fluid was still Newtonian
(viscosity η=1.28×10−3P a s at 22◦C). The overall con-
centration of bacteria was controlled such as to prepare
suspensions between 109 and 1010 bact/ml. To study the
2
FIG. 1. Identification of the swimmer populations by tracking "active" swimmers (red tracks) and "random" swimmers (blue
tracks), φA is the corresponding fraction of active swimmers. Figs. 1(a,b,g) correspond to E. coli (see inset in 1c) and Figs.
1(d,e) to Au-Pt rods (see inset in 1f). The round black circles in (a,b) are 2µm latex beads; the white small circles in (d,e) are
1µm Dynal beads. Figs. (c) and (f) display the relation between the active swimmers mean velocity VA and φA. On Fig. 1(c),
6 independent experiments with E. coli: 1N cells (brown, red and green), mixture of 1N and 2N cells (black) and 2N cells (pink,
blue). Labels (1) and (2) are for N = 100 and N = 200 cells in the observation field respectively. Fig. 1(f), for Au-Pt rods with
varying proportions of inactive Au rods (2 independent experiments). Fig. 1(g), density probability of the observed bacterial
tracks in the (Nc,(cid:104)θ(cid:105)) space. The colormap goes from blue for vanishing probability to red for high probability. Two clusters
are identified, centered at (0.9, 0.3) and (0.17, π/2), corresponding respectively to the "active" and "random" swimmers.
effect of bacterial activity on the diffusivity of passive
tracers, latex beads of 1 or 2 µm diameter (Beckman-
Coulter, density ρ=1.027 g/ml) were added to the sus-
pensions. Experiments were performed in 110 µm thick
chambers, built with two horizontal microscope cover
slips separated by a glass spacer. To avoid sticking, cover
slips were coated with PVP. The biological sample con-
sisted in a drop of liquid (20 µl) placed between the two
slides. The suspension was observed under an inverted
microscope (Zeiss-Obzerver, Z1-magnification 40X) con-
nected to a digital camera. The observation field ∆V was
96×128µm2 and 5 µm in depth. In a first series of experi-
ments, we measured the bacterial density profile through
the entire height of the chamber. We obtained profiles
similar to the ones published by Berke and coworkers [16],
namely a flat density in the bulk and a strong density in-
crease near the surfaces. However, the wild-type E. coli
we used was significantly less attracted by the surfaces
(2.5 times increase in density within 10 µm of the sur-
face) than a mutant E. coli strain that does not display
tumbling motion [17, 18]. Another series of experiments
was performed with bimetallic Au-Pt self-propelled rods
(length 1.2 µm and diameter 0.4 µm) that are very similar
in size to the E. coli cell body (1 to 2 µm long, 0.8 µm di-
ameter) but have no flagella (15 µm long for E. coli) (see
inserts in Figs. 1c and 1f). They also have a much higher
density (ρ=17 g/ml). In the presence of 1 to 10% hydro-
gen peroxide, these particles are propelled in the axial
direction towards the platinum end by the catalytic de-
composition of the peroxide fuel [8]. Recent experiments
and simulations are consistent with self-electrophoresis as
the dominant propulsion mechanism [23, 24]. Here, the
mode of propulsion is intrinsically different from the flag-
ellar one. The experiments were conducted in a similar
fashion to those involving E. coli, but in an open cham-
ber (without the upper wall), in order to allow the oxygen
bubbles produced in the reaction to escape from the cell.
The concentration of H2O2 was varied, as well as the
concentration of active rods (n=3-20×106 rods/ml). We
also used passive tracers (1µm diameter beads, Dynal-
My-one, density ρ = 1.8 g/ml) or 2 µm diameter latex
beads (Beckman-Coulter, density ρ=1.02 7g/ml)) to fol-
low the activation of the fluid by the Au-Pt rods. In all
cases, all the particles in the suspension were localized at
the bottom of the chamber due to sedimentation. Short
videos (20s duration at 20f ps) were used to track both
bacterial and self-propelled rods motion.
In the following, we only focus on bacteria and rods
moving close to the surface (less than 5µm).
In both
cases, we observed that not all swimmers display similar
trajectories. This was expected as for wild type bacteria,
the run or tumble dynamics may depend strongly on the
microenvironment or on the position in the cell cycle. For
Au-Pt rods, this is also consistent with previous observa-
tions that even within a single batch, electrochemically
grown rods have a range of catalytic activity[8]. We de-
veloped a tracking program to analyze the short videos
and obtained tracks for each swimmer present in the field.
We identified two major types of motion: a ballistic and a
random one (see Fig.1(a,b,d,e)), and the swimmers that
follow these motions are called "active" swimmers and
"random" swimmers, respectively. To discriminate sys-
tematically all the tracks, two parameters were defined.
The first parameter (cid:104)θ(cid:105) is the mean angle between two
successive steps. For example, (cid:104)θ(cid:105) = 0 for straight tra-
jectories and (cid:104)θ(cid:105) = π/2 for a purely random walk. The
second discriminating parameter is based on the minimal
circle diameter L that encompasses a given trajectory of
duration T . For an acquisition time δt (1/20s) and a
mean step size δr, the number Nc = Lδt
T δr is computed.
When Nc is close to 1, the trajectory is associated with a
straight line, whereas when Nc is small, its value points
to diffusive motion. Therefore, each track is associated
with these two numbers and in the (Nc,(cid:104)θ(cid:105)) parameter-
space, we could identify two clusters that clearly differ-
entiated the active and random swimmers (see Fig.1(g)).
Nevertheless, for very small or interrupted trajectories,
the separation procedure remained ambiguous, so we sys-
tematically discarded tracks shorter than 10 steps.
In
the case of artificial swimmers, we also managed to con-
trol a priori the fraction of active swimmers by adding
inactive rods (made only of gold) and keeping the to-
tal number of rods at the surface constant. According to
the trajectory classification, a fraction φA of active swim-
mers was determined. Thus, for a mean number (cid:104)N(cid:105) of
swimmers, identified in the field of vision, we define a
density of "active swimmers" as nA = φA(cid:104)N(cid:105)/∆V . In
Fig. 1, we display tracks during a time lag τ = 1.5 s, for
two populations of swimmers (a-b, bacteria and d-e, Au-
Pt rods), having different φA (a,d small active-fraction
and b,e high active fraction). In Figs.1(c) and 1(f), we
present the mean velocities of active swimmers VA as
a function of their fraction φA, for different experiments
with bacteria and active rods. In the case of the bacterial
suspension, we also tried several synchronization proto-
cols to select bacteria, at different position in the cell
cycle, showing different swimming characteristics. We
were able to produce "baby-bacteria" populations (1N
short cells: 1.12 µm long) which were found to have a φA
larger than the more mature bacteria populations (2N
long cells: 2.5 µm long). We took advantage of this dif-
ference to look at the influence of φA on VA in bacterial
suspensions. On Fig.1(c) we display 6 independent ex-
periments performed with populations having a majority
of 1N cells (brown,red and green), a mixture of 1N and
2N cells (black), a majority of 2N cells (pink, blue). For
each sample, φA was taken from suspensions showing an
average of 100 or 200 bacteria in the observation field
(represented by (1) and (2) on Fig. 1c, respectively). φA
shows little influence on VA but VA could be different
according to bacteria position within the cell cycle. In
the case of the Au-Pt rods, Fig.1(f) shows a stronger in-
dependence of VA on φA, which is changed by varying
the proportion of inactive Au-rods. This low dependence
3
FIG. 2. Enhanced diffusivity DP of passive tracers measured
as a function of JA ("activity flux"). Squares and circles sym-
bols represent tracer of 2µm and 1µm diameters, repsectively.
Fig.2(a) corresponds to the bacterial suspensions; 1N cells
(red and green) and unsynchronized (black, blue and pink).
Each color defines an experiment performed over a range of
bacterial dilution. Fig.2(b) corresponds to suspensions of Au-
Pt-rods: Mixture of active and inactive Au-rods (red circles,
brown squares), and various H2O2 concentration from 2,5% to
20%, (green, blue, black circles). The dashed lines are linear
fits and the error bars, standard deviations.
on φA is due to the low swimmer concentration of the
suspension, where no collective behavior is observed.
In the following, we will relate the motion of the pas-
sive tracers to the number of active swimmers and their
mean velocity. Passive tracer trajectories were analyzed
(about 10 tracks in 300-image video at 1f ps for bacteria
and 40s sequences at 8f ps for Au-Pt rods). No signif-
icant stickiness between the spheres and the swimmers
was observed and those rare cases were eliminated from
the analysis. From these tracks, the mean passive tracer
diffusion coefficient DP was extracted consistently using
two independent methods. The first one applied mean
square displacements at long times (diffusive regime [1])
of individual particles; the second one used particles as
pairs in order to eliminate residual drift.
In Fig.2 the
passive tracer diffusivities are displayed for all the exper-
iments presented in Fig.1(c,f). DP values are displayed
as a function of JA = nAVA, that we call the "activity
flux". For experiments performed with the same tracer
size, we observe a collapse of all data onto a linear curve:
DP = DB
P + βJA
(1)
where DB
P is the Brownian diffusivity of the latex par-
ticles in the vicinity of the surface, in the absence of
swimmers. Note that due to lubrication forces, this value
is smaller than the Brownian diffusivity expected in the
bulk (DB = kBT /3πηd) and, for the parallel motion,
they are related by DB
P = αDB, where α < 1 is the par-
allel drag correction factor [25–27]. The α factor depends
on the bead distance to the surface, vanishing at con-
tact and going asymptotically to one at large distances.
The beads are not at a fixed distance to the surface, but
they are distributed according to the Boltzmann's fac-
tor exp(−m∗gz/kBT ), where m∗ is the buoyant mass.
Therefore α must be averaged with this factor. In the
active rods experiments we obtained α=0.7 both for the
the buoyant d=1 µm and d=2 µm latex spheres, value
that agrees with the theoretical prediction given above
(α=0.64 for d=1µm and α=0.74 d=2 µm).
In the ex-
periments with bacteria, the suspended beads are almost
isodense but they sediment anyway. The experimental
fit gives α = 0.85. This value allows us to infer the
density mismatch ∆ρ = 0.008 g/ml, which is consis-
tent with the experimental preparation. The collapse
holds also for bacterial populations at different matura-
tion stages (1N, 2N or unsynchronized mixtures). From
dimensional analysis of expression (1), it can be seen that
the prefactor β is a length to the fourth power. It varies
from 5µm4=(1.5µm)4 for bacteria, to 13µm4=(1.9µm)4
for active rods, but seems to be almost independent of
the passive tracer size. Close to the plates the hydrody-
namic perturbations created by the swimmers decay as
the inverse cube of the distance, faster than in the bulk
[28]. Therefore, at low concentrations, the enhanced dif-
fusivity in (1), proportional to nA and VA, can be un-
derstood as a result of a series of elementary encounters
between active swimmers and the tracers: the number of
encounters per unit time is proportional to nAVA. On
the other hand, low Reynolds dynamics points out that
the tracer displacement at each encounter is independent
of the swimmer velocity and depends only on geometrical
factors: the impact parameter, the swimmer dimensions
and weakly on the tracer size through the Fax´en correc-
tion of passive transport [29, 30]. The β factor comes out
from averaging the tracer's displacements but its compu-
tation is difficult because it requires a correct modeling
of the near field interactions between the swimmer and
the tracer, taking into account the detailled swimmer ge-
ometry and the effect of the surface.
In conclusion, we have characterized active momen-
tum transfer close to solid surface for two active sus-
pensions (wild-type bacteria and artificial self-propelled
swimmers). The effect was measured using the diffusion
enhancement of a passive tracer. In spite of the a pri-
ori complexity of the hydrodynamics and essential dif-
ferences in the propulsion modes, we demonstrated that
the effect emerges quantitatively in a similar way. The
resulting diffusion coefficient is the sum of the Brownian
contribution near the wall and an active part, propor-
tional to the product of the density of active swimmers
and their mean velocity at the surface. The proportional-
ity factor, scaling as the 4th power of a micron size length,
encompasses the details of momentum transfer for each
swimmer and is found to be weakly (if at all) sensitive
to the probe diameter. Importantly, discriminating be-
tween so-called "active" and "random" swimmer trajec-
tories was crucial for predicting the induced transport
phenomenon and we have developed a protocol to make
such a distinction. The functional dependence of the en-
hanced diffusivity is explained in terms of successive in-
teractions between a single swimmer and the tracer, each
4
one producing a net displacement. Our results justify to
pursue a quantitative determination of such encounters
based on simple hydrodynamic models [29].
We thank D. Grier for discussions on the tracking
programs, financial support from PGDG Foundation,
the Alfa-SCAT program, Sesame Ile-de-France, Fonde-
cyt Grants No. 1061112, No. 1100100, Anillo Grant No.
ACT127, ECOS C07E08, and NSF 0820404.
[1] X.-L. Wu and A. Libchaber, Phys. Rev. Lett. 84, 3017
(2000).
[2] D.T.N. Chen et al., Phys. Rev. Lett. 99, 148302 (2007).
[3] D. Saintillan and M.J. Shelley, Phys. Rev. Lett. 99,
058102 (2007); Phys. Rev. Lett. 100, 178103 (2008).
[4] Y. Hatwalne et al., Phys. Rev. Lett 92, 118101 (2004).
[5] C. Dombrowski et al., Phys. Rev. Lett. 93, 098103(2004).
[6] K.C. Leptos et al., Phys. Rev. Lett. 103, 198103 (2009).
[7] S. Rafaı, L. Jibuti, and P. Peyla, Phys. Rev. Lett. 104,
098102 (2010).
[8] W.F. Paxton et al., J. Am. Chem. Soc. 126, 13431
(2004).
[9] R.A. Simha and S. Ramaswany, Phys. Rev. Lett. 89,
058101(2002).
[10] A. Sokolov and I.S. Aranson, Phys. Rev. Lett. 103,
148101 (2009); B. Haines et al., Phys. Rev. E 80, 041922
(2009).
[11] N. Darnton et al., Biophys. J. 86, 1863 (2004).
[12] A. Sokolov et al., PNAS 107, 969 (2010) and refs inside.
[13] R. Di Leonardo et al., PNAS 107, 9541, (2010).
[14] G. Gregoire, H. Chate, and Y. Yu, Phys. Rev. E 64,
011902 (2001).
[15] M. Ibele, T.E. Mallouk, and A. Sen, Angew. Chem. Int.
Ed. 48, 3308 (2009).
[16] A.P. Berke et al., Phys. Rev. Lett. 101, 038102 (2008).
[17] E. Lauga et al., Biophys. J. 90, 400 (2006).
[18] P.D. Frymier et al., PNAS 92, 6195 (1995).
[19] G. Li, L.-K.Tam, and J.X. Tang, PNAS 105, 8359 (2008).
[20] M.A.S. Vigeant and R.M. Ford, Appl. Environ. Micro-
biol. 63, 3474 (1997).
[21] B.M. Pruβ and P. Matsumura, J. Bacteriol. 179, 5602
(1997).
[22] R. Allman, T. Schjerven, and E. Boyec, J. Bacteriol. 173,
7970 (1991)
[23] Y. Wang et al., Langmuir 22, 10451 (2006).
[24] J.L. Moran, P.M. Wheat, and J.D. Posner, Phys. Rev. E
81, 065302 (2010).
[25] H. Brenner, Chem. Eng. Sci. 16, 242 (1961); A.J. Gold-
man, R.G. Cox, and H. Brenner, Chem. Eng. Sci 22, 637
(1967).
[26] P. Holmqvist, J.K.G. Dhont, and P.R. Lang, Phys. Rev.
E 74, 021402 (2006).
[27] P. Huang and K.S. Breuer, Phys. Rev. E 76, 046307
(2007).
[28] J. R. Blake and A. T. Chwang, Fundamental singularities
of viscous flow, J. Eng. Math. 8, 23 (1974).
[29] J. Dunstan, M.Sc thesis, Universidad de Chile, 2010; J.
Dunstan et al. (in preparation).
[30] J. Happel and H. Brenner, Low Reynolds Number Hydro-
dynamics: with special applications to particulate media,
(Kluwer, 2009).
|
1908.05056 | 1 | 1908 | 2019-08-14T10:15:01 | Thermodynamics of the S2-to-S3 State Transition of the Oxygen-Evolving Complex of Photosystem II | [
"physics.bio-ph",
"physics.chem-ph"
] | The room temperature pump-probe X-ray free electron laser (XFEL) measurements used for serial femtosecond crystallography provide remarkable information about the structures of the catalytic (S-state) intermediates of the oxygen-evolution reaction of photosystem II. However, mixed populations of these intermediates and moderate resolution limit the interpretation of the data from current experiments. The S3 XFEL structures show extra density near the OEC that may correspond to a water/hydroxide molecule. However, in the latest structure, this additional oxygen is 2.08 {\AA} from the Oe2 of D1-E189, which is closer than the sum of the van der Waals radii of the two oxygens. Here, we use Boltzmann statistics and Monte Carlo sampling to provide a model for the S2-to-S3 state transition, allowing structural changes and the insertion of an additional water/hydroxide. Based on our model, water/hydroxide addition to the oxygen-evolving complex (OEC) is not thermodynamically favorable in the S2 g = 2 state, but it is in the S2 g = 4.1 redox isomer. Thus, formation of the S3 state starts by a transition from the S2 g = 2 to the S2 g = 4.1 structure. Then, electrostatic interactions support protonation of D1-H190 and deprotonation of the Ca2+-ligated water (W3) with proton loss to the lumen. The W3 hydroxide moves toward Mn4, completing the coordination shell of Mn4 and moving with its oxidation to Mn(IV) in the S3 state. In addition, binding additional hydroxide to Mn1 leads to a conformational change of D1-E189 in the S2 g = 4.1 and S3 structures. In the S3 state in the population of protonated D1-E189 increases. | physics.bio-ph | physics | Thermodynamics of the S2-to-S3 State Transition of the Oxygen-Evolving Complex of Photosystem II
Muhamed Amin1,2,3*, Divya Kaur4,5, Ke R. Yang6, Jimin Wang7, Zainab Mohamed8, Gary W. Brudvig6, M.R. Gunner4,5 and Victor
Batista6
1 Center for Free-Electron Laser Science, Deutsches Elektronen-Synchrotron DESY, Notkestrasse 85, 22607 Hamburg, Germany
2 Physics Department, University College Groningen, University of Groningen, Groningen, Netherlands
3 Centre for Theoretical Physics, The British University in Egypt, Sherouk City 11837, Cairo, Egypt
4Department of Physics, City College of the City University of New York, New York, NY 10031, United States
5Ph.D. Program in Chemistry, The Graduate Center of the City University of New York, New York, NY 10016, United States
6 Department of Chemistry, Yale University, New Haven, Connecticut 06520-8107, United States
7 Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT08620-8114
8 Zewail City of Science and Technology, Sheikh Zayed, District, 6th of October City, 12588 Giza, Egypt
[email protected]
Abstract: The room temperature pump-probe X-ray free electron laser (XFEL) measurements used for serial
femtosecond crystallography provide remarkable information about the structures of the catalytic (S-state) intermediates of
the oxygen-evolution reaction of photosystem II. However, mixed populations of these intermediates and moderate
resolution limit the interpretation of the data from current experiments. The S3 XFEL structures show extra density near the
OEC that may correspond to a water/hydroxide molecule. However, in the latest structure, this additional oxygen is 2.08 Å
from the Oe2 of D1-E189, which is closer than the sum of the van der Waals radii of the two oxygens. Here, we use
Boltzmann statistics and Monte Carlo sampling to provide a model for the S2-to-S3 state transition, allowing structural
changes and the insertion of an additional water/hydroxide. Based on our model, water/hydroxide addition to the oxygen-
evolving complex (OEC) is not thermodynamically favorable in the S2 g = 2 state, but it is in the S2 g = 4.1 redox isomer.
Thus, formation of the S3 state starts by a transition from the S2 g = 2 to the S2 g = 4.1 structure. Then, electrostatic
interactions support protonation of D1-H190 and deprotonation of the Ca2+-ligated water (W3) with proton loss to the
lumen. The W3 hydroxide moves toward Mn4, completing the coordination shell of Mn4 and moving with its oxidation to
Mn(IV) in the S3 state. In addition, binding additional hydroxide to Mn1 leads to a conformational change of D1-E189 in
the S2 g = 4.1 and S3 structures. In the S3 state in the population of protonated D1-E189 increases.
Introduction
states.5 -- 10
The oxygen-evolving complex
(OEC) of
photosystem II (PSII) catalyzes the oxidation of water to
O2. With input of light, the OEC of PSII is sequentially
oxidized. As the system is oxidized protons are lost.
There are five OEC S-state intermediates along the
catalytic cycle designated S0 to S4, with S0 the most
reduced.1 -- 4 The fully oxidized S4 state catalyzes the
oxidation of two waters to O2. For understanding the
reaction mechanism, it is crucial to obtain structural
information about these catalytic intermediates and
determine the pathway for the transitions between
The OEC core contains a cluster of four Mn ions
and a Ca2+ connected through bridging oxides (i.e.,
deprotonated water molecules).11 In the dark-stable S1
state, the Mn oxidation states are (III2, IV2) with the
subscript indicating the number of Mn ions in a given
oxidation state.12,13 In the S1-to-S2 state transition, the
OEC loses an electron, oxidizing an Mn(III) to Mn(IV).
EPR measurements14 -- 17 and theoretical models9,18 show
two energetically accessible redox isomers of the S2 state.
When the dangler Mn4 (Figure 1) is oxidized, the Mn
cluster has a total spin of S = 1/2 and produces the g = 2
1
multiline EPR signal (S2,g=2). The second S2 state redox
isomer is formed when Mn1 (Figure 1) is oxidized and the
total spin is now S = 5/2, which results in the S2 state g =
4.1 EPR signal (S2,g=4.1).8,9,15,17,19,20 The S1-to-S2 state
transition is accompanied by little or no proton loss to the
lumen21 -- 23 and EXAFS measurements24,25 show no major
structural rearrangements. In contrast, the S2-to-S3 state
transition induces loss of approximately one proton/OEC
and EXAFS spectra of the S2 and S3 states show
significant structural changes.26,27
facilitating
Theoretical studies, recently confirmed by time-
resolved XFEL structures,28 show the insertion of an
additional oxo/hydroxide in the S3 state. This additional
oxygen is required to complete the coordination shell of
the only Mn(III) left in the S2 state, which has only 5
ligands,
its oxidation. Time-resolved
photothermal beam deflection measurements suggested
that this oxidation is preceded by a proton release from
the OEC.27,29 However, it is an open question which
protonatable group (amino acids/water) are releasing a
proton in this transition. Although the XFEL structures
are not affected by damage due to exposure to X-ray
radiation, there remains disagreement about the location
of this additional oxygen reported by the different XFEL
studies.
inhomogeneous
populations of the states of the different crystals and the
time point used to probe the structures.28,30 In addition,
very high resolution is required to clearly establish the
location of this oxygen near the high electron densities of
the Mn ions.31
This
results
from
the
In the calculations reported here the position of
the inserted oxygen is not fixed at the beginning of the
calculation. Rather Monte Carlo sampling is used to
generate a Boltzmann distribution of the possible binding
sites for the oxygens of water molecules or hydroxyls, and
their proton conformers in the S2 g = 2, S2 g = 4.1, and S3
states and in the intermediate state where YZ is oxidized
and its conjugate base D1-H190 is protonated (S2-
YZ
• is the electron acceptor coupled to
OEC oxidation. The use of a classical, electrostatic model
of the OEC enables sampling of the water/hydroxide
binding sites,
their protonation states and proton
positions, the protonation of the amino acids and the
oxidation states of the Mn centers. This method has been
tested against model compounds and was used to study
the proton coupled electron transfer in the Kok cycle
•/H190+).25,32 -- 34 YZ
2
previously.35
Computational Methods
(408)-G409-E413-(414); D2
The structural model that includes amino acid
residues and cofactors of PSII within a sphere with an ≈15
Å radius centered at the OEC cluster, two chloride ions
and 85 water molecules, is optimized by QM/MM in both
the g = 2 and g = 4.1 S2 states.7 The geometry optimization
is carried out using the ONIOM36 method within the
Gaussian 0937 software. The amino acid residues that are
included in the 15 Å sphere are: D1 (chain A): (57)-V58-
V67-(68), (81)-V82-L91-(92), (107)-N108-Y112-(113),
(155)-A156-I192-(193), (289)-I290-N298-(299), (323)-
A324-A344/C-terminus; CP43 (chain C):(290)-W291-
(292), (305)-G306-A314-(315), (334)-T335-L337-(338),
(341)-M342-(343), (350)-F351-F358-(359), (398)-A399-
G402-(403),
(chain
D):(311)-E312-L321-(322),
(347)-R348-L352/C-
terminus. Only backbone atoms are considered for the
capped residues highlighted in parentheses. The QM
region includes the Mn4O5Ca cluster and the amino acids
ligands D1-D170, D1-E189, D1-H332, D1-E333, D1-
D342, C-terminus of D1-A344, CP43-E354, D1-H337,
CP43-R357, D1-D61 along with ten water molecules. The
DFT/B97D38 -- 40 level of theory that includes dispersion
correction is used for the QM region while the
AMBER41,42 force field is used for the MM region. The
LanL2DZ43,44 basis set is used for calcium and manganese
metal centers while 6-31G**45,46 is used for hydrogen,
carbon, nitrogen and oxygen atoms. The capping
residues, chloride ions and oxygen atoms of water
molecules are frozen at the edge of the model in the MM
region while all other atoms in the QM region are free to
move.
Monte Carlo
(MC) sampling. All water
molecules except those that are ligands to Mn4 or Ca2+
were removed from the QM/MM optimized structure.
Then, oxygens were placed into all the cavities within 4
Å of the Mn4O5Ca cluster on a 1.0 Å grid using the
program IPECE47 (Figure 1A). In total, 451 oxygens were
added in the cavity around the OEC. Water and hydroxide
proton position conformers were generated for each
oxygen atom independently (7569 hydrogen conformers
were added). In addition, there is one conformer per
oxygen species that represents this group moving out of
the protein into solvent. Thus, each of the 451 oxygen
species on the grid can be either water or hydroxide or
moved to solvent in the microstates that will be subjected
to MC sampling.
(Multi
MCCE
Conformer
Continuum
Electrostatics)48 was used to generate rotamers for the
side chains of the amino acids. The electrostatic energies
of the allowed conformational space are then calculated
using Adaptive Poisson-Boltzmann Solver (APBS)49 by
solving the Poisson-Boltzmann equation. Based on the
calculated energies, the Metropolis Monte Carlo method
is used to generate the Grand Canonical Boltzmann
distribution for all conformers as based on their
electrostatic and van der Waals interactions at pH 7. The
dielectric constant of the proteins is set to 4, and it is 80
for the solution around the sphere. The parameters for the
OEC and ligands used here as reported previously in
Amin et al.3
where me is the mass of the electron and h is the Planck's
constant, and E0 = 6540.0 eV is the Fermi energy of Mn.
A fractional cosine-square (Hanning) window with dk =
1 was applied to the k3-weighted EXAFS data. The grid
of k points, which are equally spaced at 0.05 Å-1, was then
used for the Fourier transformation (FT) to R space. A k
range of 4.0 -- 10.5 Å-1 was employed for the FT of the
isotropic EXAFS amplitudes. In the calculation of
EXAFS for different S3 state models, we allow both the
simulated intensity of the EXAFS signal and the shift in
the k-space (edge shift) to relax and find the best fit using
a least squares fitting procedure as done previously.10
Results and Discussions
A)
Following Grand Canonical Monte Carlo
(GCMC) sampling of microstates, the structure with
occupied water/hydroxyl binding sites was optimized at
the DFT level using the 6-31G* basis set and the B3LYP
functional. All the Mn(IV) ions were defined in the high
spin state. The model included the Mn4O5Ca cluster and
the amino acids ligands (D170, E189, H332, E333, D342,
A344, CP43-E354). All amino acid residues are in the D1
protein unless otherwise noted. In addition, several other
residues that interact closely with the OEC were added
(D61, Y161, H190, H337, CP43-R357, G171, as well as
11 crystallographic waters including water ligands of the
Mn and Ca2+ centers. The pKa's of the bridging oxides
were calculated in the optimized DFT model, using
MCCE as described previously.3,18
B)
EXAFS simulations. EXAFS spectra of the S3
models were computed with the FEFF program (version
6)50 combined with the IFEFFIT code (version1.2.12).51
We included all paths with lengths up to eight scattering
legs and a Debye-Waller factor of 0.003 Å for all
calculations. The energy (E) axis was converted into the
space by usual
photoelectron wave vector
(k)
transformation, 𝑘=#2𝑚&
'ℎ2𝜋*+(𝐸−𝐸/)12+,
3
Figure 1. A) The cavity around the Mn4O5Ca cluster is
filled with 451 oxygens that represent water molecules or
hydroxyls placed on grids with 1 Å spacing. Mn are
shown in blue. B) O6 is the binding site selected by MC
sampling with the classical force field.
Water molecule-binding in the S2-to-S3 state
transition. The cavity around the Mn cluster in both the g
= 2 and g = 4.1 S2 state structures was filled with oxygen
+∆∆𝐺KHJLMJA&
∆𝐺78(999:9;)=<𝐹>𝐸?−𝐸@,ABCDE+∆∆𝐺ABCGHIJB8
where ∆𝐺78(999:9;) is the energy required to oxidize an
al.,35 ∆∆𝐺ABCGHIJB8 is the desolvation penalty of Mn4 and
∆∆𝐺KHJLMJA& is the pairwise electrostatic interaction
isolated Mn center from Mn(III) to Mn(IV), F is the
Faraday constant, Em,sol is the mid-point reduction
potential of Mn(IV) in solution as obtained in Amin et
between Mn and the surrounding residues, waters, Ca2+
and bridging oxygen atoms.
In the S2 g = 4.1 structure, Mn4 is oxidized at 0.6
V and the oxidation is coupled to binding O6, while in the
S2 g = 2 structure Mn1 has a potential for oxidation of 1.4
V, which is higher than the potential for oxidation of
P680. This suggests the advancement of the OEC to the
S3 state through the S2 g = 2 state is energetically
unfavorable, which agrees with earlier studies.8,19,32,53,55 -- 57
Thus, for the OEC to advance from the S2 g = 2 state to
the S3 state, it has to transit through the S2 g = 4.1 state.58 --
60 Experimental studies using EPR spectroscopy58,59 also
indicates the formation of the S3 state from the S2 state g
= 4.1 isomer at lower temperatures in both Ca-PSII and
Sr-PSII. Combined multiscale ab
initio DFT+U
simulations19 suggested that the oxidation of YZ stabilizes
the conversion of the open form (g = 2) to closed (g = 4.1)
S2 state isomer prior to formation of the S3 state.
The analysis thus far has used a classical force
field to extensively sample many oxygen and proton
positions of water molecules. The S3 structure with the
additional hydroxide in the O6 position, obtained by MC
sampling with only electrostatic and van der Waals
interactions, was optimized by using DFT. The
optimization adjusted the position of O6 to complete the
octahedral coordination shell of Mn4 (Figure 1B),
reducing the Mn4-O6 bond to 1.91 Å, while the Ca2+-O6
and O6-O5 distances increased to 2.46 Å and 2.74 Å,
respectively. Upon optimization, an additional hydrogen
bond was formed between the hydroxide O6 and µ-oxo
O5 with a distance of 2.30 Å.
There are two structural models of the S3 state
proposed in the literature, with "open" or "closed"
structures of the Mn4O5Ca cubane that may be relevant
for the catalytic reaction.10,61 -- 64 In the closed cubane
structure, O6-H is placed 3.57 Å away from Ca2+ and does
not form a hydrogen bond with O5. In the open structure,
atoms of water molecules to examine the possible binding
sites for an additional water/hydroxide ligand to the
Mn4O5Ca cluster (Figure 1A). Monte Carlo sampling was
used to generate a Boltzmann distribution of conformers,
as defined by electrostatic and van der Waals interactions.
In the S2 g = 2 structure, no additional binding sites are
identified nor in the S2 g = 2/YZ
•H190+ state.
However, the Mn4-Ca2+ distance is longer by
0.27 Å in the S2 g = 4.1 structure and a hydroxide (O6
Figure 1B) that bridges Mn4 and Ca2+ is occupied with a
25% probability, while the probability of the W3 water
ligand to Ca2+ being present is reduced from 100% to
75%. This indicates a strong coupling so that clusters will
have one or the other oxygen but never both. The transfer
of W3 to the hydroxide, O6 position, is coupled to the loss
of a proton. The O6 oxygen is located 2.06 Å from Mn4,
2.22 Å from Ca2+ and 1.78 Å from O5. In the S2
g = 4.1/YZ
•H190+ state, now O6 is always present and W3
is never found in accepted microstates, which suggests a
proton loss upon the formation of YZ
•H190+, before
oxidation of the Mn cluster. This sequence of events has
been observed by time-resolved photothermal beam
deflection measurements.27 The total charge of the protein
upon the formation of YZ
•H190+ is reduced by one unit,
which suggests that a proton is released to the lumen.
translates
to complete
Cluster based DFT calculations54 also suggested
that the binding of an additional water molecule takes
place at Mn1 in the g = 2 form of the S2 state, a transition
that is at high energy here. A similar mechanism in which
W3 deprotonates and
the
coordination shell of Mn4 has been proposed by Ugur et
al.52 However, they also proposed that W3 may translate
toward Mn1 in the S2 g = 2 structure to complete its
coordination shell. This is included as a possible
transition but is found to be energetically unfavorable in
our simulations. Earlier computational studies generally
agree that the water/hydroxide is inserted into the OEC
cluster from the closed, S2 state g = 4.1 structure.8,19,32,53
The S2 state g = 4.1 form is less rigid, with higher
fluctuations that allow the rearrangement needed for
binding an additional water molecule.8
The two S2 structures were then advanced to the
S3 state to oxidize the Mn centers by changing the redox
potential (Eh), at physiological pH where:
4
O6 is located at 2.97 Å from Ca2+ forming a strong
hydrogen bond with O5. The model proposed in this paper
is an intermediate between the closed and open models of
the S3 state, with O6 forming a hydrogen bond with O5
but not as strongly as in the earlier open structure model.
In addition, the O6-Ca2+ distance is shorter in our model
than the other two forms (Figure S1).
transitions.66 In addition, the recently published 2.07 Å
resolution S3 state structure by Kern et al.28 showed a
translation of E189 away from Mn1 upon the insertion of
a water molecule that is only 2.09 Å away from O5 and
1.78 Å from Mn1. Thus, to examine the role of E189, we
have generated more than 50 conformers of its sidechains
and sampled them along with 451 water molecule-binding
sites in the S2 g = 2, S2 g = 4.1 and S3 states.
The MC sampling confirms a conformational
change for E189 moving away from Mn1 that is coupled
to binding another OH -- (O7) i.e., O7 replacing the anionic
E189 as a ligand for Mn1. These conformational changes
are coupled to the oxidation of Mn1 in both the S2 state g
= 4.1 and the S3 state structures while no conformational
changes were observed for the S2 state g = 2 structure.
Although
show different
conformations of E189, the conformational changes
observed in our simulations are larger than the reported in
the XFEL structures and suggest a complete loss of
coordination of E189 from Mn1, which is replaced by a
hydroxyl.
the XFEL
structures
Figure 3. The ligand environment of Mn1 with O7
(hydroxyl anion) and protonated D1-E189. Strong
hydrogen bonds (shown in dashed lines) are formed
between D1-E189, OH7 and µ-O1.
The OH7 (i.e., the hydroxyl anion form of O7)
has a stronger dipole moment than the carboxylate group
of the amino acid and replaces the E189 ligand in the S2
g = 4.1 and in the S3 state due to the electrostatic
attraction with Mn1. However, in the S2 g = 2 state, Mn1
is in the Mn(III) state and the repulsion with the nearby
oxo bridges and negatively charged ligands is stronger
than the attraction to Mn1. Thus, in the S1 and S2 g = 2
states E189 is a Mn1ligand and no OH -- binds to Mn1. O7
binding is independent from the addition of O6, i.e. both
Figure 2. Comparison of calculated
(red) and
experimental (black) Mn EXAFS spectra of different S3
state models in k-space (A and B) and reduced distance
space (C and D) for the new model obtained in this study
(A and C) and the previous QM/MM-optimized open
form of the S3 state (B and D).10
Figure 2 compares the EXAFS spectrum of the S3
state model obtained in the current study (panels A and C)
with the corresponding spectrum of the previously
reported QM/MM S3 model (panels B and D) and the
experimental EXAFS data.65 The comparison suggests
that the S3 state model obtained in the current study does
not match the experimental EXAFS data as well as the
QM/MM-optimized open-form of the S3 state, suggesting
that the structure we obtained in this study may be an
intermediate structure during the S2-to-S3 state transition,
with the final metastable S3 state best described as the
open form of the S3 state.. The largest difference in the
Fourier Transformed spectrum is observed for the second
peak (Figure 2 C, D), i.e. the Mn-Mn/Mn-Ca2+ distances,
which are sensitive to the different positions of the
additional ligand O6 (Figure S1).
Role of D1-E189. While D1-E189 appears to be a
ligand of Mn1 in the S1 dark-adapted state, FTIR
difference spectroscopy suggested that it is not a ligand of
a Mn that undergoes oxidation between the S0 and S3 state
5
structures. Furthermore, Kern et al. have built two
conformations for E189, and in the second of these the
O7- Oe2 E189 distance is 1.60 Å. This suggests that O7
and E189 are mutually exclusive in space in this
conformation. It is relatively straightforward to detect the
binding of a water molecule in a location where nothing
is there using Fo-Fo isomorphous difference Fourier maps
regardless of its low population. However, it is much
more challenging to establish a displacement of side
chains such as E189 using the same method, which may
become undetectable when the population is low. This
may explain the uncertainties in the sidechain positions of
E189.
The S3 XFEL structure by Suga et al. at 2.25 Å
resolution suggested the insertion of an additional oxygen
(or water substrate) that is only 1.46 Å away from O5
similar to the S3 open structure.68 However, this short
distance between the two oxygens is possible only when
there is H atom trapped between them and may have
resulted from superposition of states with lower oxidation
levels.69 The O6 position observed in the Suga et al. S3
structure is far from E189. Even so, possible multiple
conformations of E189 in the S3 state have been detected
(for example, Figure S5 of Wang70 et al., 2017).
Therefore, the possibility of a third E189 conformation as
proposed by this study does not contradict the evidence of
the structure given the current uncertainty.
Table 1 shows the comparison of the Mni, Ca2+-Oj
(i = 1,4 and j = 5,6,7) distances in the different structures.
It is clear that the position of O6 is similar to the oxygen
position observed in the XFEL structure by Suga et al.,30
while the position of O7 is closer to the oxygen identified
in the S3 structure by Kern et al.28 The mismatch between
the XFEL measurements may have resulted from the
uncertainties in the oxygen positions, due to the
difficulties of accurately resolving their electron density
near the heavy Mn ions, the fact that misses in a multi-
flash experiment will produce a mixture of S states, the
different time points used to probe the structures, and the
changes induced by radiation.71,72
hydroxides may bind the OEC simultaneously. However,
when O7 has replaced E189 as the Mn1 ligand, the
protonation state of E189 is dependent on the state of O6.
Thus, E189 is ionized in the S2 g = 4.1 state because O6
is not bound yet to the cluster; instead, the Ca2+ ligand W3
is mostly occupied. In the S3 state, O6 binds to the cluster
and causes an approximately 50:50 mixture of protonated
and deprotonated E189 due to the electrostatic repulsions.
To further assess the energetics of O7 binding, we
compared the optimized DFT energies of the S3 state
having E189 ionized and bound to Mn1 (Mn1-E189) with
the S3 state having OH7 bound to Mn1 and protonated
E189 (Mn1-OH7). The X, Y, Z coordinates of the two
optimized structures, the DFT energies (in Hartree) and
the Mn spin densities are reported in the SI (Table S1).
The Mn1-OH7 structure energy is 2.4 kcal/mol lower than
the Mn1-E189 structure, which indicates that the two
states are close in energy, in agreement with the MC
sampling calculations. The mutations of E189 (E189K, R
and Q) have shown no effect on electron transfer at the
donor side of photosystem II,67 which may be explained
by the existence of an isoenergetic structure (Mn1-OH7)
that facilitates the oxidation of Mn1 through the Kok
cycle.
XFEL structures of the S3 state. The optimized
Mn1-OH7 structure shows strong hydrogen bonds formed
between O7, the carboxylate group of E189 (2.83 Å O-O
distance) and the µ-oxo that bridges Mn1 and Mn2
(Figure 3). The position of O7 is very similar to the
additional oxygen resolved in the latest S3 structure Kern
et al.28 (Table 1). However, the Kern et al. structure also
has an unusually short inter-atomic distance of 2.08 Å
between the newly inserted oxygen and Oe2 of E189,
which is much shorter than the sum of their van der Waals
radii, and is considered to be physically impossible. Such
a short distance cannot be rationalized with any known
repulsive force field parameters in molecular dynamic
simulation. If O7 is coordinated as a ligand of Mn1, E189
must move away from O7 through repulsive interactions
with E189. It is interesting that E189 always has multiple
conformations in both the Suga et al. and Kern et al. S3
6
Table 1. The distances in the optimized S2g=2, S2g=4.1 and S3 structures.
S2g=2
3.30
3.93
2.96
1.82
-
-
-
-
-
-
-
S3Kern
3.33
3.90
2.90
2.26
S2g=4.1
2.87
4.20
1.83
2.99
S3Suga
3.26
4.07
2.80
3.51
2.27
2.30
S3-DFT
Mn1-Ca2+
2.92
Mn4-Ca2+
3.93
1.84
Mn1-O5
3.35
Mn4-O5
4.26
Mn1-O6
1.97
Mn4-O6
1.79
Mn1-O7
5.74
Mn4-O7
2.65
O5-O6
2.75
O5-O7
4.29
O6-O7
All distances are in Å. No binding sites identified for O6 or O7 in the S2
g = 2 structure. The S3Suga
68 structure is resolved at 2.5 Å (PDB ID:5GTI),
while the S3Kern
28 is resolved at 2.07 Å (PDB ID:6DHO).
-
-
1.78
4.18
2.09
2.10
-
1.46
-
-
-
-
-
-
-
-
-
-
-
Figure 4. The proposed mechanism for the S2-to-S3 state transition. A, B, C and D are the S2 g = 2, S2 g = 4.1, S2
g = 4.1/YZ
•H190+ and S3 states, respectively. The reaction starts at A (S2 g = 2) and ends at D (S3). The B*, C* and D* are
possible isomers of the B, C and D states, where O7 replaces D1-E189 as a ligand for Mn1.
Conclusions
Simulations based on classical Monte Carlo sampling of
many possible oxygen positions
the
Boltzmann distribution suggest the transition from the S2
to generate
to the S3 state starts by a transition from the S2 g = 2 state
to the S2 g = 4.1 state. Then, W3 on calcium deprotonates
upon the formation of the YZ
•/H190+ intermediate prior to
the oxidation of the Mn cluster. The deprotonation of W3
is coupled to a translation of the oxygen toward Mn4,
7
which completes its coordination shell and facilitates the
oxidation of the cluster to the S3 state (Figure 4). The
EXAFS spectrum of the resulting S3 state structure
suggests it is more consistent with an intermediate
generated during the S2-to-S3 state transition prior to
formation of the S3 state in its open form.10 In addition,
the sampling of several conformers of the D1-E189
sidechain suggest that it undergoes a conformational
change that is coupled to the oxidation of Mn1 and the
binding another OH -- (O7) as suggested by the latest
XFEL S3 state structure (Figure 4, B*, C*, D*).
Supporting Information
The Supporting Information includes Mn spin densities
and the optimized atomic coordinates of the proposed
models of the S3 state.
Acknowledgements
The authors acknowledge computational resources from
NERSC (V.S.B.) and support by the U.S. Department of
Energy, Office of Science, Office of Basic Energy
Sciences, Division of Chemical Sciences, Geosciences,
and Biosciences via Grants DESC0001423 (M.R.G. and
V.S.B.), DE-FG02-05ER15646 (G.W.B.) and from
European Research Council through the Consolidator
Grant COMOTION (ERC-Küpper-614507).
References:
(1) Kok, B.; Forbush, B.; McGloin, M. Cooperation of Charges in
Photosynthetic O2 Evolution-I. A Linear Four Step
Mechanism. Photochem. Photobiol. 1970, 11 (6), 457 -- 475.
(2) Brudvig, G. W. Water Oxidation Chemistry of Photosystem II.
Philos Trans R Soc Lond B Biol Sci 2008, 363 (1494), 1211 --
1219. https://doi.org/10.1098/rstb.2007.2217.
(3) Amin, M.; Vogt, L.; Szejgis, W.; Vassiliev, S.; Brudvig, G. W.;
Bruce, D.; Gunner, M. R. Proton-Coupled Electron Transfer
during the S-State Transitions of the Oxygen-Evolving
Complex of Photosystem II. J. Phys. Chem. B 2015, 119 (24),
7366 -- 7377. https://doi.org/10.1021/jp510948e.
(4) Dau, H.; Haumann, M. The Manganese Complex of
Photosystem II in Its Reaction Cycle -- Basic Framework and
Possible Realization at the Atomic Level. Coord. Chem. Rev.
2008, 252 (3 -- 4), 273 -- 295.
https://doi.org/10.1016/j.ccr.2007.09.001.
(5) Pal, R.; Negre, C. F.; Vogt, L.; Pokhrel, R.; Ertem, M. Z.; Brudvig,
G. W.; Batista, V. S. S0-State Model of the Oxygen-Evolving
Complex of Photosystem II. Biochemistry 2013, 52 (44),
7703 -- 7706. https://doi.org/10.1021/bi401214v.
(6) Luber, S.; Rivalta, I.; Umena, Y.; Kawakami, K.; Shen, J.-R.;
Kamiya, N.; Brudvig, G. W.; Batista, V. S. S1-State Model of
the O2-Evolving Complex of Photosystem II. Biochemistry
2011, 50 (29), 6308 -- 6311.
https://doi.org/10.1021/bi200681q.
(7) Askerka, M.; Wang, J.; Brudvig, G. W.; Batista, V. S. Structural
Changes in the Oxygen-Evolving Complex of Photosystem II
8
(11) Umena, Y.; Kawakami, K.; Shen, J.-R.; Kamiya, N. Crystal
Structure of Oxygen-Evolving Photosystem II at a Resolution
of 1.9 Å. Nature 2011, 473 (7345), 55 -- 60.
https://doi.org/10.1038/nature09913.
(12) Lohmiller, T.; Ames, W.; Lubitz, W.; Cox, N.; Misra, S. K. EPR
Spectroscopy and the Electronic Structure of the Oxygen-
Evolving Complex of Photosystem II. Appl Magn Reson
2013, 44 (6), 691 -- 720. https://doi.org/10.1007/s00723-
012-0437-3.
(13) Britt, R. D.; Campbell, K. A.; Peloquin, J. M.; Gilchrist, M. L.;
Aznar, C. P.; Dicus, M. M.; Robblee, J.; Messinger, J. Recent
Pulsed EPR Studies of the Photosystem II Oxygen-Evolving
Complex: Implications as to Water Oxidation Mechanisms.
Biochim Biophys Acta Bioenerg 2004, 1655, 158 -- 171.
https://doi.org/10.1016/j.bbabio.2003.11.009.
Induced by the S1 to S2 Transition: A Combined XRD and
QM/MM Study. Biochemistry 2014, 53 (44), 6860 -- 6862.
https://doi.org/10.1021/bi5011915.
(8) Bovi, D.; Narzi, D.; Guidoni, L. The S2 State of the Oxygen-
Evolving Complex of Photosystem II Explored by QM/MM
Dynamics: Spin Surfaces and Metastable States Suggest a
Reaction Path towards the S3 State. Angew. Chem. 2013, 52
(45), 11744 -- 11749.
https://doi.org/10.1002/anie.201306667.
(9) Pantazis, D. A.; Ames, W.; Cox, N.; Lubitz, W.; Neese, F. Two
Interconvertible Structures That Explain the Spectroscopic
Properties of the Oxygen-evolving Complex of Photosystem
II in the S2 State. Angew. Chem. 2012, 51 (39), 9935 -- 9940.
https://doi.org/10.1002/anie.201204705.
(10) Askerka, M.; Wang, J.; Vinyard, D. J.; Brudvig, G. W.; Batista, V.
S. S3 State of the O2-Evolving Complex of Photosystem II:
Insights from QM/MM, EXAFS, and Femtosecond X-Ray
Diffraction. Biochemistry 2016, 55 (7), 981 -- 984.
https://doi.org/10.1021/acs.biochem.6b00041.
(14) Dismukes, G. C.; Siderer, Y. Intermediates of a Polynuclear
Manganese Center Involved in Photosynthetic Oxidation of
Water. Proc. Natl. Acad. Sci. U.S.A. 1981, 78 (1), 274 -- 278.
https://doi.org/10.1073/pnas.78.1.274.
(15) Zimmermann, J. L.; Rutherford, A. W. EPR Studies of the
Oxygen-Evolving Enzyme of Photosystem II. Biochim
Biophys Acta Bioenerg 1984, 767 (1), 160 -- 167.
https://doi.org/10.1016/0005-2728(84)90091-4.
(16) de Paula, J. C.; Brudvig, G. W. Magnetic Properties of
Manganese in the Photosynthetic Oxygen-Evolving
Complex. J. Am. Chem. Soc. 1985, 107 (9), 2643 -- 2648.
https://doi.org/10.1021/ja00295a016.
(17) Casey, J. L.; Sauer, K. EPR Detection of a Cryogenically
Photogenerated Intermediate in Photosynthetic Oxygen
Evolution. Biochim. Biophys. Acta 1984, 767 (1), 21 -- 28.
https://doi.org/10.1016/0005-2728(84)90075-6.
(18) Amin, M.; Pokhrel, R.; Brudvig, G. W.; Badawi, A.; Obayya, S. S.
A. Effect of Chloride Depletion on the Magnetic Properties
and the Redox Leveling of the Oxygen-Evolving Complex in
Photosystem II. J. Phys. Chem. B 2016, 120 (18), 4243 -- 4248.
https://doi.org/10.1021/acs.jpcb.6b03545.
(19) Narzi, D.; Bovi, D.; Guidoni, L. Pathway for Mn-Cluster
Oxidation by Tyrosine-Z in the S2 State of Photosystem II.
Proc. Natl. Acad. Sci. U.S.A. 2014, 111 (24), 8723 -- 8728.
https://doi.org/10.1073/pnas.1401719111.
(20) Ono, T.; Zimmermann, J. L.; Inoue, Y.; Rutherford, A. W. EPR
Evidence for a Modified S-State Transition in Chloride-
Depleted Photosystem II. Biochim Biophys Acta Bioenerg
1986, 851 (2), 193 -- 201. https://doi.org/10.1016/0005-
2728(86)90125-8.
(21) Rappaport, F.; Lavergne, J. Proton Release during Successive
Oxidation Steps of the Photosynthetic Water Oxidation
Process: Stoichiometries and PH Dependence. Biochemistry
1991, 30 (41), 10004 -- 10012.
https://doi.org/10.1021/bi00105a027.
(22) Lavergne, J.; Junge, W. Proton Release during the Redox Cycle
of the Water Oxidase. Photosynth Res 1993, 38 (3), 279 --
296. https://doi.org/10.1007/BF00046752.
(23) Suzuki, H.; Sugiura, M.; Noguchi, T. Monitoring Proton Release
during Photosynthetic Water Oxidation in Photosystem II by
Means of Isotope-Edited Infrared Spectroscopy. J. Am.
Chem. Soc. 2009, 131 (22), 7849 -- 7857.
https://doi.org/10.1021/ja901696m.
(24) Yano, J.; Pushkar, Y.; Glatzel, P.; Lewis, A.; Sauer, K.; Messinger,
J.; Bergmann, U.; Yachandra, V. High-Resolution Mn EXAFS
of the Oxygen-Evolving Complex in Photosystem II:
Structural Implications for the Mn4Ca Cluster. J. Am. Chem.
Soc. 2005, 127 (43), 14974 -- 14975.
https://doi.org/10.1021/ja054873a.
(25) Askerka, M.; Brudvig, G. W.; Batista, V. S. The O2-Evolving
Complex of Photosystem II: Recent Insights from Quantum
Mechanics/Molecular Mechanics (QM/MM), Extended X-
Ray Absorption Fine Structure (EXAFS), and Femtosecond X-
Ray Crystallography Data. Acc. Chem. Res. 2017, 50 (1), 41 --
48. https://doi.org/10.1021/acs.accounts.6b00405.
(26) Dau, H.; Grundmeier, A.; Loja, P.; Haumann, M. On the
Structure of the Manganese Complex of Photosystem II:
Extended-Range EXAFS Data and Specific Atomic-Resolution
Models for Four S-States. Philos Trans R Soc Lond B Biol Sci
2008, 363 (1494), 1237 -- 1244.
https://doi.org/10.1098/rstb.2007.2220.
(27) Zaharieva, I.; Dau, H.; Haumann, M. Sequential and Coupled
Proton and Electron Transfer Events in the S2 → S3
Transition of Photosynthetic Water Oxidation Revealed by
Time-Resolved X-Ray Absorption Spectroscopy.
Biochemistry 2016, 55 (50), 6996 -- 7004.
https://doi.org/10.1021/acs.biochem.6b01078.
(28) Kern, J.; Chatterjee, R.; Young, I. D.; Fuller, F. D.; Lassalle, L.;
Ibrahim, M.; Gul, S.; Fransson, T.; Brewster, A. S.; Alonso-
Mori, R.; et al. Structures of the Intermediates of Kok's
Photosynthetic Water Oxidation Clock. Nature 2018, 563
(7731), 421 -- 425. https://doi.org/10.1038/s41586-018-
0681-2.
(29) Dau, H.; Zaharieva, I.; Haumann, M. Recent Developments in
Research on Water Oxidation by Photosystem II. Curr Opin
Chem Biol 2012, 16 (1 -- 2), 3 -- 10.
https://doi.org/10.1016/j.cbpa.2012.02.011.
(30) Suga, M.; Akita, F.; Hirata, K.; Ueno, G.; Murakami, H.;
Nakajima, Y.; Shimizu, T.; Yamashita, K.; Yamamoto, M.;
Ago, H.; et al. Native Structure of Photosystem II at 1.95 Å
Resolution Viewed by Femtosecond X-Ray Pulses. Nature
2015, 517 (7532), 99 -- 103.
https://doi.org/10.1038/nature13991.
(31) Askerka, M.; Vinyard, D. J.; Wang, J.; Brudvig, G. W.; Batista, V.
S. Analysis of the Radiation-Damage-Free X-Ray Structure of
Photosystem II in Light of EXAFS and QM/MM Data.
Biochemistry 2015, 54 (9), 1713 -- 1716.
https://doi.org/10.1021/acs.biochem.5b00089.
(32) Retegan, M.; Krewald, V.; Mamedov, F.; Neese, F.; Lubitz, W.;
Cox, N.; Pantazis, D. A. A Five-Coordinate Mn(IV)
Intermediate in Biological Water Oxidation: Spectroscopic
Signature and a Pivot Mechanism for Water Binding. Chem.
Sci. 2015, 7 (1), 72 -- 84.
https://doi.org/10.1039/C5SC03124A.
9
(33) Cox, N.; Messinger, J. Reflections on Substrate Water and
Dioxygen Formation. Biochim. Biophys. Acta 2013, 1827 (8),
1020 -- 1030. https://doi.org/10.1016/j.bbabio.2013.01.013.
(34) Siegbahn, P. E. M. Substrate Water Exchange for the Oxygen
Evolving Complex in PSII in the S1, S2, and S3 States. J. Am.
Chem. Soc. 2013, 135 (25), 9442 -- 9449.
https://doi.org/10.1021/ja401517e.
(35) Amin, M.; Vogt, L.; Vassiliev, S.; Rivalta, I.; Sultan, M. M.; Bruce,
D.; Brudvig, G. W.; Batista, V. S.; Gunner, M. R. Electrostatic
Effects on Proton Coupled Electron Transfer in
Oxomanganese Complexes Inspired by the Oxygen-Evolving
Complex of Photosystem II. J Phys Chem B 2013, 117 (20),
6217 -- 6226. https://doi.org/10.1021/jp403321b.
(36) Vreven, T.; Morokuma, K. The ONIOM (Our Own N-Layered
Integrated Molecular Orbital + Molecular Mechanics)
Method for the First Singlet Excited (S1) State
Photoisomerization Path of a Retinal Protonated Schiff
Base. J. Chem. Phys. 2000, 113 (8), 2969 -- 2975.
https://doi.org/10.1063/1.1287059.
(37) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb,
M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Petersson,
G. A.; Nakatsuji, H.; et al. G09 Gaussian.Com. 2016.
(38) Becke, A. D. Density-Functional Exchange-Energy
Approximation with Correct Asymptotic Behavior. Phys.
Rev. A 1988, 38 (6), 3098 -- 3100.
https://doi.org/10.1103/PhysRevA.38.3098.
(39) Becke, A. D. Density-functional Thermochemistry. III. The Role
of Exact Exchange. J. Chem. Phys. 1993, 98 (7), 5648 -- 5652.
https://doi.org/10.1063/1.464913.
(40) Grimme, S. Semiempirical GGA-Type Density Functional
Constructed with a Long-Range Dispersion Correction. J.
Comput. Chem. 2006, 27 (15), 1787 -- 1799.
https://doi.org/10.1002/jcc.20495.
(41) Case, D. A.; Darden, T., A.; Cheathem, T. E.; Simmerling, C. L.;
Wang, J.; Duke, R. E.; Luo, R.; Walker, R. C.; Zhang, W.;
Merz, K. M.; et al. AMBER 12, University of California, San
Francisco.
(42) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Gould, I. R.; Merz, K. M.;
Ferguson, D. M.; Spellmeyer, D. C.; Fox, T.; Caldwell, J. W.;
Kollman, P. A. A Second Generation Force Field for the
Simulation of Proteins, Nucleic Acids, and Organic
Molecules. J. Am. Chem. Soc. 1995, 117 (19), 5179 -- 5197.
https://doi.org/10.1021/ja00124a002.
(43) Hay, P. J.; Wadt, W. R. Ab Initio Effective Core Potentials for
Molecular Calculations. Potentials for K to Au Including the
Outermost Core Orbitals. J. Chem. Phys. 1985, 82 (1), 299 --
310. https://doi.org/10.1063/1.448975.
(44) Wadt, W. R.; Hay, P. J. Ab Initio Effective Core Potentials for
Molecular Calculations. Potentials for Main Group Elements
Na to Bi. J. Chem. Phys. 1985, 82 (1), 284 -- 298.
https://doi.org/10.1063/1.448800.
(45) Ditchfield, R.; Hehre, W. J.; Pople, J. A. Self-consistent
Molecular-orbital Methods. IX. An Extended Gaussian-type
Basis for Molecular-orbital Studies of Organic Molecules. J.
Chem. Phys. 1971, 54 (2), 724 -- 728.
https://doi.org/10.1063/1.1674902.
(46) Hariharan, P. C.; Pople, J. A. The Influence of Polarization
Functions on Molecular Orbital Hydrogenation Energies.
Theoret. Chim. Acta 1973, 28 (3), 213 -- 222.
https://doi.org/10.1007/BF00533485.
(47) Song, Y.; Gunner, M. R. Using Multiconformation Continuum
Electrostatics to Compare Chloride Binding Motifs in α-
Amylase, Human Serum Albumin, and Omp32. J. Mol. Biol.
2009, 387 (4), 840 -- 856.
https://doi.org/10.1016/j.jmb.2009.01.038.
(48) Song, Y.; Mao, J.; Gunner, M. R. MCCE2: Improving Protein PKa
Calculations with Extensive Side Chain Rotamer Sampling. J.
Comput. Chem. 2009, 30 (14), 2231 -- 2247.
https://doi.org/10.1002/jcc.21222.
(49) Baker, N. A.; Sept, D.; Joseph, S.; Holst, M. J.; McCammon, J. A.
Electrostatics of Nanosystems: Application to Microtubules
and the Ribosome. Proc. Natl. Acad. Sci. U.S.A. 2001, 98
(18), 10037 -- 10041.
https://doi.org/10.1073/pnas.181342398.
(50) Rehr, J. J.; Albers, R. C. Theoretical Approaches to X-Ray
Absorption Fine Structure. Rev. Mod. Phys. 2000, 72 (3),
621 -- 654. https://doi.org/10.1103/RevModPhys.72.621.
(51) Newville, M. EXAFS Analysis Using FEFF and FEFFIT. J
Synchrotron Rad 2001, 8 (2), 96 -- 100.
https://doi.org/10.1107/S0909049500016290.
(52) Ugur, I.; Rutherford, A. W.; Kaila, V. R. I. Redox-Coupled
Substrate Water Reorganization in the Active Site of
Photosystem II -- The Role of Calcium in Substrate Water
Delivery. Biochim Biophys Acta Bioenerg 2016, 1857 (6),
740 -- 748. https://doi.org/10.1016/j.bbabio.2016.01.015.
(53) Capone, M.; Narzi, D.; Bovi, D.; Guidoni, L. Mechanism of Water
Delivery to the Active Site of Photosystem II along the S2 to
S3 Transition. J. Phys. Chem. Lett. 2016, 7 (3), 592 -- 596.
https://doi.org/10.1021/acs.jpclett.5b02851.
(54) Siegbahn, P. E. M. Mechanisms for Proton Release during
Water Oxidation in the S2 to S3 and S3 to S4 Transitions in
Photosystem II. Phys. Chem. Chem. Phys. 2012, 14 (14),
4849 -- 4856. https://doi.org/10.1039/C2CP00034B.
(55) Guo, Y.; Li, H.; He, L.-L.; Zhao, D.-X.; Gong, L.-D.; Yang, Z.-Z.
Theoretical Reflections on the Structural Polymorphism of
the Oxygen-Evolving Complex in the S2 State and the
Correlations to Substrate Water Exchange and Water
Oxidation Mechanism in Photosynthesis. Biochim Biophys
Acta Bioenerg 2017, 1858 (10), 833 -- 846.
https://doi.org/10.1016/j.bbabio.2017.08.001.
(56) Shoji, M.; Isobe, H.; Yamaguchi, K. QM/MM Study of the S2 to
S3 Transition Reaction in the Oxygen-Evolving Complex of
Photosystem II. Chem. Phys. Lett. 2015, 636, 172 -- 179.
https://doi.org/10.1016/j.cplett.2015.07.039.
(57) Beal, N. J.; Corry, T. A.; O'Malley, P. J. A Comparison of
Experimental and Broken Symmetry Density Functional
Theory (BS-DFT) Calculated Electron Paramagnetic
Resonance (EPR) Parameters for Intermediates Involved in
the S2 to S3 State Transition of Nature's Oxygen Evolving
Complex. J. Phys. Chem. B 2018, 122 (4), 1394 -- 1407.
https://doi.org/10.1021/acs.jpcb.7b10843.
(58) Boussac, A.; Rutherford, A. W.; Sugiura, M. Electron Transfer
Pathways from the S2-States to the S3-States Either after a
Ca 2+ /Sr2+ or a Cl− /I− Exchange in Photosystem II from
Thermosynechococcus Elongatus. Biochim. Biophys. Acta
2015, 1847 (6), 576 -- 586.
https://doi.org/10.1016/j.bbabio.2015.03.006.
(59) Boussac, A.; Ugur, I.; Marion, A.; Sugiura, M.; Kaila, V. R. I.;
Rutherford, A. W. The Low Spin - High Spin Equilibrium in
the S2-State of the Water Oxidizing Enzyme. Biochim
Biophys Acta Bioenerg 2018, 1859 (5), 342 -- 356.
https://doi.org/10.1016/j.bbabio.2018.02.010.
(60) Kaur, D.; Szejgis, W.; Mao, J.; Amin, M.; Reiss, K. M.; Askerka,
M.; Cai, X.; Khaniya, U.; Zhang, Y.; Brudvig, G. W.; et al.
Relative Stability of the S2 Isomers of the Oxygen Evolving
Complex of Photosystem II. Photosyn. Res. 2019.
https://doi.org/10.1007/s11120-019-00637-6.
(61) Blomberg, M. R. A.; Siegbahn, P. E. M. A Quantum Chemical
Study of Tyrosyl Reduction and O -- O Bond Formation in
10
Photosystem II. Mol. Phys. 2003, 101 (1 -- 2), 323 -- 333.
https://doi.org/10.1080/00268970210162781.
(62) Limburg, J.; Vrettos, J. S.; Liable-Sands, L. M.; Rheingold, A. L.;
Crabtree, R. H.; Brudvig, G. W. A Functional Model for O-O
Bond Formation by the O2-Evolving Complex in
Photosystem II. Science 1999, 283 (5407), 1524 -- 1527.
https://doi.org/10.1126/science.283.5407.1524.
(63) Siegbahn, P. E. M. O-O Bond Formation in the S4 State of the
Oxygen-Evolving Complex in Photosystem II. Chem. Eur. J.
2006, 12 (36), 9217 -- 9227.
https://doi.org/10.1002/chem.200600774.
(64) Cox, N.; Retegan, M.; Neese, F.; Pantazis, D. A.; Boussac, A.;
Lubitz, W. Electronic Structure of the Oxygen-Evolving
Complex in Photosystem II Prior to O-O Bond Formation.
Science 2014, 345 (6198), 804 -- 808.
https://doi.org/10.1126/science.1254910.
(65) Haumann, M.; Liebisch, P.; Müller, C.; Barra, M.; Grabolle, M.;
Dau, H. Photosynthetic O2 Formation Tracked by Time-
Resolved x-Ray Experiments. Science 2005, 310 (5750),
1019 -- 1021. https://doi.org/10.1126/science.1117551.
(66) Strickler, M. A.; Hillier, W.; Debus, R. J. No Evidence from FTIR
Difference Spectroscopy That Glutamate-189 of the D1
Polypeptide Ligates a Mn Ion That Undergoes Oxidation
during the S0 to S1, S1 to S2, or S2 to S3 Transitions in
Photosystem II. Biochemistry 2006, 45 (29), 8801 -- 8811.
https://doi.org/10.1021/bi060583a.
(67) Clausen, J.; Winkler, S.; Hays, A. M.; Hundelt, M.; Debus, R. J.;
Junge, W. Photosynthetic Water Oxidation in Synechocystis
Sp. PCC6803: Mutations D1-E189K, R and Q Are without
Influence on Electron Transfer at the Donor Side of
Photosystem II. Biochim. Biophys. Acta 2001, 1506 (3), 224 --
235. https://doi.org/10.1016/s0005-2728(01)00217-1.
(68) Suga, M.; Akita, F.; Sugahara, M.; Kubo, M.; Nakajima, Y.;
Nakane, T.; Yamashita, K.; Umena, Y.; Nakabayashi, M.;
Yamane, T.; et al. Light-Induced Structural Changes and the
Site of O=O Bond Formation in PSII Caught by XFEL. Nature
2017, 543 (7643), 131 -- 135.
https://doi.org/10.1038/nature21400.
(69) Pantazis, D. A. Missing Pieces in the Puzzle of Biological Water
Oxidation. ACS Catal. 2018, 8 (10), 9477 -- 9507.
https://doi.org/10.1021/acscatal.8b01928.
(70) Wang, J.; Askerka, M.; Brudvig, G. W.; Batista, V. S.
Crystallographic Data Support the Carousel Mechanism of
Water Supply to the Oxygen-Evolving Complex of
Photosystem II. ACS Energy Lett 2017, 2 (10), 2299 -- 2306.
https://doi.org/10.1021/acsenergylett.7b00750.
(71) Amin, M.; Askerka, M.; Batista, V. S.; Brudvig, G. W.; Gunner,
M. R. X-Ray Free Electron Laser Radiation Damage through
the S-State Cycle of the Oxygen-Evolving Complex of
Photosystem II. J Phys Chem B 2017, 121 (40), 9382 -- 9388.
https://doi.org/10.1021/acs.jpcb.7b08371.
(72) Amin, M.; Badawi, A.; Obayya, S. S. Radiation Damage in XFEL:
Case Study from the Oxygen-Evolving Complex of
Photosystem II. Sci Rep 2016, 6, 36492.
https://doi.org/10.1038/srep36492.
Batista6
Supporting Information for:
Thermodynamics of the S2-to-S3 State Transition of the Oxygen-Evolving Complex of Photosystem II
Muhamed Amin1,2,3*, Divya Kaur4,5, Ke R. Yang6, Jimin Wang7, Zainab Mohamed8, Gary W. Brudvig6, M.R. Gunner4,5 and Victor
1 Center for Free-Electron Laser Science, Deutsches Elektronen-Synchrotron DESY, Notkestrasse 85, 22607 Hamburg, Germany
2 Physics Department, University College Groningen, University of Groningen, Groningen, Netherlands
3 Centre for Theoretical Physics, The British University in Egypt, Sherouk City 11837, Cairo, Egypt
4Department of Physics, City College of the City University of New York, New York, NY 10031, United States
5Ph.D. Program in Chemistry, The Graduate Center of the City University of New York, New York, NY 10016, United States
6 Department of Chemistry, Yale University, New Haven, Connecticut 06520-8107, United States
7 Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT08620-8114
8 Zewail City of Science and Technology, Sheikh Zayed, District, 6th of October City, 12588 Giza, Egypt
[email protected]
1- Mn spin densities for optimized S3 state structure.
Spin densities
Mn1-O7, E1890
Table S1. The total energies and atomic spin
densities of the Mn centers in the S3 state
Mn1
Mn2
Mn3
Mn4
-6372.99484693
The difference in total energy between the 2
states is 2.4 kcal/mol. Thus, the two states
are isoenergetic.
Mn1-E189-
2.942898
3.110379
3.027931
3.016139
2.939744
2.909465
3.037126
2.972186
DFT (B97D) Energies (Hartree)
-6372.99867963
2- Figure S1. The Mn cluster in the S3 state.
a) The open S3 model. b) The closed S3 model. c) The proposed S3 model.
a) The atomic coordinates of the S3 state model with O7 bound to Mn1 and D1-E189 protonated:
C 8.31218 3.59127 -1.06653
11
C 7.64049 3.40489 0.30683
C 6.13521 3.65811 0.25782
O 5.43217 2.89152 1.09338
O 5.63343 4.50706 -0.48640
H 8.05668 4.13304 1.02305
H 8.08662 4.59153 -1.45913
H 7.83570 2.40986 0.72654
H 9.40297 3.48723 -0.97559
H 7.95388 2.85144 -1.79861
C -4.42580 6.89976 0.96293
C -5.63927 6.02215 0.58770
C -5.28077 4.83211 -0.28434
C -5.32517 3.52011 0.22169
C -4.90442 5.00220 -1.63332
C -5.03621 2.40811 -0.58269
C -4.61968 3.90232 -2.45514
C -4.70803 2.59183 -1.94179
O -4.49801 1.53268 -2.79209
H -5.09149 0.73584 -2.49237
H -3.67717 6.31314 1.51622
H -6.37617 6.65125 0.06203
H -5.61456 3.36299 1.26390
H -4.83895 6.01280 -2.04224
H -5.09250 1.39938 -0.17561
H -4.35750 4.03221 -3.50523
H -6.13190 5.66160 1.50497
C -0.98004 8.14610 2.49347
C -0.69713 6.98608 1.53967
O -0.11515 5.95181 1.93759
C -1.71244 7.64952 3.75721
H -0.00636 8.57344 2.78017
H -1.82600 8.46566 4.48374
H -1.13864 6.83850 4.22654
N -1.12686 7.14126 0.27110
C -1.84611 6.70012 -1.98557
O -2.50459 7.75002 -1.90379
C -1.07205 6.11926 -0.77262
C -1.70257 4.76655 -0.33494
C -0.77083 3.56580 -0.18377
O -1.31881 2.46756 0.07499
O 0.48854 3.78844 -0.34596
H -2.46589 4.43935 -1.05257
H -2.23245 4.88531 0.61548
N -1.76997 5.95189 -3.11090
C -2.53233 6.34275 -4.27637
C -2.42925 5.39121 -5.44854
O -1.76393 4.36961 -5.49841
H -3.04885 5.71409 -6.32009
H -2.23696 7.34720 -4.63660
12
H -3.60637 6.44435 -4.02930
H -1.18094 5.10206 -3.13476
C -6.98586 -5.33471 -2.33971
O -6.70918 -6.08019 -1.39435
C -6.10070 -5.21774 -3.58392
C -5.02928 -4.11671 -3.35990
C -4.24991 -4.37223 -2.06450
C -3.02784 -3.51553 -1.85473
O -2.44395 -2.89230 -2.76202
O -2.62033 -3.52921 -0.59358
H -3.89078 -5.41285 -2.01951
H -5.50962 -3.12681 -3.31228
H -4.89870 -4.26368 -1.18778
H -4.33611 -4.10185 -4.21251
N -8.04677 -4.46141 -2.31516
C -8.73169 -4.09484 -1.07723
C -8.88033 -2.55556 -0.97083
C -7.59176 -1.81975 -1.19497
C -7.22643 -0.85083 -2.11260
N -6.44612 -2.06844 -0.45972
C -5.45412 -1.27069 -0.93879
N -5.89716 -0.51458 -1.94230
H -8.12918 -4.48952 -0.24750
H -9.30379 -2.32156 0.01976
H -7.84246 -0.37289 -2.86771
H -4.45051 -1.24981 -0.52850
H -9.60585 -2.19631 -1.71702
C 7.72881 -2.12987 -2.17996
O 7.72126 -3.00585 -1.28588
C 7.37611 -2.49417 -3.62497
C 5.94705 -2.02223 -4.02364
C 4.83549 -2.55622 -3.16650
C 3.74105 -1.93992 -2.59121
N 4.67018 -3.90826 -2.87640
C 3.52240 -4.07376 -2.16540
N 2.94782 -2.89084 -1.97506
H 5.77251 -2.32443 -5.06979
H 3.45416 -0.89652 -2.58688
H 3.13062 -5.01938 -1.81292
H 5.90935 -0.92360 -4.00368
N 8.02274 -0.82762 -1.97230
C 8.11087 -0.23933 -0.63061
C 6.71683 -0.06315 0.01402
C 5.73436 0.55834 -0.98086
C 4.28599 0.62270 -0.55567
O 3.88018 -0.19469 0.33989
O 3.56887 1.44255 -1.21592
H 5.72057 -0.03632 -1.90507
H 8.74424 -0.88313 -0.00604
13
H 6.79156 0.56712 0.90967
H 6.03117 1.56442 -1.28855
H 8.60386 0.73418 -0.74674
H 6.34446 -1.03613 0.34158
C 8.49663 -3.71102 2.86210
C 7.30848 -4.69998 2.82744
C 6.10886 -4.12886 2.13230
C 4.78153 -4.02480 2.50020
N 6.19292 -3.54297 0.87476
C 4.98162 -3.10229 0.50484
N 4.10795 -3.38952 1.47017
H 8.82134 -3.44933 1.84355
H 7.01118 -4.97921 3.84806
H 4.27685 -4.35157 3.40061
H 4.76776 -2.53600 -0.38711
H 7.61903 -5.62930 2.32101
H 8.21742 -2.78455 3.38419
H 9.35153 -4.16317 3.38277
C -2.26391 -6.63428 0.74477
O -2.44321 -6.56495 -0.47004
C -0.90433 -7.03076 1.34760
C 0.25464 -6.56492 0.44850
C 0.42231 -5.04294 0.38303
O 0.06875 -4.37679 1.42084
O 0.92463 -4.57030 -0.69063
H 0.10384 -6.92957 -0.57440
H 1.20326 -6.98067 0.82628
N -3.25235 -6.38560 1.67450
C -4.60429 -5.96636 1.32658
C -4.92994 -4.49249 1.63203
O -6.07689 -4.05830 1.40585
H -4.75997 -6.11855 0.25032
N -3.93236 -3.74473 2.15345
C -4.00637 -2.29030 2.32982
C -2.67565 -1.67241 1.84561
O -2.66879 -0.63493 1.15219
C -4.27332 -1.90633 3.80020
O -1.64599 -2.34471 2.27222
H -4.32771 -0.81205 3.90336
H -4.81553 -1.90392 1.70143
H -3.46618 -2.28724 4.44110
H -3.00151 -4.15195 2.18393
H -5.22948 -2.34119 4.12064
C 1.67205 0.39540 7.06346
C 1.42718 0.12351 5.57124
C 2.20657 -1.12714 5.08401
C 1.91660 -1.34523 3.60790
O 1.03003 -2.24375 3.32606
O 2.52073 -0.58280 2.78631
14
H 1.90049 -2.01283 5.65798
H 2.74125 0.57052 7.25957
H 1.73915 0.98916 4.96830
H 3.28550 -0.95723 5.21516
H 1.34869 -0.45825 7.67936
H 0.35263 -0.02986 5.38398
H 1.11543 1.28498 7.39312
C -3.81623 3.42480 6.76441
C -3.25268 3.98612 5.44932
C -3.44560 3.01802 4.27055
C -2.84985 3.55277 2.95596
N -1.39510 3.74578 3.01003
C -0.50381 2.74944 2.87509
N -0.89203 1.45494 2.95503
N 0.80936 3.03884 2.75185
H 1.35278 2.34855 2.20788
H -3.67127 4.13060 7.59521
H -2.17747 4.19863 5.56091
H -4.52077 2.83475 4.10581
H -3.27752 4.54075 2.73916
H -1.02405 4.67956 2.78885
H -1.86137 1.25804 2.73685
H 1.02607 3.99327 2.47423
H -3.31604 2.47990 7.02978
H -3.74107 4.94600 5.20833
H -2.98915 2.04625 4.51947
H -3.10093 2.90824 2.09707
H -0.25686 0.74100 2.56907
H -4.89497 3.22112 6.67537
O 3.01985 3.56130 0.54927
H 4.42470 3.09736 0.94337
H 3.29415 4.35582 0.04690
O 2.57346 3.70778 -2.03891
H 1.82012 3.95108 -2.62239
O -2.35453 -0.07691 -2.33288
H -3.07714 0.54422 -2.60452
H -2.52638 -0.95744 -2.71610
O -0.50313 -2.27570 -0.21727
O 0.24930 -0.55314 1.45987
O 1.85513 -2.39800 0.77569
O 1.63754 1.36844 0.74015
O 1.49278 -0.71718 -0.97948
O 0.02031 3.76108 -3.30752
H -0.26611 3.57632 -4.21681
H 0.14743 2.88282 -2.84314
O 6.95536 1.26020 -3.70755
H 7.38553 1.67166 -4.46957
H 6.27766 1.93901 -3.41483
O 5.25740 3.24005 -3.02967
15
H 5.58142 3.83473 -2.32717
H 4.30996 3.14432 -2.78776
Mn 1.10915 -2.51061 -1.08102
Mn 0.14824 -2.33973 1.53412
Mn 1.97361 -0.36378 0.77653
Mn 1.99588 2.58241 -0.69369
Ca -0.82566 0.25786 -0.55895
O 0.74831 1.65260 -1.90653
H 1.29098 1.03253 -2.42160
H -6.34688 -2.77281 0.29467
H 7.81815 -0.15306 -2.72167
H -8.15172 -3.83407 -3.10432
H -5.61059 -6.18898 -3.73318
H -6.69409 -4.98231 -4.48164
H -0.78737 -6.62588 2.36291
H -0.87691 -8.13003 1.41715
H -5.35274 -6.58053 1.84785
H 7.47559 -3.58389 -3.72172
H 8.08795 -2.02771 -4.32126
H 5.31892 -4.64367 -3.12709
H 7.01308 -3.38813 0.25751
H -4.73412 7.74564 1.59673
H -3.93736 7.29713 0.06323
H -1.55657 8.93923 1.99401
H -9.73034 -4.55786 -1.01998
H -3.03278 -6.52421 2.65358
H -1.65668 7.97094 0.00579
H -2.71521 7.27196 3.50578
H -0.02760 5.93904 -1.06598
O 0.35896 -2.54074 -2.70219
H -0.62101 -2.69073 -2.68746
H 3.08536 -3.06936 1.36431
H -1.71266 -3.03912 -0.48479
C -3.61712 -7.98411 -0.49812
C -3.03985 -7.38027 0.79481
C -1.74094 -6.60961 0.55605
O -1.55831 -5.58735 1.37737
O -0.94095 -6.94433 -0.33235
H -2.79961 -8.18875 1.50567
H -2.86499 -8.62343 -0.97886
H -3.76259 -6.72881 1.30331
H -4.50571 -8.59272 -0.27527
H -3.90402 -7.20309 -1.21933
C 8.01049 -1.98665 -1.05358
C 8.24155 -0.88068 -2.10544
C 6.96785 -0.17633 -2.54200
C 6.76459 1.18737 -2.26352
C 5.96558 -0.85402 -3.27062
b) The atomic coordinates of the S3 state model with D1-E189 ionized and bound to Mn1:
16
C 5.62804 1.87160 -2.71888
C 4.81704 -0.19190 -3.72602
C 4.65350 1.18593 -3.47208
O 3.56393 1.82780 -4.00332
H 3.54049 2.81851 -3.72999
H 7.52801 -1.57427 -0.15414
H 8.72994 -1.33291 -2.98488
H 7.52713 1.73541 -1.70468
H 6.09232 -1.91121 -3.51065
H 5.50799 2.93811 -2.52950
H 4.06453 -0.72225 -4.30941
H 8.94554 -0.13402 -1.70544
C 7.58017 -4.20963 2.53845
C 6.18194 -3.98894 1.96469
O 5.21646 -3.64583 2.68689
C 7.54430 -4.87446 3.92542
H 8.19330 -4.79163 1.83327
H 8.55560 -4.92600 4.35137
H 7.14167 -5.89526 3.85903
N 6.05259 -4.15689 0.63052
C 4.86154 -4.50231 -1.45871
O 5.94684 -4.75347 -2.01614
C 4.81247 -3.83380 -0.06950
C 4.62889 -2.29281 -0.17495
C 3.17764 -1.84496 -0.32823
O 2.91136 -0.62662 -0.31207
O 2.35972 -2.84402 -0.41680
H 5.20773 -1.86682 -1.00584
H 5.00136 -1.82813 0.74783
N 3.67314 -4.78738 -2.03736
C 3.66420 -5.45577 -3.33349
C 4.52134 -4.70330 -4.36428
O 4.29340 -3.54694 -4.69952
H 5.35621 -5.27283 -4.82545
H 2.61905 -5.46639 -3.67390
H 4.03539 -6.48787 -3.23963
H 2.77867 -4.48531 -1.64213
C -1.01506 7.58419 -1.95121
O -1.29141 8.44567 -1.10585
C -2.01755 6.49977 -2.38036
C -1.40054 5.10668 -2.60997
C -2.47567 4.01653 -2.81066
C -1.80924 2.69526 -3.16216
O -1.03843 2.61585 -4.13220
O -2.06287 1.62285 -2.42712
H -3.12113 4.29197 -3.66164
H -0.77950 4.83841 -1.74593
H -3.09798 3.91020 -1.91606
H -0.76231 5.10159 -3.50726
17
N 0.19284 7.54807 -2.59012
C 1.31105 8.40628 -2.19844
C 2.58761 8.00084 -2.97030
C 2.90068 6.53653 -2.84467
C 3.08366 5.54123 -3.78887
N 2.99580 5.89175 -1.62114
C 3.22644 4.57310 -1.85731
N 3.29266 4.32595 -3.16418
H 1.48859 8.31415 -1.11623
H 3.41957 8.61674 -2.59273
H 3.08881 5.63528 -4.87021
H 3.31228 3.83282 -1.06920
H 2.47186 8.23184 -4.03956
C -7.05607 -3.40564 -1.28931
O -7.52258 -2.72175 -0.34446
C -7.20217 -2.93288 -2.73607
C -5.89040 -2.27583 -3.26041
C -5.36781 -1.16098 -2.40332
C -4.10881 -0.90347 -1.90491
N -6.14347 -0.08181 -1.98494
C -5.36289 0.77407 -1.27394
N -4.12665 0.28986 -1.20230
H -6.07882 -1.90371 -4.28106
H -3.18684 -1.45533 -2.03671
H -5.69987 1.69813 -0.82326
H -5.11039 -3.04599 -3.34101
N -6.38560 -4.56203 -1.12020
C -5.93809 -5.02451 0.20025
C -4.75162 -4.18508 0.72721
C -3.69810 -4.01319 -0.36803
C -2.49291 -3.15142 -0.06278
O -2.57367 -2.28097 0.87628
O -1.51776 -3.32392 -0.85624
H -4.15603 -3.54109 -1.24722
H -6.78573 -4.96977 0.89551
H -4.30911 -4.66786 1.60885
H -3.32466 -4.97901 -0.71897
H -5.64124 -6.07392 0.08145
H -5.13475 -3.21309 1.04937
C -9.35234 -0.85467 2.14347
C -8.62787 0.29688 1.40874
C -7.13926 0.24958 1.57969
C -6.22228 1.17436 2.03368
N -6.40642 -0.88941 1.26111
C -5.10572 -0.66651 1.49354
N -4.97589 0.57220 1.98085
H -9.03986 -1.82639 1.73667
H -8.98549 1.27323 1.76534
H -6.36262 2.18411 2.39680
18
H -4.27117 -1.33172 1.28735
H -8.86663 0.23706 0.33311
H -9.12990 -0.83177 3.21972
H -10.43853 -0.76086 2.00878
C -1.70133 6.27740 1.50928
O -1.23506 5.73863 0.50670
C -3.13115 6.03050 2.01258
C -3.88752 5.03792 1.11975
C -3.23370 3.65736 1.03685
O -2.40420 3.35035 1.95901
O -3.59794 2.90773 0.06296
H -3.96169 5.41827 0.09255
H -4.91690 4.91040 1.49253
N -0.94926 7.13000 2.29186
C 0.27202 7.72955 1.74550
C 1.44979 6.76467 1.54905
O 2.39622 7.11600 0.82049
H 0.06749 8.17517 0.76042
N 1.41437 5.59269 2.22986
C 2.32035 4.47727 1.95545
C 1.50592 3.26850 1.43603
O 1.92458 2.58609 0.48240
C 3.10872 4.05962 3.21447
O 0.43735 3.06268 2.15276
H 3.79096 3.22998 2.97075
H 3.01465 4.79881 1.17210
H 2.41625 3.73689 4.00513
H 0.54699 5.35375 2.70160
H 3.70531 4.90592 3.58018
C 0.38681 -1.41159 7.19023
C 0.25316 -1.00270 5.71549
C -1.12291 -0.34788 5.42949
C -1.24785 -0.00697 3.95126
O -1.19479 1.24026 3.63216
O -1.34904 -0.99530 3.15069
H -1.24915 0.56374 6.02941
H -0.39378 -2.13720 7.46698
H 0.36421 -1.88074 5.06289
H -1.91993 -1.06222 5.68750
H 0.28815 -0.53900 7.85498
H 1.05060 -0.29539 5.43929
H 1.36581 -1.87428 7.38088
C 3.91338 -0.24127 7.69247
C 4.41184 -0.69924 6.31248
C 4.42122 0.45359 5.29519
C 4.95646 0.03961 3.90916
N 4.25605 -1.10305 3.31606
C 2.96278 -1.09549 2.93761
N 2.24578 0.04767 2.96738
19
N 2.36662 -2.24731 2.59294
H 1.48572 -2.22196 2.04873
H 3.89496 -1.07624 8.40785
H 3.76818 -1.50773 5.93097
H 5.05317 1.27845 5.66362
H 6.00819 -0.26584 3.99118
H 4.73716 -2.00883 3.26690
H 2.76474 0.91470 2.98403
H 2.95123 -3.07333 2.51050
H 2.89314 0.16570 7.62117
H 5.42883 -1.11934 6.39793
H 3.39987 0.85357 5.19774
H 4.92861 0.89476 3.21001
H 1.34864 0.09891 2.44309
H 4.56430 0.54574 8.10399
O 0.55005 -4.57962 0.50720
H -0.65487 -5.08190 1.08291
H 0.70926 -5.34561 -0.07929
O 0.72585 -4.20909 -2.09679
H 0.95767 -3.66463 -2.89565
O 1.10663 1.23980 -3.14925
H 1.98961 1.39168 -3.57879
H 0.41843 1.73415 -3.65170
O -0.96795 2.25648 -0.06600
O 0.07720 0.47286 1.42256
O -2.46417 0.73089 1.29036
O 0.13370 -1.99914 0.92060
O -1.42066 -0.30148 -0.65456
O 1.57807 -2.82558 -4.35183
H 2.55437 -2.93197 -4.33947
H 1.30837 -3.21250 -5.19707
O -4.39571 -5.35978 -3.06061
H -4.52111 -5.96118 -3.80734
H -3.40634 -5.39521 -2.88695
O -1.75072 -5.68885 -2.73652
H -1.53439 -6.34213 -2.04153
H -1.10271 -4.97691 -2.54529
Mn -2.39500 1.25605 -0.54371
Mn -1.01528 1.87246 1.72775
Mn -1.18024 -0.86960 1.09644
Mn 0.45138 -3.04563 -0.66162
Ca 0.93884 0.61737 -0.86826
O 0.47756 -1.58515 -1.77157
H -0.25666 -1.53967 -2.44018
H 2.85757 6.32996 -0.69778
H -5.85114 -4.93909 -1.91689
H 0.41594 6.74667 -3.16952
H -2.76461 6.45539 -1.57847
H -2.53218 6.83725 -3.29672
20
H -3.08278 5.65786 3.04676
H -3.66939 6.99166 2.03241
H 0.59877 8.52610 2.42792
H -8.03952 -2.22220 -2.76481
H -7.44834 -3.77527 -3.39803
H -7.13786 0.01947 -2.14140
H -6.79696 -1.72715 0.76675
H 8.96527 -2.44898 -0.75560
H 7.36227 -2.77361 -1.46286
H 8.04971 -3.21320 2.61284
H 1.06354 9.45827 -2.40424
H -1.42886 7.64680 3.01967
H 6.84763 -4.41624 0.05297
H 6.89887 -4.29845 4.60161
H 3.97885 -4.24708 0.50983
O -1.22501 -0.80859 -3.73300
H -1.53551 0.01640 -3.30095
H -0.41722 -0.52436 -4.19266
H -4.01345 0.97248 2.01379
21
|
1001.5355 | 1 | 1001 | 2010-01-29T08:39:01 | Frequency-dependent electrodeformation of giant phospholipid vesicles in AC electric field | [
"physics.bio-ph"
] | A model of vesicle electrodeformation is described which obtains a parametrized vesicle shape by minimizing the sum of the membrane bending energy and the energy due to the electric field. Both the vesicle membrane and the aqueous media inside and outside the vesicle are treated as leaky dielectrics, and the vesicle itself is modelled as a nearly spherical shape enclosed within a thin membrane. It is demonstrated (a) that the model achieves a good quantitative agreement with the experimentally determined prolate-to-oblate transition frequencies in the kHz range, and (b) that the model can explain a phase diagram of shapes of giant phospholipid vesicles with respect to two parameters: the frequency of the applied AC electric field and the ratio of the electrical conductivities of the aqueous media inside and outside the vesicle, explored in a recent paper (S. Aranda et al., Biophys. J. 95:L19--L21, 2008). A possible use of the frequency-dependent shape transitions of phospholipid vesicles in conductometry of microliter samples is discussed. | physics.bio-ph | physics | Journal of Biological Physics manuscript No.
(will be inserted by the editor)
Primoz Peterlin
Frequency-dependent electrodeformation of
giant phospholipid vesicles in AC electric
field
Received: date / Accepted: date
Abstract A model of vesicle electrodeformation is described which obtains a
parametrized vesicle shape by minimizing the sum of the membrane bending
energy and the energy due to the electric field. Both the vesicle membrane and
the aqueous media inside and outside the vesicle are treated as leaky dielectrics,
and the vesicle itself is modelled as a nearly spherical shape enclosed within a
thin membrane. It is demonstrated (a) that the model achieves a good quanti-
tative agreement with the experimentally determined prolate-to-oblate transition
frequencies in the kHz range, and (b) that the model can explain a phase dia-
gram of shapes of giant phospholipid vesicles with respect to two parameters: the
frequency of the applied AC electric field and the ratio of the electrical conduc-
tivities of the aqueous media inside and outside the vesicle, explored in a recent
paper (S. Aranda et al., Biophys. J. 95:L19 -- L21, 2008). A possible use of the
frequency-dependent shape transitions of phospholipid vesicles in conductometry
of microliter samples is discussed.
Keywords electrodeformation · giant phospholipid vesicle · leaky dielectric ·
membrane bending energy · vesicle shape
PACS 87.16.D- · 41.20.Cv
1 Introduction
An increasing awareness of the effects of the electromagnetic field on biological
samples [1], as well as its use in biotechnology [2] has instigated numerous studies
P. Peterlin
University of Ljubljana, Faculty of Medicine, Institute of Biophysics,
Lipiceva 2, SI-1000 Ljubljana, Slovenia,
Tel.: +386-1-5437612
Fax.: +386-1-4315127
E-mail: [email protected]
0
1
0
2
n
a
J
9
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
5
3
5
.
1
0
0
1
:
v
i
X
r
a
2
Primoz Peterlin
Fig. 1 Morphological phase diagram of lipid vesicle shapes subjected to an AC electric field.
The two parameters spanning the diagram are the frequency of the applied AC electric field
and the ratio of the electrical conductivities of the aqueous media inside and outside the vesicle
(from ref. [6]; reproduced by permission of The Royal Society of Chemistry).
on the effects of electric field on individual biological cells. When exposed to an
AC electric field, biological cells exhibit several connected phenomena such as
orientation, dielectrophoresis, electrorotation and deformation [3,4,5]. Of these,
deformation has probably received the least attention, despite the fact that it can
be conveniently studied on giant unilamellar phospholipid vesicles (GUVs) [6,7],
which serve as models of biological cells.
Extensive studies of the behaviour of biological cells in the electric field have
been conducted by Schwan since the 1950's [8]. However, these early studies
treated cells as rigid objects. A theory of lipid bilayer elasticity [9] was needed
for the first model of vesicle electrodeformation [10]. This first model predicted
a prolate spheroidal deformation; this prediction was later confirmed both by im-
proved theoretical treatments [11,12] and by experiments [13], conducted in a
2 kHz AC electric field. However, experiments where the frequency of the field
was varied [14,15] have demonstrated that the vesicle shape undergoes a transi-
tion from prolate shapes at low frequencies to oblate shapes at higher frequencies.
Independently, such behaviour was also predicted by Hyuga and co-workers [16,
17], who described a prolate-to-oblate shape transition in certain conditions with
respect to the electrical properties of the aqueous medium inside and outside the
vesicle.
A recent paper [18] extended the studies of vesicle shape in electric field to
higher frequencies and explored the effect of varying electrical conductivity of the
internal and the external aqueous medium in a systematic manner, thus obtaining
a more complex morphological phase diagram (Fig. 1). The authors have demon-
strated that in the case when the conductivity of the external aqueous medium
exceeds the conductivity of the aqueous medium inside the vesicle, another shape
transitions exist in the MHz range in addition to the previously observed prolate-
to-oblate transition in the kHz range.
Frequency-dependent electrodeformation of GUVs
3
The rest of this paper is structured as follows: Section 2 presents the theoretical
model, which treats the vesicle as a thin shell made of leaky dielectric and derives
its shape by minimizing its total energy, consisting of membrane bending energy
and the energy due to the electric field. Section 3 describes the experiments in-
vestigating the prolate-to-oblate morphological shape transition in the kHz range.
The results -- a comparison of the model predictions with the experimental data
for the prolate-to-oblate transition on the one hand and with the experimental mor-
phological phase diagram (Fig. 1) on the other hand -- are presented in Section 4.
Section 5 discusses the limitations of the presented model, compares it with the
existing ones, and introduces a possible application of frequency-dependent tran-
sitions of vesicle shapes for conductometry. Finally, Section 6 presents the main
conclusions.
2 Theoretical analysis
In general, the model follows the approach introduced by Winterhalter and Hel-
frich [12], i.e., a parametrized vesicle shape is obtained by minimizing its total
free energy. Two terms enter the free energy: the membrane bending energy and
the energy due to the electric field. A closed-form expression for the membrane
bending energy is well-known [9], and the change of the free energy due to the
electric field is calculated as the work done by the force of the electric field while
deforming the vesicle [12].
Once an electric field E is applied, it introduces a single distinct axis into the
system, so the treatment can be limited to axially symmetric shapes. This elim-
inates the dependence on the longitude angle f , if the polar (z) axis is chosen
parallel to the applied field. Thus, the vesicle surface can be parametrized as
r(q ) = s0 + s(q ) ,
(1)
where s(q ) ≪ s0, s0 denoting the deformation independent of the polar angle
q . In an absence of deformation (s(q ) = 0), s0 equals the radius of the unde-
formed sphere (r0). The deformation is independent of the sign of the electric
field and thus proportional to E2 in the lowest order. The field itself being pro-
portional to cosq , the deformation coupled with this field can be expected to be
proportional to cos2 q . In terms of expansion into spherical harmonics, this limits
us to a sum of even terms. Retaining only the terms proportional to E2 or lower,
a quadrupolar term remains, where s(q ) equals the second Legendre polynomial:
s(q ) = 1
2 s2(3cos2 q − 1), with s2 being a measure for the extent of deformation.
Positive values of s2 indicate prolate deformation, while negative values indicate
oblate deformation.
2.1 Membrane bending energy
The requirement for a local area conservation, which assures that the membrane
stretching is independent of polar angle q , implies that a quadrupolar displacement
4
Primoz Peterlin
d rr in a radial direction is accompanied by a tangential displacement d rq [12]:
1
2
(3cos2 q − 1)s2
d rr =
d rq = −cosq sinq s2 .
(2)
(3)
The total membrane area expansion is determined by the relationship between s0
and r0. Taking into account the requirement for a constant vesicle volume, the
following relationship for s0 is obtained:
s0 = r0 −
s2
2
5r0
.
(4)
The correction for a constant volume (4) contains a higher term in the powers of s2
and thus does not affect in the lowest term either the bending energy or the energy
due to the electric field, yielding s0 = r0 an adequate approximation.
The expression for the vesicle bending energy [9]:
Gbend =
1
2
kcI (c1 + c2 − c0)2 dA .
(5)
Here, kc is the bending elastic modulus of the membrane, while c1 and c2 are
the principal curvatures of the membrane. The spontaneous curvature c0 vanishes
for a bilayer composed of two equal layers. The integration is conducted over the
total membrane area A of a quadrupolarly deformed vesicle. Up to quadratic order
terms in s2, the total bending energy of a nearly spherical vesicle can be expressed
as [9]:
Gbend = 8p kc +
48p
5
r0(cid:19)2
kc(cid:18) s2
.
(6)
This can be readily interpreted as the bending energy of a sphere plus an addition
due to the quadrupolar deformation.
2.2 Contribution of the electric field
The first step in evaluating the contribution of the electric field to the total free
energy is calculating the electric field. The present treatment is limited to small
deviations of vesicle shape from the sphere, which greatly simplifies the treat-
ment: instead of computing the electric field in the presence of the actual vesicle
shape, one can compute the electric field in the presence of a spherical shell. Both
the aqueous solution inside and outside the vesicle and the vesicle membrane are
treated as lossy dielectrics. The aqueous solutions inside and outside the vesicle
usually have identical dielectric permittivity, while their electrical conductivities
can differ (Fig. 2).
In order to compute the forces exerted by the electric field on the vesicle, one
first has to compute the electric field around a vesicle. Gauss' law (cid:209)
· D = 0 to-
gether with the requirement for an irrotational electric field (cid:209) × E = 0, which
stems from Faraday's law for electromagnetic induction, yields the Laplace equa-
tion for the electric potential U [19]:
(cid:209) 2U = 0 .
(7)
Frequency-dependent electrodeformation of GUVs
5
E
r
r
(1)
(2)
r
0
z
gm,
cm
in
ine
s ex e ex
Fig. 2 A vesicle is modelled in spherical coordinates (r,q ) as a spherical shell with a radius r
exposed to an external electric field E. The conductivity and permittivity of the aqueous solution
outside are denoted by s ex and e ex, and the corresponding quantities in the vesicle interior by
s in and e in. Membrane surface transconductance and surface capacitance are denoted by gm and
cm, respectively.
The boundary conditions on the membrane (r = r0) require that the total sur-
face charge density, including both free charge and displacement charge, must
vanish. The case where a finite potential can be supported across the membrane is
known as the series admittance limit ([3], pp. 230-232). In the spherical geometry,
this yields a system of equations
(s ex − iwe
(s ex − iwe
ex)E(1)
ex)E(1)
in)E(2)
r (r0) = (s in − iwe
r (r0) = (gm − iw cm)(cid:16)U (2)(r0) −U (1)(r0)(cid:17) .
r (r0) ,
(8)
(9)
Apart from (8) and (9), the system is constrained by two additional conditions: one
requiring that the electric field far away from the vesicle is unperturbed, and the
other requiring that the electric field is finite inside the vesicle. Membrane surface
transconductance gm and surface capacitance gm are related to membrane electric
conductivity and dielectric permittivity: gm = s m/h and cm = e m/h, h being the
membrane thickness.
In spherical coordinates, (7) readily decouples into the radial and the angular
part, and can be solved with the usual ansatz:
U (k) =
1
2" a(k)r +
b(k)
r2 ! cosq e−iw t + C.C.# .
(10)
The coefficients a(k), b(k); k = 1, 2, which can in general be complex to allow for a
phase shift, are determined from the boundary conditions (8,9).
The conditions for an unperturbed field far away from the vesicle and a finite
field inside the vesicle immediately yield two coefficients:
a(1) = −E0 ,
b(2) = 0 .
(11)
(12)
s
q
6
Primoz Peterlin
The two remaining coefficients are obtained by solving (8,9):
b(1) = −
a(2) = −
3E0r
D
E0r3
D
ex)(cid:21) ,
(cid:20)(gm − iw cm)(s ex − iwe
(cid:20)s exs in + gmr(s ex − s in) − (e exe in + cmr(e ex − e in))w 2
− i (gmr(e ex − e in) + cmr(s ex − s in) + e ins ex + e exs in)w (cid:21) ,
D = 2s exs in + gmr(2s ex + s in) − (2e exe in + cmr(2e ex + e in))w 2
− i (gmr(2e ex + e in) + cmr(2s ex + s in) + 2(e ins ex + e exs in))w
(13)
(14)
.
(15)
The surface density of the forces exerted on the boundary of dielectrics by the
electric field can be computed as a scalar product of the Maxwell stress tensor and
a vector normal to the membrane:
f = (T (1) − T (2))er .
(16)
The force vanishes in a homogeneous medium, but can in general be non-zero
on the boundaries of media with different electrical properties. The Maxwell stress
tensor is defined as
T = D ⊗ E − 1
2 (D · E)I ,
(17)
where I denotes the identity matrix.
Unlike in the case of the bending energy term (6), for which a closed-form
expression was obtained, an approach where energy difference is computed is
employed here. d Gfield denotes a small change in the energy due to the electric
field, when a sphere (s2 = 0) is perturbed by a small deformation change d s2. This
energy difference is calculated as the work done by the forces of the electric field
during the displacement of membrane elements d r (equations 2,3), integrated over
the entire membrane area [12]:
d Gfield = −I (f · d r)dA .
(18)
Substituting the coefficients (11 -- 15) into (18) yields a lengthy expression for
The integration is conducted over a sphere, which is consistent with the limit of
small deformations (s2 ≪ r0).
d Gfield, which can be written as a sum of two dispersion terms:
x 2
x 1
d Gfield = −
6p
5
e wE2
0 r2
outs2(cid:18)x ¥ +
1 + w 2t 2
1
+
1 + w 2t 2
2(cid:19) .
(19)
, x 1, and x 2, and two characteristic times t 1
Three dimensionless coefficients x ¥
and t 2 are rather lengthy expressions involving six different material constants:
e ex, e in, s ex, s in, cm, gm, and the vesicle radius r0.
Frequency-dependent electrodeformation of GUVs
7
2.3 Vesicle deformation
For small deformations, quadrupolar deformation only induces small perturbative
changes in the vesicle bending energy and the energy due to the electric field.
(20)
(21)
Gbend(s2) ≈ Gbend(s2 = 0) + 1
2
Gfield(s2) ≈ Gfield(s2 = 0) +
¶ 2Gbend
¶ s2
2
¶ Gfield
¶ s2
s2
2
(cid:12)(cid:12)(cid:12)s2=0
(cid:12)(cid:12)(cid:12)s2=0
s2
Even though the total energy of the vesicle formally depends on two parameters,
r0 and s2, the constraint requiring a constant vesicle volume eliminates one degree
of freedom, thus yielding (20 -- 21). It is also worth noting that the fact that neither
prolate (s2 > 0) nor oblate shapes (s2 < 0) have a bending energy lower than those
of a sphere means that the expansion for the bending energy (20) contains no linear
term.
Equilibrium vesicle deformation, expressed in terms of s2, can then be calcu-
lated by minimizing the total free energy over s2:
d
ds2
(Gbend + Gfield) = 0 .
(22)
Expressing the equilibrium vesicle deformation s2 from (22) at given condi-
tions yields:
s2 =
r4
0
1
16
e wE2
kc (cid:18)x ¥ +
0
x 1
1 + w 2t 2
1
+
x 2
1 + w 2t 2
2(cid:19) .
(23)
As one can see, the dependence of vesicle deformation s2 on the angular fre-
quency w = 2pn
contains two dispersion terms. It is worth emphasizing that the
only approximation used in deriving the expression (23) is that of a small defor-
mation (s2 ≪ r0). Also worth noting is the fact that the only parameter related to
membrane elasticity, kc, only scales s2; frequency-dependent electrodeformation
of vesicles is entirely governed by the electrical parameters of the membrane and
the aqueous medium and by the vesicle size.
Fig. 3 shows vesicle deformation in dependence of the angular frequency w
of the applied AC electric field. Its most prominent feature is that the deformation
exhibits a prolate-to-oblate transition at w ∼ 104 Hz (i.e., n ∼ 103 Hz) in the case
when s in < s ex (e.g., s in = 0.8s ex, solid curve). On the other hand, vesicle shape
remains prolate up to w ∼ 107 Hz (n ∼ 106 Hz) when s in > s ex (e.g., s in = 1.3s ex,
coarsely dashed curve). In both cases, a transition to spherical shape (s2 ≈ 0) is
observed at w ∼ 107 Hz (n ∼ 106 Hz). All these features are in agreement with the
experimental findings [18]. The transition to spherical shape, however, occurs at as
low as w ∼ 104 Hz when the conductivities inside and outside are equal, s in = s ex
(Fig. 3, finely dashed curve). Some of the points in Fig. 1 around s in/s ex = 1 may
indicate that such transitions have also been observed. It needs to be pointed out,
however, that this is not a distinguished property of the s in = s ex case, but merely
a coincidence at given values of material properties and at given vesicle size.
, x 1, x 2, t 1 and t 2 figuring in
(23) can be somewhat simplified. First of all, it is reasonable to assume that the
The lengthy expressions for the coefficients x ¥
8
0
r
/
2
s
5
1
.
0
0
1
.
0
5
0
.
0
0
0
.
0
5
0
.
0
−
Primoz Peterlin
in/s ex = 0.8
in/s ex = 1.0
in/s ex = 1.3
3
.
1
2
.
1
1
.
1
a
/
c
1
9
.
0
101
102
103
104
105
106
107
108
109
[Hz]
Fig. 3 Vesicle deformation as a function of the frequency w = 2pn
of the applied AC electric
field. Vesicle deformation is shown both in terms of quadrupolar deformation s2 normalized
to r0 (primary y-axis, left) and vesicle semi-axes ratio c/a ≈ (r0 + s2)/(r0 − s2/2) (secondary
y-axis, right). The three curves were computed for three different ratios between the electrical
conductivity of the external aqueous medium and the electrical conductivity of the internal aque-
ous medium. The dotted line separates prolate shapes from oblate ones. Other parameters used
for computation were r0 = 20 m m, s ex = 50 m S/cm, s m = 10−14 S/m, e w = 80e 0, e m = 2.5e 0,
h = 4 nm, kc = 1.2 × 10−19 J, E0 = 500 V/m.
dielectric permittivity of the aqueous medium inside the vesicle does not differ sig-
nificantly from the dielectric permittivity of the external medium, e ex = e in ≡ e w
(this is an approximation; it is known that the dielectric permittivity of water de-
creases with the increasing concentration of salt in it; e.g., the dielectric permit-
tivity of physiological saline is only ≈ 72 e 0, [20]). A further simplification is
possible in the case when realistic values for material parameters are taken into
account: (a) the conductivity of the aqueous solution greatly exceeds that of the
membrane, s w ≫ s m, and (b) with e w ≈ 80e 0, e m ≈ 2.5e 0, and r0/h ∼ 1000, then
e w ≪ (r0/h)e m. Retaining only the largest terms, one obtains:
x ¥ ≈
t 1 ≈
t 2 ≈
3
e w(4r0cm + e w)
(3r0cm + 2e w)2 ,
e w
s ex
r0cm
s ex
2 + x
2 + x
2x
,
.
(24)
(25)
(26)
Here, x = s in/s ex has been introduced for the ratio of the conductivities of the
aqueous medium inside and outside the vesicle.
w
s
s
s
Frequency-dependent electrodeformation of GUVs
9
1.0
0.5
0.0
-0.5
Ξ1
Ξ2
0.01
0.1
1
ΣinΣex
10
100
Fig. 4 The coefficients x 1 and x 2 figuring in (23) plotted for different values of the ratio of the
internal and the external electrical conductivity, x = s in/s ex. Other parameter values used for
computation were the same as in Fig. 3.
The two characteristic times are related to two distinct physical processes: t 1 is
the relaxation time for the Maxwell-Wagner interfacial polarization, and t 2 is the
charging time of a spherical capacitor. In the literature (see, e.g., [21,16,22,3]) the
expressions for them are often encountered in a slightly more general form, which
distinguishes between the dielectric permittivity of the internal and the external
aqueous medium.
As expected, x ¥ does not depend on conductivities, and the curves for dif-
ferent s in/s ex plotted in Fig. 3 merge into a single one at w → ¥
. One can also
see that in the high-frequency limit, the deformation is always slightly prolate,
x ¥ ≈ (h/r0)(e w/e m) . 0.01. Such subtle deformation is however in reality likely
to be subdued by the thermal fluctuations, and is therefore difficult to detect ex-
perimentally.
Fig. 4 shows how the coefficients x 1 and x 2 change when the ratio of the in-
ternal and the external electrical conductivity x = s in/s ex is varied within two
decades at given values of material parameters. On the interval x ∈ [1/100, 100], x 1
and x 2 can be approximated by rational functions of x:
x 1 ≈ −
x 2 ≈
(rocm + e w)(3r0cm + 4e w) + (rocm + e w)(3r0cm + 4e w)x
m + 13rocme w + 12e 2
m + 5rocme w + 4e 2
4(r2
w) + (3r2
0c2
0c2
w)x
m + 15rocme w + 4e 2
m + 3rocme w + 2e 2
0c2
16(r2
0c2
w) − (9r2
w)x
m + 13rocme w + 4e 2
m + 20rocme w + 24e 2
w) + 4(4r2
0c2
0c2
4(3r2
,
(27)
.
(28)
w)x
It is possible to derive an approximate expression for the critical value of x,
below which the prolate-to-oblate transition occurs. In the intermediate region,
1/t 2 < w < 1/t 1, the function x (w ; x) = x ¥ + x 1/(1 + w 2t 2
1 ) + x 2/(1 + w 2t 2
2 )
figuring in (23) approximately evaluates to x (w ; x) ≈ x ¥ + x 1. This needs to be
negative in order for an oblate deformation. From this requirement, an approxi-
10
Primoz Peterlin
a
b
c
Fig. 5 Phase-contrast micrograph of a vesicle in an AC electric field in response to the frequency
of the applied field: (a) 1 kHz, (b) 11 kHz, (c) 29 kHz. The direction of the electric field corre-
sponds to the horizontal direction. The bar represents 10 m m. The conductivity of the aqueous
medium is s in = 17 m S/cm.
mate expression for xcrit can be obtained:
xcrit ≈
27 r0cm + 93 e w
27 r0cm + 111 e w
.
(29)
Substituting realistic values for material parameters and vesicle size into (29), one
finds that xcrit ≈ 0.97 for a vesicle with r0 = 5 m m, xcrit ≈ 0.986 for a vesicle with
r0 = 10 m m, and xcrit ≈ 0.997 for a vesicle with r0 = 50 m m.
3 Experiment
3.1 Vesicle preparation
Giant unilamellar vesicles (GUVs) were prepared from commercially available
(Avanti Polar Lipids, Alabaster, AL, USA) synthetic 1-palmitoyl-2-oleoyl-sn-gly-
cero-3-phosphocholine (POPC) in double-distilled water using the electroforma-
tion method [23,24] in a chamber which allows for an easy access of the resulting
vesicle suspension [25]. In order to alter the conductivity of the solution, sodium
chloride was added to the solution in concentration of up to 0.2 × 10−3 mol/L.
3.2 Experimental procedure
The experimental chamber consisted of two 0.1 mm thick stainless steel electrodes
mounted on an object glass, leaving a 0.6 mm wide gap between the electrodes.
The cover slip was mounted with vacuum grease (Baysilone; Bayer, Leverkusen,
Germany), enclosing a few drops of vesicle suspension. The chamber was placed
onto an inverted light microscope (Zeiss/Opton IM 35, objective Zeiss Ph2 Plan
40/0.60; Zeiss, Oberkochen, Germany), and phase contrast technique was used.
The micrograph was recorded using a CCD camera (Cohu 6700; Cohu, San Diego,
USA), taped using a U-Matic VCR (Sony VO-9800P) and later digitized on a PC
with a frame grabber (Matrox Meteor II; Matrox, Dorval, Canada). The same com-
puter was used to set the voltage and frequency of the applied electric field, driving
a GPIB-controlled function generator (Iskra MA-3735; Iskra, Horjul, Slovenia).
Frequency-dependent electrodeformation of GUVs
11
increasing frequency
decreasing frequency
1.10
1.05
a
/
c
1.00
0.95
0
1
2
3
4
5
6
7
[kHz]
Fig. 6 Prolate-to-oblate transition of a phospholipid vesicle in the kHz range. Vesicle deforma-
tion is expressed as the ratio of its semi-axis in the direction along the field (c) and its semi-axis
perpendicular to the field (a). Vesicle radius r0 = 19.8 ± 0.1 m m, conductivity of the aqueous
medium s in = 4.6 m S/cm.
Using an acoustic coupler connected to RS-233 port and a telephone handset with
an audio jack, the data about the selected voltage and frequency were simultane-
ously recorded onto one audio channel, while the other audio channel was left for
the experimentalist's comments during the course of experiment [26].
In the experiment, the prolate-to-oblate transition frequency for a given vesicle
was first coarsely estimated by testing the effect of the electric field at 2 -- 3 differ-
ent frequencies. After this coarse estimate, a programmed sequence of step-wise
frequency change was applied. The frequency step ranged from 100 Hz to 1 kHz,
depending on the vesicle, and the duration of the step was approximately 3 s. This
duration was experimentally selected as being long enough for the vesicle to adapt
to a small frequency change, yet short enough to minimize dielectrophoretic drift
and allow repetitive measurements. The programmed sequence was repeated sev-
eral times with increasing and decreasing frequencies in order to observe possible
hysteresis (Fig. 6). To allow a more precise determination of transition frequency,
the range was narrowed and the frequency steps were decreased in the later runs
(symbols (cid:3), △, and ▽ in Fig. 6).
4 Results
4.1 Analysis of the experimental data
A total of 46 vesicle recordings were examined. After discarding the recordings
with a vesicle relative volume too close to 1, where no noticeable shape change
n
12
Primoz Peterlin
17. m S/cm
4.6 m S/cm
1.3 m S/cm
]
z
H
k
[
e
t
a
l
b
o
e
t
a
l
o
r
p
25
20
15
10
5
0
0
5
10
15
20
25
30
35
r0 [m m]
Fig. 7 The dependence of the prolate-to-oblate transition frequency on the vesicle size and the
conductivity of the aqueous medium. Three runs of experiments with three different conduc-
tivities of the aqueous medium were performed: 17, 4.6, and 1.3 m S/cm. The solid line corre-
, with s in = 17 m m, and the dashed line with
sponds to numerically solving s2 = 0 (Eq. 23) for w
s in = 4.6 m m. In both cases, s in/s ex = 0.9. Other parameter values used for computation were
the same as in Fig. 3.
was observed, as well as the recordings in which the experiment was interrupted
by the dielectrophoretic drift of a vesicle to a region where further observations
were not possible, 21 measurements on different vesicles were taken into con-
sideration for analysis. Of these, 8 were prepared in a medium with conductivity
17 m S/cm, 3 in a medium with conductivity 4.6 m S/cm and 10 in a medium with
conductivity 1.3 m S/cm. The transition frequency for the prolate-to-oblate transi-
tion vs. vesicle size is plotted in Fig. 7. Transition frequencies are reciprocally
related to the vesicle radius, which is consistent with the recently published find-
ings of another group (K. Antonova et al., in press).
The existence of prolate-to-oblate transitions indicates that the ratio s in/s ex
in the experiments was below xcrit, even though the compositions of the aqueous
solution in the vesicle interior and the vesicle exterior were initially identical. We
would like to propose an explanation for this phenomenon. First, we need to em-
phasize that the conductometer electrode we used requires at least 6 ml of sample,
therefore it was impossible to measure directly the conductivity of the vesicle so-
lution. Instead, the conductivity of the aqueous medium was measured before it
was used for hydrating the phospholipid film. The conductivity of the aqueous
solution may increase afterwards by the impurities introduced either at the elec-
troformation process, during storage or during sample preparation. In order to test
this hypothesis, we performed a separate experiment, where we simulated the ma-
nipulation of a vesicle sample. Two flasks were filled with 0.03 mmol/L solution
of NaCl in 0.1 mol/L sucrose solution (s = 14.5 m S/cm). The first flask was kept
n
-
Frequency-dependent electrodeformation of GUVs
13
sealed in cold storage during the course of the experiment, while the second flask
was brought to room temperature every day, and a small amount of the solution
was pipetted out for conductivity measurement before returning the flask back to
cold storage. While the conductivity of the solution in the first flask remained un-
changed, the conductivity of the solution in the second flask increased by ≈ 8%
with each consecutive pipetting, reaching 31.5 m S/cm after 11 days. We can pre-
sume that similar processes occur in vesicle suspension, where the conductivity of
the solution in the vesicle interior remains unchanged as the phospholipid mem-
brane is virtually impermeable for ions, while the conductivity of the solution in
the vesicle exterior increases with each successive pipetting. Since even for the
smallest of the vesicles shown in Fig. 7 (r0 ≈ 5 m m), xcrit ≈ 0.97 (cf. Eq. 29), it is
clear that at such low conductivities of the solution, an 8% increase of the conduc-
tivity of the external medium caused by a single pipetting is sufficient to bring the
vesicle suspension into the regime where oblate shapes exist in the intermediate
frequency range.
The two computed lines in Fig. 7 were obtained by numerical root-finding of
s2 = 0, s2 being defined by (23). The initial conductivity of the solution (17 m S/cm
and 4.6 m S/cm, respectively) was taken as the conductivity of the internal aqueous
medium, while the conductivity of the external aqueous medium is 1/0.9 s in, or
approximately 10% higher. The quantitative agreement with the measured data
at 17 m S/cm and 4.6 m S/cm is within the experimental error, while the data for
1.3 m S/cm coincide with those for 4.6 m S/cm. This may indicate that the actual
electrical conductivity s in in this latter case was not 1.3 m S/cm, but somewhat
higher (estimated around 5 m S/cm).
4.2 Morphological diagram
In order to plot a morphological phase diagram akin to the one shown in Fig. 1,
two distinct regimes have to be considered. For s in/s ex < 1, vesicle deformation
s2 (23) changes sign, and the prolate-to-oblate and the oblate-to-spherical transi-
tions can be computed by solving s2 = 0 numerically. For s in/s ex > 1, s2 remains
positive, and instead one has to rely on the characteristic times (25, 26). Here,
the relaxation term containing t 1 corresponds to prolate-to-spherical transition in
Fig. 1. The relaxation term corresponding to t 2 corresponds to an unobserved tran-
sition, i.e., a prolate vesicle is expected to become slightly less prolate (see also
the dashed curve in Fig. 3) at frequencies w & 1/t 2. Such subtle changes, how-
ever, are probably difficult to quantify experimentally, and have therefore not been
reported in [18].
The results, shown in Fig. 8, show a reasonable agreement with the experimen-
tal diagram (Fig. 1). In particular, for s in < s ex, the prolate-to-oblate transition
frequency increases with increasing s in/s ex. The prolate-to-spherical transition
for s in > s ex also increases with increasing s in/s ex, while the oblate-to-spherical
transition, observed for s in < s ex, decreases with increasing s in/s ex. All these
observations are in agreement with the experiment.
We would also like to comment on the reported coexistence of prolate and
oblate vesicle shapes at x ≈ 1 in the intermediate frequency range [18]. As we
have shown earlier (Fig. 7), the prolate-to-oblate transition frequency exhibits
a strong dependence on vesicle size: the transition frequency for a fairly large
14
Primoz Peterlin
x
e
Σ
n
i
Σ
10.0
5.0
2.0
1.0
0.5
0.2
0.1
100
1Τ2
prolate
1Τ1
l
a
c
i
r
e
h
p
s
prolate
oblate
104
106
Ω Hz
108
Fig. 8 Prolate and oblate vesicle deformation plotted for different values of the ratio of the
internal and the external electrical conductivity, x = s in/s ex. In the x < 1 region, the sign changes
in (23), and either prolate-to-oblate transition or oblate-to-spherical transitions are computed,
while for x > 1, reciprocal characteristic times 1/t 1 and 1/t 2 are plotted. Other parameter values
used in evaluation were the same as in Fig. 3. The dashed curve corresponds to the relaxation
which does not manifest itself as a morphological shape transition.
vesicle (r0 = 50 m m) is n 2 ≈ 3 kHz, while for a small vesicle (r0 = 5 m m) it is
n 2 ≈ 30 kHz. Within this frequency interval, it is thus quite feasible to observe a
small prolate vesicle simultaneously with a large oblate vesicle. While the authors
do not comment on the relative sizes of the vesicles in question, we believe this
phenomenon can be explained if: (a) the s in/s ex ratio was below xcrit (i.e., slightly
below 1), thus making oblate shapes possible in the intermediate frequency range,
and (b) the frequency was above the prolate-to-oblate transition frequency for the
larger vesicle and below the prolate-to-oblate transition frequency for the smaller
vesicle.
5 Discussion
5.1 Comparison with earlier results
While the model presented above builds on the legacy of the model developed
over 20 years ago by Winterhalter and Helfrich [12], it departs from it in two
aspects. Firstly, it allows for a different electric conductivity in the vesicle inte-
rior and the vesicle surroundings, and secondly, it is simplified by treating the
vesicle membrane as a thin shell right from the start. The first modification was
a necessity imposed by the experiments presented in [18]. The limitation to thin
shells is appropriate since any results applicable to giant vesicles fall into this
limit (g ≈ 1 in [12]). The results presented here thus do not differ significantly
from those obtained by the authors who started from a general finite-thickness
model (T. Yamamoto et al., to appear). Treating membrane thickness as finite is,
however, always required when when one attempts to tackle membrane dielectric
anisotropy [27,28,29,30,31].
Frequency-dependent electrodeformation of GUVs
15
Hyuga et al. [16] provide treatment both for the vesicle deformation in an
AC electric field and the response of the vesicle to a step-wise change in E, and
arrive at the expressions (25, 26). However, they later conclude that the fast mode
dampens very quickly and that only the effects of the slow mode are expected to
show a visible influence on deformation dynamics, and consequently focus on the
lower (i.e., prolate-to-oblate) transition.
More recently, a theoretical analysis of the experimental results presented in
[18] has been published by the same group [32], where a force-balance approach
is used instead of energy minimization. While this does not introduce any major
advantage in the problems treated here, i.e., the phase diagram of vesicle shapes,
it allowed a more correct prediction of not only the nature of deformation, but also
of its extent (see below). Even more important, by extending the treatment beyond
the equilibrium shape it opens a possibility of treating vesicle dynamics.
Finally, we find it necessary to relate the results in this paper with our own
earlier results [31], where we argued that the prolate-to-oblate transition can be
explained by the dielectric anisotropy of the membrane. In view of later exper-
imental findings [18], which put the prolate-to-oblate transition into a broader
context, as well as of the fact that even a very minute difference in the electrical
conductivities of the aqueous medium inside and outside the vesicle (cf. eq. 29)
makes prolate-to-oblate transition possible, we believe it is quite possible that the
observed prolate-to-oblate transition was caused primarily by a minute unintended
increase of the conductivity of the external aqueous medium, which may have
overshadowed other possible effects. Even though the frequency-dependent elec-
trodeformation of giant vesicles is likely to be primarily governed by other effects,
we however continue to believe that the dielectric anisotropy of the phospholipid
membrane is a phenomenon worth investigating.
5.2 Quantitative estimate of vesicle deformation
Though simple, the model presented in this paper provides a fairly accurate de-
scription of the experimentally obtained morphological phase diagram (Fig. 1),
i.e., it predicts the nature of deformation (prolate, oblate, or spherical), and the
frequencies at which the transitions between these shapes occur. However, being
limited by design to deformations close to spherical, this model, like the original
Winterhalter-Helfrich model, does not predict the extent of deformation correctly.
In the case of the Winterhalter-Helfrich model, this feature has already been noted
[18,32]. For the most part, the extent of deformation is determined by the relative
volume of the vesicle, i.e., a flaccid vesicle deforms more than a nearly spheri-
cal one. For simplicity, the treatment presented in this paper omits the effects of
thermal fluctuations, which have already been treated elsewhere [33,34].
5.3 Application in microliter conductometry
A quantitative model of vesicle shape with respect to the electrical conductivity of
the medium inside and outside the vesicle allows us to exploit this phenomenon
from the other end: a vesicle with a known internal conductivity and a measurable
size can serve as a probe for sensing the electrical conductivity of the medium
16
Primoz Peterlin
in its surroundings. For this potential application, the kHz-range prolate-to-oblate
transition seems particularly well-suited, because (a) it is an easily observable
qualitative change in the vesicle shape which occurs in a narrow frequency in-
terval, and (b) the transition frequency exhibits a strong dependence on vesicle
size, which allows for some tuning, i.e., in a heterogeneous vesicle population,
one can select the vesicle which manifests the effect most prominently. Recent
developments in vesicle preparation methods [35,36] also bring the physiological
conductivities into reach.
6 Conclusions
It has been demonstrated that a modified Winterhalter-Helfrich model [12] which
obtains a parametrized vesicle shape by minimizing the sum of the membrane
bending energy and the energy due to the electric field, can be successfully ap-
plied to (a) quantitatively explaining the experimentally observed prolate-to-oblate
shape transition in the kHz region, and (b) more generally, to the experimentally
obtained morphological phase diagram of GUVs with respect to two parameters,
the frequency of the applied AC electric field, and the ratio of the electrical con-
ductivities of the aqueous media inside and outside the vesicle [18]. The obtained
results are compared with the findings in the literature, and the limitations of the
model concerning the quantitative prediction of membrane deformation are dis-
cussed. Finally, an application of the observed effects in the conductometry of
microliter samples is proposed.
Acknowledgements The author would like to thank Prof. S. Svetina and Prof. B. Zeks for
numerous helpful discussions and V. Arrigler for the help with vesicle preparation. This work
has been supported by the Slovenian Research Agency through grant J3-2268.
References
1. Polk, C., Postow, E. (eds.): Handbook of biological effects of electromagnetic fields, 2nd
edn. CRC Press, Boca Raton, New York, London, Tokyo (1996)
2. Zimmermann, U., Neil, G.A. (eds.): Electromanipulation of Cells. CRC Press, Boca Raton
3. Jones, T.B.: Electromechanics of particles. Cambridge University Press, Cambridge, New
(1996)
York, Melbourne (1995)
4. Zimmerman, U., Friedrich, U., Mussauer, H., Gessner, P., Hamel, K., Sukhorukov, V.: Elec-
tromanipulation of mammalian cells: Fundamentals and application. IEEE Trans Plasma
Sci 28, 72 -- 82 (2000)
5. Gimsa, J.: A comprehensive approach to electro-orientation, electrodeformation, dielec-
trophoresis, and electrorotation of ellipsoidal particles and biological cells. Bioelectro-
chemistry 54(1), 23 -- 31 (2001)
6. Dimova, R., Riske, K.A., Aranda, S., Bezlyepkina, N., Knorr, R.L., Lipowsky, R.: Giant
vesicles in electric fields. Soft Matter 3, 817 -- 827 (2007)
7. Dimova, R., Bezlyepkina, N., Domange Jordo, M., Knorr, R.L., Riske, K.A., Staykova, M.,
Vlahovska, P.M., Yamamoto, T., Yang, P., Lipowsky, R.: Vesicles in electric fields: some
novel aspects of membrane behavior. Soft Matter 5, 3201 -- 3212 (2009)
8. Schwan, H.P.: Electrical properties of tissue and cell suspensions. Adv. Biol. Med. Phys. 5,
9. Helfrich, W.: Elastic properties of lipid bilayers: Theory and possible experiments. Z. Natur-
147 -- 209 (1957)
forsch. C 28, 693 -- 703 (1973)
Frequency-dependent electrodeformation of GUVs
17
10. Helfrich, W.: Deformation of lipid bilayer spheres by electric fields. Z. Naturforsch. C 29,
182 -- 183 (1974)
11. Bryant, G., Wolfe, J.: Electromechanical stresses produced in the plasma membranes of
suspended cells by applied electric fields. J. Membrane Biol. 96, 129 -- 139 (1987)
12. Winterhalter, M., Helfrich, W.: Deformation of spherical vesicles by electric field. J. Colloid
Interf. Sci. 122, 583 -- 586 (1988)
13. Harbich, W., Helfrich, W.: Alignment and opening of giant lecithin vesicles by electric
fields. Z. Naturforsch. A 34, 1063 -- 1065 (1979)
14. Mitov, M.D., M´el´eard, P., Winterhalter, M., Angelova, M.I., Bothorel, P.: Electric-field-
dependent thermal fluctuations of giant vesicles. Phys. Rev. E 48(1), 628 -- 631 (1993)
15. Peterlin, P., Svetina, S., Zeks, B.: The frequency dependence of phospolipid vesicle shapes
in an external electic field. Pflugers Arch. Eur J. Physiol. 439, R139 -- R140 (2000)
16. Hyuga, H., Kinosita Jr., K., Wakabayashi, N.: Transient and steady-state deformations of
a vesicle with an insulating membrane in response to step-function or alternating electric
fields. Jpn. J. Appl. Phys. 30(10), 2649 -- 2656 (1991)
17. Hyuga, H., Kinosita Jr., K., Wakabayashi, N.: Steady-state deformation of a vesicle in al-
ternating electric fields. Bioelectrochem. Bioenerg. 32, 15 -- 25 (1993)
18. Aranda, S., Riske, K.A., Lipowsky, R., Dimova, R.: Morphological transitions of vesicles
induced by AC electric fields. Biophys. J. 95, L19 -- L21 (2008)
19. Landau, L.D., Lifshitz, E.M., Pitaevskiı, L.P.: Electrodynamics of Continuous Media,
Course of Theoretical Physics, vol. 8, 2nd edn. Butterworth-Heineman, Oxford (1984)
20. Nortemann, K., Hilland, J., Kaatze, U.: Dielectric properties of aqueous NaCl solutions at
microwave frequencies. J. Phys. Chem. A 101, 6864 -- 6869 (1997)
21. Turcu, I., Lucaciu, C.M.: Dielectrophoresis: a spherical shell model. J. Phys. A.: Math.
Gen. 22, 985 -- 993 (1989)
22. Foster, K.R., Sauer, F.A., Schwan, H.P.: Electrorotation and levitation of cells and colloidal
particles. Biophys. J. 63(1), 180 -- 190 (1992)
23. Angelova, M.I., Dimitrov, D.S.: Liposome electroformation. Faraday Discuss. Chem. Soc.
81, 303 -- 311 (1986)
24. Angelova, M.I., Sol´eau, S., M´el´eard, P., Faucon, J.F., Bothorel, P.: Preparation of giant
vesicles by external AC electric fields. Kinetics and applications. Prog. Colloid Polym. Sci.
89, 127 -- 131 (1992)
25. Peterlin, P., Arrigler, V.: Electroformation in a flow chamber with solution exchange as a
means of preparation of flaccid giant vesicles. Colloid. Surface. B 64, 77 -- 87 (2008)
26. Sevsek, F., Sukharev, S., Svetina, S., Zeks, B.: The shapes of phospholipid vesicles in elec-
tric field as determined by video microscopy. Stud. Biophys. 138, 143 -- 146 (1990)
27. Sukhorukov, V.L., Meedt, G., Kurscher, M., Zimmerman, U.: A single-shell model for bio-
logical cells extended to account for the dielectric anisotropy of the plasma membrane. J.
Electrostat. 50, 191 -- 204 (2001)
28. Ambjornsson, T., Mukhopadhyay, G.: Dipolar response of an ellipsoidal particle with an
anisotropic coating. J. Phys. A: Math. Gen. 36, 10,651 -- 10,665 (2003)
29. Ko, Y.T.C., Huang, J.P., Yu, K.W.: The dielectric behaviour of single-shell spherical cells
with a dielectric anisotropy in the shell. J. Phys.: Condens. Matter 16, 499 -- 509 (2004)
30. Simeonova, M., Gimsa, J.: Dielectric anisotropy, volume potential anomalies and the per-
sistent Maxwellian equivalent body. J. Phys.: Condens. Matter 17, 7817 -- 7831 (2005)
31. Peterlin, P., Svetina, S., Zeks, B.: The prolate-to-oblate shape transition of phospholipid
vesicles in response to frequency variation of an AC electric field can be explained by the
dielectric anisotropy of a phospholipid bilayer. J. Phys.: Condens. Matter 19, 136220 (2007)
32. Vlahovska, P.M., Serral Graci`a, R., Aranda-Espinoza, S., Dimova, R.: Electrohydrodynamic
model of vesicle deformation in alternating electric field. Biophys. J. 96, 4789 -- 4803 (2009)
33. Kummrow, M., Helfrich, W.: Deformation of giant lipid vesicles by electric fields. Phys.
Rev. A 44(12), 8356 -- 8360 (1991)
34. Niggemann, G., Kummrow, M., Helfrich, W.: The bending rigidity of phosphatidylcholine
bilayers: Dependences on experimental method, sample cell sealing and temperature. J.
Phys. II France 5, 413 -- 425 (1995)
35. Pott, T., Bouvrais, H., M´el´eard, P.: Giant unilamellar vesicle formation under physiologi-
cally relevant conditions. Chem. Phys. Lipids 154, 115 -- 119 (2008)
36. Horger, K.S., Estes, D.J., Capone, R., Mayer, M.: Films of agarose enable rapid formation
of giant liposomes in solutions of physiologic ionic strength. J. Am. Chem. Soc. 131, 1810 --
1819 (2009)
|
1701.00712 | 1 | 1701 | 2017-01-03T15:03:29 | Direction- and Salt-Dependent Ionic Current Signatures for DNA Sensing with Asymmetric Nanopores | [
"physics.bio-ph"
] | Solid-state nanopores are promising tools for single molecule detection of both DNA and proteins. In this study, we investigate the patterns of ionic current blockades as DNA translocates into or out of the geometric confinement of such conically shaped pores. We studied how the geometry of a nanopore affects the detected ionic current signal of a translocating DNA molecule over a wide range of salt concentration. The blockade level in the ionic current depends on the translocation direction at a high salt concentration, and at lower salt concentrations we find a non-intuitive ionic current decrease and increase within each single event for the DNA translocations exiting from confinement. We use recently developed DNA rulers with markers and finite element calculations to explain our observations. Our calculations explain the shapes of the signals observed at all salt concentrations and show that the unexpected current decrease and increase are due to the competing effects of ion concentration polarization and geometric exclusion of ions. Our analysis shows that over a wide range of geometry, voltage and salt concentrations we are able to understand the ionic current signals of DNA in asymmetric nanopores enabling signal optimization in molecular sensing applications. | physics.bio-ph | physics | Direction- and Salt-Dependent Ionic Current
Signatures for DNA Sensing with Asymmetric
Nanopores
Kaikai Chen,1,2 Nicholas A. W. Bell,1,† Jinglin Kong,1 Yu Tian,2 and Ulrich F. Keyser1,*
1Cavendish Laboratory, University of Cambridge, JJ Thomson Avenue, Cambridge, CB3
0HE, United Kingdom; and 2State Key Laboratory of Tribology, Tsinghua University,
Beijing 100084, China
*Correspondence: [email protected]
†Present address: Department of Chemistry, University of Oxford, Oxford, OX1 3TA, United
Kingdom
KEYWORDS: nanopore, DNA sensing, ionic current, resistive-pulse amplitude, glass
capillary
ABSTRACT Solid-state nanopores are promising tools for single molecule detection of both
DNA and proteins. In this study, we investigate the patterns of ionic current blockades as
DNA translocates into or out of the geometric confinement of such conically shaped pores.
We studied how the geometry of a nanopore affects the detected ionic current signal of a
translocating DNA molecule over a wide range of salt concentration. The blockade level in
the ionic current depends on the translocation direction at a high salt concentration, and at
lower salt concentrations we find a non-intuitive ionic current decrease and increase within
each single event for the DNA translocations exiting from confinement. We use recently
developed DNA rulers with markers and finite element calculations to explain our
observations. Our calculations explain the shapes of the signals observed at all salt
concentrations and show that the unexpected current decrease and increase are due to the
competing effects of ion concentration polarization and geometric exclusion of ions. Our
analysis shows that over a wide range of geometry, voltage and salt concentrations we are
able to understand the ionic current signals of DNA in asymmetric nanopores enabling signal
optimization in molecular sensing applications.
INTRODUCTION
Nanopores have become a powerful technique for single molecule sensing of polynucleotides
such as DNA and RNA (1). As an alternative to biological pores such as alpha-hemolysin,
solid-state nanopores fabricated with silicon nitride, glass, graphene or molybdenum disulfide
membranes have found versatile applications in single molecule detection and for the study
of confined transport (2-11). The basic principle of nanopore sensing is that a single molecule
can be detected by measuring the ionic current change during its translocation through the
pore, named as the resistive-pulse method (12). It is intuitive that the current drops during the
1
translocation since the molecule increases the pore resistance by blocking the space which
would otherwise be filled with electrolyte ions. Indeed such reductions have been
consistently observed at high salt concentrations. However at low salt concentrations, current
increases and multi-level current blockades have been observed due to translocating particles
or molecules in a variety of nanopore and nanochannel geometries (13-21). As the salt
concentration decreases, the Debye length becomes longer and may be comparable to the
pore size, thus surface charge begins to play a more important role (22). In this case the ionic
current change is determined not only by the geometric exclusion of ions but also the change
in ion concentration due to the translocation molecule or particle (16-19).
Asymmetric conical nanopores pulled from quartz or borosilicate capillaries are frequently
used in single molecule detection, with the advantages of simple and inexpensive fabrication,
low noise and multiplexed measurement capabilities (8, 23). These conically shaped glass
nanopores have been used for the detection of a range of biomolecules such as DNA (24-26)
and proteins (27-29). In all translocation experiments we have reported so far, the negatively
charged biomolecules move into the conical nanopore from a large reservoir outside, i.e. a
positive potential was applied inside the conical nanopore. With low salt concentration, the
conical shape of the pore causes current rectification and also electroosmotic flow
rectification (30,31). It is important to study the underlying physics behind DNA
translocations and fully understand the current signatures at a variety of experimental
conditions because this provides more options for biological molecule sensing such as salt
concentrations close to the physiological environment.
In this report, we studied DNA translocations through ~15 nm diameter glass nanopores
with salt concentrations from 4 M to 20 mM. We examine the ionic current signatures caused
by the DNA entering or exiting the conical nanopore confinement. The asymmetric pore
geometry gives rise to a variety of distinct ionic current signals depending on the salt
concentration and translocation direction. At high salt concentrations exemplified by 4M
LiCl, a current reduction is exclusively observed but with a small difference in magnitude
between the two translocation directions. At ~1M LiCl or KCl there is a distinctive difference
between translocations into and out of the conical confinement. Translocations into
confinement show a uniform current blockade but translocations out of confinement create a
biphasic pattern, i. e. current decrease and increase in a single event. We use finite element
calculations to simulate the distribution of ions in the pore and the current change caused by a
charged rod representing the DNA. Our model reveals that the ion concentration is modulated
within the nanopore as the DNA passes the pore. The current change is ultimately caused by
a combination of electrolyte concentration modulation and geometric exclusion of ions by the
DNA. Our simulations account for the shapes of the observed ionic current blockades and
their dependence on salt concentration.
MATERIALS AND METHODS
Glass nanopore fabrication
Glass nanopores were pulled from quartz capillaries with outer diameter 0.5 mm and inner
diameter 0.2 mm using a commercial pipette puller (P2000, Sutter Instruments), with the
same parameters shown in ref 27 where the final tip diameter was estimated at 15±3 nm
(mean±s.d.) from scanning electron microscopy.
DNA samples and chemicals
2
Double-stranded DNA samples (NoLimits DNA fragment) were purchased from Fisher
Scientific with the following lengths: 3, 5, 7, 8, 10 and 15 kbp. The DNA ruler was
synthesized according to the method shown in ref 38. LiCl and KCl powders were purchased
from Sigma-Aldrich. Solutions with different salt concentrations were all buffered with
1×Tris-EDTA (Sigma-Aldrich, 10 mM Tris, 1 mM EDTA), except for the 20 mM KCl and
LiCl solutions for which 0.2×Tris-EDTA was used. The pH values were adjusted to 8 for all
solutions using HCl/LiOH or HCl/KOH solutions.
Setup and ionic current measurement
The ionic current was recorded by an amplifier Axopatch 200B (Molecular Devices) with the
current signal filtered at 50 kHz and then collected using a data card (PCI 6251, National
Instrument) at a sampling frequency of 250 kHz. The data writing, voltage control and data
analysis were performed with custom-written Labview programs.
Forward and backward translocation measurement
Forward and backward translocations were achieved by switching the voltage between a
positive value and the opposite one with a period of 60 s, keeping the voltage constant at each
polarity for 30 s. The DNA solutions were added in both reservoirs for the data recording and
only on one side for the verification of translocation direction.
RESULTS AND DISCUSSION
Ionic current traces and current change during translocation
Nanopores with orifice diameters 15±3 nm were pulled from quartz capillaries (see methods).
The bulk reservoir outside the nanopore tip was grounded as shown in Fig. 1 A and the DNA
molecules moved into the conical nanopore at a positive voltage and out of the conical
nanopore at a negative voltage. We define these translocation directions as "forward" and
"backward" respectively (Fig. 1 A). We use two types of salts - KCl and LiCl. KCl has been
the most widely used electrolyte in nanopore experiments due to the similar ion mobilities of
anions and cations. LiCl solution was recently shown to slow down DNA translocations
which results in a significant increase in the sensing resolution due to reduced translocation
velocities (32).
In the analysis, the current change ΔI is defined as I-Ibase (I is the real-time ionic current
and Ibase is the base current), which is positive/negative if the current increases/decreases
relative to the baseline. Fig. 1 C shows a summary of typical current traces with LiCl
solutions and exemplary, so-called unfolded events which are caused by the linear threading
of the DNA through the pore without folds or knots (33). At a concentration of 4 M, current
decreases were observed during translocations in either direction (Fig. 1 C). In 1 M LiCl
solution, the current decreased during the forward translocation but for the backward
translocation, a biphasic current change appeared with a current decrease first and then an
increase. When the concentration of LiCl decreased to 20 mM, as shown in Fig. 1 C, the
current change for backward translocation turned to a sole increase, as previously observed
for conical pores at low salt concentrations (34).
3
FIGURE 1 Schematic of DNA translocation and examples of ionic current traces. (A)
Simplified diagram of DNA translocation through a conical glass nanopore with the voltage
applied inside the conical nanopore. (B) Scanning electron microscope (SEM) image of a
typical nanopore tip showing the outer dimensions. (C) Raw current traces (left) and current
change ΔI (I-Ibase) for translocation events (right, with events marked in the green dashed
rectangles) during 8 kbp DNA translocations through ~15 nm diameter pores in 4 M, 1 M and
20 mM LiCl solutions at +600 mV (blue) and -600 mV (red). I-V curves for the pores are
shown in Fig. S1. DNA was added on both sides with concentrations of 0.95 nM in 4 M and 1
M LiCl solutions and 0.19 nM in 20 mM LiCl solution.
Direction dependence of event depth with high salt concentration
Forward and backward translocations of 8 kbp DNA through the ~15 nm-diameter nanopores
were performed in 4 M LiCl, 2 M LiCl, 4 M KCl and 2 M KCl solutions and typical current
changes for unfolded events are shown in Fig. 2 A. For 4 M and 2 M LiCl we observe that the
magnitude of the current decrease is different for the different directions despite using the
same magnitude of voltage. At 2 M KCl solution the ionic current blockade for a backward
4
translocation shows a biphasic shape. To show the current change quantitatively, we
calculated the unfolded event level ("1-level current" ΔI1-level) by fitting the histograms of
current change for all events using Gaussian functions as the folded parts did not contribute
to the fitting of the unfolded parts (Fig. S2). As shown in Fig. 2 B, the levels were different at
+600 mV and -600 mV in 4 M LiCl solution (histograms at 400~800 mV and -400~-800 mV
are shown in Fig. S2). Since there was a slight slope for the current blockade level for conical
pores, we studied the length dependence of the event shape and 1-level current (Fig. S3). Our
results show that the direction-dependent event depth was more significant for longer DNA,
and the derived ΔI1-level did not change significantly further when the length was above 7 kbp,
so the slight slope did not have a significant effect on the derived 1-level current for the 8 kbp
DNA. The voltage dependence of the current blockades for forward and backward
translocations is shown in Fig. S2 and the difference was more obvious at higher voltage.
This illustrates that even in 4 M LiCl solution, with a Debye length of ~0.15 nm which is
much smaller than the pore diameter ~15 nm, the event depth shows a direction dependence.
Similar direction-dependent event depth profiles have been seen for particles of diameter
above 100 nm translocating asymmetric nanopores with final tip diameters of several hundred
nanometers in salt concentrations (100 mM or 10 mM) where the Debye length is much less
than the pore diameter (19,35).
The voltage dependence of the normalized current change in both directions in 4 M LiCl
and 2 M LiCl solutions are shown in Fig. 2 C and Fig. 2 D. Results for more pores in Fig. S4
and Fig. S5 also show directional dependence of the current change but with slight
differences in the voltage-dependence of the normalized current change. As the LiCl
concentration went down from 4 M to 2 M, the Debye length increased and also the effective
surface charge on the DNA increased (32), both enhancing the effect of ion concentration
polarization - a well-known property of nanochannels resulting from the preferred transfer of
one ion over another and causing local ion concentration depletion or enrichment of co- and
counterions, respectively (36). Experiment for particles showed the trends were not the same
for different geometries and surface charge states, and the voltage dependence was
sometimes even non-monotonic (35), in accordance with our experimental results for
nanopores and DNA.
5
FIGURE 2 Event depths for forward and backward translocations in 2 and 4 M solutions. (A)
Typical current changes for the unfolded events during 8 kbp DNA translocations in 4 M
LiCl, 2 M LiCl, 4 M KCl and 2 M KCl solutions at ±600 mV. (B) An example for calculating
the "1-level current" ΔI1-level with 491 events at 600 mV and 396 events at -600 mV in 4 M
LiCl. (C) and (D) show the absolute values of normalized ΔI1-level (ΔI1-level/Ibase) during 8 kbp
DNA translocations in 4 M and 2 M LiCl solutions. Error bars in (C) and (D) represent
standard deviations of the Gaussian fit to the ionic current distributions.
Direction-dependent event shapes with medium salt concentration
For the events with high salt concentration shown above, the current decreased in both
translocation directions (except for the 2 M KCl solution). Fig. 3 shows examples of the
unfolded events with 1 M KCl and 1 M LiCl solutions (more events are shown in Fig. S6).
For the backward events, the current decreased at the beginning, increased slowly in the
middle part, and ended with a peak. The negative and positive peak amplitudes for backward
translocations were much smaller than the event depths for the forward translocations at the
same voltage amplitude. Similar profiles were reported for particle or polymer translocations
at low salt concentrations in former reports where they are attributed to ion concentration
polarization (19,20). From these findings we expect that at concentrations of 1 M the
electrical double layer (EDL) plays a more important role. The current increase during
backward translocation in 1M KCl solution was more significant than that in 1 M LiCl
solution. Also, the current increase at the end of an event was observed with 2 M KCl
solution but not with 2 M LiCl solution. This difference can be explained by the expected
dependence of DNA effective surface charge densities on the type of counter ion (32). At the
same molarity, the effective charge of DNA is significantly higher in KCl compared to LiCl
which results in a stronger ion concentration polarization effect as the DNA translocates
through. This point is discussed further in the modelling section.
6
Fig. 3 B shows the current change for unfolded events at voltages ranging from 500 mV to
800 mV and from -500 mV to -800 mV in 1 M KCl solution. The derived normalized 1-level
current for forward translocations as a function of voltage is shown in Fig. 3 C, where the
values increased monotonically with voltage for the 3 pores.
FIGURE 3 Direction-dependent event shapes for DNA translocations with 1 M salt
concentration. (A) ΔI for unfolded events during 8 kbp DNA translocations in 1 M LiCl
solution at ±600 mV. (B) ΔI for unfolded events during 8 kbp DNA translocations at different
voltages in 1 M KCl solution. (C) Absolute values of normalized current change for forward
translocations of 8 kbp DNA in 1 M KCl solution against voltage for 3 pores. Error bars in
(C) show standard deviations of the Gaussian fit to the ionic current distributions.
Current increase during backward translocations with low salt concentration
With the salt concentration going further down to 100 mM, the current increased during the
backward translocations (Fig. 4 A), while forward translocations were not observed for our
~15 nm pores. The absence of translocation in the forward direction is likely due to electro-
osmotic flow which opposes the electrophoretic motion of the DNA. This electro-osmotic
7
flow was extensively characterized by Laohakunakorn et al. and shown to increase strongly
with decreasing salt concentration (31). Furthermore the flow is asymmetric with respect to
voltage reversal with significantly larger flow rates measured at positive voltages. We note
that for conical nanopores with larger diameters of around 50 nm, forward translocations are
sometimes observed at low salt concentrations (34).
Similar to the event shapes for forward translocations with high salt concentration, one can
clearly identify unfolded events, as shown in Fig. 4 A and Fig. 4 B. Using the same method
shown in Fig. 2 B, we analyzed the normalized 1-level current for 8 kbp DNA translocations
in 100 mM KCl and 20 mM KCl solutions, with the results shown in Fig. 4 C and Fig. 4 D.
The normalized current change increased monotonically or showed a non-monotonic
behavior, depending on the pore conductance determined by the pore diameter and conical
angle. It is notable that for translocations in the 20 mM KCl solution, the normalized current
rise was up to 15%, which was much higher than the normalized current drop for
translocations with high salt concentration (37).
8
FIGURE 4 Current increase during backward translocations with 100 and 20 mM
concentrations. (A) and (B) show the ΔI for backward translocations of 8 kbp DNA in 100
mM and 20 mM KCl solutions. (C) and (D) show the corresponding normalized 1-level
current as a function of voltage with results for 3 pores at each salt concentration. Error bars
in (C) and (D) show standard deviations of the Gaussian fit to the ionic current distributions.
Translocations using DNA rulers
In order to understand the origin of the diverse event shapes, we conducted experiments using
our recently developed DNA rulers (38). The ruler consists of a 7.2 kbp backbone of double-
stranded DNA with six equally spaced zones of hairpin loops protruding from the backbone,
as shown in Fig. 5 A. The ruler is of practical importance for assessing DNA translocation
and it helps to determine where the DNA is positioned in the pore when a particular ionic
current level is detected (38). Fig. 5 B shows typical ionic current blockades of unfolded
events caused by the ruler in 1 M LiCl in forward and backward directions. Previous analysis
for our nanopores, based on charge exclusion, has suggested that ionic current signatures
above baseline noise are caused only by full translocations (39). The DNA ruler confirms that
the ionic blockade is due to the DNA fully translocating the nanopore and also allows us to
relate the position of the DNA in the nanopore to the ionic blockade at a certain time during
the translocation. For the backward translocation the largest current decrease occurred at
approximately the same time as the first marker exits the nanopore, i. e. approximately 1/7 of
the way through the translocation. The largest point of the current increase occurred after the
last marker passed through indicating that the current increase is in the last section of the
translocation where >6/7 of the DNA has already exited the nanopore.
FIGURE 5 Schematic of a DNA ruler (A) and ΔI for DNA ruler translocations at ±600 mV
with 1 M LiCl solution (B).
Numerical simulation
Having investigated the experimental characteristics of the ionic current blockades, we
conducted finite element analysis numerical simulations to investigate the mechanisms
behind the observed phenomena. Our simulation solves the Poisson−Nernst−Planck and
Navier−Stokes equations (PNP-NS equations) (40) using COMSOL Multiphysics 4.4 with an
9
axially symmetric geometry shown in FiG. S8 which is an average from nanopores measured
by SEM (more model details are given in the Supporting Material). The DNA is simulated as
a cylindrical rod of diameter 2.2 nm and length 2720 nm. The mesh size was refined to be 0.1
nm at the boundaries of the pore and the rod representing DNA. Fig. 6 A shows the calculated
ion concentration profiles at both ±600 mV with 1 M KCl. For direct comparison, Fig. 6 B
shows the profile at -600 mV with 20 mM KCl. We assume a surface charge density of -0.01
C/m2 on the pore walls and -0.018 C/m2 on DNA which is in the same order of magnitude as
values estimated in 1 M KCl solution in the literature (41, 42). The surface charge is an
important parameter in our model which is known to vary with salt type and concentration
(43). The uncertainty of the surface charge prevents any quantitative analysis as well as other
factors such as molecule trajectory and interaction. The latter also cause variations in the
calculated current levels. While a model incorporating a surface related molecular drag would
give a better quantitative explanation on the current blockade (44,45), we concentrate here
on explaining the event shape rather than an absolute quantitative value. In Fig. 6 A and Fig.
6 B we systematically vary the position of the DNA thereby reflecting the passage of the
DNA during a translocation. The calculated ionic current-DNA position traces are displayed
in Fig. 6, C-E, in accordance with the experimental data shown above (note the translocation
direction).
FIGURE 6 Simulation for the direction-dependent current blockades. (A) and (B) show the
sum of potassium and chloride ion concentrations (C(K+)+C(Cl-)) with particular DNA
10
positions with 1 M and 20 mM bulk KCl concentrations respectively. Definition of the DNA
position zDNA is shown in the top right corner. (C) and (D) show the current change as a
function of DNA position at ±600 mV in 1 M KCl solution. (E) Current change as a function
of DNA position at -600 mV in 20 mM KCl solution. Arrows show directions of DNA
translocation. Pore surface charge density σpore -0.01 C/m2 and DNA surface charge density
σDNA -0.018 C/m2 are used in the simulation.
Our simulations reveal a difference between the concentrations of ions inside the conical
nanopore and those in the bulk reservoir, which is known as ion concentration polarization
(36). In our experiment the selectivity arises from the negative charge of the pore walls and
DNA which induce a cloud of positively charged counterions. The magnitude of ion
concentration around the nanopore tip changes depending on the position of the DNA in the
nanopore. There is 5 to 10 percent difference in Fig. 6 A and this concentration modulation
extends to regions far into the conical nanopore (Fig. S9). The shape of the ionic current
blockade is then mainly determined by the combined effects of geometrical exclusion of ions
and ion concentration polarization. At a positive voltage, the ion concentration was depleted,
so the effects of geometrical exclusion of ions and ion concentration polarization were
cooperative, both decreasing the current. At a negative voltage, their effects were
competitive, with the geometrical exclusion effect decreasing the current and ion
concentration polarization effect increasing the current. The measured current is determined
by the combination of these two effects. For instance, at -600 mV, the negative peak occurred
at position e2 in Fig. 6 A and Fig. 6 D when the geometrical exclusion effect had the largest
extent and the ion concentration enrichment was not so significant. The positive peak
occurred before the end of the translocation - b2 in Fig. 6 A and Fig. 6 D coincident with
significant ion concentration enrichment and small correction due to the geometrical
exclusion. Fig. S10 shows a similar simulated current profile with a lower DNA charge
density. If the salt concentration was low enough, the effect of ion concentration
enhancement overtook the geometrical exclusion effect, resulting in a current increase only,
as shown in Fig. 6 B and Fig. 6 E. The simulated event shapes for translocations in 100 mM
KCl and 20 mM KCl solutions are very similar to the experimental results, as shown in Fig. 4
A and Fig. 4 B, Fig. S11 and Fig. 6 E respectively.
We also studied the effect of the pore geometry with 20 mM KCl solution by simulation
(Fig. S12). The simulations showed that the ion concentration modulation was more
significant for 12 nm tip diameters than that of 18 nm tip diameter. This result is expected
due to a reduction in the relative importance of surface charges, in accordance with our
experimental results. Overall our numerical simulations account for the variety of ionic
current blockades observed in experiments over a large salt concentration range.
CONCLUSIONS
In summary, we have presented a comprehensive study of the direction, salt and voltage
dependence of ionic current signatures for DNA translocations through conical asymmetric
nanopores. At salt concentrations of ~4M, translocations in both directions caused current
drop but with different amplitudes. With an intermediate salt concentration of ~1M, the
current decreased during the forward translocations while it decreased at the beginning and
increased at the end in a backward translocation event. At lower salt concentrations closer to
physiological ionic strengths, only translocations from inside the pore to outside were
11
observed due to the increasing influence of electroosmotic flow. Numerical simulations were
conducted to explore the current change by solving the PNP-NS equations, and revealed that
the ion concentration inside the pore was depleted at a positive voltage and enriched at a
negative voltage. The direction-dependent ionic current blockades were found to be caused
by differences in the effects of ion concentration modulation and geometrical exclusion of
ions, with a cooperative action at a positive voltage and a competitive action at a negative
voltage. Our results pave the way for optimizing biological molecule sensing with
asymmetric conical nanopores by tuning voltage and salt concentrations.
SUPPORTING MATERIAL
Supplementary figures and discussion are available in the supporting material.
AUTHOR CONTRIBUTIONS
K. Chen, N. A. W. Bell, J. Kong and U. F. Keyser designed the research. K. Chen performed
the nanopore measurements. N. A. W. Bell synthesized the DNA rulers. K. Chen., N. A. W.
Bell, J. Kong, Y. Tian and U. F. Keyser wrote and edited the manuscript. U. F. Keyser
oversaw the research.
ACKNOWLEDGMENT
The authors thank K. Misiunas for discussion on COMSOL simulations. K. Chen
acknowledges funding from Chinese Scholarship Council (201506210147). J. Kong
acknowledges funding from Chinese Scholarship Council and Cambridge Trust. N. Bell and
U. Keyser acknowledge funding from an ERC consolidator grant (Designerpores 647144).
REFERENCES
1. Kasianowicz, J. J., Brandin, E., …, D. W. Deamer. 1996. Characterization of individual
polynucleotide molecules using a membrane channel. P. Natl. Acad. Sci. USA. 93:13770-
13773.
2. Storm, A. J., Chen, J. H., …, C. Dekker. 2003. Fabrication of solid-state nanopores with
single-nanometre precision. Nat. Mater. 2:537-540.
3. Li, J., Gershow, M., …, J. A. Golovchenko. 2003. DNA molecules and configurations in a
solid-state nanopore microscope. Nat. Mater. 2:611-615.
4. White, R. J., Ervin, E. N., …, H. S. White. 2007. Single ion-channel recordings using glass
nanopore membranes. J. Am. Chem. Soc. 129:11766-11775.
5. Balan, A., Chien, C. C., …, M. Drndic. 2015. Suspended solid-state membranes on glass
chips with sub 1-Pf capacitance for biomolecule sensing applications. Sci. Rep. 5, 17775.
6. Garaj, S., Hubbard, W., …, J. A. Golovchenko. 2010. Graphene as a subnanometre trans-
electrode membrane. Nature. 467:190-193.
12
7. Liu, K., Feng, J., …, A. Radenovic. 2014. Atomically thin molybdenum disulfide
nanopores with high sensitivity for DNA translocation. ACS Nano. 8:2504-2511.
8. Steinbock, L. J., Bulushev, R. D., …, A. Radenovic. 2013. DNA translocation through
low-noise glass nanopores. ACS Nano. 7:11255-11262.
9. Li, W., Bell, N. A. W., …, U. F. Keyser. 2013. Single protein molecule detection by glass
nanopores. ACS Nano. 7:4129-4134.
10. Tsutsui, M., Hongo, S., …, T. Kawai. 2012. Single-nanoparticle detection using a low-
aspect-ratio pore. ACS Nano. 6:3499-3505.
11. Lan, W. J., Holden, D. A. …, H. S. White. 2011. Nanoparticle transport in conical-shaped
nanopores. Anal. Chem. 83:3840-3847.
12. Coulter, W. H. Means for counting particles suspended in a fluid. 1953. U.S. Patent.
2656508, October 20.
13. Smeets, R. M. M., Keyser, U. F. …, C. Dekker. 2006. Salt dependence of ion transport
and DNA translocation through solid-state nanopores. Nano Lett. 6:89-95.
14. Kowalczyk, S. W., and C. Dekker. 2012. Measurement of the docking time of a DNA
molecule onto a solid-state nanopore. Nano Lett. 12:4159-4163.
15. Vlassarev, D. M., and J. A. Golovchenko. 2012. Trapping DNA near a solid-state
nanopore. Biophys. J. 103:352-356.
16. Zanjani, M. B., Engelke, R. E., ..., M. Drndic. 2015. Up and down translocation events
and electric double-layer formation inside solid-state nanopores. Phys. Rev. E. 92:022715.
17. Goyal, G., Freedman, K. J., and M. J. Kim. 2013. Gold Nanoparticle translocation
dynamics and electrical detection of single particle diffusion using solid-state nanopores.
Anal. Chem. 85:8180-8187.
18. Menestrina, J., Yang, C., …, Z. S. Siwy. 2014. Charged particles modulate local ionic
concentrations and cause formation of positive peaks in resistive-pulse-based detection. J.
Phys. Chem. C. 118:2391-2398.
19. Lan, W.-J., Kubeil, C., …, H. S. White. 2014. Effect of surface charge on the resistive
pulse waveshape during particle translocation through glass nanopores. J. Phys. Chem. C.
118:2726-2734.
20. Cabello-Aguilar, S., Abou Chaaya, A., …, S. Balme. 2014. Experimental and simulation
studies of unusual current blockade induced by translocation of small oxidized PEG through
a single nanopore. Phys. Chem. Chem. Phys. 16:17883-17892.
21. Ivanov, A. P., Actis, P., …, J. B. Edel. 2015. On-demand delivery of single DNA
molecules using nanopipets. ACS Nano. 9:3587-3595.
22. Schoch, R. B., Han, J., and P. Renaud. Transport phenomena in nanofluidics. 2008. Rev.
Mod. Phys. 80:839-883.
23. Bell, N. A., Thacker, V. V., …, U. F. Keyser. 2013. Multiplexed ionic current sensing
with glass nanopores. Lab Chip. 13:1859-1862.
13
24. Freedman, K. J., Otto, L. M., …, J. B. Edel. 2016. Nanopore sensing at ultra-low
concentrations using single-molecule dielectrophoretic trapping. Nat. Commun.7.
25. Fraccari, R. L., Ciccarella, P., …, T. Albrecht. 2016. High-speed detection of DNA
translocation in nanopipettes. Nanoscale. 8:7604-7611.
26. Fraccari, R. L., Carminati, M., …, T. Albrecht. 2016. High-bandwidth detection of short
DNA in nanopipettes. Faraday Discuss.
27. Bell, N. A., and U. F. Keyser. 2015. Specific protein detection using designed DNA
carriers and nanopores. J. Am. Chem. Soc. 137:2035-2041.
28. Bell, N. A., and U. F. Keyser. 2016. Digitally encoded DNA nanostructures for
multiplexed, single-molecule protein sensing with nanopores. Nat. Nanotechnol.
29. Kong, J. Bell, N. A., and U. F. Keyser. 2016. Quantifying nanomolar protein
concentrations using designed DNA carriers and solid-state nanopores. Nano Lett. 16:3557-
3562.
30. Siwy, Z.S. Ion-current rectification in nanopores and nanotubes with broken symmetry.
2006. Adv. Func. Mater. 16:735-746.
31. Laohakunakorn, N., and U. F. Keyser. 2015. Electroosmotic flow rectification in conical
nanopores. Nanotechnology. 26:275202.
32. Mihovilovic, M., Hagerty, N., and D. Stein. 2013. Statistics of DNA capture by a solid-
state nanopore. Phys. Rev. Lett. 110:028102.
33. Steinbock, L. J., Lucas, A., …, U. F. Keyser. 2012. Voltage-driven transport of ions and
DNA through nanocapillaries. Electrophoresis. 33:3480-3487.
34. Qiu, Y., Vlassiouk, I., …, Z. S. Siwy. 2016. Direction dependence of resistive-pulse
amplitude in conically shaped mesopores. Anal. Chem. 88:4917-4925.
35. Kowalczyk, S. W., Wells, D. B., …, C. Dekker. 2012. Slowing down DNA translocation
through a nanopore in lithium chloride. Nano Lett. 12:1038-1044.
36. Hlushkou, D., Perry, J. M., …, U. Tallarek. 2011. Propagating concentration polarization
and ionic current rectification in a nanochannel-nanofunnel device. Anal. Chem. 84:267-274.
37. Smeets, R.M., Keyser, U.F., …, C. Dekker. 2008. Noise in solid-state nanopores. P. Natl.
Acad. Sci. USA. 105:417-421.
38. Bell, N. A.. and U. F. Keyser. Direct measurements reveal non-Markovian fluctuations of
DNA threading through a solid-state nanopore. arXiv preprint arXiv:1607.04612.
39. Bell, N. A., Muthukumar, M., and Keyser, U. F. 2016. Translocation frequency of
double-stranded DNA through a solid-state nanopore. Phys. Rev. E. 93:022401.
40. Chen, K., Shan, L., …, Y. Tian. 2015. Biphasic resistive pulses and ion concentration
modulation during particle translocation through cylindrical nanopores. J. Phys. Chem. C
119:8329-8335.
41. Keyser, U. F., Koeleman, B. N., …, C. Dekker. 2006. Direct force measurements on
DNA in a solid-state nanopore. Nat. Phys. 2:473-477.
14
42. Lu, B., Hoogerheide, D. P., …, D. Yu. 2012. Effective driving force applied on DNA
inside a solid-state nanopore. Phys. Rev. E. 86:011921.
43. Schellman, J.A., and D. Stigter. 1977. Electrical double layer, zeta potential, and
electrophoretic charge of double-stranded DNA. Biopolymers. 16:1415-1434.
44. Tsutsui, M., He, Y., …, T. Kawai. 2015. Particle trajectory-dependent ionic current
blockade in low-aspect-ratio pores. ACS Nano. 10:803-809.
45. Kesselheim, S., Muller, W., and C. Holm. 2014. Origin of current blockades in nanopore
translocation experiments. Phys. Rev. Lett. 112:018101.
15
|
1404.7816 | 2 | 1404 | 2017-09-08T18:27:15 | Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System | [
"physics.bio-ph",
"cond-mat.mes-hall",
"cond-mat.other",
"quant-ph"
] | We study theoretically the noise-assisted quantum exciton (electron) transfer (ET) in bio-complexes consisting of a single-level electron donor and an acceptor which has a complicated internal structure, and is modeled by many electron energy levels. Interactions are included between the donor and the acceptor energy levels and with the protein-solvent noisy environment. Different regions of parameters are considered, which characterize (i) the number of the acceptor levels, (ii) the acceptor \band-width", and (iii) the amplitude of noise and its correlation time. Under some conditions, we derive analytical expressions for the ET rate and efficiency. We obtain equal occupation of all levels at large times, independently of the structure of the acceptor band and the noise parameters, but under the condition of non-degeneracy of the acceptor energy levels. We discuss the multi-scale dynamics of the acceptor population, and the accompanying effect of quantum coherent oscillations. We also demonstrate that for large number of levels in the acceptor band, the efficiency of ET can be close to 100%, for both downhill and uphill transitions and for sharp and at redox potentials. | physics.bio-ph | physics |
Multi-Scale Exciton and Electron Transfer in
Multi-Level Donor-Acceptor System
Shmuel Gurvitz
E-mail: [email protected]
Department of Particle Physics and Astrophysics, Weizmann Institute, 76100,
Rehovot, Israel
E-mail: [email protected]
Alexander I. Nesterov
Departamento de F´ısica, CUCEI, Universidad de Guadalajara, Av. Revoluci´on 1500,
Guadalajara, CP 44420, Jalisco, M´exico
E-mail: [email protected]
Gennady P. Berman
Theoretical Division, T-4, Los Alamos National Laboratory, and the New Mexican
Consortium, Los Alamos, NM 87544, USA
E-mail: [email protected]
Abstract. We study theoretically the noise-assisted quantum exciton (electron)
transfer (ET) in bio-complexes consisting of a single-level electron donor and an
acceptor which has a complicated internal structure, and is modeled by many electron
energy levels. Interactions are included between the donor and the acceptor energy
levels and with the protein-solvent noisy environment. Different regions of parameters
are considered, which characterize (i) the number of the acceptor levels, (ii) the
acceptor "band-width", and (iii) the amplitude of noise and its correlation time. Under
some conditions, we derive analytical expressions for the ET rate and efficiency. We
obtain equal occupation of all levels at large times, independently of the structure of the
acceptor band and the noise parameters, but under the condition of non-degeneracy
of the acceptor energy levels. We discuss the multi-scale dynamics of the acceptor
population, and the accompanying effect of quantum coherent oscillations. We also
demonstrate that for large number of levels in the acceptor band, the efficiency of ET
can be close to 100%, for both downhill and uphill transitions and for sharp and flat
redox potentials.
PACS numbers: 87.15.ht, 05.60.Gg, 82.39.Jn
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
2
1. Introduction
Exciton (and electron) transfer (ET) in photosynthetic bio-complexes, including plants,
eukaryotic algae and cyanobacteria, has a large range of characteristic time-scales, from
tens of femtoseconds to milliseconds and more. In particular, ET in light-harvesting
complexes (LHCs) and primary charge separation processes in the reaction centers (RCs)
of the photosystem I (PSI) and photosystem II (PSII) take place on very short time-
It was recently experimentally discovered that the exciton
scales, of order 1 − 5ps.
dynamics in this type of systems can involve quantum coherent effects (quantum
Brownian motion) [1, 2, 3, 4, 5]. These results have generated significant interest in
creating adequate mathematical tools for describing and modeling of quantum coherent
processes in these systems [6, 7, 8, 9, 10, 11].
When considering analytically the ET dynamics in LHCs, usually one uses the
Forster resonant perturbation theory with different generalizations [12, 13, 14].
In
spite of this approach, based on application of the Fermi's Golden Rule (FGR), is
very straightforward, physically visible, and useful in many applications, it does not
describe the multi-scale ET dynamics, which usually is the case for donor and acceptor
with the finite energy band-widths. (See our results below.) Indeed, a single ET rate,
ΓF GR = 2πV 2ρ/, which does not depend on time and formally is valid for t ∈ [0,∞],
occurs in the well-known Weisskopf-Wigner model [15], where V is the matrix element of
interaction between a single electron energy level (donor) with the acceptor, modeled by
an infinite energy band with the density of homogeneously distributed electron states,
ρ. In the case of the initially populated donor, the acceptor is populates in time with
the probability: PA(t) = 1− exp(−ΓF DRt). Generally, the FGR approach and, based on
it the Forster resonant perturbation theory, are valid only for some region of parameters
and for some intermediate times, and don't describe the multi-scale ET dynamics.
Then, it becomes an important issue to develop an approach which (i) can describe
analytically the ET rates in the LHCs on different time-scales, (ii) is valid for finite
energy bands of both donor and acceptor, and (iii) can easily be implemented in
numerical analysis. The approach, developed below, contributes in resolution of these
issues, which is the main subject of our paper.
In real situations, the LHCs and RCs networks consist of many chlorophyll,
carotenoids, and other complex organic molecules which include the corresponding
electron and exciton quantum states with different energy levels. So, for modeling of the
electron and exciton transfer in these complexes, an extremely large Hilbert space should
be used. Then, to make the corresponding models useful and predictive, different parts
of the LHCs and RCs can be considered, under some conditions, as interacting clusters
with particular sets of molecules, geometries, and structures of exciton and electron
energy levels. This approach, to some extent, is equivalent to the well-known coarse-
graining procedure, which is usually used for these purposes. (See, for example [16],
and references therein.) Then, the questions arise: How do the structures of individual
clusters, the interactions between different clusters, and their interactions with noisy or
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
3
thermal protein environment(s) affect the electron and exciton transfer?
In this paper, the ET in a simple donor-acceptor (two-cluster) model is considered,
which allowed us to derive many useful preliminary results in these directions. Namely,
we analytically and numerically study the ET between a single-level electron donor
interacting with an acceptor, which is modeled by a finite number of electron energy
levels.
In particular, this situation takes place when a single excited energy level
(say,
in Chlb) is initially populated, and the acceptor (excited electron state of
Chla) has a complex structure due to the contribution of the Chla vibrational levels
[17, 18, 19, 20, 21, 22, 23].
Similar situation takes place, for example, when a coarse-graining procedure can
be applied, and a bio-complex with many energy levels can be considered as an electron
"donor" and/or an "acceptor" with complicated internal structures. We assume that an
external (classical) diagonal noise interacts with both the donor and the acceptor energy
levels. This approach is often used for modeling the protein-solvent environment under
non-equilibrium conditions. (See, for example, [26, 27, 28, 29, 30, 31], and references
therein.) Note, that our model can be easily generalized to include (i) both multi-level
donor and acceptor, (ii) complex band structures, (iii) interactions inside the bands, (iv)
more than two interacting donor-acceptor type clusters, (v) thermal (instead of noisy)
environment(s).
Our main results include:
1) We derived analytical expressions for the ET rates and efficiencies in different
regions of parameters, which are important for understanding a complicated quantum
dynamics of the system. We also provided the numerical simulations which confirm our
analytical results.
2) We showed that generally the dynamics of the acceptor population is
characterized by multi-scale processes, accompanied by quantum coherent oscillations.
We estimated the corresponding ET rates and the period of these oscillations.
3) We obtained the equal occupation of all levels at large times, independent of
the structure of the acceptor band, but under the condition of non-degeneracy of the
acceptor energy levels. The case of a degenerate acceptor band is analyzed in details.
4) We demonstrated that the efficiency of population of the acceptor can be close
to 100 %, in relatively short times, for both sharp and flat redox potentials.
5) We demonstrated the possibility of the efficient uphill population of the acceptor,
due to the "entropy factor" (large number of levels in the acceptor band).
In Sec.
The paper is organized as follows.
II, we consider a single-level donor
interacting with a multi-level acceptor, in the absence of noise. Some general features
of this system are described for this case. In Sec. III, we introduce an approach based
on the master equation for the reduced density matrix, in the presence of noise. In Sec.
IV, we present analytical and numerical results for the ET rates for different regions
of parameters, for a single-level donor and single-level acceptor system. In Sec. V, we
extend the results to the N-level acceptor. In Appendix A, we derive the mathematical
expressions needed in Sec. V. In the Conclusion, we summarize our results, and discuss
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
4
possible generalizations of our approach.
2. Single-level donor interacting with multi-level acceptor
Consider a single level donor coupled with a N-level band of acceptor. This is shown
schematically in Fig. 1. We describe this system by the following tunneling Hamiltonian,
Donor
EN
E1
E0
V
N
Acceptor
En
Figure 1. (Color online) Schematic of our simplified model consisting of an single-level
donor and of a N -level acceptor.
N
N
HS = E00ih0 +
Ennihn +
VN (nih0 + 0ihn),
(1)
Xn=1
Xn=1
where E0 is the energy level of the donor and En is the n-th level of the acceptor-band.
We assume that the interaction (tunneling coupling), VN , is the same for all levels of the
acceptor, but depends on the number of levels (N) as VN = V /√N. This dependence
results from the normalization factor of the acceptor states, ni. Note, that the same
Hamiltonian describes either the exciton (energy) transfer or the electron tunneling [24],
and therefore the corresponding dynamics will be the same.
Consider the donor, 0i, is initially populated by an exciton. Due to its coupling
with the acceptor, the exciton makes transitions to the acceptor's states, described by
the time-dependent wave function, Ψ(t)i. The latter can be written as,
N
Ψ(t)i = b0(t)0i +
bn(t)ni,
(2)
with the initial condition, Ψ(0)i = 0i. Substituting Eq. (2) into the Schrodinger
equation, i∂tΨ(t)i = HSΨ(t)i, we obtain [25],
Xn=1
N
Xn=1
ib0(t) = E0b0(t) +
VN bn(t),
ibn(t) = Enbn(t) + VN b0(t).
(3)
(4)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
5
(Here and below we choose = 1.) In order to solve these equations, we apply the
0 b(t)eiEtdt. As a result we obtain,
Laplace transform, b(E) =R ∞
(E − E0)b0(E) −
(E − En)bn(E) − VNb0(E) = 0.
Xn=1
VN
N
bn(E) = i,
(5)
(6)
The r.h.s. of these equations reflects the initial condition.
Substituting bn(E) from the second equation into the first one, we obtain for the
amplitude, b0(E),
b0(E) =
i
N
E − E0 −
Xn=1
.
V 2
N
E − En
(7)
The time-dependent amplitude of the donor population, b0(t), is given by the inverse
Laplace transform,
b0(t) =
1
2π
∞
Z−∞
b0(E) e−iEt dE.
(8)
The probability of finding the donor occupied (survival probability) is, P0(t) =
b0(t)2. Respectively, the occupation probability of the acceptor' n-th level, Pn(t) =
bn(t)2, is obtained through the inverse Laplace transform of the amplitude, bn(E) =
VN b0(E)/(E − En), Eq. (6).
For the infinite acceptor' band (E1 → −∞, EN → ∞, and N → ∞) we can
replace, Pn → R N (En)dEn, where N (En) is the density of states. It can be written
as, N (En) = N, where we assume that is independent of En. Then we obtain,
∞
V 2
N
E − En →
Xn
Z−∞
V 2
N N dEn
En − E
= iπV 2.
(9)
Substituting this result into Eqs. (7) and (8), one can easily perform integration over
E by closing the integration contour in the lower half-plane. As a result, b0(t)2 =
exp(−Γt), where Γ = 2πV 2. Therefore, the donor is totally depopulated in the limit
of t → ∞, in agreement with the Weisskopf-Wigner approach [15].
This is not the case for a finite N. In particular, when the donor' energy is far
outside the acceptor band, E0 − EN ≫ V , the transition probability is very small, so
the donor remains almost totally populated in the limit of t → ∞. This can be easily
illustrated by the example of N = 1. Evaluating the integral (8) for this case, we find
for occupation of the donor,
ǫ2 + 4V 2 cos2 ωRt
P0(t) =
ǫ2 + 4V 2
where ǫ = E0 − E1, and ωR = 1
It
follows from Eq. (10) that the occupation of donor at t → ∞ oscillates in time and is
proportional to 4V 2/ǫ2 ≪ 1.
√4V 2 + ǫ2 is the frequency of Rabi oscillations.
2
(10)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
6
The same occurs for any finite N, and when E0 − EN ≫ V .
Eqs. (5)-(6) by iterations (in powers of interaction), we find,
Indeed, solving
b0(E) =
bn(E) =
i
+ · · ·
iVN
E − E0
(E − E0)(E − En)
Then, performing the inverse Laplace transform, Eq. (8), we obtain for the acceptor'
population, Pa(t),
+ · · · .
(11)
Pa(t) =
bn(t)2 ∼ N
V 2
N
ǫ2 ∼
V 2
ǫ2 ≪ 1.
(12)
N
Xn=1
An example of the acceptor' population, obtained from the exact numerical solution
of Eqs. (3)-(4), is shown in Fig. 2 for N = 10 and for acceptor' levels being equally
distributed inside the band,
En =
2n − 1 − N
2(N − 1)
δa,
(13)
where δa is the band-width of the acceptor. The energy of the band-center is set to zero,
so that E1 = −δa/2, EN = δa/2 and the donor's energy is E0 = ǫ.
Pat
0.05
0.04
0.03
0.02
0.01
0
20
40
60
80
t
100
Figure 2. (Color online) Occupation of the acceptor as a function of time; N = 10,
ǫ = 10, V = 1, and the acceptor's band-width, δa = 1.
The parameters used in Fig. 2 correspond to δa = 1, V = 1, and ǫ = 10, in arbitrary
units. (For instance, it is convenient to choose the energy parameters in units of ps−1,
where time is measured in ps. Then, the values of our parameters in energy units should
be multiplied values by ≈ 6.58 × 10−13meVs, so that ǫ = 10 ps−1 ≈ 6.58meV).
One finds that the results shown in Fig. 2 confirm our estimates, Eq. (12), based on
the perturbative calculations (assuming small V ), that the acceptor cannot be populated
if the energy levels of the donor are outside the acceptor band. This situation can change
very drastically in the presence of noise.
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
7
3. Density matrix approach and inclusion of noise
Let us consider the donor and acceptor interacting with an external environment that
perturbs their energy levels. The total Hamiltonian can be written as,
H = HS + Henv + Hint,
(14)
where HS is the Hamiltonian of the donor-acceptor system, Eq. (1), and the two other
terms describe the environment and its interaction with the system, correspondingly.
By solving the Schrodinger equation, i∂tΨ(t)i = HΨ(t)i, we can find the total wave
function,
In order to determine how the environment affects the system,
we take the trace of the total density matrix over all variables of the environment,
ρ(t) = Trenv [Ψ(t)ihΨ(t)]. The resulting (reduced) density matrix describes the system'
behavior under the influence of the environment.
Ψ(t)i.
Now we specify the interaction term. We assume it as a product of two operators,
Hint = VUenv, where V acts on the system and Uenv on the environment. One can
find many examples of this interaction. For instance, the electron of the system can
be coupled capacitively to a nearby fluctuator, generated by a current flowing through
it (as in a single-electron transistor), or by any other mechanism [33, 34], belonging to
the environment. In biological systems, when considering exciton and electron transfer,
noise is mainly caused by the protein environment [26, 27, 28, 29, 30, 31]. Then, one
can write for the ET,
VUenv = Ud0ih0 + Ua
nihn! c†c,
(15)
N
Xn=1
where Ud,a is the Coulomb interaction between the electron on the donor (or on the
acceptor) and the fluctuator, and c†c is the electron density operator of the fluctuator.
Note, that when applying our approach to a donor-acceptor system which describes the
exciton transfer in bio-complexes, instead of the Coulomb interaction, the dipole-dipole
interaction is usually used. In this case, instead of the density operator, c†c, in Eq. (15),
k + h.c.), is used (see, for example [11, 32],
and references therein), where gk is a form factor related to the protein environment,
and a†
a bosonic operator of the type, ϕ =Pk(gka†
k and ak are the creation and annihilation operators of the k-th mode.
We assume that Ud 6= Ua, since the donor and the acceptor are at different spatial
locations. This model describes telegraph noise, because c†c can have only two values:
1 or 0 for an occupied or unoccupied fluctuator. Since the effect of noise on the system
is not sensitive to the particular microscopic origin of noise, we restrict our study to this
model only. This allows us to investigate the effect of the environment on the system
in the simplest way. The problems of telegraph noise have been considered in many
publications [33, 34, 35, 36, 37, 38]. We choose the approach of Ref. [38], which results
in Bloch-type master equations, which have a very transparent physical meaning. In
addition, these equations can be rigorously derived for certain microscopic models of
telegraph noise [34], or by averaging the time-dependent fluctuating energy levels, by
using a very effective method of Shapiro and Loginov [39, 40].
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
8
In this paper, we do not present a detailed microscopic treatment of these equations
for a particular Henv, but simply replace the operator c†c by a random variable,
[ζ(t) + 1]/2, where ζ(t) jumps randomly from 1 to -1 (or from −1 to 1) at a rate γ+ (or
γ−) [35]. This procedure is valid whenever there is no back-action from the system on
the noise spectrum [34].
The noise distribution is determined by the probabilities, p±, of finding ζ(t) at the
values ±1. Detailed balance then implies the relation, p−γ+ = p+γ−, and therefore
p± = γ±/(2γ), where γ = (γ+ + γ−)/2 is the inverse time associated with noise
(γ = 2/S(0), where S(ω) is the noise spectrum [38]).
The symmetric noise, γ+ = γ−, corresponds to p+ = p−, and therefore it formally
corresponds to the infinite temperature regime. Respectively, an asymmetric noise,
γ+ 6= γ−, would be considered as one corresponding to the finite temperature regime
(see, for example, [32, 41, 42].) However, we have to point out that in our consideration
noise is not an equilibrium sub-system. Indeed, noise we are dealing with is sustained
in its steady state by an external source and its spectrum, and it is not affected by the
system. The latter, therefore cannot be in a thermal equilibrium with the noise, as is
demonstrated explicitly below.
It is interesting to compare our treatment of noise with that of Haken, Reineker
and Strobl
[42, 43, 44, 45], which on the first sight look similar. However, the
main assumption of their treatment is that the phonon dynamics is infinitely quick
as compared to that of the exciton. This implies that noise is δ-correlated in time
[42]. This is not the case of our approach, which has not such restrictions. Moreover,
most important effects of noise in exciton (electron) transport take place when noise
dynamics is comparable with that of the exciton (electron) one [40]. Another, interesting
descriptions of random effects in the ET have been proposed by use of the Random
Matrix Theory technique and Keldysh non-equilibrium Green's function approach
[46, 47].
Although we use a particular model for the noise, the main goal of our approach
is to demonstrate analytically and numerically the multi-scale ET dynamics, and not
the dependencies on the characteristic parameters of an external reservoir. Indeed, in
this context, the main function of the external noise in our approach is to assist the ET
when the donor level (band) does not overlap with the acceptor band. We also note that
because the ET rates in LHCs are relatively large (∼ ps−1, or even more), no consensus
exists on what are the main contributions in the non-equilibrium ET dynamics (with
non-zero reduced density matrix elements), noise or thermal fluctuations, or both.
Let us obtain these equations for the reduced density matrix of the system, ρ(t).
First consider the case of no interaction with the environment, Hint = 0. Then, the
density matrix of the system, defined as ρjj ′(t) = bj(t)b∗
j ′(t), satisfies the following
Bloch-type equations,
ρ00 = iVN
N
Xn=1
(ρ0n − ρn0),
(16)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System
9
ρnn′ = i(En′ − En)ρnn′, +iVN (ρn0 − ρ0n′),
ρ0n = i(En − E0)ρ0n + iVN (ρ00 −Xn′
ρn′n).
These equations are derived straightforwardly from Eqs. (3)-(4).
(17)
(18)
In the presence of interaction, Eq. (15), the energy levels of the donor and acceptor
in Eqs. (16)-(18) are replaced by E0,n → E0,n + 1
2Ud,aζ(t), where the constant terms,
Ud,a/2, were included in the definition of the energy levels. As mentioned above, in the
case of the exciton energy transfer, the operator, c†c, should be replaced by the bosonic
operator, ϕ, which in our case is reduced to a random variable, ζ(t)/2, with the ensemble
average, ζ(t) = 0. So, the constant terms, Ud,a/2, do not appear in the renormalization
of the donor and acceptor energy levels.
Now it is quite natural to extend Eqs. (16)-(18) to include noise by replacing the
density matrix ρjj ′(t), Eqs. (16)-(18), by the two-component vector, {ρ(+)
jj ′ (t)}.
For the redox potential (energy gap), E0 − En → E0 − En ± D in Eq. (18), where
D = (Ud − Ua)/2. In addition, the stochastic hopping terms, ρ(+)
jj ′ (t), with
rates, γ±, should be included in the equation of motion. Thus, we replace Eqs. (16)-(18)
by the following equations of motion, which now include noise,
jj ′ (t) ←→ ρ(−)
jj ′ (t), ρ(−)
N
(19)
(20)
(21)
(22)
ρ(+)
00 = iVN
ρ(−)
00 = iVN
N
Xn=1
Xn=1
0n − ρ(+)
(ρ(+)
n0 ) − γ−ρ(+)
00 + γ+ρ(−)
00 ,
(ρ(−)
0n − ρ(−)
n0 ) − γ+ρ(−)
00 + γ−ρ(+)
00 ,
nn′ + iVN (ρ(+)
n0 − ρ(+)
0n′ )
nn′ + γ+ρ(−)
nn′,
nn′ = i(En′ − En)ρ(+)
ρ(+)
− γ−ρ(+)
nn′ = i(En′ − En)ρ(−)
ρ(−)
ρ(+)
0n = i(En − E0 − D) ρ(+)
ρ(−)
0n = i(En − E0 + D) ρ(−)
0n (t), ρ(±)
jj ′ (t) + ρ(−)
nn′ + iVN (ρ(−)
N
nn′ + γ−ρ(+)
nn′,
0n′ ) − γ+ρ(−)
n0 − ρ(−)
n′n) − γ−ρ(+)
ρ(+)
0n + iVN (ρ(+)
Xn′=1
00 −
Xn′=1
ρ(−)
n′n) − γ+ρ(−)
0n + iVN (ρ(−)
00 −
N
0n + γ+ρ(−)
0n ,(23)
0n + γ−ρ(+)
0n ,(24)
n′n(t) = ρ(±)∗
and ρ(±)
n0 (t) = ρ(±)∗
nn′ (t). Finally, one has to average over the noise, so
that, ρjj ′(t) = ρ(+)
jj ′ (t). For more detailed arguments leading to Eqs. (19)-(24)
and also for their exact microscopic quantum mechanical derivation for particular noise
models, see Refs. [34, 38].
4. One-level acceptor
4.1. Steady state
First, assume the acceptor is a one-level system, N = 1. We introduce the variables,
ρ(t) = ρ(+)(t) + ρ(−)(t) and ξ(t) = ρ(+)(t) − ρ(−)(t). Note, that ρ00(t) + ρ11(t) = 1. In
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 10
these variables, Eqs. (19)-(24) for N = 1 are,
ρ00 = iV (ρ01 − ρ10),
ρ11 = iV (ρ10 − ρ01),
ρ01 = −iǫ ρ01 + iV (ρ00 − ρ11) − iDξ01,
ξ00 = iV (ξ01 − ξ10) − 2γ ξ00 + 2ηγ ρ00,
ξ11 = iV (ξ10 − ξ01) − 2γ ξ11 + 2ηγ ρ11,
ξ01 = −(iǫ + 2γ)ξ01 + iV (ξ00 − ξ11) + (2ηγ − iD)ρ01,
(25)
(26)
(27)
(28)
(29)
(30)
where, γ = (γ+ + γ−)/2 and η = (γ+ − γ−)/(γ+ + γ−).
Consider the density matrix in the asymptotic limit, ρ(t → ∞) ≡ ¯ρ and ξ(t →
∞) ≡ ¯ξ. If the density matrix reaches its steady-state in this limit, then ¯ρ = 0. In this
case, Eqs. (25)-(30) can be easily solved. Indeed, it follows from (25) that Im ¯ρ01 = 0.
Substituting this into (27), we find Im ¯ξ01 = 0. From the real parts of Eqs. (28)-(30) we
find,
¯ξ00 = η ¯ρ00, ¯ξ11 = η ¯ρ11, Re ¯ξ01 = η Re ¯ρ01.
Taking the imaginary parts of Eqs. (27), (30), we have,
− ǫRe ¯ρ01 + V (¯ρ00 − ¯ρ11) − D Re ¯ξ01 = 0,
− ǫ Re ¯ξ01 + V ( ¯ξ00 − ¯ξ11) − D Re ¯ρ01 = 0.
Using Eqs. (31) one can rewrite these equations as,
− ǫRe ¯ρ01 + V (¯ρ00 − ¯ρ11) − η D Re ¯ρ01 = 0,
Re ¯ρ01 = 0.
− ǫ Re ¯ρ01 + V (¯ρ00 − ¯ρ11) −
D
η
(31)
(32)
(33)
It immediately follows from these equations that, Re ¯ρ01 = Re ¯ξ01 = 0 and ¯ρ00 = ¯ρ11 =
1/2. This implies equal distribution of the donor and acceptor in the asymptotic limit
for any initial conditions. This result is drastically different from the no-noise case
(D = 0 or η = ±1 in Eqs. (33)), considered in the previous section, where there is
no steady-state, and the population of the acceptor at t → ∞ remains very small if
ǫ ≫ V . In the case of noise, however, the system always reaches equal distribution in
the steady-state, no matter how small the noise is.
Note, that the equal distribution between the donor and the acceptor populations is
always reached in the asymptotic limit, irrespectively of the relative position of the donor
and the acceptor levels. This implies that the up and down-hill transitions, generated
by noise, proceed with the same probabilities. Usually such a behavior is considered
as taking place in the high-temperature limit. However, this is not necessarily the
case. Indeed, as we proved above, the equal populations of the donor and the acceptor
in the asymptotic limit, takes place even for η 6= 0 (see Eqs. (33)), corresponding to
p+ 6= p−. Thus, the equal population of the donor and the acceptor is related to the
non-equilibrium effect of noise.
Indeed, noise is sustained in its steady state by an
external source and therefore it is not affected by the system. As a result, the average
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 11
probabilities for the system of loosing and gaining energy from noise will be the same,
resulting in the same occupation of the system states.
Now a natural question can be asked, what is a role of the system and noise
parameters in the transition to equal distribution, since the later takes place for any
values of these parameters. The answer is in the relaxation times which, for instance,
can by very long, if noise is weakly coupled to the system. The analysis of the relaxation
times is a main subject of this paper. Now we are going to evaluate the transition rates
by analyzing the time-dependent Eqs. (25)-(30).
4.2. Transition time (rate)
Consider, for simplicity, η = 0. In order to solve Eqs. (25)-(30), we apply the Laplace
0 ρ(t) exp(iEt) dt, and correspondingly, ξ(t) → ξ(E). Then,
transform, ρ(E) = R ∞
Eqs. (25)-(30) become (see Ref. [34, 38]),
iE ρ00 + iV (ρ01 − ρ10) = −i,
iE ρ11 + iV (ρ10 − ρ01) = 0,
(iE − iǫ)ρ01 + iV (ρ00 − ρ11) − iD ξ01 = 0,
(iE − 2γ) ξ00 + iV ( ξ01 − ξ10) = 0,
(iE − 2γ) ξ11 + iV ( ξ10 − ξ01) = 0,
(iE − iǫ − 2γ) ξ01 + iV ( ξ00 − ξ11) − iD ρ01 = 0,
01(−E). Equations (34)-(39) can be rewritten in matrix form as,
(iE I + M) R(E) = −R(0),
and ρ10(E) = ρ∗
where the density matrix, R(E) (R(t)), is written as an 8-vector,
(34)
(35)
(36)
(37)
(38)
(39)
(40)
R = {ρ00, ρ11, ρ01, ρ10, ξ00, ξ11, ξ01, ξ10} ,
(41)
M is an 8× 8 matrix corresponding to the r.h.s. part of Eqs. (34)-(39), and I is an 8× 8
unit matrix. Solving these equations, we obtain rational expressions for the Laplace
transformed density matrix elements,
Rk(E) =
det[mk(E)]
det[iE I + M]
,
(42)
where k = {1, 2, . . . , 8}, and mk is the corresponding minor determinant.
The density matrix as a function of t is finally obtained via the inverse Laplace
transform, Eq. (8),
Rk(t) =
∞
Z−∞
R(E) e−iEt dE
2π
.
(43)
The secular and minor determinants can be represented by polynomials in powers of E.
One finds for the secular determinant the following expression,
det[iE I + M] = E(E + 2iγ)
ApEp ,
6
Xp=0
(44)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 12
where
A0 = −16D2V 2γ2,
A1 = 2iγ[(D2 − ǫ2 + 4V 2)2 + 16V 2ǫ2 + 4γ2(4V 2 + ǫ2)],
A2 = (D2 − ǫ2 + 4V 2)2 + 16V 2ǫ2 + 4γ2(12V 2 + 3ǫ2 + 2D2),
A3 = −8iγ(cid:0)γ2 + D2 + 4V 2 + ǫ2(cid:1) ,
A4 = −2(cid:0)6γ2 + D2 + 4V 2 + ǫ2(cid:1) ,
A5 = 6iγ, A6 = 1.
(45)
(46)
The integral in (43) can be calculated analytically by closing the integration contour
in the lower half E-plane over the poles of denominator (E = Er). One finds,
ρk(t) = −i
e−iU (r)t−Γ(r)t,
8
Xr=1
det[mk(Er)]
Qr′6=r(Er − Er′)
where Er = U (r) − iΓ(r) are the zeros of the secular determinant,
det[iEr I + M] = 0,
(47)
(48)
in the complex E-plane. The first pole, at U (1) = Γ(1) = 0, produces a finite occupation
in the asymptotic limit. The second pole, at U (2) = 0 and Γ(2) = 2γ, produces a decay
of the corresponding term with the rate, 2γ. The remaining poles are obtained from the
equation,
A0 + A1E + A2E2 + · · · + A6E6 = 0.
(49)
The asymptotic transition rate, Γ1, is given by the pole with the smallest imaginary
part, Γ1 = min{Γ(r)}.
Now we will find an approximate analytical solution of Eq. (49). Since we are
looking for the pole with minimal value of the energy, we can keep only the first
two terms in Eq. (49), neglecting the higher powers in E. This yields the transition
rate, Γ1 = Im[A0/A1]. Using Eq. (46), one can write explicitly for the transition time,
τ1 = 1/Γ1,
τ1 =
(D2 − ǫ2 − 4V 2)2 + 4γ2ǫ2
8γD2V 2
+
2(D2 + γ2)
γD2
.
The accuracy of this procedure is determined by the parameter,
κ =
A0A2
A2
1
=
A2
γ (cid:19)2
16D2V 2 (cid:18)Γ1
,
which is expected to be less than one.
(50)
(51)
It follows from Eq. (50), that the transition time is minimal at D2 = ǫ2 + 4V 2 for
γ ≪ ǫ. This has a simple physical meaning: due to the influence of noise, the donor
level fluctuates between E0 ± D. This makes it partially in resonance with the acceptor'
levels. As a result, the rate of the donor-acceptor transitions increases. This effect
is illustrated in Fig. 3, where we show the occupation of the donor as a function of
time for different values of the noise amplitude. The solid lines show ρ00(t), obtained
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 13
from the numerical solution of Eqs. (25)-(30). The dashed lines are the asymptotic rate
approximation,
ρ00(t) =
1
2
(1 + e−Γ1t),
(52)
and Γ1 = 1/τ1 is given by Eq. (50). The following parameters are used (in arbitrary
units): ǫ = 5, V = 1, γ = 1. The noise amplitudes are: D = 1 (red lines), D = 5 (blue
lines), D = 10 (black lines).
Ρ00t
1.0
0.9
0.8
0.7
0.6
0.5
0
100
200
300
400
t
500
Figure 3. (Color online) Donor occupation a function of time, for N = 1, η = 0,
ǫ = 5, V = 1, γ = 1, for 3 values of noise amplitude: D = 1 (red lines), D = 5 (blue
lines) and D = 10 (black lines). Solid lines show the results of the numerical solution of
Eqs. (25)-(30), and dashed lines correspond to the asymptotic approximation, Eq. (52).
It follows from this figure, that the asymptotic limit, given by Eq. (52), describes
the behavior of ρ00(t) quite well. As expected, the shortest transition time corresponds
to, D ≃ ǫ.
Now we consider an asymmetric noise, γ+ 6= γ− (or η 6= 0), corresponding to
different noise probabilities, p+ 6= p−. Obviously, in the extreme case of very large
asymmetry, p+ = 1 and p− = 0 (or η = 1), there is no noise effect on the donor-acceptor
transition. Then, the donor (acceptor) occupation stays very far from equal distribution,
and does not reach the steady-state limit, Fig. 2. However, for any other values of η 6= 1,
the probabilities of the donor and the acceptor occupations become equal and reach the
steady state limit. Therefore, we anticipate very small transition rate when η is very
close to 1.
This is illustrated in Fig. 3, which shows the donor occupation as a function of
time, obtained from the numerical solution of Eqs. (25)-(30), for ǫ = 5, D = 10, V = 1,
γ = 1, and three values of the noise asymmetry: η = 0 (solid black line), η = 0.5 (dashed
blue line) and η = 0.9 (dot-dashed red line), as a result of the numerical solution of
Eqs. (25)-(30). As expected, the donor reaches the asymptotic limit very slowly for
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 14
Ρ00t
1.0
0.9
0.8
0.7
0.6
0.5
0
50
100
150
t
200
Figure 4. (Color online) Donor occupation as a function of time for asymmetric noise,
for N = 1, ǫ = 5, D = 10, V = 1, γ = 1, for three values of noise asymmetry: η = 0
(solid black line), η = 0.5 (dashed blue line) and η = 0.9 (dot-dashed red line), as a
result of the numerical solution of Eqs. (25)-(30).
η = 0.9. However, otherwise it is not very different from the symmetric noise, η = 0.
Therefore, in the rest of this paper we concentrate only on η = 0, mainly because the
analytical formulas are most simple for interpretation, without loosing a generality.
4.3. Reduced master equations
Equations (25)-(27) resemble Bloch-type equations for the two-state density matrix,
ρ(t), except for the last term in Eq. (27), depending on ξ01(t). In fact, ξ(t) is a function
of ρ(t), so that equations for ρ(t) can be written in a closed form. This can be done by
resolving Eqs. (37)-(39) for the Laplace transformed amplitudes, ξ(E). Consider again,
for simplicity, the case of γ+ = γ− = γ. One finds,
ξ01 = D
[(E + 2iγ)(E + 2iγ + ǫ) − 2V 2]ρ01 + 2V 2 ρ10
(E + 2iγ)[(E + 2iγ)2 − ǫ2 − 4V 2]
.
(53)
The time-dependent amplitude, ξ01(t), is obtained through the inverse Laplace
transform (8) of the amplitude, ξ01(E), by closing the integration contour over the poles
in the complex E-plane. Since we are interested in the asymptotic regime (t → ∞),
only the pole which is closest to zero survives. We therefore can replace E → 0 in the
prefactors of the amplitudes, ρ, in Eq. (53), thus obtaining,
(2γ2 + V 2 − iγǫ)ρ01(t) − V 2ρ10(t)
(54)
.
ξ01(t) = −iD
γ(ǫ2 + 4V 2 + 4γ2)
Substituting this result into Eq. (27), we obtain,
ρ00 = iV (ρ01 − ρ10),
ρ01 = −iǫ′ρ01 + iV (2ρ00 − 1) − γ1ρ01 − γ2(ρ01 − ρ10),
(55)
(56)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 15
where,
ǫ′ = ǫ(cid:18)1 −
D2
ǫ2 + 4V 2 + 4γ2(cid:19) ,
is a renormalized donor energy, and
γ1 =
2γD2
ǫ2 + 4V 2 + 4γ2 ,
γ2 =
V 2
2γ2 γ1,
are the damping rates.
(57)
(58)
Equations (55)-(56) have the form of Bloch equations for spin precession in a
magnetic field in the presence of the environment. This can be seen by mapping
the density matrix, ρ(t), to the "polarization" vector, ~S(t) = {Sx(t), Sy(t), Sz(t)}, via
ρ(t) = [1 + ~τ · ~S(t)]/2, where τx,y,z are the Pauli matrices. Thus, we define Sz = 2ρ00 − 1,
Sy = i(ρ01 − ρ10) and Sx = ρ01 + ρ10. We find,
Sz = 2V Sy,
Sy = −2V Sz − (γ1 + 2γ2)Sy + ǫ′Sx,
Sx = −ǫ′Sy − γ1Sx.
(59)
These equations coincide with the Bloch equations, where γ1,2 are related to the two
damping times, T1 = γ−1
1 and T2 = (γ1 + 2γ2)−1.
Similar equations can be derived when the interaction, V (tunneling coupling)
fluctuates, instead of fluctuating donor and acceptor energy levels [34, 38]. However,
the redox potential, ǫ, would not be renormalized, as occurs in the case of the energy-
level fluctuations, Eq. (57). This difference is essential for the electron transfer. Indeed,
fluctuations of the energy levels can drive the donor and acceptor into resonance, which
can greatly increase the transfer rate, Eq. (50), Fig. 3.
In contrast, in the case of
fluctuating coupling (V ), resonance cannot occur, even though both of these noise-
assisted processes are described by similar Bloch-type equations.
In the weak interaction limit, V ≪ D, γ, one finds that T1 = T2. Then, Eqs. (55)-
(56) become further simplified. Solving these equations in this limit, we find for the
asymptotic transition rate,
Γ1 =
8γD2V 2
(D2 − ǫ2)2 + 4γ2ǫ2 ,
which coincides with Eq. (50) in the same limit.
(60)
The weak interaction limit can be very useful for the multi-level case, since it greatly
simplifies the treatment, without losing any physical features of the process. We show
in Fig. 5 the occupation of donor, ρ00(t), for different values of the noise amplitude.
The solid lines correspond to Eqs. (25)-(30), whereas the dashed lines are obtained
from the reduced master equations (59) in the limit of weak interaction: γ2 = 0 and
γ1 = 2γD2/(ǫ2 + 4γ2). The parameters are the same as in Fig. 3. It follows from this
figure that the reduced Bloch-type master equations describe the asymptotic limit very
well, even for V ∼ γ.
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 16
Ρ00t
1.0
0.9
0.8
0.7
0.6
0.5
0
100
200
300
400
t
500
(Color online) Donor occupation as a function of time, for N = 1, for
Figure 5.
parameters the same as in Fig. 3. Solid lines show the result of the numerical solution
of Eqs. (25)-(30), and dashed lines correspond to Eqs. (55)-(56) in the limit of small
V .
5. N -level acceptor
Consider Eqs. (19)-(24). As in the previous section, we rewrite these equations in the
variables, ρ = ρ(+) + ρ(−) and ξ = ¯ρ(+) − ρ(−). Then, these equations can be rewritten
as,
N
ρ00 = iVN
(ρ0n − ρn0),
Xn=1
where ǫn′n = En′ − En.
5.1. Degenerate case
ǫnn′ = 0 for
Consider the case when all energy levels of the acceptor coincide:
n, n′ = 1, . . . , N. Then, one can sum over the acceptor states, n, in Eqs. (61)-(66),
N
ρn′n(cid:17) − iD ξ0n,
Xn′=1
(ξ0n − ξn0) − 2ηγ ρ00,
ρnn′ = −iǫnn′ ρnn′ + iVN (ρn0 − ρ0n′),
ρ0n = iǫn0 ρ0n + iVN(cid:16)ρ00 −
ξ00 = −2γ ξ00 + iVN
Xn=1
ξnn′ = (iǫn′n − 2γ) ξnn′ + iVN (ξn0 − ξ0n′) + 2ηγ ρnn′,
ξ0n = (iǫn0 − 2γ) ξ0n + iVN(cid:16)ξ00 −
N
N
Xn′=1
ξn′n(cid:17) + (2ηγ − iD) ρ0n,
(61)
(62)
(63)
(64)
(65)
(66)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 17
thus obtaining,
ρ00 = iV (01 − 10),
11 = iV (10 − 01),
01 = −iǫ 01 + iV(cid:16)ρ00 − 11(cid:17) − iD ζ01,
ξ00 = iV (ζ01 − ζ10) − 2γ ξ00 − 2ηγ ρ00,
ζ11 = −2γ ζ11 + iV (ζ10 − ζ01) + 2ηγ 11,
ζ01 = −(iǫ + 2γ) ζ01 + iV(cid:16)ξ00 − ζ11(cid:17) + (2ηγ − iD) 01,
where
11 =
ζ11 =
1
N
1
N
N
N
Xn,n′=1
Xn,n′=1
ρnn′,
01 =
ξnn′,
ζ01 =
1
√N
1
√N
ρ0n,
ξ0n.
N
N
Xn=1
Xn=1
(67)
(68)
(69)
(70)
(71)
(72)
(73)
(74)
One finds that Eqs. (67)-(72) coincide with Eqs. (25)-(30), describing time-evolution
of the density matrix, ρ(t), for the case of one-level acceptor. As follows from these
equations, ρ00(t) + 11(t) =const. However, in contrast with the one-level acceptor, this
constant is not unity, in general, and its value depends on the initial conditions. Indeed,
11(t) is not the occupation probability of acceptor, since it includes the off-diagonal
density matrix elements. It can be easily seen by rewriting it explicitly,
11(t) =
1
N
ρnn(t) +
1
N
N
Xn=1
N
Xn6=n′
Re ρnn′(t).
(75)
Consider now the asymptotic limit, where we denote, ¯ρ00 = ρ00(t → ∞) and
¯11 = 11(t → ∞). Using the same transformations as in Sec.
IV A, we arrive at
¯ρ00 = ¯11, Eqs. (33). Thus, the probability of finding the donor occupied in the steady-
state is,
¯ρ00 =
1
2
[ρ00(0) + 11(0)] =
1 + (N − 1)ρ00(0)
2N
+
1
2N
Re ρnn′(0),
(76)
N
Xn6=n′
where we used the normalization condition, ρ00(0) +PN
In contrast with the case of one-state acceptor, the off-diagonal density matrix
n=1 ρnn(0) = 1.
elements (coherences) do not vanish in the steady state limit. One finds,
N
Xn6=n′
Re ¯ρnn′ = (N + 1)¯ρ00 − 1.
(77)
This is the case of partial decoherence [38, 53], which takes place when the quantum
system possesses a symmetry, that cannot be destroyed by the environment.
It follows from Eq. (76), that the steady-state distribution depends on the initial
state. In particular, ¯ρ00 = 1/2 for the initially occupied donor, ρ00(0) = 1. However,
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 18
from the previous study of the electron transport in a similar system [52], one would
expect equal occupation of all levels in the limit of t → ∞, for any value of the noise
amplitude and level spacing. This would be drastically different from the degenerate
level case. To understand this problem, we investigate the non-degenerate case in detail.
5.2. Asymptotic state
Consider now the asymptotic limit, ρ(t → ∞) = ¯ρ, and ξ(t → ∞) = ¯ξ, where the
acceptors energy levels are not degenerate. Since in this limit ρ(t) → 0 and ξ(t) → 0,
we can rewrite Eqs. (61)-(66) as,
N
ρ00 +
ρnn = 1,
Xn=1
ǫn′n ρnn′ + VN (ρn0 − ρ0n′) = 0,
ǫn0 ρ0n + VN(cid:16)ρ00 −
Xn′=1
N
ρn′n(cid:17) − D ξ0n = 0,
2VN
Im ξ0n + 2γ ξ00 = 0,
N
Xn=1
(iǫn′n − 2γ )ξnn′ + iVN (ξn0 − ξ0n′) = 0,
(iǫn0 − 2γ) ξ0n + iVN(cid:16)ξ00 −
ξn′n(cid:17) − iDρ0n = 0,
where ǫjj ′ = Ej − Ej ′. For simplicity, we assume γ+ = γ− = γ.
Xn′=1
N
Since ǫnn′ 6= 0, we obtain from Eq. (79)-(81),
Im ρ0n = Im ρnn′ = Im ξ0n = ξ00 = 0.
Then, one finds from Eqs. (82) and (83),
Re (ξ0n − ξn′0)
ξn′n = iVN
,
iǫnn′ − 2γ
(iǫn0 − 2γ) Re ξ0n + V 2
N
Taking the real part of Eq. (86) we find,
N
Xn′=1
Re (ξ0n − ξn′0)
iǫnn′ − 2γ
= iDRe ρ0n.
(78)
(79)
(80)
(81)
(82)
(83)
(84)
(85)
(86)
(87)
Xn′ = 0,
Cnn′
1 +Pn′ Cnn′
Xn −Xn′
n′=1 A(N )
N /(ǫ2
where Xn = Re ξ0n and Cnn′ = V 2
nn′ + 4γ2). This is a system of coupled linear
equations, PN
nn′ Xn′ = 0, where det A(N ) > 0. (See Appendix A for details.) As a
result, Xn = Re ξ0n = 0, for all n. Inserting this into Eq. (86), we find that Re ρ0n = 0.
Substituting this result into Eq. (79), one finds Re ρnn′ = 0 for n 6= n′. Then, using
Eqs. (78) and (80), we find equal occupation of all levels in the asymptotic limit of
t → ∞,
.
(88)
ρ00 = ρnn =
1
N + 1
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 19
This implies that for N ≫ 1, the probability of finding the acceptor occupied is close to
unity. So, the efficiency of the electron (exciton) transfer can be close to 100%.
It is remarkable that the above result has been derived for any value of spacing
between the acceptor' levels (ǫnn′), no matter how small. However, for the case of the
exact degenerate acceptor' levels, ǫnn′ = 0, this proof is not valid, since we cannot
use Eq. (79) to obtain Re ρnn′ = 0 for n 6= n′.
Indeed, this quantity is not zero in
the degenerate case, as follows from Eq. (77). Moreover, the asymptotic distribution
depends on the initial conditions, Eq. (76). One finds, ¯ρ00 = 1/2, for the initial condition
corresponding to the occupied donor, instead of 1/(N + 1) for the non-degenerate case,
no matter how small the acceptor' band-width, δa, Eq. (13).
Thus the above result displays discontinuity of the steady-state occupation with
the acceptor's bandwidth, δa. The only possible explanation of this discontinuity can
be found in an analysis of the donor-acceptor transfer dynamics. We can anticipate
that the system reaches first its equal donor-acceptor occupation (1/2). Then, it is
distributed inside the acceptor's states, finally approaching equal occupation for all
levels, 1/(N + 1). One expects that the second transition rate would depend on the
level spacing, and that decreases to zero when the acceptor's levels are degenerate. In
this case, the equal distribution, 1/(N + 1), can never be reached, so that the system
stays asymptotically in the equal donor-acceptor occupation (1/2). Such an explanation
can resolve the discontinuity problem, and it will be confirmed by the following analysis
of the transfer dynamics.
5.3. Reduced master equations
In order to determine the transition rate, we first reduce Eqs. (61)-(66) to simplified
equations that involve only the density matrix, ρ(t). This is similar to the case of Sec.
5.3, for N = 1. For this reason, we will express ξ(t) in terms of ρ(t) using Eqs. (64)-(66).
For simplicity, we assume η = 0, and consider the limit of VN ≪ γ, D. Then, we can
neglect the term proportional to VN in Eq. (66), thus obtaining,
ξ0n(t) = −
iD
iǫ0n + 2γ
ρ0n(t).
Substituting this result in Eq. (63), we find,
(ρ0n − ρn0),
ρ00 = iVNXn=1
ρnn′ = −iǫnn′ ρnn′ + iVN (ρn0 − ρ0n′),
0n + γ0n) ρ0n + iVN(cid:16)ρ00 −
ρ0n = −(iǫ′
where VN = V /√N and,
γ0n =
2γD2
0n + 4γ2 and ǫ′
ǫ2
0n = ǫ0n(cid:18)1 −
N
Xn′=1
ρn′n(cid:17),
D2
0n + 4γ2(cid:19) .
ǫ2
(89)
(90)
(91)
(92)
(93)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 20
These equations are similar to the Bloch-type equations (59) for a two-state system,
but now they have been extended to include a multi-level acceptor. The most
pronounced feature of these equations is the renormalization of the donor-acceptor
energies, ǫ0n → ǫ′
0n (Eq. (93)), due to noise. This reconstruction of the redox potential
is given by: ∆ǫ0n/ǫ0n = −D2/(ǫ2
0n +4γ2). This is analogous to the reconstruction energy
in the Marcus ET rate, but for a multi-level acceptor, and a noisy (instead of thermal)
environment [32, 41]. As a result, the transition rate can be greatly increased if ǫ′
0n ≃ 0,
when the donor and acceptor are effectively in resonance.
We show in Fig. 6 the donor occupation as a function of time, obtained from
Eqs. (61)-(66) (black lines), in comparison with the same quantity obtained from the
reduced master equations (59) (red lines) for N = 2 and N = 10. The parameters are:
ǫ = 5, V = 1, γ = 1, and D = 5, corresponding to the effective "resonance conditions".
The acceptor levels, En, are given by Eq. (13), where the acceptor bandwidth, δa = 1,
is the same for N = 2 and N = 10. The dashed lines show the asymptotic limit,
1/(N + 1). The results shown in Fig. 6 clearly demonstrate that the reduced master
equations are a very good approximation, even beyond the condition, VN ≪ γ, D, used
for their derivation.
Ρ00t
1.0
0.8
0.6
0.4
0.2
0
100
200
300
400
t
500
(Color online) Donor occupation for the two cases of acceptor levels,
Figure 6.
N = 2, 10, but with the same bandwidth, δa = 1. The noise amplitude, D = ǫ = 5.
The other parameters are the same as in Fig 3. The dashed lines correspond to the
asymptotic limit, 1/(N + 1).
It follows from Fig. 6, that the characteristic time, τN , needed to approach equal
occupation of all states of the system is much longer for N = 10, than for N = 2,
although the bandwidth, δa, is the same for both cases. (See also [54], where the case
N = 2 was considered for a thermal environment.) This dependence on N is not trivial,
and at first glance, even counter-intuitive. Indeed, if the bandwidth, δa, of the acceptor is
fixed and N increases, then the acceptor density of states increases. So, intuition tells us
that the ET rate may increase. According to Fig. 6, it indeed increases, but only for the
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 21
initial stage of the acceptor' population. At later times, other factors become important.
For the parameters chosen in Fig. 6, the bandwidth, δa is 5 times smaller than ǫ. In
this case, two characteristic regimes can be expected. The first one is a relatively rapid
population of the acceptor with the population close to 1/2, as for two-level system.
The ET rate for this stage is greater for N = 10 compared with N = 2. The second
regime involves a re-population of the acceptor levels. This requires additional time,
and is accompanied by oscillations which are clearly seen in Fig. 6. The period of these
oscillations can be estimated to be: T = 2πN/δ. So, even if the asymptotic efficiency
(probability) of the acceptor' population, Pa(t → ∞) = N/N + 1, increases with N
increasing, the time of approaching this asymptotic regime can increase with N. Below,
we analyze this dependence on both N and δ, and show that the following scaling exists:
τN ∝ (N/δ)2, for large N. These results are very important for engineering of this type
of bio-complexes in order to achieve optimal ET rates and efficiencies. Below, we derive
analytical formulas for the transition time as a function of level spacing.
5.4. Transition rates
We now evaluate the transfer rates, using Eqs. (90)-(92). We start with N = 2. Then,
the Laplace transform of these equations can be written:
iE ρ00 + iV2 (ρ01 − ρ10) + iV2 (ρ02 − ρ20) = −i,
iE ρ11 + iV2 (ρ10 − ρ01) = 0,
iE ρ22 + iV2 (ρ20 − ρ02) = 0,
(iE − iδ)ρ12 + iV2 (ρ10 − ρ02) = 0,
(cid:16)iE − iǫ′
(cid:16)iE − iǫ′
01 − γ01(cid:17)ρ01 + iV2 (ρ00 − ρ11 − ρ21) = 0,
02 − γ02(cid:17)ρ02 + iV2 (ρ00 − ρ22 − ρ12), = 0,
(94)
(95)
(96)
(97)
(98)
(99)
where δ = ǫ12, V2 = V /√2, and ǫ′
01(2), γ01(2) are given by Eq. (93), with ǫ01(2) = ǫ ∓ δ/2.
The asymptotic transition rate, Γ2, is obtained from Eq. (48) by expanding the
secular determinant in powers of E, Eq. (49), and keeping only the lowest power
term, which dominates the asymptotic behavior. By neglecting the higher order terms
proportional to δ4 and V 2δ2, we obtain for the transition time τ2 = 1/Γ2,
τ2 =
16
3
τ1 +
2V 2 (4γ2 + ǫ2)
3 γ D2 δ2
,
(100)
where τ1 is the transition time for N = 1, Eq. (50).
It follows from Eq. (100) that in the limit of δ → 0, the time required for equal
occupation of all acceptor' states, Eq. (88), diverges as 1/δ2. This is not surprising,
since this limit corresponds to the degenerate case, considered in Sec. 5.1. Then, all
levels of the acceptor effectively become a single level. As a result, the acceptor is only
partially occupied in the asymptotic limit. Indeed, its occupation approaches 1/2, with
the corresponding occupation rate Γ1, Eq. (60). This implies that for δ 6= 0, the acceptor
occupation (depletion of donor) can be described effectively by two rates. The first, Γ1,
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 22
populates the acceptor up to an occupation of 1/2. Then, the process proceeds with the
second rate, Γ2 ≪ Γ1, to the asymptotic population of 2/3. Neglecting the first term in
Eq. (100), we can write,
Γ2 =
3 γ D2 δ2
2V 2 (4γ2 + ǫ2)
.
(101)
In fact, both transition rates, given by Eqs. (60) and (101), can be obtained from
the same Eq. (48), by keeping the higher order terms in the expansion of the secular
determinant in powers of E.
Ρ00t
1.0
0.8
0.6
0.4
0.2
0
100
200
300
400
t
500
Figure 7. (Color online) Donor occupation as a function of time, for N = 2, ǫ = 5,
V = 1, γ = 1, δa = 1, for 3 values of the noise amplitude: D = 1 (red lines), D = 5
(blue lines) and D = 10 (black lines). Solid lines show the results of the numerical
solution of Eqs. (61)-(66), and the dashed lines correspond to Eq. (102).
As in Eq. (52), but now with two exponents, one can effectively represent the
donor-acceptor transitions as,
ρ00(t) =
1
6
(1 + e−Γ1t)(2 + e−Γ2t).
(102)
This simple formula describes the dynamics of the donor-acceptor transition reasonably
well, as demonstrated in Fig. 7. We display there the donor occupation, given by
Eq. (102) (dashed lines), for different values of the noise amplitude, D, in comparison
with the exact calculations, Eqs. (61)-(66), shown by the solid lines.
It is remarkable that the second rate, Γ2,
in
contrast with Γ1, Eq. (60). This supports the two-step dynamics of the donor-acceptor
transitions.
Indeed, during the first step the transferred electron does not "discern"
individual levels of the acceptor. Therefore, one can anticipate that the corresponding
rate, Γ1, is independent of the number of acceptor states (N) and has a Lorentzian-type
shape as a function of the noise amplitude.
is not of the Breit-Wigner type,
At the next step, the electron is redistributed among the acceptor' states. Since the
energy spread of these states is narrow, there is no reason to expect any Lorentzian-type
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 23
dependence of the second rate Γ2 on the noise amplitude, D. It it natural to expect
the second rate Γ2 would always increase with D, as it is demonstrated by Eq. (101).
This point could be very important, since a large noise amplitude can compensate the
decrease of Γ2 with a decrease of the level splitting.
The proposed two-step dynamics for ET suggests a natural extension of our results
to any number of acceptor states, N. This can be done by replacing the bandwidth,
δa, in Eq. (101) for N = 2 by the ratio, δ/(N − 1), for any N ≥ 2, which represents
the level spacing (inverse density of states). Then, the corresponding transition time
for occupation of all acceptor states (N) would be τN = 1/ΓN . The partial occupation
of n < N acceptor states requires less time, of course. The transition time can be
evaluated using the same expression, obtained from Eqs. (60), (100), by replacing δa by
δa/(n − 1). Finally, the total transition time becomes,
V 2 (4γ2 + ǫ2)
(D2 − ǫ2)2 + 4γ2ǫ2
+
γD2V 2
τn ∝
γ D2 δ2
a
(n − 1)2.
(103)
Here the first term is a "coarse-grained" time, ∝ 1/Γ1, Eq. (60), when the acceptor can
be considered as a single level. The maximal acceptor occupation in this case can reach
only 1/2. Subsequently, the electron is redistributed among the acceptor states. The
corresponding time is given by the second term in Eq. (103). ‡
Interesting "scaling" regime occurs when the noise amplitude, D = ǫ, and γ ≪ ǫ.
In this case, the transition time is given by,
τN ∝
4γ
V 2 +
V 2
γ (cid:16) N − 1
δa (cid:17)2
.
(104)
It is quite remarkable that in this case the transition time does not depend on the redox
potential, ǫ, and on the noise amplitude, D, but only on its spectral width, γ. Moreover,
it scales with γ/V 2.
This prediction can be verified by a direct evaluation of the donor occupation,
ρ00(t), for different values of γ, by keeping V , and under the condition, D = ǫ. The
results of the calculations, using the exact equations, Eqs. (61-66), are shown in Fig. 8,
for N = 20.
The red and black curves correspond to V = 1, γ = 1 and V = 0.5, γ = 0.25, so
that the ratio of V 2/γ remains the same. These curves display the same transfer time
at large t, in agreement with the scaling, Eq. (104). For comparison, we display ρ00(t)
for V = 0.5, γ = 1, which are out of the scaling (dashed-blue curve). This clearly shows
a quite different transfer time.
In Fig. 9, the asymptotic transition rate, ΓN = 1/τN , as a function of the noise
amplitude, D, is demonstrated for N = 1, 5, 10, 20. As it follows from Eq.(103), the
transition rate, ΓN (D) experiences a resonant behavior with a maximum at D = DN ,
‡ Note, that the coarse-grained dynamics, described by Eq. (103), cannot be valid for a "Markovian"
acceptor (with infinite band-width and constant density of state), which generates a pure exponential
decay, (see Sec. 2). Formally it follows from Eq. (100), obtained as an expansion in powers of
δ2/(V 2 + ǫ2).
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 24
(Color online) Donor occupation as a function of time, for N = 20:
Figure 8.
ǫ = D = 5 and δa = 1. Red and black curves correspond to V = 1, γ = 1 and
V = 0.5, γ = 0.25, respectively. Dashed-orange curve corresponds to values of V and
γ out of the scaling: V = 0.5, γ = 1. The asymptotic limit is shown by green line.
Figure 9. (Color online) The asymptotic transition rate, ΓN , as a function of the
noise amplitude, D, for ǫ = 5, V = 1, γ = 1, δa = 1 and N = 1, 5, 10, 20 (from up to
bottom).
noise amplitude is in resonance with the redox potential, if
DN =(cid:16)(ǫ2 + 4γ2)(cid:16)ǫ2 +
V 4
δ2
a
(N − 1)2(cid:17)(cid:17)1/4
.
(105)
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 25
(a)
(b)
Figure 10. (Color online) Donor occupation as a function of time. The asymptotic
limit is shown by green line. Parameters: V = 1, D = 5, γ = 1, δa = 1; a) N = 20,
E0 = 5 (blue curve), E0 = −5 (red curve). Inset: occupation, ρ11(t), of level, n = 1,
of the acceptor. (b) N = 50, E0 = 10 (blue curve), E0 = −10 (red curve). Inset:
occupation, ρ11(t), of level, n = 1, of the acceptor. The occupation, ρ5050(t), of level,
n = 50, is depicted by orange dashed curve.
Using this results, one can rewrite Eq. (103) as,
τN ∝
D4 − 2ǫ2D2 + D4
N
γD2V 2
.
(106)
The transition time, τN , reaches the minimum value at D = DN , which corresponds to
maximum value of the transition rate, Γn, in Fig. 9,
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 26
N = 1/Γmax
τ min
N ∝
D2
N − ǫ2
γV 2
.
(107)
Uphill ET:
In this case, the energy level of donor, E0, is positioned below the energy
levels of the acceptor band. In Fig. 10, we presented the results of numerical simulations
on comparison of downhill and uphill ET. In Fig. 10a, the blue curve demonstrates the
downhill ET (E0 = 5), and the red curve demonstrates the uphill ET (E0 = −5), for
In the insert the dynamics of ρ11(t) is shown for the same parameters. The
ρ00(t).
asymmetry in the behavior of blue and red curves occurs because the energy level E1 is
the lowest level of the acceptor band for the downhill ET (blue curve) and the upper level
for the uphill ET (red curve). In Fig. 10b, we present similar results as in Fig. 10a, but
for N = 50 and E0 = ±10. As one can see from the insert in Fig. 10b, the symmetry
is restored for the functions, ρ11(t) (blue dashed curve) and ρ50,50(t) (orange dashed
curve).
Now, we will analyze the multi-scale time given by Eq. (103). It shows that the total
depletion time of the donor τN is proportional to 1/δ2
a. Thus, τN will strongly decrease
with increasing bandwidth, δa, in particular when it approaches ǫ. This is illustrated in
Fig. 11. Here, too we used the exact equations Eqs. (61)-(66), for evaluation of ρ00(t).
(Color online) Donor occupation as a function of time, for N = 20,
Figure 11.
ǫ = D = 5, V = γ = 1, and different values of the bandwidth: δa = 40 (orange),
δa = 5 (black), δa = 1 (red), δa = 0.1 (blue), δa = 0.01 (red-dashed). The asymptotic
limit is shown by green line.
As one can see from Fig. 11, an increase of the acceptor bandwidth, δa, significantly
decreases the time of population of the acceptor.
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 27
5.5. Non-equidistant acceptor' energy spectrum
As one can see from the results shown in Figs. 6, 8, and 10, the quantum coherent
damping oscillations are observed during the process of the acceptor' population. In
our model, these oscillations result from a re-population of the acceptor' states during
the ET process, and they are especially pronounced under the condition: ǫ = D. For
the equidistant energy spectrum of the acceptor' band, given by Eq. (13), the period of
these oscillations can be estimated as: T = 2πN/δ. The question arises if these quantum
coherent oscillations will be observed for non-equidistant acceptor' energy spectrum, and
up to what extent the destructive quantum interference effects could suppress them.
To clarify this issue, we performed numerical simulations for the non-equidistant
acceptor' energy spectrum given by,
En =
(108)
2n − 1 − N
2(N − 1)
δa +
κδa
N
sin(cid:16) πn
√3(cid:17).
The first term in this expression describes the equidistant energy spectrum given by
Eq. (13), The second term in Eq. (108) was chosen to model the non-equidistant part of
the spectrum. In Fig. 12, we present the results of numerical simulations of the donor
(Color online) Donor occupation as a function of time, for N = 50,
Figure 12.
δa = 30, ǫ = D = 40, V = 10, γ = 10, and different values of the parameter κ: κ = 0
(blue), κ = 0.25 (orange dashed curve), κ = 5 (red dot-dashed curve). The asymptotic
limit is shown by green line.
population, for the non-equidistant energy spectrum of the acceptor band, Eq. (108),
and for different values of the parameter, κ, which describes the displacement from non-
equidistance. The value, κ = 0, corresponds to the equidistant energy spectrum (blue
curve in Fig. 12). One can see from Fig. 12 that quantum coherent oscillations survive
even for non-equidistant spectrum (κ 6= 0), if the non-equidistance is small enough.
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 28
However, with increasing κ, these oscillations become quasi-periodic (involving many
frequencies). The amplitudes of these oscillations become smaller, and the characteristic
decay time decreases. These coherent quantum oscillations could be used for extracting
characteristic spectroscopic parameters in natural bio-complexes and for engineering
artificial bio-nano devices.
6. Conclusion
We studied analytically and numerically noise-assisted quantum exciton (and electron)
transfer (ET) in bio-complexes consisting of a single-level electron donor interacting with
a multi-level acceptor. This situation takes place, in particular, when a single excited
energy level of the donor is populated, and it decays into a multi-level acceptor. Our
approach can also be used in a coarse-graining procedure, for a bio-complex with many
energy levels considered as an electron "donor" and/or"acceptor" with complicated
internal structures. All energy levels are assumed to interact with the protein-solvent
environment, modeled by a diagonal classical noise, which corresponds (between other
assumptions) to the high-temperature regime. Our approach can be applied for both
the exciton transfer in the light harvesting complexes and for the electron transfer in the
reaction centers. We vary the number of the acceptor levels, the acceptor bandwidth, the
strength of the donor-acceptor interaction, and the noise amplitude and the correlation
time. Under certain conditions, we derive analytical expressions for the ET rate and
efficiency. We demonstrate that, for a relatively wide acceptor band, the efficiency of the
ET from donor to acceptor can be close to 100% for a broad range of noise amplitudes,
for both sharp and flat redox potentials. We show that generally the dynamics of
the acceptor population can be characterized by multi-scale processes, which display a
coarse-graining structure. We would like to note here that the multi-scale ET dynamics
may result in decreasing the ET efficiency (probability of the acceptor population) due to
such processes as fluorescence and recombination of the exciton through the interaction
with the environment, which usually take place of the time-scales ∼ ns.
We also estimate the corresponding ET rates in a multi-scale regime. For our model,
we obtain equal occupation of all levels at large times, independent of the structure of
the acceptor band. This implies the possibility of optimizing the efficiencies of the
acceptor population by engineering the donor-acceptor complexes.
Our approach demonstrates a possibility of the efficient uphill population of the
acceptor, due to the "entropy factor" (large number of levels in the acceptor band). This
result can be useful in many applications for controlling of the ET in bio-complexes.
It would be important to experimentally verify this result, for example, in chlorophyll
based heterodimer.
It is very remarkable that the coarse-graining dynamics for the electron transfer
naturally occurs from our microscopic derivations, without any ad hoc assumptions.
Thus, our approach, developed in this paper, can also be applied as a consistent coarse-
graining procedure that describes exciton and electron transfer in large bio-complexes.
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 29
Indeed, the coarse-graining procedure usually suggests that connected clusters, which
include many electron energy levels, are replaced by interacting effective "donors" and
"acceptors", which are characterized by electron bandwidths with finite numbers of
energy levels. Then, according to our results, the ET rates and efficiencies will strongly
depend on the electron bandwidths of the corresponding clusters and on the number
of electron energy levels localized in these clusters. By engineering in a controlled way
the bandwidths of these clusters, one can significantly increase the efficiency of the ET
in bio-complexes. To do this, one must take into account the structure of energy levels
and their interactions inside the individual bio- clusters; interactions between different
clusters; and their interactions with local and collective environments. The quantum
coherent oscillations, which we discussed, are analogous to the quantum coherent effects
observed in experiments on exciton transfer in photosynthetic complexes, such as the
Fenna-Matthews-Olson (FMO) bacteriochlorophyll complex, which is found in green
sulphur bacteria [3], and in photosynthetic marine algae [4], at ambient temperature.
These oscillations could be used in the spectroscopic experiments, such as dynamic
fluorescence, for resolving the structures of energy spectra inside the bio-complexes. At
the same time, we should mention that presently, there is no consensus about the origin
of these oscillations, as they can have even completely classical vibrational origin. All
this will require further analytical, numerical, and experimental studies.
Acknowledgments
We are thankful to A. Aharony, O. Entin-Wohlman, and G.D. Doolen for useful
comments. S.G. acknowledges the Beijing Computational Science Research Center for
supporting his visit, where a part of this work was done. A.I.N. acknowledges the
support from the CONACyT. G.P.B. thanks the Einstein Center and the Goldschleger
Center for Nanophysics, at the Weizmann Institute, for supporting his visit to Israel,
where a part of this work was done.
Appendix A.
We consider the system of algebraic equations (87), written in the form:
N
Xn=1
A(N )
mn Xn = 0,
where A(N )
mn = δmn − Amn and,
.
Amn =
Cmn
1 + Fm
(A.1)
(A.2)
Here, Cmn = V 2
N /(ǫ2
mn + 4γ2) and Fm =Pn Cmn.
Let λi be the eigenvalues of the matrix, A. Employing the Perron-Frobenius
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 30
theorem [55], one can show that,
1 + Fmo < 1.
To evaluate det(I − B), we use the following formula:
Amn} = maxn Fm
λi ≤ max{Xn
det(I − A) = exp (Tr ln(I − A)).
Expanding the logarithm in the series, we find,
det A(N ) = det(I − A) = exp(cid:16) −
∞
Xn=1
1
n
trAn(cid:17).
Then, using the equation, trAn =PN
Xn=1
Xi
det A(N ) = exp(cid:16) −
i=1 λn
λn
i
∞
N
n (cid:17).
From here it follows,
i , we obtain,
det A(N ) ≥ exp(cid:16) −
N
Xi
∞
Xn=1
λin
n (cid:17).
One can recast inequality (A.7) as,
det A(N ) ≥
(1 − λi) > 0.
N
Yi=1
max{Xn
Amn} =
.
1 + F0
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
(A.9)
Let F0 be the value of Fm (m = 1, 2, . . . N) yielding the maximum of Pn Amn,
F0
Then, the estimate (A.8) can be simplified as follows:
det A(N ) ≥
1
(1 + F0)N .
References
(A.10)
[1] M. Mohseni, Y. Omar, G.S. Engel, and M.B. Plenio (eds.), Quantum Effects in Biology, (Cambridge
University Press, Cambridge, 2014).
[2] K.L.M. Lewis, F.D. Fuller, J.A. Myers, C.F. Yocum, S. Mukamel, D. Abramavicius, and J. P.
Ogilvie, Simulations of the Two-Dimensional Electronic Spectroscopy of the Photosystem II
Reaction Center, J. Phys. Chem. A, 117, 34 (2012).
[3] G. Engel, T. Calhoun, E. Read, T. Ahn, T.Mancal, Y. Cheng, R. Blankenship, and G. Fleming,
Evidence for wavelike energy transfer through quantum coherence in photosynthetic systems,
Nature Letters, 446, 782 (2007).
[4] E. Collini, C. Wong, K. Wilk, P. Curmi, P. Brumer, and G. Scholes, Coherently wired light-
harvesting in photosynthetic marine algae at ambient temperature, Nature Letters, 463, 644
(2010).
[5] G. Panitchayangkoon, D. Hayes, K. Fransted, J. Caram, E. Harel, J. Wenb, Long-lived quantum
coherence in photosynthetic complexes at physiological temperature, R. Blankenship, and
G. Engel, PNAS, 107, 12766 (2010).
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 31
[6] A. Ishizaki and G.R. Fleming, On the adequacy of the Redfield equation and related approaches
to the study of quantum dynamics in electronic energy transfer, J. Chem. Phys., 130, 234110
(2009).
[7] A. Ishizaki and G. Fleming, Theoretical examination of quantum coherence in a photosynthetic
system at physiological temperature, PNAS, 106, 17255 (2009).
[8] A. Ishizaki and G.R. Fleming, On the interpretation of quantum coherent beats observed in two-
dimensional electronic spectra of photosynthetic light harvesting complexes, J. Phys. Chem. B,
115, 6227 (2011).
[9] P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd, and A. Aspuru-Guzik, Environment-assisted
quantum transport, New J. Phys., 11(3), 033003 (2009).
[10] G. Celardo, F. Borgonovi, M. Merkli, V.I. Tsifrinovich, and G.P. Berman, Superradiance transition
in photosynthetic light-harvesting complexes, J. Phys. Chem., 116, 22105 (2012).
[11] M. Merkli, G.P. Berman, and S.T. Sayre, Electron transfer reactions: Generalized spin-boson
approach, J. Math. Chem., 51, 890 (2013).
[12] V. May and O. Kuhn, Charge and Energy Transfer Dynamics in Molecular Systems, (WILEY-VCH
Verlag GmbH & Co. KGaA, 2011).
[13] J. Ma and J. Caoa, Forster resonance energy transfer, absorption and emission spectra in
multichromophoric systems. I. Full cumulant expansions and system-bath entanglement, J.
Chem. Phys., 142, 094106 (2015).
[14] Al. Govorov, P. L. H. Martnez, and H.i V. Demir, Understanding and Modeling Forster-type
Resonance Energy Transfer (FRET). Introduction to FRET, Vol. 1., (Springer, 2016).
[15] S. Mukamel, Principles of Nonlinear Optical Spectroscopy (Oxford University Press, New York,
1995).
[16] D.I.G. Bennett, K. Amarnath, and G.R. Fleming, A structure-based model of energy transfer
reveals the principles of light harvesting in photosystem II supercomplexes, JACS, 135, 9164
(2013).
[17] R.L. Fulton and M. Gouterman, Vibronic coupling. I. mathematical treatment for electronic states,
J. Chem. Phys., 35, 1059 (1961).
[18] M. Lutz, Resonance Raman spectra of chlorophyll in solution, J. Raman Spec., 2, 497 (1974).
[19] V. Butkus, D. Zigmantas, L. Valkunas, and D. Abramavicius, Vibrational vs. electronic coherences
in 2D spectrum of molecular systems, Chem. Phys. Lett., 545, 40 (2012).
[20] A. Kolli, E.J. OReilly, G.D. Scholes, and A. Olaya-Castro, The fundamental role of quantized
vibrations in coherent light harvesting by cryptophyte algae, J. Chem. Phys., 137, 174109
(2012).
[21] A.G. Dijkstra, C. Wang, J. Cao, and G.R. Fleming, Coherent exciton dynamics in the presence of
under-damped vibrations, J. Phys. Chem. Lett., 6, 627 (2015).
[22] H. Dong, N.H.C. Lewis, T.A.A. Oliver, and G.R. Fleming, Determining the static electronic and
vibrational energy correlations via two-dimensional electronic-vibrational spectroscopy, J. Chem.
Phys., 142, 174201 (2015).
[23] Y. Fujihashi, G.R. Fleming, and A. Ishizaki, Impact of environmentally induced fluctuations on
quantum mechanically mixed electronic and vibrational pigment states in photosynthetic energy
transfer and 2D electronic spectra, J. Chem. Phys., 142, 212403 (2015).
[24] S. Jang, M.D. Newton, and R.J. Silbey, Multichromophoric Forster Resonance Energy Transfer,
Phys. Rev. Lett., 92, 218301 (2004).
[25] P.W. Milonni, J.R. Ackerhalt, H.W. Galbraith, and M.L. Shih, Exponential decay, recurrences,
and quantum-mechanical spreading in a quasicontinuum model, Phys. Rev. A, 28, 32 (1983).
[26] T.G. Dewey and J.G. Bann, Biophys, Protein dynamics and 1/f noise, J. Biophys Society, 63, 594
(1992).
[27] B.H. McMahon, P.W. Fenimore, and M.X. LaButea, Fluctuations and Noise in Biological,
Biophysical, and Biomedical Systems, S. M. Bezrukov, H. Frauenfelder, F. Moss, Eds.,
Proceedings of SPIE, 5110, 10 (2003).
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 32
[28] S.S. Skourtis, D.H.Waldeck, and D.N. Beratan, Fluctuations in biological and bioinspired electron-
transfer reactions, Annu. Rev. Phys. Chem., 61, 461 (2010).
[29] A.J. Das, S. Mukhopadhyay, and K.S. Narayana, Characteristic noise features in light transmission
across membrane protein undergoing photocycle, J. Chem. Phys., 134, 075101 (2011).
[30] A.I. Nesterov, G.P. Berman, and A.R. Bishop, Non-Hermitian approach for modeling of noise-
assisted quantum electron transfer in photosynthetic complexes, Fortschritte der Physik, 61, 95
(2013).
[31] A.I. Nesterov, G.P. Berman, J.M. S´anchez M´artinez, and R. Sayre, Noise-assisted quantum electron
transfer in photosynthetic complexes, J. Math. Chem., 51, 1 (2013).
[32] X. Hu, A. Damjanovic, T. Ritz, and K. Schulten, Architecture and mechanism of the light-
harvesting apparatus of purple bacteria, Proc. Natl. Acad. Sci., 95, 5935 (1998).
[33] Y.M. Galperin, B.L. Altshuler, J. Bergli, D. Shantsev, and V. Vinokur, Non-Gaussian dephasing
in flux qubits due to 1/f noise, Phys. Rev. B, 76, 064531 (2007).
[34] S.A. Gurvitz and D. Mozyrsky, Quantum mechanical approach to decoherence and relaxation
generated by fluctuating environment, Phys. Rev. B, 77, 075325 (2008).
[35] M. Blume, Stochastic Theory of Line Shape: Generalization of the Kubo-Anderson Model, Phys.
Rev., 174, 351 (1968).
[36] T. Itakura and Y. Tokura, Dephasing due to background charge fluctuations, Phys. Rev. B, 67,
195320 (2003).
[37] J. Bergli, Y.M. Galperin, and B.L. Altshuler, Decoherence in qubits due to low-frequency noise ,
New J. Phys., 11, 025002 (2009).
[38] A. Aharony, S. Gurvitz, O. Entin-Wohlman, and S. Dattagupta, Retrieving qubit information
despite decoherence, Phys. Rev. B82, 245417 (2010).
[39] V.E. Shapiro and V.M. Loginov, "Formulae of differentiation" and their use for solving stochastic
equations, Physica 90A, 563 (1978).
[40] S. Gurvitz, A. Aharony and O. Entin-Wohlman, Temporal evolution of resonant transmission
under telegraph noise, Phys. Rev. B94, 075437 (2016).
[41] R. Marcus and N. Sutin, Electron transfers in chemistry and biology, Biochimica et Biophysica
Acta, 811, 265 (1985).
[42] V. Capek, Generalized Haken-Strobl-Reineker model of excitation transfer, Z. Phys. B Condensed
Matter, 60, 101 (1985).
[43] H. Haken and G. Strobl, In: The triplet state, (Cambridge: Cambridge University Press, 1967).
[44] H. Haken and P. Reineker, Z. Phys., 249, 253 (1972).
[45] P. Reineker, In: Exciton Dynamics in Molecular Crystals and Aggregates, pp. 111-226, (Berlin,
Heidelberg, New York: Springer 1982).
[46] E. Gudowska-Nowak, G. Papp, and J. J. Brickmann, Bridged-assisted electron transfer. Random
matrix theory approach, Phem. Phes., 232, 247 (1998).
[47] Z. Bihary and M.A. Ratner, Dephasing effects in molecular junction conduction: An analytical
treatment, Phys. Rev. A., 72, 115439 (2005).
[48] H. van Amerongen, L. Valkunas, and R. van Grondelle, Photosynthetic Excitons, (World Scientific,
Singapore New Jersey London Hong Kong, 2000).
[49] R.E. Blankenship, Molecular Mechanisms of Photosynthesis, (Wiley Blackwell, 2014).
[50] D. Xu, K. Schulten, Coupling of protein motion to electron transfer in a photosynthetic reaction
center: investigating the low temperature behavior in the framework of the spin-boson model,
Chem. Phys., 182, 91 (1994).
[51] L.D. Zusman, Outer-Sphere electron transfer in polar solvents, Chem. Phys., 49, 295 (1980).
[52] S.A. Gurvitz, Delocalization in the Anderson model due to a local measurement, Phys. Rev. Lett.
85, 812 (2000).
[53] A. Aharony, S. Gurvitz, Y. Tokura, O.Entin-Wohlman, and S. Dattagupta, Partial decoherence in
mesoscopic systems, Physical Scripta, T151, 014018 (2012).
[54] M. Merkli, H. Song, and G.P Berman, Multiscale dynamics of open three-level quantum systems
Multi-Scale Exciton and Electron Transfer in Multi-Level Donor-Acceptor System 33
with two quasi-degenerate levels, J. Phys. A: Math. Theor., 48, 275304 (2015).
[55] M. Marcus and H. Minc, A Survey of Matrix Theory and Matrix Inequalities (Dover, New York,
1992).
|
1412.2371 | 5 | 1412 | 2016-09-12T20:59:44 | Anisotropic membrane curvature sensing by amphipathic peptides | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.BM"
] | Many proteins and peptides have an intrinsic capacity to sense and induce membrane curvature, and play crucial roles for organizing and remodelling cell membranes. However, the molecular driving forces behind these processes are not well understood. Here, we describe a new approach to study curvature sensing, by simulating the direction-dependent interactions of single molecules with a buckled lipid bilayer. We analyse three amphipathic antimicrobial peptides, a class of membrane-associated molecules that specifically target and destabilize bacterial membranes, and find qualitatively different sensing characteristics that would be difficult to resolve with other methods. These findings provide new insights into the curvature sensing mechanisms of amphipathic peptides and challenge existing theories of hydrophobic insertion. Our approach is generally applicable to a wide range of curvature sensing molecules, and our results provide strong motivation to develop new experimental methods to track position and orientation of membrane proteins. | physics.bio-ph | physics | Anisotropic membrane curvature sensing by amphipathic peptides
Jordi G´omez-Llobregat,1, 2, ∗ Federico El´ıas-Wolff,3, † and Martin Lind´en4, ‡
1Center for biomembrane research, Department of Biochemistry and Biophysics,
Stockholm University, SE-106 91 Stockholm, Sweden.
2Present address: Escola T´urbula, Carretera de Matar´o,
26 08930 Sant Adri`a del Bes`os, Barcelona, Spain.
3Center for biomembrane research, Department of Biochemistry and Biophysics,
Stockholm University, SE-106 91 Stockholm, Sweden
4Department of Cell and Molecular Biology, Uppsala University, Box 596, 751 24 Uppsala, Sweden
(Dated: October 2, 2018)
Many proteins and peptides have an intrinsic capacity to sense and induce membrane curvature,
and play crucial roles for organizing and remodelling cell membranes. However, the molecular driv-
ing forces behind these processes are not well understood. Here, we describe a new approach to
study curvature sensing, by simulating the direction-dependent interactions of single molecules with
a buckled lipid bilayer. We analyse three amphipathic antimicrobial peptides, a class of membrane-
associated molecules that specifically target and destabilize bacterial membranes, and find qualita-
tively different sensing characteristics that would be difficult to resolve with other methods. These
findings provide new insights into the curvature sensing mechanisms of amphipathic peptides and
challenge existing theories of hydrophobic insertion. Our approach is generally applicable to a wide
range of curvature sensing molecules, and our results provide strong motivation to develop new
experimental methods to track position and orientation of membrane proteins.
Published version available at doi:10.1016/j.bpj.2015.11.3512.
INTRODUCTION
Curvature sensing and generation by membrane pro-
teins and lipids is ubiquitous in cell biology, for exam-
ple to maintain highly curved shapes of organelles, or
drive membrane remodelling processes [1]. Membrane
curvature sensing occurs if a molecule's binding energy
depends on the local curvature [2]. For proteins, the
presence of multiple conformations with different curva-
ture preferences can couple protein function to membrane
curvature [3], with interesting but largely unexplored bi-
ological implications.
Curvature sensing by lipids is often rationalized by
a lipid shape factor, classifying lipids as 'cylindrical' or
'conical' when they prefer flat or curved membranes, re-
spectively [1, 2]. Membrane proteins offer a wider range
of sizes, shapes, and anchoring mechanisms [4], and thus
potentially more diverse sensing mechanisms. In particu-
lar, shape asymmetry implies that the binding energy de-
pends on the protein orientation in the membrane plane
[5], and thus cannot be a function of only mean and Gaus-
sian curvature, which are rotationally invariant. This
calls for more complex descriptions, and one natural ex-
tension is to model the binding energy in terms of the lo-
cal curvature tensor Cij in a frame rotating with the pro-
tein [5 -- 11], which allows different curvature preferences
in different directions. For example, a preference for lon-
gitudinal curvature is generally associated with proteins
∗ [email protected]
† [email protected]
‡ [email protected]
that are curved in this direction, such as BAR domains
[12, 13], whereas amphipathic helices [14] are expected
to sense transverse curvature, since their insertion into
the membrane-water interface is energetically favored if
the membrane curves away in the transverse direction
[15 -- 17].
Anisotropic curvature sensing is potentially complex,
and theoretical investigations have demonstrated a wide
range of qualitative behavior in local curvature mod-
els [5 -- 11], but the models have not been rigorously
tested. In principle, the curvature-dependent binding en-
ergy landscape E(Cij) could be determined by measur-
ing the Boltzmann distribution of protein configurations
on curved membranes of known shape. However, cur-
rent experimental techniques track only protein positions
[18 -- 24], and hence orientational information is averaged
out. Here, we track both position and orientation of sin-
gle molecules, using a computational approach based on
simulated membrane buckling.
The method is applied to three amphipathic antimicro-
bial model peptides: magainin, which is found in the skin
of the African clawed frog [25], melittin, an active com-
ponent in bee venom [26], and LL-37, a peptide derived
from the human protein cathelicidin which is involved
in the innate immune defense system [27]. As shown
in Fig. 1, the peptides vary in length and shape, and
can thus be expected to display different sensing char-
acteristics. Many antimicrobial peptides are believed to
work by mediating membrane disruption [28]. The pep-
tides studied here are thought to mediate the formation
of toroidal membrane pores with a highly curved inner
surface partly lined with lipids [29 -- 34], although the ev-
idence appears less clear for LL-37 [35]. The ability to
stabilize highly curved membrane structures suggests an
6
1
0
2
p
e
S
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
5
v
1
7
3
2
.
2
1
4
1
:
v
i
X
r
a
2
model [42, 46, 47], and a relative dielectric constant of 2.5
(as recommended [47]). We used standard lipid param-
eters for 1-palmitoyl-2-oleoyl phosphatidylethanolamine
(POPE) [48], 1-palmitoyl-2-oleoyl phosphatidylglycerol
(POPG) [49], and peptides [50]. The peptide struc-
tures for magainin (PDB ID:1DUM), melittin (PDB
ID:2MLT), and LL-37 (PDB ID: 2K6O) were obtained
from the Protein Data Bank, and coarse-grained with
the martinize script provided by the MARTINI devel-
opers. Constant temperature was maintained with the
velocity rescaling thermostat [51] with a 1.0 ps time con-
stant, and pressure was controlled with the Berendsen
barostat [52] using a time constant of 12 ps and a com-
pressibility of 3 × 10−4 bar−1. Peptide (when present),
lipids and solvent were coupled separately to the temper-
ature bath. Coulomb interactions were modelled with the
particle mesh Ewald method [53] setting the real-space
cut-off to 1.4 nm and the Fourier grid spacing to 0.12
nm. Lennard-Jones interactions were shifted to zero be-
tween 0.9 and 1.2 nm. A time step of 25 fs was used in
all simulations.
b. System assembly and membrane buckling We as-
sembled and equilibrated three rectangular (Lx = 2Ly)
bilayer patches of 1024 lipids each, with 70% POPE and
30% POPG, solvated with ∼ 21000 coarse-grained water
beads and neutralized with sodium ion beads. POPG
is negatively charged, which promotes peptide binding.
These patches were equilibrated for 25 ns in an NPT en-
semble at 300 K and 1 bar, with pressure coupling applied
semi-isotropically.
After equilibration, all systems were laterally com-
pressed in the x direction by a factor γ = (L − Lx)/L =
0.2, where L is the linear size of the flat system, and Lx
the size of the compressed simulation box, in the x di-
rection. This was done by scaling all x-coordinates, and
the box size Lx, by a factor 1− γ = 0.8 at the end of the
equilibration run, yielding Lx = 20.88, 20.81 and 20.89
nm for the three patches, respectively. After rescaling,
the compressibilities were set to 0 in the x and y direc-
tions to keep the system size constant in those directions
for subsequent simulations. Pressure coupling was then
applied anisotropically in the z direction only. We then
performed an energy minimization and a short equilibra-
tion run (25 ns) to let the bilayer buckle.
Next, we added one peptide to each system, using the
three independent patches to create three independent
replicas for each peptide. The peptide was initially placed
about 3 nm above the membrane surface, but quickly at-
tached to the bilayer. After the binding event, we equili-
brated the system for another 5 µs before starting a pro-
duction run of 15 µs, where we collected data every 5 ns.
All peptides remained essentially parallel to the mem-
brane surface as expected , in agreement with experimen-
tal results for low peptide concentrations [35, 40, 41, 54].
c. Membrane alignment and peptide tracking The
buckled membrane profile diffuses as a traveling wave the
simulation (movie S1), but curvature sensing by a pep-
tide is reflected in its distribution relative to the buckled
FIG. 1. Structures of magainin [40], melittin [41] and LL-37
[39]. The melittin and LL-37 structures contain two α-helices
that form an angle β (not the same for both structures). The
α-helices used in the analysis are colored in blue (N-terminal)
and orange (C-terminal), with the limiting amino acids la-
beled on the structure. Side chain and non-helical residues
are colored in gray.
intrinsic preference for curved membranes, as is generally
expected for amphipathic peptides.
Our method uses
simulated membrane buckling
single
to sample the unconstrained interaction of
biomolecules with a range of membrane curvatures, and
extends previous simulation studies of buckling mechan-
ics [36, 37], curvature-dependent folding and binding of
amphipathic helices [38], and lipid partitioning [39]. We
obtain joint distributions of peptide positions and orien-
tations that yield new biophysical insights about curva-
ture sensing. The three model peptides display similar
rotation-averaged curvature preferences but differ in ori-
entational preferences, which demonstrates the value of
directional information. The asymmetry of the position-
orientational distributions challenges continuum models
of amphipathic helices as cylindrical membrane inclu-
sions [15, 16]. We speculate that such asymmetry is im-
portant for certain modes of antibacterial activity, and
argue that it might be common also for larger curva-
ture sensing proteins. Finally, we discuss the limitations
of characterizing curvature sensing mechanisms from as-
says with zero Gaussian curvature, and conclude that
this uncertainty affects the overall binding energy, but
not the orientational preferences. These results motivate
efforts to track positions and orientations of membrane
proteins experimentally, and to develop assays with a
broader range of local curvatures.
METHODS
To study curvature sensing by single peptides, we sim-
ulate their interactions with a buckled membrane using
the coarse-grained Martini model [42], and track their
position and orientation, as shown in Fig. 2. On a mi-
croscopic level, curvature sensing by amphipathic helices
is associated with the density and size of bilayer surface
defects [38, 43], which are well described by the Martini
model [44].
a. Simulation parameters We performed molecular
dynamics simulations using Gromacs 4.6.1 [45], and the
coarse-grained Martini force-field with polarizable water
shape. Hence, the buckled configurations must be aligned
in order to extract useful information. To do this, we fit
the xz-profile of the membrane by the ground state of the
Helfrich model with periodic boundary conditions, which
is one of the Euler buckling profiles of an elastic beam
[36, 37]. This shape depends only on the dimensionless
buckling parameter γ (γ = 0 is the flat state). Hence,
if we compute the shape for some reference system, the
general case can be obtained by shifting and scaling. We
chose Lx = 1 as reference, and write the buckling profile
as a parametric curve x = s + ξ(s, γ), z = ζ(s, γ), param-
eterized by a normalized arclength coordinate 0 < s < 1
(the absolute arclength is given by sL = sLx(1 − γ)−1).
For fast evaluation, we expanded ξ(s, γ) and ζ(s, γ) in
truncated Fourier series in s, and created look-up tables
for Fourier coefficients vs. γ. We defined s to give the
curve z(x) a maximum at s = 0.5, minima at s = 0, 1,
and inflection points at s = 0.5 ± 0.25, and aligned the
buckled shapes by fitting the bilayer in each frame to the
buckling profile and aligning the inflection points (Fig. 2,
movies S2-S3). Specifically, we fit the rescaled buckling
profile to the innermost tail beads of all lipids in each
frame using least-squares in the x and z directions, i.e.,
(cid:0)s+ξ(si, γ)(cid:1)−xi)2+(z0+Lxζ(si, γ)−zi)2 (1)
minimizing(cid:88)
(x0+Lx
i
with respect to γ, the translations x0, z0, and the nor-
malized arc-length coordinates si of each bead (xi, zi are
bead positions). The time-averaged bilayer shape, after
alignment, agrees well with the theoretical buckled shape
(Fig. 3a).
The normalized arclength coordinate s of the peptide
was computed by projecting the peptide center of mass
onto the buckled profile fitted to the membrane midplane.
The in-plane orientation θ was then computed by fitting
a line through the backbone particles of the α-helical part
of the peptide, projecting it onto the tangent plane at s,
and computing the in-plane angle to the tangent vector
t (see Fig. 2a). The local curvature at s, in the tangent
direction of the buckled shape, is given by
C(s) =
1 − γ
Lx
dψ
ds
,
(2)
where ψ is the bilayer mid-plane tangent angle of
Fig. 2a,b [55]. (Note that the opposite sign convention is
also common [1]) . In the theoretical analysis, we neglect
small shape and area fluctuations (std(γ) ≈ 0.005) and
use the nominal value γ = 0.2.
d. Fitting We used least-squares routines in MAT-
LAB (MathWorks, Natick, MA) to fit the Boltzmann
distributions e−E(si,θi)/Z of the EC (Eq. 8) and E2 (Ta-
ble S1) models to (s, θ)-histograms built from the aggre-
gated data with 50 bins for each coordinate. Both data
and model histograms were normalized numerically. Er-
ror bars in Fig. 4d are boot-strap standard deviations
from 1000 bootstrap realizations, using blocks of length
100 (500 ns) as the elementary data unit for resampling
[56].
3
RESULTS
e. Preferred curvature and orientations We simu-
lated single peptides interacting with a buckled bilayer,
using three independent production runs of 15 µs for each
peptide, and tracked their normalized arc-length coor-
dinates s ∈ [0, 1] and in-plane orientations θ (Fig. 2a).
Aggregated (s, θ)-histograms are shown in Fig. 3b-d, and
convergence is discussed in Sec. S1.
All three peptides prefer the concave high curvature
regions with a maximum at s = 0.5, as expected for hy-
drophobic insertion mechanisms [15 -- 17, 38, 44, 58]. Re-
garding the angle distributions, the three peptides behave
differently. Magainin displays a rather uniform angle dis-
tribution, probably because its short α-helical segment
creates a fairly symmetric insertion footprint. For melit-
tin, the joint between the N- and C-terminal helices ap-
pears very flexible, resulting in a broad distribution of the
internal angle β (Fig. 3e). Both helices prefer directions
nearly parallel to the x-axis, the direction of maximum
curvature, but the preference is stronger and slightly off-
set (θmax ≈ −15◦, 165◦) for the C-terminal helix shown
in Fig. 3c, while the N-terminal helix is more symmetri-
cally oriented (Fig. S2).
structure, and its θ-
distribution displays two sharp maxima near θ = 70◦
and θ = −110◦ (Fig. 3d). This is remarkable since, by
reflection symmetry around s = 0.5, the curvatures in
those directions are the same as along −70◦ and 110◦,
orientations that are clearly not preferred. As we will ar-
gue below, this can be understood as curvature sensing
along directions different from that of the peptide itself.
These sensing directions adopt θ = 0, 90◦, and thus map
onto themselves under reflection. Notably, none of the
peptides orient directly along the flat direction θ = 90◦
as commonly assumed in mechanical models [15, 16].
LL-37 maintains a linear
f. Orientation-averaged binding free energy Next,
we look at the orientation-averaged binding free en-
ergy, corresponding to the curvature-dependent enrich-
ment measured in many in vitro assays [18 -- 24]. To ex-
tract the curvature dependence of the binding energy,
we analyse center-of-mass positions along the buckled
shape. These should follow a Boltzmann distribution,
proportional to e−G(s), where G(s) is the orientation-
averaged binding free energy in units of kBT . We model
this as depending on the local curvature only, and hence
set G(s) = G(C(s)), and extract G(C) from curvature
histograms, weighted according to the change-of-variable
transformation that relates the density of curvatures,
ρ(C), to the density of positions ρ(s). Indeed, dropping
normalization constants, we have
ρs(s)ds ∝ e−G(C(s))ds ∝ e−G(C)dC/ds−1dC ∝ ρC(C)dC,
(3)
4
FIG. 2. Buckled simulation and analysis. (a) The position s of a peptide is defined by the projection of the center-of-mass (blue
dot) onto the midplane surface (yellow). The in-plane orientation θ is defined by projecting the peptide backbone direction
(green arrow, pointing towards the C-terminal end) onto the local tangent plane (gray) at s. The local tangent and normal
vectors are indicated by t and n, respectively. (b) Side and (c) top view of a simulation snapshot with peptide position and
orientation indicated using the notation and local coordinate system in (a). The system size is Lx = 20.88 nm and Ly = 13.05
nm. The peptide (LL-37 in this case) is shown in green, and lipids in gray (tails), light red (phosphate groups) and blue
(innermost tail beads). The side view (b) also shows the Euler buckling profile (red line) fitted to the bilayer mid-plane, and
the inflection points at s = 0.5 ± 0.25 (yellow crosses) used to align the buckled configurations. Molecular graphics generated
with VMD [57].
from which it follows that
G(C) = − ln(cid:0)ρC(C)dC/ds(cid:1) + const.
(4)
The weights dC/ds can be understood as compensating
for the fact that not all curvatures have equal arclength
footprints along the buckled profile. To estimate G(C),
we estimated ρ(C) using a simple histogram, and the
weights as the mean of dC/ds for all contributions to
each bin.
Fig. 3f shows the binding free energy profiles G(C)
for the different peptides, which are more similar than
the (θ, s)-distributions, and well fit by quadratic curves.
Note that Eq. (4) does not yield absolute binding energies
of the peptides, and the G(C) curves are instead offset
vertically for easy visualization. Experimental binding
free energies of these peptides to flat membranes with
anionic lipids range from -15 to -10 kBT [59].
g. Quantitative models We now turn to quantitative
models of the peptides' curvature sensing. As described
in the introduction, we model the binding energy of a
peptide as a function of the local curvature tensor in a
frame rotating with the peptide, and treat the bilayer it-
self as having fixed shape and thus a fixed deformation
energy which we neglect. Generally, if the principal cur-
vatures and directions are c1,2 and (cid:126)e1,2, the curvature
tensor, or second fundamental form, in a frame rotated
by an in-plane angle θ relative to (cid:126)e1, is given by
(cid:20) H+D cos 2θ D sin 2θ
D sin 2θ H−D cos 2θ
(cid:21)
=
(cid:21)
(cid:20) C(cid:107) CX
CX C⊥
, (5)
Cij =
where H = (c1 + c2)/2 and D = (c1 − c2)/2 are the mean
and deviatoric curvatures, and the Gaussian curvature is
given by K = c1c2 = C(cid:107)C⊥ − C 2
X . Note the symme-
try under rotations by 180◦, since the curvature of a line
is the same in both directions. For the buckled surface,
c1 = C(s), c2 = 0 (and hence K = 0, H = D = C(s)/2),
(cid:126)e1 = t, (cid:126)e2 = y. As shown in Fig. 2, we define the ro-
tating frame using the peptide's center of mass and the
direction of the α-helical parts, and thus θ is the peptide
in-plane orientation, and (cid:107),⊥ denote the longitudinal (θ)
and transverse (θ + 90◦) directions.
The simplest models are linear in Cij, but can be ruled
out since they cannot reproduce the convex binding free
energies in Fig. 3f. To see this, we write a general linear
model in the form E1 = aH + bD cos(2(θ − α)) [5], and
integrate out the angular dependence to get
G1 = − ln
e−E1dθ = aH − ln I0(bD) + const.
(cid:90) 2π
(6)
0
Since H = D = C(s)/2 on the buckled surface, and the
modified Bessel function I0 is convex, G1 will be either
downward convex (if b (cid:54)= 0) or linear and direction insen-
sitive (when b → 0), in disagreement with Fig. 3.
Moving on to quadratic terms, Akabori and Santangelo
[10] explored a model of the form
k(cid:107)
2
k⊥
2
EX =
(C(cid:107)− C(cid:107)0)2+ kX (CX − CX0)2+
(C⊥− C⊥0)2,
(7)
where C(cid:107)0, CX0 and C⊥0 are preferred curvatures. Fur-
ther simplifications kX =0 and kX =k⊥=0 have also been
studied [6 -- 9]. While these models can all display non-
trivial behavior, EX is not the most general quadratic
model, which would include all 9 linear and quadratic
combinations of the three independent curvature tensor
components. In particular, EX does not contain a sim-
ple preferred mean curvature as a special case, because
H = (C(cid:107) + C⊥)/2, and hence (H − H0)2 contains a term
C(cid:107)C⊥ which is absent in Eq. (7).
However, the general quadratic model is not identifi-
able on surfaces with only one non-zero principal cur-
vature. This is because the Gaussian curvature K is
5
2 , and ±π. However,
this means angular dependence only in the form cos 2θ,
which is symmetric around θ = 0, ± π
the orientational distributions in Fig. 4a do not display
this symmetry, although the statistics is not quite clear
in the case of melittin (see Fig. S2). Apparently, the cur-
vature sensing directions are not generally aligned with
the actual helices. This resembles results for α-synuclein,
where peptides and induced membrane deformations ap-
pear similarly misaligned [60]. A simple quadratic model
incorporating these observations is
(2H − C0)2 + bD cos(cid:0)2(θ − α)(cid:1) + κGK,
(8)
EC =
κ
2
where the Gaussian curvature coefficient κG is unidenti-
fiable since K = 0 in our data. As shown in Fig. 4, EC
describes all peptides reasonably well, and using the full
quadratic model does not significantly improve the fit.
To better understand the physical meaning of this
model, we explore some alternative formulations. First,
using Eq. (5) to trade H, D for the Cij, and rearranging
the terms, we find an equivalent formulation that resem-
bles the EX model,
E(cid:48)
C =
κ
2
(cid:0)C(cid:107) − C0 +
(cid:0)C⊥ − C0 − b
cos 2α(cid:1)2
cos 2α(cid:1)2
b
2κ
+ κ(cid:0)CX +
2κ
+
κ
2
sin 2α(cid:1)2
b
2κ
+ (κ + κG)K.
(9)
Continuing, we can rotate the basis attached to the pep-
tide by α, and thus generate a transformed curvature
tensor with elements C (α)
C (α)(cid:107) + C (α)⊥ = 2H, C (α)(cid:107) − C (α)⊥ = 2D cos(cid:0)2(θ − α)(cid:1).
ij (θ) = Cij(θ + α) satisfying
(10)
In this basis, there is an EX -like equivalent model that
lacks 'off-diagonal' elements,
(cid:0)C (α)(cid:107) −C0+
(cid:1)2
b
2κ
+
κ
2
(cid:0)C (α)⊥ −C0− b
(cid:1)2
2κ
E(cid:48)(cid:48)
C =
κ
2
(11)
i.e., sensing curvature along two orthogonal directions
that are rotated by an angle α with respect to the pep-
tide backbone. Note that since Gaussian curvature is
rotationally invariant, the unidentifiable Gaussian curva-
ture term only affects the overall affinity to membranes
with Gaussian curvature, and not the orientational pref-
erences of the peptides.
As a consistency check, we integrated out θ from EC.
Proceeding as for G1 in Eq. (6) and setting H = D =
C(s)/2, K = 0, we get
+(κ+κG)K,
GC = − ln
dθe−EC =
κ
2
(C − C0)2 − ln I0
(12)
which we compare with G(C) in Fig. 3f using the parame-
ters of Fig. 4d. Magainin and melittin shows good agree-
ment, but not LL-37, whose (s, θ)-distribution (Fig. 3d)
is also less symmetric around s = 0.5 than expected from
(cid:0)bC/2(cid:1),
(cid:90) 2π
0
FIG. 3. Distributions of peptide positions and orientations
in the buckled bilayer. (a) Average buckled shape in terms of
densities of inner lipid tail beads (blue) and phosphate groups
(gray) after alignment, for one production run with LL-37.
Green dots show representative peptide center-of-mass posi-
tions. Dashed red lines indicate the average fitted mid-plane
±2.15 nm offsets in the normal direction. (b-d) Aggregated
(s, θ)-histograms for (b) magainin, (c) melittin (using the ori-
entation of the C-terminal helix), and (d) LL-37.
(e) Dis-
tributions of internal angle β (see Fig. 1) for melittin and
LL-37. (f) Orientation-averaged binding free energy vs. cur-
vature at the peptide center-of-mass (Eq. (4)) for the three
peptides. Error bars show max and min values from three
independent simulations. Dashed lines are guides to the eye
(fits to quadratic curves), and solid lines are results for the
EC model (Eq. (8)) using the fit parameters in Fig. 4.
zero, and hence the model can only be specified up to
a term proportional to K. Also, EX can then be made
to behave as a mean curvature sensor, since all angu-
lar dependence cancels if k(cid:107) = k⊥ = kX , C(cid:107)0 = C⊥0,
and CX0 = 0. These limitations apply to our buckled
surface, as well as to tubular and plane-wave geometries
used experimentally [18, 19, 22 -- 24]. A curvature sensing
mechanism therefore cannot be completely characterized
using such surfaces, but some conclusions can be drawn.
In particular, setting kX = 0 in Eq. (7) yields an in-
tuitive model with curvature sensing only along the lon-
gitudinal and transverse directions [6 -- 9]. From Eq. (5),
(a)LL-37(b)(c)MAGMEL Cθ [ o ](d)C [nm-1]-0.2-0.10G(C) / kBT012345MAGMELLL-37xzs=0s=0.5s=1(e)(f)β [ o ]freq. [a.u.]0090180MELLL-370.5θ [ o ]θ [ o ] 6
FIG. 5. Curvature sensing site and (s, θ)-correlations. (a)
A freely rotating peptide whose midpoint (black) is fixed at
s = 0. (b) Resulting correlations sC,N ∝ ± cos θ for the N-
and C-terminal ends (red,blue).
Our data is well described by modelling the bind-
ing energy in terms of local curvatures, yielding more
complex models than commonly used to fit orientation-
averaged data [18 -- 21], and also less symmetric than some
theoretical suggestions [6 -- 9]. The observed asymmetry
also seems difficult to reconcile with continuum elasticity
models of hydrophobic insertion in terms of cylindrical
membrane inclusions [15, 16]. Recently, continuum elas-
ticity models were found to underpredict the induced cur-
vature of a hydrophobic insertion compared to atomistic
simulations [17]. Our data shows an additional qualita-
tive effect of molecular detail, which we believe reflect the
fact that the mirror symmetry of cylinder-shaped inclu-
sions is absent from the peptide structures. Instead, our
data can be described in terms of curvature sensing di-
rections that are not aligned with the inserted α-helices.
Since amphipathic helices are common curvature sens-
ing motifs [14] and mirror symmetry is generally absent
also in multimeric proteins [61], such asymmetric sensing
might be common.
These results should motivate efforts to track the po-
sition and orientation of membrane proteins experimen-
tally, for example using polarization-based optical tech-
niques [62] or electron microscopy [63].
It would also
be valuable to vary mean and Gaussian curvatures inde-
pendently in order to probe Gaussian curvature sensing,
for example by extending supported bilayer assays with
plane-wave surfaces[22] to shapes with non-zero Gaussian
curvature. Another possibility might be to combine as-
says with cylindrical geometries (K = 0), such as plane
waves [22] or membrane tethers[18 -- 21, 23] with spherical
geometries (K = H 2) such as vesicles [3, 58] or deposited
nanoparticles [24].
An interesting aspect of the EC model is that it pre-
dicts a free energy minimum, i.e., a preferred curvature,
at least when K = 0 (Eq. 12). The preferred curvature
radii C−1
min of our peptides, listed in Fig. 4c, are well above
the monolayer thickness of about 2.2 nm (Fig. 3a) where
the bilayer folds back on itself, but below the lowest ra-
dius in our simulations (about 4.5 nm), meaning that this
prediction is somewhat speculative, since higher order
terms might become important at very high curvatures.
FIG. 4. Fitting quadratic models to data. (a,b) marginal po-
sition and angle distributions, showing data (gray) and nearly
identical curves from the EC and general quadratic model
(E2, see Table S1). (c) (s, θ)-distributions for the EC model.
(d) EC fit parameters ± bootstrap SEM [56] due to finite
sampling (see Sec. S1), with α indicating the preferred ori-
entation. Cmin is the preferred curvature, from minimizing
Eq. 12.
the symmetry of the buckled shape. A simple explana-
tion is that the effective "sensing site" does not coincide
with the center of mass used to define s. This is illus-
trated in Fig. 5 by a hypothetical peptide which is fixed
at s = 0.5 but free to rotate. As a result, the N-terminal
end shows (s, θ)-correlations resembling those seen for
the LL-37 center of mass, indicating that its "sensing
site" is located in the C-terminal part. Numerical exper-
iments in Sec. S2 agree qualitatively with this geometric
argument, and both symmetry and consistency improves
when tracking the LL-37 C-terminal helix instead (but
the fit parameters do not change significantly).
DISCUSSION
We describe a simulation approach to study membrane
curvature sensing by tracking positions and orientations
of single molecules interacting with a buckled lipid bi-
layer. This approach is widely applicable, and the utility
of angular information is obvious from the observation
that the three peptides show distinct orientational dis-
tributions, but very similar orientation-averaged binding
energy curves (Fig. 3).
s00.51-180-90090180s00.51s00.51(a)(b)(c)magaininmelittinLL-37freq.s00.250.50.751s00.250.50.751s00.250.50.75100.10.05freq.θ[o]-180-90090180θ[o]-180-90090180θ[o]-180-900901800.050ECE2(d) [ o ]72±26150±50332±140 [kBT nm2]-3.4±0.7-4.3±0.5-4.2±0.6C0-1 [nm]1.1±0.36.4±0.48.9±0.6b [kBT nm]-40±8-16±269±2 [ o ]-3.4±0.7-3.8±0.6-4.1±0.5Cmin-1 [nm]melittinLL-37magainin(a)(b)s0.5 [o]-90090midpointC-term.N-term.C On a molecular level, curvature sensing by amphipathic
peptides is thought to reflect an affinity for packing de-
fects in the membrane-water interface [38, 44, 58]. It is
not clear that this mechanism predicts a preferred curva-
ture at all. Testing this seems like an interesting question
for future work.
Our results also have biophysical implications. At
high concentrations, the three peptides are thought
to mediate the formation of membrane pores with
highly curved inner surfaces [29 -- 34]. The orientational
preferences we see in single peptides are consistent
with atomistic [31] and coarse-grained [34] simulations
of multi-peptide pores.
In particular, the asymmetric
curvature preference of LL-37 should help select for a
single handedness of the resulting tilted pore structure
[34], which might facilitate pore formation by reducing
frustration. This mechanism may represent a general
way for membrane proteins to induce a particular orien-
tation or handedness in patterns on curved surfaces [64].
7
h. Acknowledgments We thank Astrid Graslund,
Oksana V. Manyuhina, Christoph A. Haselwandter,
and two anonymous reviewers for helpful comments
and discussions.
Simulations were performed on re-
sources provided by the Swedish National Infrastruc-
ture for Computing (SNIC) at the National Supercom-
puter Centre (NSC) and the High Performance Comput-
ing Center North (HPC2N). Financial support from the
Wenner-Gren Foundations and the Swedish Foundation
for Strategic Research (SSF) via the Center for Biomem-
brane Research are gratefully acknowledged.
i. Author contributions JG and ML designed re-
search. JG and FEW performed research. JG and ML
analysed data. JG, FEW, and ML wrote the paper.
[1] Zimmerberg, J., and M. M. Kozlov, 2006. How proteins
produce cellular membrane curvature. Nat. Rev. Mol.
Cell Bio. 7:9 -- 19.
[2] Baumgart, T., B. R. Capraro, C. Zhu, and S. L. Das,
2011. Thermodynamics and mechanics of membrane cur-
vature generation and sensing by proteins and lipids.
Annu. Rev. Phys. Chem. 62:483 -- 506.
[3] Tonnesen, A., S. M. Christensen, V. Tkach, and D. Sta-
mou, 2014. Geometrical membrane curvature as an al-
losteric regulator of membrane protein structure and
function. Biophys. J. 106:201 -- 209.
[4] Engelman, D., 2005. Membranes are more mosaic than
fluid. Nature 438:578 -- 580.
[5] Fournier, J. B., 1996. Nontopological saddle-splay and
curvature instabilities from anisotropic membrane inclu-
sions. Phys. Rev. Lett. 76:4436 -- 4439.
[6] Perutkov´a, S., V. Kralj-Iglic, M. Frank, and A. Iglic,
2010. Mechanical stability of membrane nanotubular pro-
trusions influenced by attachment of flexible rod-like pro-
teins. J. Biomech. 43:1612 -- 1617.
[7] Ramakrishnan, N., P. B. Sunil Kumar, and J. H. Ipsen,
2010. Monte Carlo simulations of fluid vesicles with in-
plane orientational ordering. Phys. Rev. E 81:041922.
[8] Ramakrishnan, N., P. B. S. Kumar, and J. H. Ipsen,
2011. Modeling anisotropic elasticity of fluid membranes.
Macromol. Theor. Simul. 20:446 -- 450.
[9] Ramakrishnan, N., P. B. Sunil Kumar, and J. H. Ipsen,
2013. Membrane-mediated aggregation of curvature-
inducing nematogens and membrane tubulation. Bio-
phys. J. 104:1018 -- 1028.
[10] Akabori, K., and C. D. Santangelo, 2011. Membrane
morphology induced by anisotropic proteins. Phys. Rev.
E 84:061909.
[11] Walani, N., J. Torres, and A. Agrawal, 2014. Anisotropic
spontaneous curvatures in lipid membranes. Phys. Rev.
E 89:062715.
[12] Peter, B. J., H. M. Kent, I. G. Mills, Y. Vallis, P. J. G.
Butler, P. R. Evans, and H. T. McMahon, 2004. BAR
domains as sensors of membrane curvature:
the am-
phiphysin BAR structure. Science 303:495 -- 499.
[13] Blood, P. D., and G. A. Voth, 2006. Direct observation
of Bin/amphiphysin/Rvs (BAR) domain-induced mem-
brane curvature by means of molecular dynamics simu-
lations. Proc. Natl. Acad. Sci. U.S.A. 103:15068 -- 15072.
[14] Drin, G., J.-F. Casella, R. Gautier, T. Boehmer, T. U.
Schwartz, and B. Antonny, 2007. A general amphipathic
α-helical motif for sensing membrane curvature. Nat.
Struct. Mol. Biol. 14:138 -- 146.
[15] Campelo, F., H. T. McMahon, and M. M. Kozlov, 2008.
The hydrophobic insertion mechanism of membrane cur-
vature generation by proteins. Biophys. J. 95:2325 -- 2339.
[16] Campelo, F., and M. M. Kozlov, 2014. Sensing mem-
brane stresses by protein insertions. PLoS Comput. Biol.
10:e1003556.
[17] Sodt, A. J., and R. W. Pastor, 2014. Molecular model-
ing of lipid membrane curvature induction by a peptide:
more than simply shape. Biophys. J. 106:1958 -- 1969.
[18] Zhu, C., S. L. Das, and T. Baumgart, 2012. Nonlinear
sorting, curvature generation, and crowding of endophilin
N-BAR on tubular membranes. Biophys. J. 102:1837 --
1845.
[19] Sorre, B., A. Callan-Jones, J. Manzi, B. Goud, J. Prost,
P. Bassereau, and A. Roux, 2012. Nature of curvature
coupling of amphiphysin with membranes depends on its
bound density. Proc. Natl. Acad. Sci. U.S.A. 109:173 --
178.
[20] Aimon, S., A. Callan-Jones, A. Berthaud, M. Pinot,
G. E. S. Toombes, and P. Bassereau, 2014. Membrane
shape modulates transmembrane protein distribution.
Dev. Cell 28:212 -- 218.
[21] Shi, Z., and T. Baumgart, 2015. Membrane tension
and peripheral protein density mediate membrane shape
transitions. Nat. Commun. 6:5974.
[22] Hsieh, W.-T., C.-J. Hsu, B. R. Capraro, T. Wu, C.-M.
Chen, S. Yang, and T. Baumgart, 2012. Curvature sort-
ing of peripheral proteins on solid-supported wavy mem-
branes. Langmuir 28:12838 -- 12843.
[23] Ramesh, P., Y. F. Baroji, S. N. S. Reihani, D. Stamou,
L. B. Oddershede, and P. M. Bendix, 2013. FBAR syn-
dapin 1 recognizes and stabilizes highly curved tubular
membranes in a concentration dependent manner. Sci.
Rep. 3:1565.
[24] Black, J. C., P. P. Cheney, T. Campbell, and M. K.
Knowles, 2014. Membrane curvature based lipid sorting
using a nanoparticle patterned substrate. Soft Matter
10:2016.
[25] Zasloff, M., 1987. Magainins, a class of antimicrobial
isolation, characterization
peptides from Xenopus skin:
of two active forms, and partial cDNA sequence of a pre-
cursor. Proc. Natl. Acad. Sci. U.S.A. 84:5449 -- 5453.
[26] Habermann, E., 1972. Bee and wasp venoms. Science
177:314 -- 322.
[27] Gudmundsson, G. H., B. Agerberth, J. Odeberg,
T. Bergman, B. Olsson, and R. Salcedo, 1996. The hu-
man gene FALL39 and processing of the cathelin precur-
sor to the antibacterial peptide LL-37 in granulocytes.
Eur. J. Biochem. 238:325 -- 332.
[28] Melo, M. N., R. Ferre, and M. A. R. B. Castanho, 2009.
Antimicrobial peptides:
linking partition, activity and
high membrane-bound concentrations. Nat. Rev. Micro-
biol. 7:245 -- 250.
[29] Ludtke, S. J., K. He, W. T. Heller, T. A. Harroun,
L. Yang, and H. W. Huang, 1996. Membrane Pores In-
duced by Magainin†. Biochemistry 35:13723 -- 13728.
[30] Yang, L., T. A. Harroun, T. M. Weiss, L. Ding, and H. W.
Huang, 2001. Barrel-stave model or toroidal model? A
case study on melittin pores. Biophys. J. 81:1475 -- 1485.
[31] Leontiadou, H., A. E. Mark, and S. J. Marrink, 2006.
Antimicrobial peptides in action. J. Am. Chem. Soc.
128:12156 -- 12161.
[32] Henzler Wildman, K. A., D.-K. Lee, and A. Ramamoor-
thy, 2003. Mechanism of lipid bilayer disruption by
the human antimicrobial peptide, LL-37. Biochemistry
42:6545 -- 6558.
[33] Anthony G., L., 2011. Biological membranes: the impor-
tance of molecular detail. Trends in Biochemical Sciences
36:493 -- 500.
[34] Sun, D., J. Forsman, and C. E. Woodward, 2015. Am-
phipathic membrane-active peptides recognize and stabi-
lize ruptured membrane pores: exploring cause and effect
with coarse-grained simulations. Langmuir 31:752 -- 761.
[35] Wang, G., B. Mishra, R. F. Epand, and R. M. Epand,
2014. High-quality 3D structures shine light on antibacte-
rial, anti-biofilm and antiviral activities of human cathe-
licidin LL-37 and its fragments. BBA - Biomembranes
1838:2160 -- 2172.
[36] Noguchi, H., 2011. Anisotropic surface tension of buckled
fluid membranes. Phys. Rev. E 83:061919.
[37] Hu, M., P. Diggins, and M. Deserno, 2013. Determining
the bending modulus of a lipid membrane by simulating
buckling. J. Chem. Phys. 138:214110 -- 214110 -- 13.
[38] Cui, H., E. Lyman, and G. A. Voth, 2011. Mechanism of
membrane curvature sensing by amphipathic helix con-
taining proteins. Biophys. J. 100:1271 -- 1279.
[39] Wang, G., 2008. Structures of human host defense cathe-
licidin LL-37 and its smallest antimicrobial peptide KR-
12 in lipid micelles. J. Biol. Chem. 283:32637 -- 32643.
PDB: 2K6O.
[40] Hara, T., H. Kodama, M. Kondo, K. Wakamatsu,
A. Takeda, T. Tachi, and K. Matsuzaki, 2001. Effects
of peptide dimerization on pore formation: Antiparal-
lel disulfide-dimerized magainin 2 analogue. Biopolymers
58:437 -- 446. PDB: 1DUM.
8
[41] Terwilliger, T. C., L. Weissman, and D. Eisenberg, 1982.
The structure of melittin in the form I crystals and its
implication for melittin's lytic and surface activities. Bio-
phys. J. 37:353 -- 361. PDB: 2MLT.
[42] Marrink, S. J., H. J. Risselada, S. Yefimov, D. P. Tiele-
man, and A. H. de Vries, 2007. The MARTINI force
field: coarse grained model for biomolecular simulations.
J. Phys. Chem. B 111:7812 -- 7824.
[43] Hatzakis, N. S., V. K. Bhatia, J. Larsen, K. L. Mad-
sen, P. Bolinger, A. H. Kunding, J. Castillo, U. Gether,
P. Hedegard, and D. Stamou, 2009. How curved mem-
branes recruit amphipathic helices and protein anchoring
motifs. Nat. Chem. Biol. 5:835 -- 841.
[44] Vanni, S., H. Hirose, H. Barelli, B. Antonny, and R. Gau-
tier, 2014. A sub-nanometre view of how membrane cur-
vature and composition modulate lipid packing and pro-
tein recruitment. Nat. Commun. 5:4916.
[45] Pronk, S., S. P´all, R. Schulz, P. Larsson, P. Bjelkmar,
R. Apostolov, M. R. Shirts, J. C. Smith, P. M. Kas-
son, D. van der Spoel, B. Hess, and E. Lindahl, 2013.
GROMACS 4.5: a high-throughput and highly parallel
open source molecular simulation toolkit. Bioinformatics
29:845 -- 854.
[46] Monticelli, L., S. K. Kandasamy, X. Periole, R. G. Lar-
son, D. P. Tieleman, and S.-J. Marrink, 2008. The MAR-
TINI coarse-grained force field: extension to proteins. J.
Chem. Theory Comput. 4:819 -- 834.
[47] Yesylevskyy, S. O., L. V. Schafer, D. Sengupta, and
S. J. Marrink, 2010. Polarizable water model for the
coarse-grained MARTINI force field. PLoS Comput. Biol.
6:e1000810.
[48] Marrink, S. J., A. H. de Vries, and A. E. Mark, 2004.
Coarse grained model for semiquantitative lipid simula-
tions. J. Phys. Chem. B 108:750 -- 760.
[49] Baoukina, S., L. Monticelli, M. Amrein, and D. P. Tiele-
man, 2007. The molecular mechanism of monolayer-
bilayer transformations of lung surfactant from molecular
dynamics simulations. Biophys. J. 93:3775 -- 3782.
[50] de Jong, D. H., G. Singh, W. F. D. Bennett, C. Arnarez,
T. A. Wassenaar, L. V. Schafer, X. Periole, D. P. Tiele-
man, and S. J. Marrink, 2013. Improved parameters for
the Martini coarse-grained protein force field. J. Chem.
Theory Comput. 9:687 -- 697.
[51] Bussi, G., D. Donadio, and M. Parrinello, 2007. Canoni-
cal sampling through velocity rescaling. J. Chem. Phys.
126:014101.
[52] Berendsen, H. J. C., J. P. M. Postma, W. F. van Gun-
steren, A. DiNola, and J. R. Haak, 1984. Molecular dy-
namics with coupling to an external bath. J. Chem. Phys.
81:3684 -- 3690.
[53] Essmann, U., L. Perera, M. L. Berkowitz, T. Darden,
H. Lee, and L. G. Pedersen, 1995. A smooth particle
mesh Ewald method. J. Chem. Phys. 103:8577 -- 8593.
[54] Lee, M.-T., T.-L. Sun, W.-C. Hung, and H. W. Huang,
2013. Process of inducing pores in membranes by melit-
tin. Proc. Natl. Acad. Sci. U.S.A. 110:14243 -- 14248.
[55] Kreyszig, E., 1991. Differential geometry. Dover Publi-
cations, New York.
[56] Kunsch, H. R., 1989. The Jackknife and the Bootstrap
for general stationary observations. Ann. Stat. 17:1217 --
1241.
[57] Humphrey, W., A. Dalke, and K. Schulten, 1996. VMD:
Visual molecular dynamics. J. Mol. Graphics 14:33 -- 38.
9
[58] Hatzakis, N. S., V. K. Bhatia, J. Larsen, K. L. Madsen,
P.-Y. Bolinger, A. H. Kunding, J. Castillo, U. Gether,
P. Hedegard, and D. Stamou, 2009. How curved mem-
branes recruit amphipathic helices and protein anchoring
motifs. Nat. Chem. Biol. 5:835 -- 841.
[59] He, Y., and T. Lazaridis, 2013. Activity determinants
of helical antimicrobial peptides: A large-scale computa-
tional study. PLoS ONE 8:e66440.
[60] Braun, A. R., E. Sevcsik, P. Chin, E. Rhoades,
S. Tristram-Nagle, and J. N. Sachs, 2012. α-Synuclein
induces both positive mean curvature and negative Gaus-
sian curvature in membranes.
J. Am. Chem. Soc.
134:2613 -- 2620.
[61] Goodsell, D. S., and A. J. Olson, 2000. Structural sym-
metry and protein function. Annu. Rev. Biophys. Biomol.
Struct. 29:105 -- 153.
[62] Rosenberg, S. A., M. E. Quinlan, J. N. Forkey, and Y. E.
Goldman, 2005. Rotational motions of macromolecules
by single-molecule fluorescence microscopy. Acc. Chem.
Res. 38:583 -- 593.
[63] Davies, K. M., C. Anselmi, I. Wittig, J. D. Faraldo-
G´omez, and W. Kuhlbrandt, 2012. Structure of the
yeast F1Fo-ATP synthase dimer and its role in shaping
the mitochondrial cristae. Proc. Natl. Acad. Sci. U.S.A.
19:13602 -- 13607.
[64] Mim, C., H. Cui, J. A. Gawronski-Salerno, A. Frost,
E. Lyman, G. A. Voth, and V. M. Unger, 2012. Struc-
tural basis of membrane bending by the N-BAR protein
endophilin. Cell 149:137 -- 145.
[65] Ge, C., J. G´omez-Llobregat, M. J. Skwark, J.-M. Ruyss-
chaert, A. Wieslander, and M. Lind´en, 2014. Membrane
remodeling capacity of a vesicle-inducing glycosyltrans-
ferase. FEBS J. 281:3667 -- 3684.
10
ANISOTROPIC MEMBRANE CURVATURE SENSING BY AMPHIPATHIC PEPTIDES
-- SUPPORTING INFORMATION.
S1. CONVERGENCE AND INDIVIDUAL REPLICAS
Simulations of proteins interacting with mixed bilayers can be challenging to converge due to slow lipid diffusion
and long-lived protein-lipid interactions [65]. For this reason, we run three independent replicas rather than one long
simulation for each peptide, and use them as a simple control of the robustness of our conclusions. Figures S1-S3
show histograms of center-of-mass positions, orientations, and joint positions-orientations of both the three individual
production runs for each peptide, as well as aggregated histograms. In the case of melittin (Fig. S2), orientations of
both the N- and C-terminal helices are shown.
While the results for individual trajectories are obviously noisier than the aggregated statistics, it is clear that the
same qualitative features are present in all replicas. In particular, two well-separated orientational states of melittin
and LL-37 are clearly visible (albeit not equally populated) in all trajectories, strongly indicating that our simulations
are long enough to capture the major low-energy states of these systems. However, the sampling is still limited enough
to induce significant statistical uncertainty in fit parameters, as seen Fig. 4d.
FIG. S1. Results for magainin, from (a) three independent production runs, and (b) aggregated.
11
FIG. S2. Results for melittin, from (a) the three independent production runs, and (b) aggregated. Both N- and C-terminal
results are shown.
TABLE S1. Fit parameters (fit± bootstrap std.) for the E2 model for the curves shown in Fig. 4, rounded to two significant
2 (C⊥ − a6)2 +
digits, in appropriate units of kBT and nm. This model is given by E2 = a1
a7CX (C(cid:107) + C⊥) + a8CX (C(cid:107) − C⊥), i.e., with a C(cid:107)C⊥-term omitted for identifiability.
2 (C(cid:107) − a2)2 + a3(CX − a4)2 + a5
a1
a5
a2
a6
a3
a7
MAG 71 ± 30 −0.30 ± 0.26 73 ± 28 0.024 ± 0.028
MEL 170 ± 52 −0.23 ± 0.03 130 ± 52 0.002 ± 0.02
LL-37 300 ± 130 −0.24 ± 0.07 300 ± 140 −0.02 ± 0.03
a4
MAG 68 ± 28 −0.30 ± 0.27 −24 ± 19
27 ± 28
MEL
LL-37 310 ± 150 −0.25 ± 0.73 68 ± 59
90 ± 55 −0.3 ± 1.3
a8
5.3 ± 3.3
2.4 ± 4.2
1 ± 7
12
FIG. S3. Results for LL-37, from (a) three independent production runs, and (b) aggregated.
13
S2. LOCATION OF THE CURVATURE SENSING SITE ON LL-37
LL-37 shows indications of asymmetry around s = 0.5 that is incompatible with the symmetry of the curvature
tensor elements (Fig. 3d), and the fitted EC model is also less consistent with the orientation averaged binding
energy (Fig. 3e) than the other peptides. Here, we explore the hypothesis that these effects are caused by using the
center-of-mass of the peptide for defining the position s, which might be inappropriate if the sensitivity is unequally
distributed along the peptide. Our rationale for this hypothesis is that a correlation between position and orientation,
as indicated in the LL-37 data in Fig. 3d might come about if the effective curvature sensing site is different than the
center-of-mass which we tracked to extract that data, as sketched in Fig. 5.
In Fig. S4, we show the corresponding analysis for LL-37 assuming a few alternative effective curvature sensing
sites, with the center-of-mass in the middle row. The correlation between θ and s around each peak clearly becomes
more pronounced and N-terminal-like (c.f. Fig. 5) when the tracking site moves towards the N-terminal end. However,
the asymmetry almost disappears when one assumes the effective curvature sensing site to be the center of mass of
the C-terminal helix, and appears again with the opposite C-terminal-like trend when tracking the C-terminal end.
Of these cases, the center-of-mass of the C-terminal helix is most consistent with the symmetries of curvature tensor
elements and thus acts as an effective "sensing site", which indicates that this part of the peptide is more important
for curvature sensing. Fitting the EC model to this data yields κ = 323 ± 127 kBT nm, C−1
0 = −4.1 ± 0.5 nm,
b = 8.2 ± 0.6 kBTnm, and α = 69 ± 2◦, not significantly different from the parameters shown in Fig. 4.
However, all distributions are still slightly asymmetric around s = 0.5, with average s-values ranging from about
0.52 to 0.51 for the N- and C-terminal ends respectively, corresponding to an average displacement of 0.5 nm to 0.35
nm from the mid point. A closer examination of the significance of this observation would require substantially better
statistics, perhaps from using some enhanced sampling method, as well as more systematic studies using a larger
range of curvatures. This is outside the scope of this study.
14
FIG. S4. Analysis of LL-37 using different definitions of s and θ, namely (a) the first residue and orientation of the N-terminal
helix, (b) the center-of-mass and orientation of the N-terminal helix, (c) the center-of-mass and orientation of the whole peptide
(same as shown in the main text), (d) the center-of-mass and orientation of the C-terminal helix, and (e) the last residue and
orientation of the C-terminal helix. relevant curvature sensing site. The columns show (left) the (s, θ)-histogram, (mid) a fit of
the EC model, and (right) the orientation-averaged binding free energy, obtained from the model fit (line) or using weighted
histograms, Eq. 4, (dots) with error bars as in Fig. 3. The fit and histogram curves are vertically aligned by least-squares fit
of the points at C ≤ −0.15 nm−1.
|
1411.2454 | 3 | 1411 | 2015-01-10T14:06:43 | Solitary Shock Waves and Adiabatic Phase Transition in Lipid Interfaces and Nerves | [
"physics.bio-ph"
] | This study shows that the stability of solitary waves excited in a lipid monolayer near a phase boundary requires positive curvature of the adiabats, a known necessary condition in shock compression science. It is further shown that the condition results in a threshold for excitation, saturation of the wave amplitude and the splitting of the wave at the phase boundaries. Splitting in particular confirms that a hydrated lipid interface can undergo condensation on adiabatic heating thus showing retrograde behavior. Finally, using the new theoretical insights and state dependence of conduction velocity in nerves, the curvature of the adiabatic state diagram is shown to be closely tied to the thermodynamic blockage of nerve pulse propagation. | physics.bio-ph | physics | http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
Solitary Shock Waves and Adiabatic Phase Transition in Lipid Interfaces and
Nerves.
Shamit Shrivastava, Kevin Heeyong Kang, Matthias F. Schneider*
Department of Mechanical Engineering, Boston University, Boston MA 02215
Abstract
This study shows that the stability of solitary waves excited in a lipid monolayer near a
phase boundary requires positive curvature of the adiabats, a known necessary
condition in shock compression science. It is further shown that the condition results in
a threshold for excitation, saturation of the wave’s amplitude and the splitting of the
wave at the phase boundaries. Splitting in particular confirms that a hydrated lipid
interface can undergo condensation on adiabatic heating thus showing retrograde
behavior. Finally, using the new theoretical insights and state dependence of conduction
velocity in nerves, the curvature of the adiabatic state diagram is shown to be closely
tied to the thermodynamic blockage of nerve pulse propagation.
Introduction
Dense networks of hydrated membrane interfaces populate cellular environments [1].
The elastic properties of these quasi-2D systems have been a subject of extensive
research [2]. However, most studies usually consider quasi-static processes [3] and/or
small displacements [4]. We have recently demonstrated that the elastic properties of
such interfaces support the propagation of acoustic pulses, which has implications for
cell communication from single cells to action potentials in the nervous system [5,6].
1
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
Particular interest arises from the observation that the cellular machinery spends
considerable resources in fine tuning the thermodynamic (TD) state diagrams of
biological membranes, usually adapting in the vicinity of a phase transition, where the
state diagram exhibits clear nonlinearities [7,8]. For dynamic processes the curvature of
the adiabatic state diagram is of fundamental importance [9,10]. Indeed, a non-linearity
in the elastic properties of the plasma membrane is believed to be crucial for the
phenomenon of nerve pulse propagation [11–13] which is known to be adiabatic [14].
Therefore an understanding of the adiabatic state diagrams of lipid interfaces near
similar nonlinearities is crucial for an improved thermodynamic understanding of nerve
pulse propagation. Interestingly, spontaneous mechanical perturbations have been
observed in a variety of biological systems [15] and are found to propagate along with
nerve pulse propagation as well, analogous to sound waves [16,17]. In a recent study
[6] we showed that thermodynamically coupled perturbations (electrical-optical-
mechanical) during 2D sound waves in a lipid monolayer, a simple model system for
plasma membrane, are strikingly similar to those observed during nerve pulse
propagation. Based on experimentally determined sound velocities 𝑐(𝜋), it appeared
that solitary waves only exist for a positive curvature (
𝜕2𝑎
𝜕𝜋2) of the state diagram (𝜋 𝑎𝑛𝑑 𝑎
represent the lateral pressure and specific area at the lipid interface). This results in a
threshold for excitation if the initial equilibrium state is within a regime of negative
curvature and changes to positive during excitation/propagation
Here we investigate the velocity of propagation of such waves as a function of
amplitude and show that locally the condition (
𝜕2𝑎
𝜕𝜋2)
𝑆
> 0 (S representing the interfacial
entropy) is preserved for the 2D sound waves at the liquid expanded/ liquid condensed
2
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
phase boundary, even though the isothermal compression (
𝜕2𝑎
𝜕𝜋2)
< 0. Furthermore we
𝑇
show that the maximum amplitude saturates at a value that is significantly less than
expected from isothermal compression. The evolution of these waves over distance
shows splitting into a non-dispersive forerunner wave and a slower dispersive wave.
These results are in accordance with classical shock theory where the condition
𝜕2𝜐
(
𝜕𝑃2)
𝑆
> 0 [9,18] or (
𝜕2𝑃
𝜕𝜐2)
𝑆
> 0 [10] is associated with the existence of compression
shocks and its violation is associated with rarefaction shock waves, where 𝑃 and υ are
pressure and specific volume. Finally we discuss the implications of the curvature of
the state diagram of the nerve membrane and its possible relation to reversible
thermodynamic blockage of nerve pulse propagation.
A lipid monolayer easily self assembles by adding a lipid and fluorophore mixture at the
air/water interface of a Langmuir trough. Lipid monolayers are not only accessible and
robust, but their molecular composition and thermodynamic state and hence the
mechanical properties can also be precisely controlled, monitored, and characterized
[19]. Thus they provide an excellent platform to study 2D interfacial sound waves, both
experimentally and theoretically [5,6]. The opto-mechanical setup has been described
in detail elsewhere [6,20]. A cantilever excites the monolayer (containing the lipids
(DPPC), the donor (NBD-PE) and acceptor dye molecules (Texas Red-DHPE)
(100:1:1)) longitudinally with a piezo controlled deflection producing 2D sound waves. A
microscope records these sound waves by observing the ratiometric Forster resonance
energy transfer (FRET), simultaneously at two wavelengths (535 and 605nm) between
3
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
a pair of lipid-conjugated fluorophores, using
∆𝜃
𝜃
=
Δ𝐼535
𝐼535
−
Δ𝐼605
𝐼605
. The distance between the
cantilever and the objective can be controlled by a screw meter.
Results and Discussion
Threshold, Saturation, Velocity and Pulse Width
For the linear case (infinitesimal amplitude), the velocity of sound is related to state
variables according to 𝑐2 =
1
𝜌𝑘𝑆
= (
𝜕𝜋
𝜕𝜌
)
𝑠
, where 𝜋 is the lateral pressure, 𝜌 (kg/m2) is the
density of the quasi-2D interfacial region and 𝑘𝑆 is the isentropic compressibility of the
interface [5]. For a nonlinear system, such as a lipid monolayer near a phase transition,
𝑘𝑆 strongly depends on the density change ∆𝜌 and can undergo significant changes
within a single pulse ∆𝜌(𝑡). Therefore the velocity varies within a single pulse resulting
in evolving pulse shapes [6,12]. Figure 1(a) shows this behavior for different pulses as
solitary waves of different amplitudes, obtained by varying the stimulus (i.e. the
mechanical impulse from the piezo device) and measured via FRET, arrive at different
times for a given mean equilibrium state (𝜋 =
and 𝑇 = 293.15𝐾) (Fig. 1b). Note that
4.3𝑚𝑁
𝑚
Δ𝑎
𝑎
compression amplitudes (∆𝜌/𝜌0) - with
∆𝜌
𝜌0
= −
- can be estimated from variations in
FRET parameter Δ𝜃/𝜃 using the characteristic curve (𝑎 ↔ 𝜃)𝑇 obtained during
isothermal compression (Fig 1b and c) [6].
4
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
𝑚
4.3𝑚𝑁
Figure 1 (Color Online) Dependence on excitation strength and amplitude (color
online). (a) Measured pulse shapes at 0.84cm from the excitation blade as a function of
blade amplitude (mm). The pulses are excited at t=0. The fixed equilibrium state
𝑎𝑛𝑑 𝑇 = 293.15𝐾) is indicated by the dot on (b) the isothermal state diagram
(𝜋 =
and the isothermal compressibility plot in the inset. (c) The characteristic isothermal
curve relating FRET parameter and surface area (𝑎 ↔ 𝜃)𝑇 obtained during quasi-static
compression at two different spots along the propagation path. The arrows in (a)
indicate the splitting of the pulse on increasing the excitation strength beyond saturation
limit.
Fluorescent probes that depend on dipole reorientation for sensing voltage changes (as
observed here in lipid monolayer [6,20]), have been shown to report them without
discernible time lag during nerve pulse propagation [21–23]; which indicates that as far
as our overall goal of understanding thermodynamics of nerve-pulse propagation is
concerned the isothermal opto-mechanical coupling is a reasonable approximation
5
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
during the observed pulses as well. This allows immediate extraction of three key
relations from these experiments; (i) The response (∆𝜌/𝜌0) is highly nonlinear with a
clear threshold and an asymptotic saturation of amplitude as a function of excitation
strength (Fig 2a). (ii) The velocity calculated from the time of arrival varies linearly with
relative compression (∆𝜌/𝜌0) upto ∆𝜌/𝜌0=0.15 (Fig. 2b), which coincides with the
beginning of the saturation (indicated by dashed lines Fig. 2a and b) of the nonlinear
response curve. (iii) The half-width of a pulse as a function of relative compression also
follows the exact same trend as velocity and diverges from a linear dependence near
maximum amplitude (Fig 2b). Notably, the observed saturation of (∆𝜌/𝜌0)𝑚𝑎𝑥=0.15 is
approximately 20% of the value expected from the relative compression during a quasi-
static phase transition (
∆𝜌
𝜌0
= −
Δ𝑎
𝑎
= 0.75) (Fig 1b). Further increase in the excitation
strength results into splitting of the pulse (indicated by arrows) as will be discussed
below. All these factors clearly indicate that a fundamental understanding of these
processes has to be tied to the dynamic properties of the system.
Positive curvature of the Adiabatic State Diagram
It is useful to write the curvature (
𝜕2𝑎
𝜕𝜋2)
𝑆
in its non-dimensional form Γ =
𝑐4
2𝑎3 (
𝜕2𝑎
𝜕𝜋2)
,
𝑠
because then it is directly related to the well-known acoustic parameter of nonlinearity
B/A as Γ =
𝐵
2𝐴
+ 1 . Within a thermodynamic treatment of acoustics, A and B are directly
related to the coefficients of the isentropic Taylor expansion for the lateral pressure
𝜋(𝜌, 𝑆) [24];
𝜋𝑆 = 𝜋0 + 𝐴 (
2
∆𝜌
𝜌0
) +
𝐵
2
(
∆𝜌
𝜌0
)
with
+ ⋯
(1)
6
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
𝐴 = 𝜌0 (
𝜕𝜋
𝜕𝜌
)
𝑆
= 𝜌0𝑐0
2; 𝐵 =
2
𝜌0
2!
(
𝜕2𝜋
𝜕𝜌2)
𝑆
.
(2)
Figure 2 (Color Online) Threshold, Saturation, Velocity and Pulse Width. (a)
Relative compression (
∆𝜌
𝜌0
) extracted from fig. 1a using the quasi-static response curve
of fig 1b, as a function of excitations strength (blade’s displacement amplitude in mm)
(b). The velocity as obtained from the time of arrival of the peak amplitude is plotted with
respect to relative compression amplitude (
∆𝜌
𝜌0
) . (a,b) The limit of linear dependence on
amplitude is indicated by dotted lines.
Furthermore a first order approximation for the relation between c and ∆𝜌/𝜌0 can be
written as [24,25];
𝑐 = 𝑐0 [1 +
1
𝐵
2
𝐴
(∆𝜌/𝜌0)]
(3)
which can be directly compared to the observed dependence of velocity on relative
compression (∆𝜌/𝜌0) in Fig. 2b. From the y-intercept of the linear fit 𝑐0 = 0.236𝑚/𝑠 can
be directly extracted as the velocity for infinitesimal amplitude (linear limit). Further, the
7
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
slope of the fit allows the determination of B/A (eq.3) to be 0.5 for our system and using
Γ =
1
𝐵
2
𝐴
+ 1 we get Γ = 1.25 which indeed shows that the curvature of the adiabatic state
diagram is positive locally. However this is in stark contrast with the negative curvature
(
𝜕2𝑎
𝜕𝜋2)
𝑇
4.3
𝑚𝑁
𝑚
of the isothermal state diagram (Fig.2) at the given equilibrium state (𝜋 =
𝑎𝑛𝑑 𝑇 = 293.15𝐾) and hence announces
the
failure of a quasi-static
approximation. In fact, we believe that the discrepancy between the isothermal and
adiabatic curvature is closely related to existence of the observed threshold for
excitation: A stable nonlinear wave-front exists only for (
𝜕2𝑎
𝜕𝜋2)
𝑠
> 0 (see below). When
the interface is prepared, however, to exhibit a negative curvature in it’s equilibrium
state (
𝜕2𝑎
𝜕𝜋2)
𝑇
, only excitations, which provide sufficient power to transfer the state of the
interface from a negative (
𝜕2𝑎
𝜕𝜋2)
𝑇
< 0 (in the quasi-static limit) to a positive (
𝜕2𝑎
𝜕𝜋2)
𝑠
> 0 (in
the adiabatic limit) while decoupling the interface from the bulk, will result in the
formation of stable nonlinear pulses (Fig.3).
Shocks near Phase Transition – Saturation of Amplitude and Splitting
Note that a blade displacement of 1.3mm with the rise time of ~5ms gives a maximum
particle velocity of ~0.26m/s which is comparable to the velocity of sound in the lipid
monolayer in the given state. Hence, the observed pulses can be treated as shock
waves and we can learn from classical shock theory. Indeed, in shock compression
science (
𝜕2𝜐
𝜕𝑃2)
𝑆
> 0 is a necessary condition for stable shocks, which is usually satisfied
as in most cases the compressibility (~ −
𝜕𝜐
𝜕𝑃
) decreases with pressure. However the
8
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
exception (
𝜕2𝜐
𝜕𝑃2)
𝑆
28].
< 0 can occur near a phase transition or critical points (fig. 3) [18,26–
excited with enough power to move along a paths in the diagram of (
Figure 3 (Color Online) Representation of the phenomenon on (𝝅 → 𝒂) State
Diagrams. Isothermal (dashed) and (shock) adiabatic (solid) state changes are shown.
Initially the system is at 1 (equilibrium) prepared right at the equlibirum phase boundary
(σ). Starting from this point, the non-equilibrium state diagram during a pulse can
proceed along various paths determined by the blade velocity. Only if the pulse is
𝜕2𝑎
𝜕𝜋2) > 0 (see text),
a solitary wave appears. When observed at a fixed distance (fig.1 and fig.2) the different
amplitudes lie on the non-equilibrium adiabat as represented by the solid curve between
𝜕2𝑎
𝜕𝜋2) > 0. Since the system was prepared at the equilibrium phase
boundary these adiabats extend into the meta-stable (shaded) region, given the positive
curvature. For fig. 1 and 2 the end states corresponding to different amplitudes are
shown clearly in the inset. The slope of the straight dotted lines is directly related to the
shock velocity and represents the jump condition (Rayleigh line) during the shocks. The
observed splitting of the pulse (fig. 1 and 4) represents a discontinuity in velocity and
hence a discontinuity in the slope of the Rayleigh lines (compare (𝟏 → 𝟐) vs (𝟐 → 𝟑′)).
This also indicates crossing over the spinodal condition (Wilson line, W) or adiabatic
1 and 𝟐 with (
9
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
phase transition. If we follow a pulse in fig. 4, it’s amplitude and hence velocity
decreases till the pulse eventually splits at a critical condition represented by (𝟏 → 𝟐 →
𝟑). Prior to splitting (closer to the source) the pulse strength may be sufficient to induce
a complete phase transition (𝟏 → 𝟐′). Figure is not to scale. CP indicates the critical
point.
The consequences of a discontinuous and/or negative (
𝜕2𝜐
𝜕𝑃2)
𝑆
in the phase
transition region for the stability of a shock-front have been investigated in detail, both
theoretically as well as experimentally
[18,27,29,30]. One direct observable
consequence of (
𝜕2𝜐
𝜕𝑃2)
𝑆
< 0, which can occur at phase transition, is the splitting of a
compressive shock into a non-dispersive forerunner wave and a slower more dispersive
condensation wave near a phase boundary [27]. The emergence of a second wave-
front can in fact already be seen in fig. 1a at maximum excitation and indicates a
discontinuity within the adiabatic state diagram.
10
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
Figure 4 (Color Online) Distance dependence, evolution and splitting. Pulse
shapes are plotted at various distances from the excitation blade (indicated on each plot
in mm) for a fixed excitation displacement of 1.3mm and equilibrium state (fig. 1b). After
initial solitary propagation up to 7mm, the front runner wave begin to emerge at 9.8mm
(indicated by the arrow) and is completely evolved at 12.6mm, while the residual wave
that follows shows strong dispersion and loss of amplitude. The frontrunner wave
however continues
for
approximately another 1cm up to 21mm (dotted line shows stable amplitude), at which
point it starts to disappear as well. To emphasize the repeatability of these experiments
pulses from three different experiments and their average is plotted.
to propagate with stable amplitude and pulse shape
In Fig. 4 the splitting process of the waves – indeed a very rare phenomenon in
shock science - is observed in detail. As the pulse shape evolves, it reaches a
maximum amplitude at a distance 7mm followed by a decay, which as seen here can
result in splitting beginning at 9.8mm. Post splitting, the forerunner wave grows at the
11
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
cost of the slower more dispersive wave and propagates much further. This along with
the fact that the splitting is observed only during decay is consistent with similar
observations for vapor-compression shocks in other systems near phase transitions
[27]. In order to compare their interpretation with our data it is helpful to assume that
the solitary wave profile (Fig. 1 and 4) can be decomposed into a compression wave-
front, discussed below, followed by an expansion wave-front both related via the
continuity condition1 (note: in contrast to a compression wave-front, the expansion
wave-front is stable for (
𝜕2𝑎
𝜕𝜋2)
𝑆
< 0 [10].) As the excitation strength is increased beyond
threshold, the interface first remains in a regime of positive curvature (hence the
increase in c shown in Fig. 2b). We likely cross the metastable regime of the phase
transition (see the path in Fig. 3) before the termination of the path at the spinodal, W
(wave-front (𝟏 → 𝟐) in Fig.1 and Fig. 3) or if near the source (𝟏 → 𝟐′) which then decays
(Fig. 3 and Fig. 4). Eventually the wave becomes instable and splits ((𝟏 → 𝟐 → 𝟑) in
Fig.3) indicating that a new regime of the adiabatic diagram is entered. Thus instead of
simple adiabatic heating during compression, the large amplitude causes nucleation
and a phase transition beyond point 2 in Fig.3 (retrograde behavior [27]). Due to the
resulting discontinuity at the spinodal line, the “combined” wave (𝟏 → 𝟐 → 𝟑) splits into a
forerunner shock (𝟏 → 𝟐) that propagates – in our case - with a velocity of the liquid-
expanded state and a condensation wave (𝟐 → 𝟑′), propagating at a slower velocity
determined by the properties of the coexistence region of the lipid interface. The
1 In order to follow the lines of Thomson’s work [26–28] we imagine our biphasic [6] shock consisting
of a compression shock followed by one of rarefaction. In this case shocks begin to “interact” and can for
instance weaken each other thereby supporting splitting [27]. However the stability of the rarefaction tail
and its interaction with the front will require further analysis and experiments, which are outside the
scope of this work and will be treated elsewhere.
12
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
interpretation of a pulse-induced nucleation is consistent with the simultaneous
observations of (i) a saturation of amplitude, (ii) an abrupt change in curvature and (iii)
broadening of pulse shape (increased dispersion), observed in Fig. 1 and 2.
It is most likely that similar waves were observed during electrically induced
critical de-mixing of multicomponent lipid monolayers by McConell et. al [31,32] and
were believed to be shock waves. In these experiments, nucleation resulting from an
electrical impulse that propagated as condensed domains, dispersed and dissolved
rapidly. Although this non-equilibrium phenomenon was mentioned only qualitatively,
these experiments indicate that similar phenomenon can also be observed near
complex phase boundaries in multi component systems and can be excited electrically
at lipid interfaces.
Nonlinear behavior in terms of a relation between amplitude and velocity has
been treated theoretically in the context of models for soliton propagation in isolated
lipid bilayers [12,33] and non-equilibrium phase transition in liquid crystals [34].
However, the dispersion relation intrinsic to such models depends not only on state but
also on boundary conditions and/or geometry and therefore can be very different, even
qualitatively, between a lipid monolayer at the air/water interface and a biological
membrane with all its structural complexities. For example, by allowing a quadratic
negative nonlinearity in pressure and assuming an adhoc positive linear dispersion,
postulated based on measurements of phase velocity in lipid vesicles, Heimburg and
Jackson derived a nonlinear wave equation to predict a decrease in velocity and pulse
width with increasing amplitude for solitons in lipid membranes [12,35]. This is in
complete contrast with the current observations in lipid monolayers where the width and
13
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
velocity increase with amplitude indicating positive nonlinearity B/A and a highly
nonlinear dispersion [36], which is also consistent with the phase transition [37].
We stress, that the line of argument above, relating the curvature of adiabatic
state diagrams to the stability and instability of propagating shocks, is universal. This
means it will not only hold for mono- and bilayers, but also for an interface in a living
system with all its structural complexities. It will be exciting to see experiments on the
state diagrams of “living interfaces” and the propagation of nonlinear pulses within them.
An intriguing example of a negative curvature (
𝜕2𝑎
𝜕𝜋2)
𝑠
< 0 in living systems may
be found in nerves near temperature induced reversible blocks for pulse propagation
(Fig.5). When approximating the dimensionless curvature Γ =
𝑐4
2𝑎3 (
𝜕2𝑎
𝜕𝜋2)
𝑠
using Maxwell
relations of a simple thermodynamic system2, an independent thermodynamic equation
relating
𝐵
𝐴
(and hence Γ) to experimentally accessible variables can be written as [9,38]
(𝛼𝑇 being the isothermal expansion and 𝑐𝑝 the heat capacity at constant pressure);
2(Γ − 1) =
𝐵
𝐴
= 2𝑐0[𝜌0𝐴 (
𝜕𝑐
𝜕𝜋
)
𝑇
+
𝑇𝛼𝑇
𝜌0𝐴𝑐𝑝
(
𝜕𝑐
𝜕𝑇
)
]
𝜋
(4)
2 Here we assume a mechanically system, i.e. the state is characterized by a, and T only.
However, further couplings (Maxwell relations) may play an important role, e.g. electrical (U-q),
thermal (E-T) or chemically (-N) couplings. These would alter the presented relation.
14
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
Figure 5. Heat block of the nerve impulse in squid giant axon (data adapted from
[39]). All known action potentials, including those in human, animals or even algae,
exhibit a heat block [40]. The temperature varies and can be adapted by changing
growth conditions [41] . At the heat block a transition from (
𝜕𝑐
𝜕𝑇
)
𝑃
> 0 to (
𝜕𝑐
𝜕𝑇
)
𝑃
< 0 takes
place. This condition leads to an instability of the shock wave excited near a phase
transition.
For a lipid monolayer all the variables are experimentally accessible except 𝜌0𝐴, which
in fact can be calculated using eq. 4, allowing subsequent calculation of ∆𝜋 using eq. 2
(see appendix A). In order to estimate the curvature Γ, for a nerve fiber, we use the
experimentally available [39,42] relations for the bulk pressure and temperature
dependence of c , i.e. (
𝜕𝑐
𝜕𝑃
)
𝑇
and (
𝜕𝑐
𝜕𝑇
)
𝑃
. While experiments demonstrate (
𝜕𝑐
𝜕𝑃
)
𝑇
is usually
15
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
negative3 [42,43], the quantity (
𝜕𝑐
𝜕𝑇
)
𝑃
starts of positive but gets increasingly negative as
we approach the temperature corresponding to heat block [39,44]. Indeed negative
values of (
𝜕𝑐
𝜕𝑃
)
𝑇
and (
𝜕𝑐
𝜕𝑇
)
𝑃
imply a negative value of B/A and – if less than -2 – a negative
Γ as well, which would lead to the above mentioned cessation of the shock wave.
In conclusion we have demonstrated the application of shock compression
science at a soft interface near phase transition and its implications for biological
systems, especially nerves. In particular we extracted the curvature (
𝜕2𝑎
𝜕𝜋2)
𝑠
of the
adiabatic state diagram from solitary waves observed in lipid monolayer and tied it to
the observations of excitation threshold, amplitude saturation and stability of solitary
waves (against splitting). Since this is the first attempt to apply shock compression
science to soft interfaces, several questions remain open and predictions to be tested:
For example our approach predicts the culmination of metastable regimes into a critical
point near heat-block in nerves [11,45]. Furthermore the saturation in amplitude, which
in our case originates from crossing over the spinodal condition, should correspond to a
“dynamic” phase transition in nerve pulse propagation, which remains to be observed
[46]. Another aspect arises from the fact, that opto-mechancial coupling deserves some
attention, as the calibration is done for the isothermal case [20]. Obviously, the
interpretation of
the experiments
in
the present work has been
intentionally
oversimplified as for a more quantitative description, further analysis, both theoretical
and experimental is required. New insights will arise from challenging the quasi-2D
3 precise measurements of (
𝜕𝑐
)
near the heat block to clarify the sign are currently not available. It seems plausible,
𝜕𝑃
𝑇
however, to assume a decrease in c with pressure even at temperatures near heat block.
16
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
nature of the propagation, as the role of thickness of the hydration layer as well its
interaction with the bulk or other interfaces nearby needs to be investigated. In addition,
the coupling of further thermodynamic variables (lipid dipole, bulk pH, charge) and the
role of boundary conditions will be important for a deeper understanding of these pulses
in biology4. This would also lead to a better understanding of dissipation in our system,
since even though we find regimes where the pulse can cover significant distance with
constant amplitude and width (fig. 4), dissipation of our solitary waves remains to be a
crucial unresolved issue when comparing them to nerve impulses. Finally the interaction
of propagating shock waves with the complex chemistry of biological interfaces (e.g.
enzymes) might lead to completely new phenomena that can now be studied
systematically [47]. We believe this work opens new doors for the physics community
to contribute to life sciences, in particular we imagine, to the understanding of inter- and
intra-cellular communication by combining nonlinear acoustics and the physics of critical
phenomenon at interfaces.
APPENDIX A
It is assumed that the lipid monolayer along with a few hydration layers form the
propagation medium that is adiabatically decoupled from the bulk. Hence 𝜌0𝐴with its
dimensions of kg/m2 essentially represents the mass of this medium projected on a 2D
interface. In order to calculate 𝜌0𝐴we employ the following independent thermodynamic
relation for B/A;
𝐵
𝐴
= 2𝑐0[𝜌0𝐴 (
𝜕𝑐
𝜕𝜋
)
𝑇
+
𝑇𝛼𝑇
𝜌0𝐴𝐶𝑝
(
𝜕𝑐
𝜕𝑇
)
𝜋
]
(5)
𝜕2𝑣
4 The condition
𝜕𝑃2 > 0 results from ∆𝑆 > 0 in a purely mechanical system [18] which implies entropy is a function
of only the volume. However in general entropy is a multi-dimensional function of all the extensive variables of the
interface.
17
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
While B/A is known from the experimental amplitude-velocity relation, all the other
parameters are accessible except 𝜌0𝐴. For example, the state 𝜋 = 4.3𝑚𝑁/𝑚 and
T=293.15K corresponds to the edge of the transition region and we have previously
shown [5] that velocity decreases from 0.4m/s to 0.2m/s on increasing pressure by
1mN giving
Δ𝑐
Δπ
= −0.2 𝑚/𝑚𝑁𝑠 at constant T. An increase in temperature at constant
pressure on the other hand, moves the system away from the transition region resulting
in an increase in velocity, here found experimentally to be
Δ𝑐0
ΔT
= 0.009𝑚/𝐾𝑠 at constant
𝜋=4.3mN/m. Finally, the phenomenon takes place across a LE-LC phase transition,
allowing the approximation
𝑇𝛼𝑇
𝜌0𝐶𝑝
=
𝑑𝑇𝑡𝑟
𝑑𝜋𝑡𝑟
[48] (the subscript indicates that the pressure
and temperature are taken at the transition). Plugging in the values gives 𝜌0𝐴 =
0.012𝑘𝑔/𝑚2 completing the equation of state Eq. 1, defined locally for 𝜋 = 4.3𝑚𝑁/𝑚
T=293.15K. As a result, ∆𝜋 can be calculated and is found to be 0.12 mN/m, which is in
good agreement with the previously reported values of pressure pulses in the transition
region [5]. Note that the obtained value of 𝜌0𝐴is a gross overestimate of the actual
value as it also accounts for charge/dipole effects that have been completely ignored in
Eq.1 and Eq.5.
Acknowledgments: We thank Dr. Konrad Kauffman (Göttingen) who introduced us to
the thermodynamic origin of nerve pulse propagation and its theoretical
explanation. We would also like to thank him for numerous seminars and
discussions. We would like to thank Dr. Christian Fillafer and Prof. Glynn Holt for
helpful discussions and critical reading of the manuscript. Financial support by
BU-ENG-ME is acknowledged. MFS appreciates funds for guest professorship
from the German research foundation (DFG), SHENC-research unit FOR 1543.
18
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
References
[1] E. Sackmann, Handbook of Biological Physics 1, (1995).
[2] M. Bloom, E. Evans, and O. L. E. G. Mouritsenj, Quarterly Reviews of Biophysics
3, 293 (1991).
[3] D. Needham and E. Evans, Biochemistry 27, 8261 (1988).
[4] R. Lipowsky, in Handbook of Biological Physics, edited by R. Lipowsky and E.
Sackmann (Elsevier B.V., 1995), pp. 521–602.
[5]
J. Griesbauer, S. Bössinger, A. Wixforth, and M. F. Schneider, Physical Review
Letters 108, 198103 (2012).
[6] S. Shrivastava and M. F. Schneider, Journal of Royal Society Interface 11,
20140098 (2014).
[7]
J. Hazel, Annual Review of Physiology 19 (1995).
[8] D. Georgescauld, J. Desmazes, and H. Duclohier, Molecular and Cellular
Biochemistry 27, 147 (1979).
[9] P. A. Thompson, Physics of Fluids 14, 1843 (1971).
[10] H. Bethe, in Classic Papers in Shock Compression Science, edited by J. Johnson
and R. Cheret (Springer New York, 1998), pp. 421–495.
[11] K. Kaufmann, Action Potentials and Electrochemical Coupling in the Macroscopic
Chiral Phospholipid Membrane, 1st ed. (Caruaru, Brasil, 1989).
[12] T. Heimburg and A. D. Jackson, Proceedings of the National Academy of
Sciences of the United States of America 102, 9790 (2005).
[13] Y. Kobatake, I. Tasaki, and A. Watanabe, Advances in Biophysics 2, 1 (1971).
[14] J. Howarth and R. Keynes, The Journal of Physiology 249, 349 (1975).
[15] A. E. Pelling, S. Sehati, E. B. Gralla, J. S. Valentine, and J. K. Gimzewski,
Science (New York, N.Y.) 305, 1147 (2004).
[16]
I. Tasaki, Biochemical and Biophysical Research Communications 215, 654
(1995).
[17] G. H. Kim, P. Kosterin, a L. Obaid, and B. M. Salzberg, Biophysical Journal 92,
3122 (2007).
19
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
[18] L. Landau and E. M. Lifshitz, in Fluid Mechanics, edited by J. Sykes and W. Reid,
Second (Pergamon Press Ltd., 1987), pp. 327–329.
[19] O. Albrecht and H. Gruler, Journal De Physique 39, 301 (1978).
[20] S. Shrivastava and M. F. Schneider, PloS One 8(7), e67524.
doi:10.1371/journal.pone.0067524 (2013).
[21] F. Conti, Annual Review of Biophysics and Bioengineering 4, 287 (1975).
[22] L. B. Cohen and B. M. Salzberg, Reviews of Physiology, Biochemistry and
Pharmacology 83, 35 (1978).
[23] F. Conti, R. Fioravanti, F. Malerba, and E. Wanke, Biophysics of Structure and
Mechanism 1, 27 (1974).
[24] R. Beyer, Nonlinear Acoustics (Academic Press, San Diego, CA, n.d.).
[25] M. F. Hamilton and D. T. Blackstock, Journal of Acoustical Society of America 83,
74 (1988).
[26] P. a. Thompson, Physics of Fluids 26, 3211 (1983).
[27] B. P. A. Thompson, H. Chaves, G. E. A. Meier, and Y. Kim, Journal of Fluid
Mechanics 185, 385 (1987).
[28] P. a. Thompson, G. C. Carofano, and Y.-G. Kim, Journal of Fluid Mechanics 166,
57 (2006).
[29] H. Bethe, in Classic Papers in Shock Compression Science, edited by J. Johnson
and R. Cheret (Springer New York, 1998), pp. 421–495.
[30] Y. . Zel’dovich and Y. P. Raizer, Physics of Shock Waves and High-Temperature
Hydrodynamic Phenomena (Academy of Sciences, Moscow, 1967).
[31] K. Y. C. Lee, J. F. Klingler, and H. M. McConnell, Science 263, 655 (1994).
[32] K. Y. Lee and H. M. McConnell, Biophysical Journal 68, 1740 (1995).
[33] P. Das and W. Schwarz, Physical Review E 51, 3588 (1995).
[34] W. Xin-Yi, Physical Review A 32, 3126 (1985).
[35] B. Lautrup, A. Jackson, and T. Heimburg, The European Physical Journal. E, Soft
Matter 34, 57 (2011).
20
http://arxiv.org/abs/1411.2454
Shrivastava, Kang and Schneider
[36] U. Eichmann, A. Ludu, and J. Draayer, Journal of Physics A: Mathematical and
General 35, 6075 (2002).
[37] L. D. Mosgaard, A. D. Jackson, and T. Heimburg, arXiv:1203.1248v1 1 (n.d.).
[38] R. Beyer, A. B. Coppens, M. B. Seiden, J. Donohue, F. Guepen, R. Hodson, and
C. Townsend, Journal of Acoustical Society of America 38, 797 (1965).
[39] R. A. Chapman, Nature 1143 (1967).
[40] C. Fillafer and M. F. Schneider, PloS One 8, e66773 (2013).
[41] J. J. Rosenthal and F. Bezanilla, The Biological Bulletin 199, 135 (2000).
[42] Y. Grossman and J. Kendig, Journal of Neurophysiology 52, 692 (1984).
[43] C. Spyropoulos, Am. J. Physiol 189, 214 (1957).
[44] A. Hodgkin and B. Katz, The Journal of Physiology 109, 240 (1949).
[45] R. FitzHugh, Biophysical Journal 1, 445 (1961).
[46] K. Cole, R. Guttman, and F. Bezanilla, Proceedings of the National Academy of
Sciences 65, 884 (1970).
[47] D. L. Chapman, “On the Rate of Explosion in Gases” Philosophical Magazine and
Journal of Science (Taylor and Francis, London, Edinburgh, Dublin, 1899), pp.
90–103.
[48] D. Steppich, J. Griesbauer, T. Frommelt, W. Appelt, A. Wixforth, and M.
Schneider, Physical Review E 81, 1 (2010).
21
|
1806.10740 | 1 | 1806 | 2018-06-28T02:23:59 | Fe3O4@astragalus polysaccharide core-shell nanoparticles for iron deficiency anemia therapy and magnetic resonance imaging in vivo | [
"physics.bio-ph",
"physics.med-ph"
] | Fe3O4@astragalus polysaccharide core-shell nanoparticles (Fe3O4@APS NPs) were demonstrated to be an efficient therapeutic drug for treating iron deficiency anemia (IDA) in vivo. The Fe3O4@APS NPs have been synthesized using a two steps approach involving hydrothermal synthesis and subsequent esterification. Transmission electron microscopy (TEM) and Fourier transform infrared (FTIR) spectroscopy studies show that APS are attached on the surfaces of the highly monodisperse Fe3O4 NPs. Dynamic light scatting (DLS) and magnetic characterizations reveal that the Fe3O4@APS NPs have outstanding water solubility and stability. Cytotoxicity assessment using Hela cells and pathological tests in mice demonstrate their good biocompatibility and low toxicity. The IDA treatment in rats shows that they have efficient therapeutic effect, which is contributed to both the iron element supplement from Fe3O4 and the APS-stimulated hematopoietic cell generation. Moreover, the Fe3O4@APS NPs are superparamagnetic and thus able to be used for magnetic resonance imaging (MRI). This study has demonstrated the potential of nanocomposites involving purified natural products from Chinese herb medicine for biomedical applications. | physics.bio-ph | physics | Fe3O4@astragalus polysaccharide core-shell nanoparticles for iron
deficiency anemia therapy and magnetic resonance imaging in vivo
Kai Wang, Lina Li, Xiaoguang Xu*, Liying Lu, Jian Wang, Shuyan Wang, Yining
Wang, Zhengyu Jin, Jin Zhong Zhang, Yong Jiang *
K. Wang, Prof. X. G. Xu, Prof. Y. Jiang. Beijing Advanced Innovation Center for
Materials Genome Engineering, School of Materials Science and Engineering,
University of Science and Technology Beijing, Beijing 100083, China
E-mail: [email protected] (X.G. Xu), [email protected] (Y. Jiang)
Prof. L. N. Li, S. Y. Wang. School of Chinese Medicine, Beijing University of Chinese
Medicine, Beijing 100029, China
Prof. L.Y. Lu. School of Chemistry and Biological Engineering, University of Science
and Technology Beijing, Beijing 100083, China
Dr. J. Wang, Prof. Y. N. Wang, Prof. Z. Y. Jin. Department of Radiology, Peking Union
Medical College Hospital, Chinese Academy of Medical Sciences, Beijing 100730,
Prof. J. Z. Zhang. Department of Chemistry & Biochemistry, University of California,
China
Santa Cruz, CA 95064, USA
1
ABSTRACT:
Fe3O4@astragalus
polysaccharide
core-shell
nanoparticles
(Fe3O4@APS NPs) were demonstrated to be an efficient therapeutic drug for treating
iron deficiency anemia (IDA) in vivo. The Fe3O4@APS NPs have been synthesized
using a two steps approach involving hydrothermal synthesis and subsequent
esterification. Transmission electron microscopy (TEM) and Fourier transform
infrared (FTIR) spectroscopy studies show that APS are attached on the surfaces of
the highly monodisperse Fe3O4 NPs. Dynamic light scatting (DLS) and magnetic
characterizations reveal that the Fe3O4@APS NPs have outstanding water solubility
and stability. Cytotoxicity assessment using Hela cells and pathological tests in mice
demonstrate their good biocompatibility and low toxicity. The IDA treatment in rats
shows that they have efficient therapeutic effect, which is contributed to both the iron
element supplement from Fe3O4 and the APS-stimulated hematopoietic cell
generation. Moreover, the Fe3O4@APS NPs are superparamagnetic and thus able to
be used for magnetic resonance imaging (MRI). This study has demonstrated the
potential of nanocomposites involving purified natural products from Chinese herb
medicine for biomedical applications.
2
Iron is one of the most important trace elements in human body and is essential for
the normal function of organisms, particularly for the metabolism and immune
functions. [1] As reported by World Health Organization, 2 billion people have anemia,
with nearly 1 billion suffering from iron deficiency anemia (IDA).[2] Therefore, IDA
is a common nutritional disease in modern society.[3] Oral iron supplementation, such
as ferrous salts, is common in the treatment and prevention of iron deficiency in
human body.[4] However, it usually causes side effects, such as epigastric pain,
diarrhea and constipation.[5] Therefore, it is highly desired to develop new iron
supplements with no or few side effects.[6]
Ferumoxide, known as Feridex in America and Endorem in Europe, is a colloid of
ultra-small superparamagnetic iron oxide (USPIO) nanoparticles (NPs) coated by
dextran.[7] It is a FDA-cleared drug for the treatment of IDA in adult chronic kidney
disease patients. [8] [9] After intravenous administration, the USPIO NPs are cleared
from blood by phagocytosis.[10] The USPIO NPs are metabolized in the lysosomes
into a soluble, nonmagnetic form of iron that becomes part of the normal iron pool
after the intracellular uptake. Among the USPIO NPs, Fe3O4 NPs have been proved to
exhibit good superparamagnetic property below 15 nm and can be used as both an
iron supplement and a magnetic resonance imaging (MRI) contrast agent candidate.
[11]
MRI is a powerful noninvasive diagnostic technique for visualizing the fine
structure of the human body with high spatial resolution. To obtain accurate diagnosis,
contrast agents are usually used to increase the MRI image quality by exerting an
3
influence on the longitudinal (T1), transverse (T2) relaxation time and T2-star (T2*), a
relaxation parameter arising principally from local magnetic field in homogeneities
that are increased with iron deposition.[12] Due to the unique superparamagnetic
property, Fe3O4 NPs have been widely used as ultrasensitive negative contrast agents
for early detection of tumors due to their strong T2 and T2* shortening effect.[13]
Unfortunately, the exposed Fe3O4 NPs have poor solubility and biocompatibility.
Therefore, the surface of Fe3O4 NPs needs to be suitably engineered to acquire
improved biocompatibility.[14, 15] The most popular methods are introducing
hydrophilic groups and modification of bioinorganic shell on the surface of Fe3O4
NPs, which can effectively optimize the properties of Fe3O4 NPs to satisfy the
requirements of applications, such as MRI contrast agents, [16] magnetic targeting drug,
biomolecule separation, and hyperthermal cancer
therapy.[17]
For example,
Fe3O4-based nanocomposites, such as Fe3O4@Polydopamine,[18] WS2@Fe3O4,[19]
Fe3O4@Au[20] and C-Fe3O4 quantum dots (QDs)[21], have been studied as MRI
contrast agents and drug carriers in preclinical and clinical setting.[22]
Among the modification materials of Fe3O4 NPs, polysaccharides have the
advantages of stability, water solubility and few side effects on the organism.
Meanwhile, some natural polysaccharides can promote
the
formation of
hematopoietic cell.[23] Astragalus membranaceus, a traditional Chinese herb medicine
containing many active components, such as polysaccharides, flavonoids, saponins,
amino acids and trace elements, has been used in China for medicine for more than
2,000 years to improve human immunity and treat cardiovascular disorders.
4
Astragalus polysaccharide (APS) is one of the main active components of Astragalus
membranaceus extracted from Astragalus roots, which is described in the 2005
version of the ''Chinese Pharmacopoeia".[24] APS is comprised of uniform
polysaccharide fraction, including rhamnose, arabinose and glucose.[25] APS is stable
and biocompatible in vivo, which makes it an ideal material for biomedical
applications. Many reports on the biological activities of APS indicate that it possesses
potent anti-inflammatory, organ protective, anti-tumor activities and growth of red
blood cell stimulative.[26] Therefore, APS is a promising material for modifying Fe3O4
NPs.
To develop a potential IDA therapeutic agent combining iron supplements and
contents of promoting the hematopoietic cell formation, we report a novel material of
Fe3O4@APS Core-Shell NPs fabricated by a two steps approach involving
hydrothermal synthesis[27] and esterification. As shown in Figure 1a，the water
soluble Fe3O4 NP cores were obtained in sodium citrate solution, in which sodium
citrate replaces the oleic acid molecular on the surface of the Fe3O4 NPs. Then, The
Fe3O4 NPs were bonded with the APS backbone, and the sodium citrate reacted with
the hydroxyl groups of the APS chains to form the Fe3O4@APS NPs. Finally, the
Fe3O4@APS NPs were used in mice intragastric administration for IDA treatment and
MRI
testing
followed by pathological
tests, after
the structural stability
characterization and cytotoxicity assessment, as depicted in Scheme 1. As expected,
the Fe3O4@APS NPs show unique therapeutic effect for IDA treatment, which is
much better than that of water soluble Fe3O4 NPs. This is the result of the combined
5
contribution of the hematopoietic cell generation stimulated by APS and the
simultaneous iron element supplement from Fe3O4. Besides this, the Fe3O4@APS NPs
are also potential MRI contrast agents in vivo, which are safe and good for health. The
study opens a new way to design IDA medicine with USPIO NPs and natural
polysaccharides, which can serve as a contrast agent at the same time.
The crystal structural of the Fe3O4 NPs and Fe3O4@APS NPs were investigated by
X-ray diffraction (XRD) (Figure S2, Supporting Information). The XRD patterns of
both NP samples show sharp diffraction peaks of Fe3O4 (JCPDS75-0033)[28] with
Fd-3m space group, indicating good crystallinity. Figure 1b shows the transmission
electron microscope (TEM) image of the oil-soluble Fe3O4 NPs, which are spherical
with a uniform diameter of ~10 nm. High-resolution TEM images of the Fe3O4 NPs
exhibit (220) and (331) crystal facets as indexed in Figure 1c. The hydrodynamic
diameter of the water soluble Fe3O4 NPs was measured to be 11 nm by dynamic light
scattering (DLS) (Figure 1e), which is slightly larger than that measured by TEM
(Figure 1d).
For the Fe3O4@APS NPs, the diameter does not show obvious increase in TEM
image (Figure 1f), revealing clear cores of Fe3O4 NPs but insignificant evidence of the
APS shells. This is because APS molecules are composed by C, H, O and N elements,
which are invisible in TEM. However, the DLS results show a mean size of 29.5 nm
for the Fe3O4@APS NPs (Figure 1g), which demonstrates the existence of the APS
shells coated on the surface of Fe3O4 NPs. The dispersion stability of the oleic
acid-coated Fe3O4 NPs shows good water stability and biocompatibility after the
6
ligand exchange reaction on the surface from oleic acid to sodium citrate and APS.
The zeta potential of the water soluble Fe3O4 NPs and Fe3O4@APS NPs are -36.8 mV
and -28.8 mV, respectively, which suggests good water dispersity of the NPs (Figure
S3, Supporting Information).[29]
Fourier transform infrared (FT-IR) absorption spectra were measured to confirm the
presence of the APS around the Fe3O4 NPs (Figure 1h). Compared to the FT-IR
spectrum of Fe3O4 around with oleic acid, the one of Fe3O4@APS NPs shows new
strong absorption bands at 1383 cm-1 and 1633 cm-1, which can be attributed to the
C-O-C and C=O stretch of APS and sodium citrate vibration of polysaccharide,
respectively. It is evident that the oleic acid on the surface of Fe3O4 NPs was
efficiently replaced by APS and sodium citrate. Moreover, the Fe3O4@APS NPs show
a strong absorption band at 586 cm-1, attributed to Fe-O bond, while the APS
copolymer has the stretching bands at 2921 cm-1 and 3424 cm-1, corresponding to
hydroxyl groups, which are a further evidence of the ligand exchange on the surface
of Fe3O4 NPs.
To study the magnetic properties of Fe3O4@APS NPs, the magnetization curves
have been measured, as shown in Figure 1i. All the curves exhibit no hysteresis,
suggesting that the Fe3O4 NPs are superparamagnetic, with or without APS coating.
The Fe3O4 NPs have a higher saturation magnetization (Ms) (about 62.2 emu/g) than
that of the Fe3O4@APS NPs (about 36.7 emu/g), which further demonstrates the
existence of the APS shells on the surface of Fe3O4 NPs. From the change of Ms, the
mass percentage of the Fe3O4 NPs can be estimated to be around 60% in the
7
Fe3O4@APS NPs. With a magnet placed outside the cuvette, the Fe3O4@APS NPs in
the water solution can rapidly accumulate near the magnet (See the inset of Figure 1i).
When the magnet is removed, the Fe3O4@APS NPs disperse in water again, even
after 6 months, suggesting the Fe3O4@APS NPs are superparamagnetic with good
stability. To further investigate the magnetic behavior of the Fe3O4@APS NPs, the
temperature dependence of their magnetization was investigated by zero-field cooling
(ZFC) and field cooling (FC) respectively (Figure S4, Supporting Information). The
MZFC-T curves show broad peaks with a clear maximum (Tmax) at ~226 K for the
Fe3O4 NPs and ~167 K for the Fe3O4@APS NPs, above which temperature the
particles are in superparamagnetic regime. Thus, both kinds of the NPs are
superparamagnetic at 300 K.[14]
Since the Fe3O4@APS NPs should be fed to the Institute of Cancer Research (ICR)
mice and Wistar rats, it is possible that the Fe3O4@APS NPs might be degraded
during digestion in their stomachs. Thus, we have checked the stability of the
Fe3O4@APS NPs in imitated gastric acid by stirring in HCl solution with pH = 1.5 for
5 h. The magnetization curve of the Fe3O4@APS NPs collected after the imitating
digestion still shows good superparamagnetic property (as shown in Figure 1i). The
Ms of the Fe3O4@APS NPs increases to 45.6 emu/g, which is larger than that of the
as-prepared Fe3O4@APS NPs. This suggests that a certain amount of APS molecules
could desorb from the NPs surfaces in the stomach due to the acid environment.
However, comparing with the Ms of the Fe3O4 NPs, we confirmed that there is still
sufficient amount of APS molecules attached to the Fe3O4 NPs cores. Moreover, the
8
FT-IR spectrum of the Fe3O4@APS NPs (Figure 1h) shows no significant change
relative to that of the as-prepared Fe3O4@APS NPs, consistent with the results of the
magnetic studies. The iron concentration in the Fe3O4@APS NPs was also
quantitatively determined by
inductively coupled plasma optical emission
spectroscopy (ICP-OES). The weight percentage of Fe increases from 60% to 75%
after stirring in HCl solution, which is consistent with the change of the magnetization
curves. Accordingly, only a small amount of APS molecules were degraded by HCl
solution. Therefore, the Fe3O4@APS NPs are stable enough and can be absorbed into
blood as whole NPs, instead of the separate Fe3O4 NPs and APS.
The cytotoxicity evaluation is necessary for a new biomaterial before in vivo
applications. We assessed the biocompatibility of the Fe3O4@APS NPs with
mammalian cells by a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide
(MTT) assay using a HeLa cell line. As shown in Figure 2a, the cell viabilities are
higher than 75% after 24 h, 48 h and 72 h exposure to the iron concentration of 0-50
μg/mL. There is no decrease in cell viability when the exposure time was prolonged to
72 h. Therefore, the Fe3O4@APS NPs have a high biocompatibility and low
cytotoxicity.
The MRI performance of the Fe3O4@APS NPs as contrast agents has been studied
both in vitro and in vivo. For comparison, the Fe3O4 NPs have also been investigated.
Figure 2b shows the T2* MRI images of the Fe3O4 NPs and Fe3O4@APS NPs
suspensions at different Fe concentrations. The low signal intensities in the T2* MRI
images increase with Fe concentration for both kinds of nanoparticles. Moreover, the
9
intensities are similar when the Fe concentration in the Fe3O4 NPs and Fe3O4@APS
NPs suspensions are approximately equal. As a negative MRI contrast agent, the
superparamagnetic Fe3O4 core can shorten the T2* values of water, resulting in a
hyperintense signal on T2* imaging. In addition, the T2* relaxation rate (r2* = 1/T2*)
of the Fe3O4 NPs and Fe3O4@APS NPs was calculated to be 483 mM-1s-1 (Figure 2c)
and 400 mM-1s-1 (Figure 2d), respectively. The observed decrease in the r2* value can
be attributed to the APS on the surface of Fe3O4 NPs, which weakens the magnetic
properties of the NPs. However, the Fe3O4@APS NPs remain a good contrast agent
for MRI applications.
The preliminary in vitro MRI signal contrast enhancement of the Fe3O4@APS NPs
led us to evaluate its possibility to serve as a negative MRI contrast agent for MRI in
vivo imaging. Fe3O4@APS NPs were gastrically infused into the stomach of live
healthy male ICR mice at a dosage of 10 mg/kg. Figure 2e shows the T1 and T2
weighted MRI images of the reference group (normal mice) and the mice after gastric
infusion for 15 min. Comparing to the reference group, the T1 and T2 images for the
mice after the gastric infusion show a positive and a negative contrast enhancement in
the stomach and the bowels, respectively. Therefore, the Fe3O4@APS NPs have a
significant signal enhancement effect as a MRI contract agent. The MRI images of the
main organs have no specific changes even after being perfused for 4 h, 8 h and 16 h
(Figure S5, Supporting Information), indicating the safety during Fe3O4@APS NPs
metabolism.
In vivo toxicity and possible side effects of nanomedicines have to be carefully
10
studied before practical applications in clinic. In order to further demonstrate the
safety of the Fe3O4@APS NPs, the stomachs of healthy male ICR mice were perfused
with the Fe3O4@APS NPs at a dosage of 1.0 mg/kg/d and 2.0 mg/kg/d, respectively.
On the 15th and 30th day, the mice were tested by MRI to examine the effect of the
Fe3O4@APS NPs on the organs such as liver and kidney. Figure 3a and Figure S6a
(Supporting Information) show the T2-weighted images of the livers and kidneys of
the mouse treated by 1.0 mg/kg/d and 2.0 mg/kg/d of Fe3O4@APS NPs on the 30th
and 15th day, together with that of the normal mouse as contrast group. Comparing
with the contrast group, there is no obvious damage in the groups treated by
Fe3O4@APS NPs. The T2* values shown in Figure 3b and Figure S6b (Supporting
Information) also reveal no particular change in T2* values of the kidney and the liver
in the range of the organ regions with different dosage and time. These results
demonstrate that no pathological change happened in the organs of the ICR mice after
a long time supplementation of the Fe3O4@APS NPs.
The histopathological studies were carried out on the main organs of the ICR mice
after the in vivo MRI testing. The histopathological images of the livers, kidneys and
brains are presented in Figure 3c and Figure S7 (Supporting Information), which
correspond to the ICR mice after the intragastric administration of Fe3O4@APS NPs
for 30 and 15 days, respectively. As expected, no particles were observed in the
organs, and no sign of organ damage or inflammation was observed, demonstrating
the minimal side effects of the Fe3O4@APS NPs in vivo after 30 days intragastric
administration.
11
To study the effect of the Fe3O4@APS NPs on IDA, low-iron diet Wistar rats were
separated into different groups and fed with the designed ferralia dosage. The rats
were weighted every week. As shown in Figure 3d, the mean body weights of all
groups increase almost linearly with time. As expected, the model group shows the
smallest slope of the weight growth due to serious IDA. The group treated with the
Fe3O4 NPs has a weight growth rate higher than the model group but lower than the
other treated and normal groups without IDA. However, the groups treated with the
Fe3O4@APS NPs have weight growth rates comparable to that of the normal group,
and a higher Fe3O4@APS NPs dosage results in a larger weight growth rate.
Therefore, the Fe3O4@APS NPs treatment can improve the health of the low-iron diet
rats.
The Hemoglobin (HGB), Hemoglobin (RBC) and Hematocrit (HCT) are important
parameters for the diagnosis and therapy of IDA. Therefore, the blood analyses were
carried out for the rats studied, and the results are presented in Figure 3e-3g. Usually,
the rats are considered as anemic when the HGB level is reduced to 75% of the
original level. As shown in Figure 3e, the HGB values of the low-iron diet groups are
only half of that of the normal group before ferralia treatment, which indicates a
serious IDA of the low-iron diet rats. After the ferralia treatment, all the treated groups
have significant increase of HGB values, while only the group supplied with the
Fe3O4@APS NPs at a dosage of 2.0 mg/kg/d has a HGB value (136 g/L) comparable
to the normal group (143.1 g/L). The group treated with the 1.0 mg/kg/d Fe3O4@APS
NPs (117.1 g/L) has a curative effect similar to the group of the 2.0 mg/kg/d Fe3O4
12
NPs (117.2 g/L), demonstrating that the Fe3O4@APS NPs have more positive effect
on HGB than the Fe3O4 NPs. The other two important parameters, RBC and HCT,
have the trends similar to that of HGB. All blood analyses suggest that the health
condition of the group supplied by the 2.0 mg/kg/d Fe3O4@APS NPs recovered to the
level of the normal group. Moreover, the superoxide dismutase (SOD), glutathione
(GSH), lactic dehydrogenase (LDH), hepcidin (HEP), malondialdehyde (MDA) and
serum iron (SI) data were also measured after 30 days of the treatment. All results
(Table S1, Supporting Information) demonstrate the high therapeutic effect of the
Fe3O4@APS NPs on IDA in vivo. Considering the structure of the Fe3O4@APS NPs,
their high therapeutic effect is attributed to both iron element supplement from Fe3O4
and the APS stimulated hematopoietic cell generation. As a result, the Fe3O4@APS
NPs are highly promising as an IDA drug with the additional benefit of serving as a
MRI contrast agent.
In conclusion, we have designed and experimentally demonstrated the Fe3O4@APS
NPs for the targeting IDA treatment and the MRI contract agent. The stability of the
Fe3O4@APS NPs was determined by the imitating digestion in imitated gastric acid.
Both in vitro and in vivo toxicity studies were carried out using the cell experiments
and intragastric administration of animal model. The blood analyses show that the
Fe3O4@APS NPs have potent therapeutic effect on IDA, evidenced by the HGB, RBC
and HCT values of the 2.0 mg/kg/d Fe3O4@APS NPs treated IDA rats. Moreover,
strong MRI contrast enhancement has been observed for the Fe3O4@APS NPs.
Therefore, the Fe3O4@APS NPs are promising candidates for IDA drugs with the
13
additional functionality as MRI contract agent.
Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.
Acknowledgements
K. Wang and L. N. Li contributed equally to this work. This work was partially
supported by
the National Basic Research Program of China (Grant No.
2015CB921502), the National Science Foundation of China (Grant Nos. 51671019,
51471029, 51731003, 61471036) and Beijing science and technology Nova cross
program (Z171100001117136).
Conflict of Interest
The authors declare no conflict of interest.
Keywords
Fe3O4 nanoparticles, astragalus polysaccharides, iron deficiency anemia, magnetic
resonance imaging.
14
References
[1] S. Denic, M. M. Agarwal, Nutrition 2007, 23, 603.
[2] T. Vos, C. Allen, M. Arora, R. M. Barber, Z. A. Bhutta, A. Brown, ... & C. J. L.
Murray, Lancet 2016, 388, 1545.
[3] A. Lopez, P. Cacoub, I. C. Macdougall, L. Peyrin-Biroulet, Lancet 2016, 387, 907.
[4] C. Camaschella, N. Engl. J. Med. 2015, 372, 1832.
[5] G. D. Lewis, R. Malhotra, A. F. Hernandez, S. E. McNulty, A. Smith, G. M. Felker,
W. H. W. Tang, S. J. LaRue, M. M. Redfield, M. J. Semigran, M. M. Givertz, P.
Van Buren, D. Whellan, K. J. Anstrom, M. R. Shah, P. Desvigne-Nickens, J.
Butler, E. Braunwald, JAMA 2017, 317, 1958.
[6] S. D. Anker, G. Filippatos, R. Willenheimer, K. Dickstein, H. Drexler, ... & P.
Ponikowski, N. Engl. J. Med. 2009, 361, 2436.
[7] P. R. Ros, P. C. Freeny, S. E. Harms, S. E. Seltzer, P. L. Davis, T. W. Chan, A. E.
Stillman, L. R. Muroff, V. M. Runge, M. A. Nissenbaum, Radiology 1995, 196,
481.
[8] D. W. Coyne, Expert Opin. Pharmacother. 2009, 10, 2563.
[9] J. T. Ferrucci, D. D. Stark, AJR, Am. J. Roentgenol. 1990, 155, 943.
[10] A. Khurana, H. Nejadnik, F. Chapelin, O. Lenkov, R. Gawande, S. Lee, S. N.
Gupta, N. Aflakian, N. Derugin, S. Messing, G. Lin, T. F. Lue, L. Pisani, H. E.
Daldrup-Link, Nanomedicine 2013, 8, 1969.
[11] a) S. Zanganeh, G. Hutter, R. Spitler, O. Lenkov, M. Mahmoudi, A. Shaw, J. S.
Pajarinen, H. Nejadnik, S. Goodman, M. Moseley, L. M. Coussens, H. E.
15
Daldrup-Link, Nat. Nanotechnol. 2016, 11, 986; b) Z. Gao, Y. Hou, J. Zeng, L.
Chen, C. Liu, W. Yang, M. Gao, Adv. Mater. 2017, 29, 1701095.
[12] a) Y. X. J. Wang, S. M. Hussain, G. P. Krestin, Eur. Radiol. 2001, 11, 2319; b) Y.
Wang, K. Zhou, G. Huang, C. Hensley, X. Huang, X. Ma, T. Zhao, B. D. Sumer, R.
J. DeBerardinis, J. Gao, Nat. Mater. 2014, 13, 204.
[13] R. Hao, R. Xing, Z. Xu, Y. Hou, S. Gao, S. Sun, Adv. Mater. 2010, 22, 2729.
[14]C. Scialabba, R. Puleio, D. Peddis, G. Varvaro, P. Calandra, G. Cassata, L. Cicero,
M. Licciardi, G. Giammona, Nano Res. 2017, 10, 3212.
[15] J. Zeng, L. Jing, Y. Hou, M. Jiao, R. Qiao, Q. Jia, C. Liu, F. Fang, H. Lei, M. Gao,
Adv. Mater. 2014, 26, 2694.
[16] J. Yu, C. Yang, J. Li, Y. Ding, L. Zhang, M. Z. Yousaf, J. Lin, R. Pang, L. Wei, L.
Xu, F. Sheng, C. Li, G. Li, L. Zhao, Y. Hou, Adv. Mater. 2014, 26, 4114.
[17] S. Tong, C. A. Quinto, L. Zhang, P. Mohindra, G. Bao, ACS nano 2017, 11, 6808.
[18] L. S. Lin, Z. X. Cong, J. B. Cao, K. M. Ke, Q. L. Peng, J. Gao, H. H. Yang, G.
Liu, X. Chen, ACS nano 2014, 8, 3876.
[19] G. Yang, H. Gong, T. Liu, X. Sun, L. Cheng, Z. Liu, Biomaterials 2015, 60, 62.
[20] a) C. Li, T. Chen, I. Ocsoy, G. Zhu, E. Yasun, M. You, C. Wu, J. Zheng, E. Song,
C. Z. Huang, W. Tan, Adv. Funct. Mater. 2014, 24, 1772; b) W. Dong, Y. Li, D.
Niu, Z. Ma, J. Gu, Y. Chen, W. Zhao, X. Liu, C. Liu, J. Shi, Adv. Mater. 2011, 23,
5392; c) L. S. Lin, X. Yang, Z. Zhou, Z. Yang, O. Jacobson, Y. Liu, A. Yang, G.
Niu, J. Song, H. H. Yang, X. Chen, Adv. Mater. 2017, 29, 1606681.
16
[21] X. Liu, H. Jiang, J. Ye, C. Zhao, S. Gao, C. Wu, C. Li, J. Li, X. Wang, Adv. Funct.
Mater. 2016, 26, 8694.
[22] a) Z. Wang, R. Qiao, N. Tang, Z. Lu, H. Wang, Z. Zhang, X. Xue, Z. Huang, S.
Zhang, G. Zhang, Y. Li, Biomaterials 2017, 127, 25. b) X. Wang, D. Niu, P. Li, Q.
Wu, X. Bo, B. Liu, S. Bao, T. Su, H. Xu, Q. Wang, ACS nano 2015, 9, 5646.
[23] a) Z. Liu, Y. Jiao, Y. Wang, C. Zhou, Z. Zhang, Adv. Drug Deliv Rev. 2008, 60,
1650; b) M. Swierczewska, H. S. Han, K. Kim, J. H. Park, S. Lee, Adv. Drug
Delivery Rev. 2016, 99, 70.
[24] a) B. M. Shao, W. Xu, H. Dai, P. Tu, Z. Li, X. M. Gao, Biochem. Biophys. Res.
Commun. 2004, 320, 1103; b) X. Q. Ma, Q. Shi, J. A. Duan, T. T. X. Dong, K. W.
K. Tsim, J. Agric. Food Chem. 2002, 50, 4861.
[25] J. H. Xie, M. L. Jin, G. A. Morris, X. Q. Zha, H. Q. Chen, Y. Yi, J. E. Li, Z. J.
Wang, J. Gao, S. P. Nie, P. Shang, M. Y. Xie, Crit. Rev. Food Sci. Nutr. 2016, 56,
S60.
[26] Q. Lu, L. Xu, Y. Meng, Y. Liu, P. Li, Y. Zu, M. Zhu, Int. J. Biol. Macromol. 2016,
93, 208.
[27] S. Si, C. Li, X. Wang, D. Yu, Q. Peng, Y. Li, Cryst. Growth Des. 2005, 5, 391.
[28] M. E. Fleet, Acta Crystallogr., Sect. C: Cryst. Struct. Commun. 1984, 40, 1491.
[29] S. Bhattacharjee, J. Controlled Release 2016, 235, 337.
17
Figure Captions
Scheme 1. Schematic illustration of the design and application of Fe3O4@APS NPs
for IDA therapy and MRI.
Figure 1. a) Schematic drawing the synthesis of the Fe3O4@APS NPs. b) TEM and c)
high-resolution TEM images of the oil soluble Fe3O4 NPs. d) TEM image of
water soluble Fe3O4 NPs and e) their hydrodynamic profiles. f) TEM image
of Fe3O4@APS NPs and g) their hydrodynamic profiles. h) FT-IR spectra
and i) magnetization curves of the Fe3O4 NPs, Fe3O4@APS NPS and
Fe3O4@APS NPs after stirring in HCl solution. Inset of i) is the image of
the Fe3O4@APS NPs in water dispersion with (right) and without (left)
external field.
Figure 2. a) Cell viability of Hela cells after treatment with the Fe3O4@APS NPs at
different iron concentrations after 24h, 48h and 72h. b) T2* images of the
Fe3O4 NPs and Fe3O4@APS NPs water solution with different Fe
concentrations. Corresponding T2* relaxation rate of the Fe3O4 NPS c) and
the Fe3O4@APS NPs d) with respect to iron concentration. e) T1-weighted
and T2-weighted MRI images of contrast group (normal mouse) and the
Fe3O4@APS NPs gastric infused Male ICR mouse after 15 min.
Figure 3. a) In vivo T2-weighted images of the male ICR mice liver and kidney for
18
contract group and the group treated with the Fe3O4@APS NPs at different
Fe concentrations after 30 days. b) Signal intensity of T2* values of liver
and kidney. c) H&E stained organ slices of the mice after 30 days
Fe3O4@APS NPs treatment. Scale bars for all images are 100 µm. d) Body
Weight increase with time, e) HGB concentration, f) RBC concentration and
g) HCT values of male Wistar rats (8 per group) before and after 30 days
treatment with the Fe3O4@APS NPs intragastric administration.
19
Scheme 1. (Wang K. et al.)
20
Figure 1. (Wang K. et al.)
21
Figure 2. (Wang K. et al.)
22
Figure 3. (Wang K. et al.)
23
|
1806.02995 | 1 | 1806 | 2018-06-08T07:25:48 | Spatial Awareness of a Bacterial Swarm | [
"physics.bio-ph"
] | Bacteria are perhaps the simplest living systems capable of complex behaviour involving sensing and coherent, collective behaviour an example of which is the phenomena of swarming on agar surfaces. Two fundamental questions in bacterial swarming is how the information gathered by individual members of the swarm is shared across the swarm leading to coordinated swarm behaviour and what specific advantages does membership of the swarm provide its members in learning about their environment. In this article, we show a remarkable example of the collective advantage of a bacterial swarm which enables it to sense inert obstacles along its path. Agent based computational model of swarming revealed that independent individual behaviour in response to a two-component signalling mechanism could produce such behaviour. This is striking because independent individual behaviour without any explicit communication between agents was found to be sufficient for the swarm to effectively compute the gradient of signalling molecule concentration across the swarm and respond to it. | physics.bio-ph | physics | Spatial Awareness of a Bacterial Swarm
Harshitha S. Kotian[1][2], Shalini Harkar[2], Shubham Joge[3], Ayushi Mishra[3], Amith Zafal[1], Varsha
Singh[3], Manoj M. Varma[1][2]
[1]Centre for Nano Science and Engineering, Indian Institute of Science, Bangalore, India
[2] Robert Bosch Centre for Cyber Physical Systems, Indian Institute of Science, Bangalore, India
[3]Molecular Reproduction, Development and Genetics, Indian Institute of Science, Bangalore, India
Abstract
Bacteria are perhaps the simplest living systems capable of complex behaviour involving sensing and
coherent, collective behaviour an example of which is the phenomena of swarming on agar surfaces.
Two fundamental questions in bacterial swarming is how the information gathered by individual
members of the swarm is shared across the swarm leading to coordinated swarm behaviour and
what specific advantages does membership of the swarm provide its members in learning about
their environment. In this article, we show a remarkable example of the collective advantage of a
bacterial swarm which enables it to sense inert obstacles along its path. Agent based computational
model of swarming revealed that independent individual behaviour in response to a two-component
signalling mechanism could produce such behaviour. This is striking because independent individual
behaviour without any explicit communication between agents was found to be sufficient for the
swarm to effectively compute the gradient of signalling molecule concentration across the swarm
and respond to it.
Introduction
Bacteria are perhaps the simplest living systems capable of complex behaviour involving sensing and
coherent, collective behaviour an example of which is the phenomena of swarming, where bacteria
colonize a solid surface (typically, nutrient loaded agar) in geometric patterns characteristic of
different species (Ben-Jacob, 1997; Kearns, 2010). A natural and fascinating question in this regard is
how the information gathered by individual members of the swarm through sensing their respective
local environments is shared across the swarm leading to coordinated swarm behaviour. Specifically,
does membership of the swarm provide its members any advantage in learning more about their
environment? For instance, are they able to extract more information about their surroundings
collectively than acting alone? Swarming involves several, possibly collective, decision making steps
such as quorum sensing (Daniels et al., 2004). The swarming patterns produced by the Gram-
negative bacteria Pseudomonas aeruginosa (PA) are special due to the presence of long straight
segments (tendrils) in its swarming pattern [Figure 1 a)]. The swarming pattern produced by PA is
unique in two major aspects. Firstly, the typical tendril length (~ cm) is nearly 4 orders of magnitude
compared to the micron-sized individual swarm member. Considering the large stochastic
fluctuations likely to be experienced by a micron sized individual member, it is remarkable that
collective motion of these individuals produces such long straight segments. The long directional
persistence leading to this highly anisotropic collective motion is worthy of detailed mathematical
modelling. Indeed, there have been several attempts to model the swarming pattern of PA based on
a highly mechanistic multiscale model(Du et al., 2011), as a population dispersal phenomenon using
spatial kernels(Deng et al., 2014) and based on Marangoni forces(Du et al., 2012). However, so far
none of these have succeeded in producing patterns with similar branching statistics similar to the
one seen in Figure 1 a) [SI Section1 Figure S2]. While generic models such as the Vicsek model
[(Vicsek et al., 1995)] can explain large directional persistence collections of stochastically moving
particles, one would like to have a more detailed model involving experimentally accessible
parameters of the system which produces statistically identical patterns as seen in the experimental
system. While this is an open problem, in this paper we focus on a different shortcoming of the
existing models which is that these models have primarily been employed to study the pattern
formation and do not describe other aspects of swarming such as the question we posed earlier
concerning sensing of the environment by swarms.
The tendrils produced by swarming PA are typically of the order of a centimetre with a width of 4-5
mm and their motion during swarming can be easily tracked with time-lapse imaging using regular
digital cameras. This enables us to study the motion of the tendrils and their response to various
environmental perturbations with relative ease compared to isotropically swarming bacteria (Kearns
& Losick, 2003) and other bacterial species which form dense fractal-like patterns (Ben-Jacob, 1997).
The motion of single bacterial cells on the agar surface can also be observed using GFP expressing
bacteria leading to the generation of motility data spanning multiple spatial (microns to cm) and
temporal (seconds to days) scales. Such multi-scale imaging data helps to link individual behaviour to
the collective behaviour of the swarm. The observation and analysis of the response of the tendrils
to perturbations introduced into the swarming medium (agar) reveals a remarkable ability of PA
swarm to sense its spatial environment and respond to the presence of co-swarming sister tendrils
(other tendrils from the same swarm initiating colony) [Figure 1 b)], approaching boundary of the
petri-plate and even inert obstacles (Poly DiMethyl Siloxane (PDMS) and glass objects), several
millimetres away along the swarming path. The ability to sense inert objects at millimetre scale
distances is a particularly striking example of the spatial awareness of the PA swarm considering that
the swarm is able to sense objects at distances three orders of magnitude (mm scale) further than
body length of the individual members (micron scale). This fact is indicative of the collective
advantage provided by the swarm as it is inconceivable that individual bacterium can sense inert
objects at such long distances. We emphasize that the swarm is spatially aware of the object, i.e. the
object does not actively secrete any molecules which are sensed by the bacterium.
The observation of spatial awareness by the swarm leads to the question of causative mechanisms.
Using a continuous, fluid dynamic model involving secretion of a signalling molecule by the bacteria
comprising the swarm, we show that a concentration gradient of the signalling molecule emerges
within the swarm. This leads to two possible scenarios, namely, one in which the bacteria in different
locations within the swarm behave differently leading to the collective response, or the other where
the global gradient is implicitly computed by the swarm and to which it responds to. The latter
scenario requires possibly complex information exchange within the swarm. To explore this further,
we developed a multi-agent model, based on attractive-repulsive interaction arising purely from local
information of the concentration of signalling molecules, which was able to replicate the spontaneous
retraction (reversal of direction) of the swarm from an inert boundary. While this model is not yet a
comprehensive representation of the swarming phenomenon, the ability to reproduce retraction
suggests that the remarkable examples of spatial awareness seen in the PA swarm may not require
active exchange of information between agents.
Results
General aspects of swarming pattern
The PA swarming system is interesting due to the highly specialized branching pattern [Figure 1 a)]
and the remarkable spatial awareness and response of the swarm to its environment. A short survey
of swarming patterns by Paenibacillus dendritiformis, Paenibacillus vortex, Escherichia coli, Proteus
mirabilis , Rhizobium etli, Serratia marcescens and Salmonella typhimurium (Verstraeten et al.,
2008)[SI Section 1 Figure S1] species of bacteria suggests that the PA swarming pattern occupies a
unique parameter space relative to the other swarm patterns. Many bacterial species such as E. coli
and P. mirabilis produce dense patterns which expand isotropically from the point of inoculation.
Bacillus subtilis (Fujikawa and Matsushita 1989) and Paenibacillus dendritiformis under some
conditions produce fractal like patterns which can be described by DLA models (Ben-Jacob, 1997). In
contrast, PA swarms expand in straight segments (tendrils) from the inoculated region. The expansion
of each tendril is highly anisotropic along a straight line with constant speed [SI Section 1 Figure S3].
The overall pattern is characterized by a robust statistical distribution of branch lengths and
divergence angles [SI Section 1 Figure S2].
Awareness of the Presence of Sister Tendrils
The swarming tendrils display self-avoidance (Caiazza et al. 2005)(Tremblay et al., 2007). This
behaviour is also seen in growing Bacillus subtilis tendrils (James et al., 2009) and in P..
dendritiformis by Avraham Be'er et al (Be'er et al., 2009) who used the term sibling rivalry to
describe this behaviour. In this case swarm fronts emerging from two locations in the same swarm
plate lead to the formation of a zone of inhibition unpopulated by either of the advancing swarm
fronts which can be viewed as a form of self-avoidance. However, there is a crucial difference
between the self-avoidance seen in the PA system compared with the Paenibacillus system. As
(Be'er et al., 2010) showed the zone of inhibition is due to mutually lethal secretions produced by
the swarming P. dendritiformis resulting in the death of bacteria in the region between the
advancing fronts producing the zone of inhibition. In the case of PA, the advancing sister tendrils
sense each other and typically one of them retracts back (complete reversal) or changes its
direction. [SI Section 2] This is a form of true self-avoidance and not a consequence of lethal
secretions killing off the bacteria in the advancing front (James et al., 2009). In other words, the
bacteria on one of the advancing tendrils sense the presence of another tendril advancing towards it
and initiate changes in the motility profile which result in a change in the swarming direction. Thus,
the self-avoidance phenomena in PA is significantly more complex and requires coordination far
more in extent to that described previously in the case of P. dendritiformis and requires deeper
study. We do not yet have a quantitatively accurate dynamical model of swarm direction reversal
associated with the sensing of sister tendrils. However, the continuous model, based on sensing of
the concentration gradient of a signalling molecule, presented later in the article is able to explain
certain features such as the expected change in direction and why generally only one tendril
retracts.
Another behaviour related to intra-species sensing is the avoidance of non-swarming mutant strains
by advancing tendrils of swarming strains as shown in Figure 1 c) [SI Section 2 Figure S5 and
video].The flgM mutant of PA is nonswarming due to the absence of flagella, yet it induces
avoidance response in wild type PA. Thus this mutant likely indicates its presence, through
secretions, sensed by the wildtype swarm which subsequently changes the direction of the
advancing tendril.
Figure 1., a) Representative labelled picture of a PA swarm, b) Avoidance behaviour among multiple colonies of
wild type on the same petri-plate, c) Avoidance exhibited by the swarm colony of the wild type in the presence
of flgM mutant, d) Avoidance exhibited by the swarm colony in the presence of a passive obstacle made of
PDMS.
Awareness of the Presence of Petri-plate Boundary
As the tendrils approach the edge of the petri-plate, we often observe a complete reversal of
swarming direction. Closer observation reveals that this effect happens most likely because one of
the tendrils reaches the edge of the petri-plate and releases the swarming bacteria which move
along the gap between the agar and the petri-plate wall. The signalling molecule(s) from these set of
bacteria are sensed by the other tendrils which have not yet reached the edge and causes direction
reversal in response to this sensory information. There is further support for this hypothesis as the
advancing tendrils reverse their directions only after at least one of the tendrils has reached the
edge of the petri-plate [SI Section 3 for video].
Awareness of the Presence of Inert Objects
The examples of spatial awareness presented above are all related by the fact that the target for
spatial awareness was of biological origin be it the tendrils of its own colony or that of a sister
colony. The sensing mechanism in this case could be hypothesized to arise out of signalling
molecules secreted by the targets. We investigated if the advancing tendrils could sense the
presence of inert obstacles which would not secrete signalling molecules. Interestingly, we found
strong evidence of the advancing tendril detecting the presence of the obstacle and changing its
direction as it approached the obstacle. The bulk of our studies were conducted with the inert
polymeric material PDMS. However, in order to check the material dependence of the obstacle, we
also conducted more limited studies with obstacles made of glass [SI Section 4 Figure S6]. We did
rigorous statistical analysis of experimental data to quantify the detection of inert obstacles by
advancing swarm tendrils. We analysed around 120 experiments involving sensing of inert objects of
various shapes with negative control (no obstacle) and positive control (flgM mutant) which causes
the wild type strain to avoid it as described earlier. The positive control clearly shows that the
swarming pattern is perturbed by the presence of the flagellar mutant. For the inert obstacle
experiments we first evaluated the possibility that the perturbation seen in the swarming pattern
indeed arises due to the presence of the obstacle. This was done by computationally scanning the
obstacle shape across the image and finding the best fit position. A unique fit at the exact obstacle
position or the best fit being around the obstacle position supports the hypothesis that the
perturbation is due to the obstacle and not due to the natural stochasticity in the swarming pattern.
This analysis revealed that about 80% of experiments, a significant majority, indicate successful
detection of the inert obstacle by the swarm from as far as 5mm [see SI Section 4 for video]. The
detection of obstacles was also observed with change in nutrient media (From PGM to M9) as with
change of obstacle material. [See SI section 4 for videos]
In addition to this analysis, we quantified the asymmetry in the swarming pattern. This is motivated
by the fact that the branch distribution in the case of a negative control can be assumed to be
unbiased hence symmetric. However, the presence of an inert obstacle or the positive control
creates asymmetry in the swarming pattern. To factor out the asymmetry due to the presence of the
obstacles itself (as opposed to the perturbation due to the obstacle) we digitally replicated the
image of the obstacle (inert or positive control) in all the four quadrants and subtract the obstacle
region from the original image. The pattern obtained after subtracting the digitally created obstacles
is used to calculate the variance of the swarm coverage in each quadrant. Both positive control and
the experiments on inert obstacles showed significantly larger inter-quadrant variance indicating
strong perturbation of the baseline swarming pattern. More details of the data analysis are
provided in the SI text [Section 5]. The sensing of inert obstacles from a distance represents the
strongest evidence of the advantages conferred by membership of the swarm as it is inconceivable
that an individual bacterium can detect an inert object from such a long distance. In the subsequent
section we propose a unified model to explain all these examples of spatial awareness in PA.
Figure. 2, a) Swarm plate with PDMS obstacle, b) Heat map showing available positions of the given obstacle to
occupy ( blue: Obstacle will not overlap with the pattern , yellow: Obstacle overlaps with the pattern) ,c)
Thresholded image pattern obtained by subtracting the obstacle for asymmetry analysis (See SI Section 5 for
more details), d) Statistics of different types of perturbations to the pattern due to presence of the obstacle
(Most probable: Patterns encircle the obstacle while avoiding it, Undecidable: Patterns have restricted growth
and/ or are unable to indicate the effect of the obstacle), e) Cumulative Distribution Function of inter-quadrant
variance in the distribution of the area occupied by the swarming colony
Towards a Mathematical Model of Spatial Awareness in the PA Swarm
The main idea of the mathematical model is that the individual bacterium and consequently the
swarm can sense a signalling molecule (or a cocktail of molecules) to decide its direction. For the
purpose of this model it is enough to consider a single molecule even though in practice several
molecules may be involved. We perform quantitative modelling to understand a) coarse-grained
dynamics of swarm direction changes induced by perturbations of the various kinds mentioned
above and b) what specific advantages are available to the individual members of the swarm by
virtue of membership in the swarm. The requirement of a signalling molecule leading to self-
avoidance of sister tendrils as well as the other instances of spatial awareness mentioned in this
article is well supported by previous reports describing the role of rhamnolipids (RL) in PA swarming
[(Caiazza et al., 2005)(Tremblay et al., 2007)]. We suppose that the advancing swarm tip contains
active (motile) bacteria which secrete signalling molecules at a specified rate f. The signalling
molecules diffuse with diffusion constant D. The governing equations then become
(1)
With f = 0 in the region outside the swarm tip. Inert obstacles or petri-plate edges are represented
by reflecting (non-diffusive) boundaries. Parameters such as the production rate and diffusion
constant of the signalling molecule, number of bacteria in a swarm are required to draw meaningful
inferences from this model. Although these parameters have not been explicitly measured, we can
estimate these from experiments if available, or by order of magnitude calculations [See SI Section 6
for more details]
Simulations proceed by initializing a circular disk of fixed radius representing the advancing swarm
tip with bacteria secreting the signalling molecule. The concentration field of the signalling molecule
is calculated using Eq. (1) above over the entire region and the "swarm tip" is advanced to a new
position representing the motion of the swarming tendril. The concentration field is updated and the
process continues. Firstly, we see that for the best estimates we have for the model parameters, a
gradient in the order of M/mm emerges within the swarm which is comparable to the gradients
which bacterial species such as E. coli (Jeon et al., 2009)(Diao et al., 2006) have been reported to
sense. We found that this gradient of the signalling molecule concentration accurately predicts the
future direction of the swarm. Specifically, the swarm will move in the direction of the steepest
negative gradient although inertia and stochastic effects would induce some deviations around this
expected direction. These effects are not included in this model currently. However, this model
serves to demonstrate that for reasonably realistic parameter values, measurable concentration
differences of the signalling molecule emerge within the swarm which regulate the direction of the
swarming tendril. In the case of sensing of sister tendrils, we see that M/mm gradients appear
typically only in the smaller swarm tip which then changes its direction of motion [Figure 3(a)(f)]. In
the case of inert obstacles, we again see that the presence of a reflecting boundary results in
measurable gradients which predict the future direction of the swarm [Figure 4]. For this case, we
find the strongest argument for a bacterium's requirement of membership in a swarm because
measurable gradients from inert reflecting boundaries will only form if the source strength is large
enough. The large source strength required is provided by the swarm whereas individual bacterium
would never be able to produce it on its own.
Figure. 3, Signalling molecule gradient which emerges within the swarm tip when encountered by other
approaching swarm tips of different sizes at a separation of 5mm.
Figure 4, Signalling molecule gradient which emerges within the swarm tip due to an inert obstacle at a
separation of around 1 mm.
The model discussed above is a coarse-grained model which assures us of the strong likelihood of
measurable gradients of a signalling molecule forming within the swarm. Ultimately, we would like
to understand collective spatial awareness of the swarm emerging from individual behaviour and
cell-cell interactions including its inherent stochasticity. Insights derived from such studies can guide
robust distributed control strategies for future robotic swarms (Rubenstein et al., 2012)] and other
collective sensing phenomena. We constructed a multi-agent model [Wilensky, U. (1999). NetLogo]
based on two-component signalling system. The two signalling molecules produced by the agents
have different diffusion constants with each of them either invoking an attracting or repelling
response among the agents. The spatial distribution of the signalling molecules govern the
behaviour of the agent [See SI Section 8 for a detailed description of the multi-agent model], which
spontaneously shows the emergence of branching as observed in experiments [Figure 5]. The agent
based system also exhibited spontaneous emergence of spatial awareness similar to the biological
system, most notably, the detection of boundaries (analogous to detection of inert obstacles
presenting non-diffusive reflective boundaries) and consequent retraction of the advancing swarm
tendril as observed in experiments. The spontaneous emergence of these features in the simulations
suggest that the remarkable spatial awareness seen in the bacterial swarm may arise from simple
attractive-repulsive
indirectly
leads to the effective
computation of the global gradient predicted by the fluid dynamic model.
interactions between bacteria which
Figure. 5, Illustration of agent based simulation showing branched pattern and retraction from boundary [ Red
pixels = Active swarmers in state 0, Yellow pixels = Active swarmers in state 1, White pixels = water film formed
or colony spread (See SI section 7 for more details)]. Regions i and ii highlights the retraction phenomena from
the boundary and other tendril respectively.
Discussion and open questions
Detailed spatial awareness expectedly increases the fitness of an organism to survive a complex
environment. In this article, we showed examples of a primitive organism displaying sophisticated
spatial awareness such as long-range sensing of an inert obstacle. The response of the swarm to
these perturbations is coherent and highly coordinated unlike previous descriptions of sensing
neighbouring swarming colonies by lethal secretions. The role of a relatively fast diffusing signalling
molecule is expected and strongly suggested by the mathematical models. In particular, the coarse-
grain model provides strong evidence to the fact that measurable differences in the concentration of
signalling molecules can arise from our expectations on the model parameters which will induce
difference in behaviour in individual bacteria at the swarm boundaries. Indeed, a single cell resolved,
multi-agent model based on signalling molecule concentration dependent attractive-repulsive
potential exhibits spontaneous emergence of branching and some aspects of spatial awareness
observed in the actual biological system. Presently the agreement between experiments and models
is largely qualitative. The patterns observed in the multi-agent simulations do not resemble the
patterns observed in the experiments indicating that the present model is not complete. Another
open question is how the swarming tendril avoids self-inhibition from its own secretions while being
able to inhibit the advancement of a neighbouring tendril at a much longer distance away. Although
these questions are still open, this work presents some essential ingredients likely to lead to a
quantitatively rigorous agent based model for PA swarming which can reproduce not only the
pattern formation aspects but also its collective sensing abilities of the swarm. Such a model would
also enable one to study robustness of the collective behaviour in the presence of defective
individuals and other perturbations. Insights related to robust behaviour would be of exceptional
value to the emerging field of swarm robotics from the perspective of robust decentralized control.
Methods
Swarming Motility Assay
For swarming assay, we used Peptone growth medium (PGM). Composition of PGM 0.6% agar
plates are 6 grams of bacteriological agar (Bacto agar), 3.2 grams of peptone, and 3 grams of
sodium chloride (NaCl) added in 1 litre of distilled water. The medium was autoclaved at 121°C for
30 minutes. After autoclaving the media, 1 mL of 1M CaCl2 (Calcium chloride), 1 mL of 1 M MgSO4
(Magnesium sulphate), 25 ml of 1M KPO4 and 1 mL of 5 mg/mL cholesterol were added into the
medium and mixed properly. 25 mL PGM were poured in each 90mm Petri-plates and allowed them
to solidify at room temperature (RT) for a half an hour under the laminar hood flow with the lid
opened. And all the plates were kept at room temperature for 16-18 hours for further drying.
Swarming
2ul of a planktonic culture of Wild type PA or flgM mutant with OD >2.8 is inoculated at the centre of
90 mm petri-plate containing PGM-0.6% agar. The wild type PA14 forms the pattern as shown in
Figure 1 a) over a period of 24 hours in a 90 mm petri-plate. FlgM is a transposon insertion mutant in
the flgM gene of PA14 and is part of P. aeruginosa transposon insertion library (Liberati et al., 2006).
Preparation of PDMS (Poly DiMethyl Siloxane) obstacle
We have used Sylgard 184 from Dow Corning. It has two parts: an elastomer part and the curing
agent. The two parts i.e. elastomer: curing agent is mixed in the ratio of 10:1. This mixture is stirred
well. The air bubble cause due to stirring is removed by degassing the PDMS mixture in a desiccator
connected to vacuum pump. An acrylic template is made to obtain different shapes of the obstacle.
The air bubble free PDMS mixture is then poured into the template and cured for 12 hours. The
cured PDMS solidifies and is removed from the acrylic template. These PDMS obstacles are then
sterilised in autoclave.
The sterilised obstacle blocks are placed in an appropriate position in the petri-plate. The nutrient
agar is then poured around the obstacle such that the obstacle is half immersed in the nutrient agar
while held intact in its original position. The nutrient agar with the obstacle is allowed to dry under
the laminar hood.
Acknowledgements
We gratefully acknowledge Robert Bosch Centre for Cyber Physical Systems at Indian institute of
Science, Bangalore, India for funding this research. We also acknowledge the use of facilities at
Centre for Nano Science and Engineering, Indian Institute of Science, Bangalore, India.
Reference
Be'er, A., Ariel, G., Kalisman, O., Helman, Y., Sirota-Madi, A., Zhang, H. P., … Swinney, H. L. (2010).
Lethal protein produced in response to competition between sibling bacterial colonies.
Proceedings of the National Academy of Sciences of the United States of America, 107(14),
6258–6263. http://doi.org/10.1073/pnas.1001062107
Be'er, A., Zhang, H. P., Florin, E.-L., Payne, S. M., Ben-Jacob, E., & Swinney, H. L. (2009). Deadly
competition between sibling bacterial colonies. Proceedings of the National Academy of
Sciences of the United States of America, 106(2), 428–433.
http://doi.org/10.1073/pnas.0811816106
Ben-Jacob, E. (1997). From snowflake formation to growth of bacterial colonies II: Cooperative
formation of complex colonial patterns. Contemporary Physics, 38(3), 205–241.
http://doi.org/10.1080/001075197182405
Caiazza, N. C., Shanks, R. M. Q., & Toole, G. A. O. (2005). Rhamnolipids Modulate Swarming Motility
Patterns of Pseudomonas aeruginosa. Journal of Bacteriology, 187(21), 7351–7361.
http://doi.org/10.1128/JB.187.21.7351–7361.2005
Daniels, R., Vanderleyden, J., & Michiels, J. (2004). Quorum sensing and swarming migration in
bacteria. FEMS Microbiology Reviews, 28, 261–289.
http://doi.org/10.1016/j.femsre.2003.09.004
Deng, P., Roditi, L. D. V., Ditmarsch, D. Van, & Xavier, J. B. (2014). The ecological basis of
morphogenesis : branching patterns in swarming colonies of bacteria. New Journal of Physics.
http://doi.org/10.1088/1367-2630/16/1/015006
Diao, J., Young, L., Kim, S., Fogarty, E. a, Heilman, S. M., Zhou, P., … DeLisa, M. P. (2006). A three-
channel microfluidic device for generating static linear gradients and its application to the
quantitative analysis of bacterial chemotaxis. Lab on a Chip, 6, 381–388.
http://doi.org/10.1039/b511958h
Du, H., Xu, Z., Anyan, M., Kim, O., Leevy, W. M., Shrout, J. D., & Alber, M. (2012). High density waves
of the bacterium pseudomonas aeruginosa in propagating swarms result in efficient
colonization of surfaces. Biophysical Journal, 103, 601–609.
http://doi.org/10.1016/j.bpj.2012.06.035
Du, H., Xu, Z., Shrout, J. D., & Alber, M. (2011). MULTISCALE MODELING OF PSEUDOMONAS
AERUGINOSA SWARMING. Mathematical Models and Methods in Applied Sciences, M3AS(21
Suppl 1), 939–954. http://doi.org/10.1142/S0218202511005428
James, B. L., Kret, J., Patrick, J. E., Kearns, D. B., & Fall, R. (2009). Growing Bacillus subtilis tendrils
sense and avoid each other: Research letter. FEMS Microbiology Letters, 298, 12–19.
http://doi.org/10.1111/j.1574-6968.2009.01665.x
Jeon, H., Lee, Y., Jin, S., Koo, S., Lee, C. S., & Yoo, J. Y. (2009). Quantitative analysis of single bacterial
chemotaxis using a linear concentration gradient microchannel. Biomedical Microdevices, 11,
1135–1143. http://doi.org/10.1007/s10544-009-9330-8
Kearns, D. B. (2010). A field guide to bacterial swarming motility. Nature Reviews Microbiology, 8,
634–644. http://doi.org/10.1038/nrmicro2405
Kearns, D. B., & Losick, R. (2003). Swarming motility in undomesticated Bacillus subtilis. Molecular
Microbiology, 49(3), 581–590. http://doi.org/10.1046/j.1365-2958.2003.03584.x
Liberati, N. T., Urbach, J. M., Miyata, S., Lee, D. G., Drenkard, E., Wu, G., … Ausubel, F. M. (2006). An
ordered, nonredundant library of Pseudomonas aeruginosa strain PA14 transposon insertion
mutants. Proceedings of the National Academy of Sciences of the United States of America, 103,
2833–2838. http://doi.org/10.1073/pnas.0511100103
Rubenstein, M., Ahler, C., & Nagpal, R. (2012). Kilobot: A low cost scalable robot system for collective
behaviors. Proceedings - IEEE International Conference on Robotics and Automation, 3293–
3298. http://doi.org/10.1109/ICRA.2012.6224638
Tremblay, J., Richardson, A., Lépine, F., & Déziel, E. (2007). Self-produced extracellular stimuli
modulate the Pseudomonas aeruginosa swarming motility behaviour. Environmental
Microbiology, 9(10), 2622–2630. http://doi.org/10.1111/j.1462-2920.2007.01396.x
Verstraeten, N., Braeken, K., Debkumari, B., Fauvart, M., Fransaer, J., Vermant, J., & Michiels, J.
(2008). Living on a surface: swarming and biofilm formation. Trends in Microbiology, 16(10),
496–506. http://doi.org/10.1016/j.tim.2008.07.004
Vicsek, T., Czirok, A., Ben-Jacob, E., Cohen, I., & Shochet, O. (1995). Novel Type of Phase Transition in
a System of Self-Driven Particles. Physical Review Letters, 75(6), 1226–1229.
Section 1
Supplementary information
Different swarming patterns of different bacteria- sparsity compactness analysis
Each bacterial species produces a unique swarming pattern. To characterize the uniqueness of each
pattern we define two factors: Sparsity and Compactness.
Compactness helps in differentiating a branched pattern from a circularly expanding one.
Compactness is defined as the ratio of the area to the square of the perimeter. Circle has
compactness equal to 1. A highly branched pattern has higher perimeter for a given area hence
compactness is very low.
Compactness
C=
Sparsity factor shows how well separated the branches are. The closely packed branches are less
sparse.
Sparsity
S=
Figure S1. Sparsity vs. log(compactness) for swarming patterns of different bacteria
The graph shows pseudomonas aeruginosa clusters are well separated from the patterns of the rest
of swarming bacteria. This indicates the uniqueness of the pattern and the need for a unique model
to describe the pattern as the existing models that explain the other patterns may not hold good.
To characterise the swarming pattern of the Pseudomonas aeruginosa, we calculate the branch
length, branch width and branch angles. The statistics of the pattern parameters have been
summarised in the Figure S2.
Figure S2. a) Swarming pattern of PA on a PGM-0.6% agar with labelled branch parameters. Statistics
Probability density function of different branch parameters have been obtained with b) mean tendril length is
16.85 mm c) mean tendril angle is 54.04 degrees and d) mean tendril width is 4.4 mm.
The tip of the branch has been tracked in the time lapse video recorded over 24 hours. The velocity
of the branch TIP has been calculated for 10 such branches and the average was found to be about
55.48 µm/min
Figure S3., a) distance covered by the tip of the tendril at different time points from the starting point, b) speed
of the tip of the tendril is almost constant at 50 µm/min
Section 2
Awareness of the Presence of Sister Tendrils
Figure S4., a) Swarming pattern respond to a sister colony on the same petri-plate, b) swarm tip position (3.35
pixels/mm) and density with increasing time(0.67 frames/min)(t1 and t2 are time points at which the swarm tip
passes a point ' p' along the tendril, one during forward swarming and the other during retraction ), c)
example of swarm tip progression and retraction with time
Along with awareness of the edge, the tendrils can sense the sister colonies and the branches of
their own colony. The Figure S4 a) shows five colonies on the same petri-plate. They show two types
of behaviours, either change in direction to avoid colliding with the other branches or active cells
stop and retract when they find no other suitable direction to progress.
When the wild type is allowed to swarm on the same plate as a mutant like flgM which do not
swarm, we again see interesting behaviour of avoidance as illustrated in the Figure S5 .[ See video
PA_mutant_interaction.avi]
Figure S5. a) Swarming pattern respond to a non swarming mutant on the same petri-plate b) variation of
distance between swam tip and the center of the non swarming mutant with time (0.56 frames/min).
Section 3
The video shows wild type PA swarming on a 90mm petri-plate and the swarm tip retracts from the
edge of the petri-plate. [See video Edge_retraction.avi]
Section 4
Obstacle made of different materials
To test the independence of the avoidance behaviour to the material used as the obstacle, we have
used obstacles made of glass and steel. We see the similar phenomena for the obstacles made of
glass and steel, indicating the material independence of the avoidance behaviour. This strengthens
our hypothesis of the obstacle being just a reflecting boundary and not a source of any signalling
molecule that could be sensed by the bacteria.
Figure S6. Swarming patterns in the presence of obstacle made of a) steel(s), glass(g), PDMS(p), b) all of
glass(g), c) two of them made of glass(g) and one of steel(s)
Obstacle videos
The time lapse videos of the wild type PA in PGM media and M9 media in the presence of obstacles
made of PDMS.
[See videos Obstacle_PGMmedia.avi , Obstacle_M9media.avi]
Section 5
Asymmetry Analysis to quantify perturbation to swarm pattern
To test the effect of the obstacle on the swarm pattern of the wild type, we take wildtype-flgM plate
as a positive control to get a baseline for a definitely biased swarm pattern and a normal swarming
plate as negative control to get a baseline for unbiased swarm pattern. We expect the system with
an inert obstacle to lie in between these extremes.
The average size of the obstacles is comparable to the mutant spread on the plate. This size is small
compared to the area available in-between branches. Since the pattern is quite sparse, we need to
rule out the possibility that the branch might have avoided the obstacle by chance. So in our
analysis, we translate the obstacle along the pattern and check for regions where the obstacle might
occur. Each pixel position represents the centroid position of the obstacle and the intensity of the
pixel represent the degree of overlap of the obstacle with the pattern. Yellow pixel represents the
position of maximum overlap with the pattern and the dark blue would indicate that obstacle would
fit perfectly in the position and would not intersect the swarm pattern. We analysed about 120
pattern with different shapes of obstacles like triangle, rectangle and a circle .
Figure S7. Obstacle translation method to generate a heat map representing the suitable positions the
obstacle can occupy without touching the swarm pattern
Figure S8. (a),(b) Example of case of most probable detection and its corresponding heat map (c),(d) Example
of Undecidable cases and its corresponding heat map
The heat maps of 120 swarms with obstacles show open positions on each petri-plate that a swarm
could occupy without touching the tendril(s). This allowed us to categorise the obstacle sensing into
2 distinct patterns- 1) Most probable detection- The obstacle is present in the largest open position
in the swarm pattern. This indicates that the presence of the obstacle somehow perturbed the
branches to move away from the obstacle leading to shift in patterns, 2) Undecidable case - the
swarm does not progress till the end of the petri-plate, hence it is difficult to say if the swarm
detected the obstacle or not.
In the presence of the obstacle, we see that about 80 % of the plates showed most probable
detection. This indicates the presence of obstacle indeed perturbs the PA swarms and their
behaviour. This can be a new example of the social communication and decision making that PA is
known to exhibit.
Figure S9: step1: Replicating the obstacle in all four quadrants to obtain a symmetric distribution step2:
Subtracting symmetric obstacle distribution from the perturbed swarm pattern induced by the obstacle
To quantify the obstacle induced perturbation of the swarm, the symmetry breaking of the pattern
needs to be quantified. The branch distribution in the case of a negative control (no obstacle) can be
assumed to be unbiased hence symmetric. But the symmetry was statistically (80 out of 120 plates)
affected in the tendrils around the obstacle. The presence of obstacle may create asymmetry in the
nutrient distribution thus inducing change/asymmetry in the swarm behaviour and pattern. To
negate this asymmetry, we digitally created images of obstacles in four quadrants(Figure S9 step1)
and subtract the obstacle region from the pattern region(Figure S9 step2). This will remove the
asymmetry in the pattern due to asymmetry in the nutrient distribution itself. Now the pattern
obtained after subtracting the digitally created obstacle distribution is subjected to symmetry
analysis (Figure S9).
Figure S10: a) Inter-quadrant asymmetry analysis of swarms with obstacles. b) Minimum bounding circle
analysis of swarms with obstacles.
The pattern is divided into four quadrants. The white pixels represent the bacterial swarm and black
pixels represent the part of the image excluding the swarm. The number of the white pixels in each
quadrant indicates the extent of the colony spread in that quadrant. By calculating the variance of
the number of white pixel distribution over the four quadrants, extent of symmetry in the pattern
can be quantified. Higher the inter-quadrant variance, higher is the difference in swarm area
distribution among quadrants and hence higher the asymmetry. Figure S10 shows the mean variance
of the pattern in different quadrants. The mean variance is clearly higher for obstacle experiments
and is comparable with the experiments with flgM mutants on the same plate. This indicates that
there is asymmetry in the pattern when physically or chemically perturbed.
A control plate of swarming PA has tendrils reaching the edge of the petri-plate in 24 hours. For the
asymmetry argument to hold, the swarms should have reached the periphery of the petri-plate in
the direction of no perturbation. To verify this, a minimum bounding circle for each plate inoculated
with the bacteria at the centre is considered. This can be quantified by measuring the minimum
radius of the circle with its centre fixed at the point of inoculation (which is usually the center of the
plate) and can completely bound the pattern. The radius of the minimum bounding circle is equal to
the radius of the petri-plate in control plates. The petri-plate considered for the experiments are of
40 mm radius and from the histogram more than 75% of the plates have a minimum bounding
radius at 40 mm thus confirming the pattern swarm till the end of the petri-plate.
Section 6
Experimental parameters
In a swarm on PGM-0.6% agar, tendrils start at the swarm centre and move away with an average
speed of 1 µm/s after some lag period. The mass of a bacteria is approx. 1pg. Assuming, it produces
signalling molecules equal to its body weight in its complete life time. The cell division/binary fission
time of the bacteria is about 60- 90 minutes in minimal media (Badalà et al.,2008).we measure the
cell density at the tendril tip to be 1.5e6 – 2e6 cell/mm^2. The rate of production of signalling
molecule is thus
For simulation, secretion rate is taken to be 1ng/mm^2.s
The active swarm cells have been observed to change direction by sensing the presence of the other
branch from distances as far as 5mm. Assuming the communication is happening only based on the
gradients of the self-produced signalling molecules, the long range sensing observed in the
experiments is possible only if the signalling molecules diffused faster than the distance covered by
the moving sources. This constraint can be represented mathematically as
L= length at which sensing happens
D=diffusion constant of signalling molecule
V= velocity of swarm tip
Taking L=5mm and velocity of about 1 µm/s (experimentally determined)
Diffusion constant of D=1e-8 m^2/s is considered for simulations. This diffusion constant implies a
small signalling molecule with the Stokes radius of 25.5 pm (This corresponds to the radius of a
hydrogen atom assuming diffusion is happening in water).
Figure S11: a) distribution of signalling molecules at different time stamps b) probe points of the swarm tip c)
Percentage concentration difference between the center and peripheral point of the swarmtip in each direction
The simulation involves the swarm tip of about 1mm in radius. It moves about 1mm in 1000s which
is equivalent to 1µm/s (experimentally, each position within the swarm tip produces 1ng/mm^2.s of
the signalling molecules).
The simulation results indicate that the concentration difference between pairs of orthogonal points
shown in Figure S11, assuming an agar height of about 5mm, ranges from 0.2 mg/L – 2mg/L. If the
signalling molecule was to be Rhamnolipid (molecular weight of about 650 g/mol) , we get the
concentration gradient in the range of 0.3 µM/mm – 3 µM/mm.
Section 7
Multi-agent Modelling
Swarm states
The model considers 500 active swarmers initially at the center of the 81x81 patch as shown in the
Figure S12 a). All the patches initially contain no chemical or water in them. Quorum sensing is
necessary for the active swarmers to know if the local density of the swarmers has reached the
critical density enough to produce sufficient surfactant to ease their movement across the surface.
The active swarmers in the model are hence in 2 states.
State 0: Swarmers know that the quorum has not been reached
State 1: Swarmers know that the quorum has been reached and change their behaviour
Quorum signal(attractant) production
Initially all the active swarmers are in state 0 and each headed in random directions. Each swarmer
in state 0 produces quorum signalling molecule at rate 0.2 mass units/second (1 second is equivalent
to a tick in simulations) and looks for the direction in which the quorum signal molecule is highest.
The search is restricted to the patches defined by check length l (set to 1 in our simulation) and
check angle theta (set to 45 degree). The check length l is measured from the current position of the
swarmer and check angle theta is measured from the current direction of heading as shown in Figure
S12 b). The direction of swarmer is updated to the best among the patches after the check process.
If none of the left or right patches are better than the patch ahead, the swarmer continues to move
in the original headed direction. The swarmer checks the current patch value of quorum molecule at
each step and changes its state accordingly. Let Cqi be the value of quorum signalling molecule in
patch in which the ith swarmer is present and Si be the state variable of the ith swarmer
If (Cqi < Tq) Si =0 else Si =1
Where Tq = quorum threshold
Repellent production
If the swarmers are in quorum (state 1), they produce surfactant. As a result a thin water film is
formed in the current patch. Each swarmer in state 1 increases the water content of its current
patch position by 0.1 mass-units/second. Along with the ability to create water film, they also
produce repellent. Swarmers produce repellent at a rate of 0.2 mass-units/second. Along with
producing repellent, they get away from regions with high concentration of the repellent. They use
the same checking mechanism discussed above and choses the minimum repellent direction for the
next time step.
Water edge detection
The swarmers move to the edge of the water film created by them to expand their colony into niche
nutrient rich zones. They move with different speeds depending on the presence of water in the
patch where they are located. If the patch has water, it moves 0.05 step length per time and if the
patch has no water, it moves 0.01 step length per time.
The direction of the motion is decided by the current headed direction which gets updated
depending on the swarmer's state and the values of quorum signal, repellents and water around the
swarmer.
Diffusion
The quorum signal and repellent molecule diffuse to the neighbouring patch positions using rules
described by the following equations
∑
Where c(x,y,t) is the concentration of a chemical in patch (x,y) at time t
n is the number of neighbours to a patch at position (x,y)
ci(x,y,t-1) is the concentration of the chemical in the ith neighbour of the patch (x,y)
D is the diffusion constant of the chemical
The value of D of quorum signal is 0.2 and D of repellent is 0.8. We observed that the diffusion
constant of the quorum signal had to be less than that of repellent to observe branching patterns in
the simulations, qualitatively similar to those observed in experiments.
Evaporation
The quorum signal molecule evaporates and reduces its concentration at each time step by 10%.
Noise
Noise is added to the final direction chosen by the swarmer. It is in the range -45 to +45 degrees.
The video of the simulation of the swarm based on the above rules is available at the following link.
[See video swarm_simulation.avi]
Figure S12., a) Initial state of the active swarmers, b) Search strategy (solid black line – current headed
direction of the swarmer, green patch – check patches for next direction of movement), c) Simulated Branch
pattern with agent represented in red or yellow depending on the state being 0 or 1 and water in white.
References
LaBauve, A. E., & Wargo, M. J. (2012). Growth and Laboratory Maintenance of Pseudomonas
aeruginosa. Curr Protoc Microbiol., 25(1), 6E.1.1‐6E.1.8.
http://doi.org/10.1002/9780471729259.mc06e01s25
|
1808.02752 | 1 | 1808 | 2018-08-04T08:51:42 | Deciphering the relative contribution of vascular inflammation and blood rheology in metastatic spreading | [
"physics.bio-ph",
"physics.flu-dyn",
"physics.med-ph"
] | Vascular adhesion of circulating tumor cells (CTCs) is a key step in cancer spreading. If inflammation is recognized to favor the formation of vascular metastatic niches, little is known about the contribution of blood rheology to CTC deposition. Herein, a microfluidic chip, covered by a confluent monolayer of endothelial cells, is used for analyzing the adhesion and rolling of colorectal (HCT 15) and breast (MDA MB 231) cancer cells under different biophysical conditions. These include the analysis of cell transport in a physiological solution and whole blood; over a healthy and a TNF alpha inflamed endothelium; with a flow rate of 50 and 100 nL/min. Upon stimulation of the endothelial monolayer with TNF alpha (25 ng/mL), CTC adhesion increases by 2 to 4 times whilst cell rolling velocity only slightly reduces. Notably, whole blood also enhances cancer cell deposition by 2 to 3 times, but only on the unstimulated vasculature. For all tested conditions, no statistically significant difference is observed between the two cancer cell types. Finally, a computational model for CTC transport demonstrates that a rigid cell approximation reasonably predicts rolling velocities while cell deformability is needed to model adhesion. These results would suggest that, within microvascular networks, blood rheology and inflammation contribute similarly to CTC deposition thereby facilitating the formation of metastatic niches along the entire network, including the healthy endothelium. In microfluidic based assays, neglecting blood rheology would significantly underestimate the metastatic potential of cancer cells. | physics.bio-ph | physics | DECIPHERING THE RELATIVE CONTRIBUTION OF VASCULAR INFLAMMATION
AND BLOOD RHEOLOGY IN METASTATIC SPREADING
Hilaria Mollica1,6, Alessandro Coclite6, Marco E. Miali2,6, Rui Pereira6, Laura Paleari3,4, Chiara
Manneschi6, Andrea DeCensi3,5, Paolo Decuzzi6,§
1 DIBRIS, University of Genova. Via Opera Pia 13, Genoa 16145, Italy
2 Dipartimento di Meccanica, Matematica e Management, DMMM, Politecnico di Bari, Via Re David,
200-70125, Bari, Italy.
3 Division of Medical Oncology, Galliera Hospital, Via Volta 6, Genoa 16128, Italy
4 A.Li.Sa, Public Health Agency, Piazza della Vittoria 15, Genoa 16121, Italy
5 Wolfson Institute of Preventive Medicine, Queen Mary University of London, Charterhouse Square,
London EC1M 6BQ, United Kingdom
6 Laboratory of Nanotechnology for Precision Medicine, Fondazione Istituto Italiano di Tecnologia,
Via Morego 30, Genoa 16163, Italy
§ To whom correspondence should be addressed: Paolo Decuzzi, PhD. Phone: +39 010 71781 941, Fax:
+39 010 71781 228, E-mail: [email protected]
1
ABSTRACT
Vascular adhesion of circulating tumor cells (CTCs) is a key step in cancer spreading. If inflammation is
recognized to favor the formation of vascular 'metastatic niches', little is known about the contribution
of blood rheology to CTC deposition. Herein, a microfluidic chip, covered by a confluent monolayer of
endothelial cells, is used for analyzing the adhesion and rolling of colorectal (HCT-15) and breast (MDA
MB-231) cancer cells under different biophysical conditions. These include the analysis of cell transport
in a physiological solution and whole blood; over a healthy and a TNF-a inflamed endothelium; with a
flow rate of 50 and 100 nL/min. Upon stimulation of the endothelial monolayer with TNF-a (25 ng/mL),
CTC adhesion increases by 2 to 4 times whilst cell rolling velocity only slightly reduces. Notably, whole
blood also enhances cancer cell deposition by 2 to 3 times, but only on the unstimulated vasculature. For
all tested conditions, no statistically significant difference is observed between the two cancer cell types.
Finally, a computational model for CTC transport demonstrates that a rigid cell approximation
reasonably predicts rolling velocities while cell deformability is needed to model adhesion. These results
would suggest that, within microvascular networks, blood rheology and inflammation contribute
similarly to CTC deposition thereby facilitating the formation of metastatic niches along the entire
network, including the healthy endothelium. In microfluidic-based assays, neglecting blood rheology
would significantly underestimate the metastatic potential of cancer cells.
KEYWORDS
Circulating cancer cells; vascular adhesion; microfluidic chips; computational modeling; Lattice
Boltzmann;
2
INTRODUCTION
The formation of distant metastasis from a primary neoplastic mass is a very inefficient biological
process.(Talmadge and Fidler 2010; Nguyen, Bos, and Massague 2009; Wirtz, Konstantopoulos, and
Searson 2011; Joyce and Pollard 2009) Spreading of cancer cells evolves following a precise cascade of
events -- the metastatic cascade -- requiring cell migration away from the primary mass and intravasation
into blood or lymphatic vessels, following the epithelial to mesenchymal transition; circulation within
the blood stream, where cells have to survive hemodynamic stresses and immune cell recognition;
extravasation, migration and proliferation at the secondary sites. Radioactive assays documented that
only 1% of circulating tumor cells (CTCs) can successfully overcome all these sequential steps and
eventually establish distant metastases.(Fidler 1970) Despite the inefficiency and complexity of the
process, the vast majority of cancer patients who relapse eventually succumb because of metastases,
disseminated at different secondary sites, rather than for the uncontrolled growth of the original
malignancy.(Chaffer and Weinberg 2011)
CTC arrest within different vascular districts is a key step in the metastatic cascade and is primarily
mediated by two mechanisms: vascular occlusion, which generally occurs in the small capillary beds of
the brain and lungs(Kienast et al. 2010); and vascular adhesion, which is regulated by specific
interactions between receptor molecules on the endothelium and ligand molecules on CTCs.(Schluter et
al. 2006; Witz 2008) A wide range of vascular molecules are involved in this specific adhesion process,
including E- and P-selectins, avb3 and avb5
integrins, VCAM-1 and ICAM-1 adhesion
molecules.(Burdick et al. 2003; Barthel et al. 2007; Myung et al. 2011) These receptors can bind several
different ligands expressed on the CTC membrane, making target therapies against metastasis practically
impossible. This picture is further complicated by the fact that platelets, leukocytes and CTCs tend to
form in the circulation stable aggregates that favor blood longevity and vascular deposition of malignant
3
cells.(Gay and Felding-Habermann 2011; Borsig et al. 2002) In this context, pro-inflammatory
cytokines(Solinas et al. 2010; Kim et al. 2009), such as TNF-a, IL-1b and IL-6; tumor-derived
exosomes(Hoshino et al. 2015; Hood, San, and Wickline 2011) and hematopoietic cells(Kaplan et al.
2005; Shiozawa et al. 2011) have been shown to modulate the expression of adhesion molecules in
specific vascular districts thus priming the formation of so called 'pre-metastatic niches' where CTCs
more efficiently, and in a larger number, accumulate.
Cell-cell adhesion is strongly modulated by external forces and, as such, static assays may not always
reproduce the complex interactions developing under flow within the vasculature. Intravital microcopy
has been extensively employed to document cell migration within vascular and extravascular
compartments,(Kienast et al. 2010; Provenzano, Eliceiri, and Keely 2009) however these in vivo analyses
lack a precise control on the governing parameters. On the other hand, microfluidic chips allow to
precisely control blood vessel sizes, flow rates and the expression of vascular adhesion molecules and
are amenable for high through-put systematic characterizations. A variety of microfluidic chips are being
developed for studying the different steps in the metastatic cascade. For instance, the group of Kamm
designed flow devices for assessing transvascular migration of cancer cells in different extravascular
matrices.(Bersini et al. 2014; Jeon et al. 2015; Niu et al. 2014; Zervantonakis et al. 2012) The vascular
adhesion and transmigration of individual and clustered CTCs was studied under chemokine stimulation
(exposure to CXCL12 and SDF-1a) by various groups.(Song et al. 2009; Zhang, Liu, and Qin 2012;
Roberts, Waziri, and Agrawal 2016) Studies of cancer cell migration within the lymphatic system were
presented by Swartz and collaborators.(Pisano et al. 2015) The group of Jiang focused on investigating
the role of endothelial cell mechanical (cyclic shear stresses) and biochemical (exposure to TNF-a)
stimulation on CTC vascular adhesion.(Huang et al. 2015) Huang and collaborators developed cellulose-
4
based tubular artificial blood vessels for reproducing the intravasation, vascular adhesion and
extravasation of cancer cells.(Wang et al. 2015)
Although red blood cells (RBCs) are known to affect the dynamics of leukocytes and CTCs, at authors'
knowledge, no studies have addressed the relative roles of vascular inflammation and RBC dynamics on
the vascular deposition of malignant cells. In this work, a microfluidic chip is used to study the rolling
and firm adhesion of breast (MDA-MB-231) and colorectal cancer (HCT-15) cells on a confluent layer
of human vascular endothelial cells (HUVECs). The hematocrit of the working solution ranges from 0
to 40% and TNF-a is used for stimulating HUVECs. The rolling velocity and number of firmly adhering
tumor cells are measured under different conditions. A Lattice-Boltzmann computational model is also
included to interpret and reproduce the vascular adhesion dynamics of cells.
RESULTS AND DISCUSSIONS
A continuously growing body of evidence documents that vascular inflammation supports the firm
adhesion of circulating tumor cells (CTCs) and facilitate the distant colonization of otherwise healthy
tissues with the consequent formation of tumor metastases.(Solinas et al. 2010; Kim et al. 2009; Hoshino
et al. 2015; Hood, San, and Wickline 2011) In this context, human vascular endothelial cells (HUVECs)
were stimulated with the pro-inflammatory cytokine TNF-α and the adhesion propensity of cancer cells
(HCT-15 and MDA-MB-231) was assessed under static and dynamic conditions. The two cell lines are
among
the most metastatic and aggressively growing colon and breast cancer cells,
respectively.(Flatmark et al. 2004; Holliday and Speirs 2011)
Cancer cell adhesion on inflamed endothelial cells under dynamic conditions. HUVECs were seeded
in multiwell plates and, after reaching confluency, were stimulated with TNF-α (10 ng/mL, 25 ng/mL 50
5
ng/mL) for 6 hours. Cancer cells were added to the multiwell plates and left interacting with the
endothelial cells up to 4 hours, under static conditions. In agreement with a large body of literature, these
static experiments continue to confirm that endothelial stimulation with a pro-inflammatory cytokine
(TNF-α) favors CTC vascular adhesion in a dose dependent manner (Supporting Figure.1 and 2).
Moving from static to dynamic experimental conditions, a PDMS single-channel microfluidic chip was
used for monitoring the interaction of cancer and endothelial cells under flow (Figure.1a). The
microfluidic channel was 2.7 cm long and had a 210 µm wide by 42 µm high rectangular cross section.
The working fluid was introduced in the PDMS chip continuously for about 15 minutes at two different
flow rates, namely 50 and 100 nL/min. These flow rates reproduce wall shear rates (13.49 and 26.99 s-1)
and mean blood velocities (94.48 and 188.9 µm/s) typically found in the microcirculation.(Popel and
Johnson 2005) The PDMS channel was covered by a confluent layer of endothelial cells mimicking the
blood vessel walls; whereas the cancer cells were dispersed within the working fluid consisting of either
cell culture media or whole blood. Again, two different malignant cell lines were considered, namely
colon (HCT-15) and breast cancer (MDA-MB-231) cells. In order to reproduce an inflamed endothelium,
the HUVEC monolayer was stimulated with the pro-inflammatory cytokine TNF-α. Representative
confocal fluorescent images of the experimental set-up with cells are shown in Figure.1b. Red
fluorescent cancer cells (CM-DIL staining of the membrane) are spotted firmly adhering over blue
fluorescent HUVECs (DAPI staining of the nucleus). The same images show in green VE-cadherin
molecules decorating the boundary between two adjacent endothelial cells and demonstrating the high
level of confluency of the endothelial monolayer deposited on the microfluidic channel surface.
Via fluorescent microscopy, the number of adhering cells was quantified, within five different regions
of interest (ROIs) along the channel, and normalized by the total number of injected cells (ninj =106) and
the ROI area. This was performed for twelve different working conditions depending on the types of
6
cancer (colon and breast); flow rates (50 and 100 nL/min) and levels of HUVEC inflammation (un-
stimulated: -TNF-a; 6 hours stimulation: +TNF-a 6h; and 12 hours stimulation: +TNF-a 12h). Results
are provided in Figure.2b and d, respectively, for a flow rate Q = 50 and 100 nL/min, and for breast
cancer (blue bars) and colon cancer (red bars) cells. On the left hand side, Figure.2a and c, representative
fluorescent microscopy images are shown for unstimulated, 6 hour stimulated, and 12 hour stimulated
HUVECs. These images are snapshots taken from full movies available as Supporting Information.
Notably, for all twelve different working conditions, no statistically significant difference was depicted
when comparing breast and colon cancer cells. Conversely, significant differences arose when
considering different flow rates and levels of TNF-a stimulation. At Q = 50 nL/min, the normalized
number of adhering HCT-15 and MDA-MB-231 cells was, respectively, 9.952 ± 1.803 and 10.24 ± 2.841
#/m2 in control experiments, 29.09 ± 2.219 and 28.54 ± 5.038 #/m2 after 6 hours of TNF-a stimulation;
40.37 ± 9.205 and 40.26 ± 3.521 #/m2 after 12 hours of TNF-a stimulation. At Q = 100 nL/min, the
normalized number of adhering HCT-15 and MDA-MB-231 cells was, respectively, 6.698 ± 1.452 and
7.30 ± 1.088 #/m2 in control experiments, 11.87 ± 0.899 and 13.78 ± 1.716 #/m2 after 6 hours of TNF-a
stimulation; 34.05 ± 1.427 and 26.69 ± 2.780 #/m2 after 12 hours of TNF-a stimulation.
As compared to the healthy vasculature, cancer cells adhered 2 and 3-times more avidly to a 6h- and 12h-
inflamed endothelium. Maximum cell adhesion is observed under static conditions (Q = 0, Supporting
Figure.1c), followed by Q = 50 and 100 nL/min. Thus, as expected, the number of adhering cells reduces
as the flow rate increases. Indeed, this is related to the corresponding increase of the hydrodynamic
dislodging forces that would decrease the likelihood of firm CTC adhesion on HUVECs.
Cancer cell rolling on inflamed endothelial cells under dynamic conditions. A subset of circulating
tumor cells was observed to interact with the endothelial monolayer without firmly adhering but rather
rolling steadily. The cancer cells exposed to a dynamic conditions are transported within the microfluidic
7
chip at two different flow rates (50 and 100 nL/min). The solution is injected into the microfluidic chip
using a syringe pump for 15 minutes Thus, the rolling velocity uroll of tumor cells was quantified by
monitoring the displacement of the cell centroid over time. Movies for rolling cells are provided in the
Supporting Information under different flow rates, HUVEC inflammation levels and cell types. By
imaging post-processing, uroll of the metastatic colon (HCT-15) and breast (MDA-MB-231) cancer cells
was quantified at 50 and 100 nL/min, and under different HUVEC conditions, namely unstimulated
HUVECs (- TNF-α), 6h-stimulated HUVECs (+TNF-α 6h), and 12h-stimulated HUVECs (+TNF-α 12h).
Data are charted in Figure.3a and b, respectively, for 50 and 100 nL/min. At 50 nL/min, the rolling
velocity of HCT-15 cells was of 113.9 ± 4.132, 103.4 ± 2.880 and 98.00 ± 4.552 µm/sec for unstimulated
HUVECs (- TNF-α), 6h-stimulated HUVECs (+TNF-α 6h), and 12h-stimulated HUVECs (+TNF-α 12h),
respectively. Under the same conditions, for the MDA-MB-231, the rolling velocities were 118.6 ± 1.349
µm/sec 105.68 ± 3.340 µm/sec 102.1 ± 5.288 µm/sec (Figure.3a). Even in the case of rolling velocities,
no statistically significant difference was observed between the two cell lines. A 10% and 20%
statistically significant decrease in rolling velocities between the control groups and the 6 and 12 hours
TNF-α stimulated groups was observed. Under TNF-a stimulation, endothelial cells express a larger
number of adhesion molecules, which would reduce the rolling velocity and favor the firm deposition of
CTCs. Note that, this is in agreement with what was documented by Navarro and collaborators (Ríos-
Navarro et al. 2015) in the case of polymorphonuclear (PMNCs) and peripheral blood mononuclear
(PBMCs) cells.
As expected, the rolling velocity slightly but steadily decreased as the level of TNF-α stimulation
increased. At 100 nL/min, the rolling velocities for the HCT-15 cells were 163.6 ± 20.10 µm/sec (-TNF-
α), 157.4 ± 4.531 µm/sec (TNF-α 6h) and 158.06 ± 1.187 µm/sec (TNF-α 12h). For the MDA-MB-231,
the same physical quantity took the values 170.9 ± 11.03 µm/sec (-TNF-α); 151.8 ± 8.182 µm/sec (6h
TNF-α) and 144.9 ± 1.500 µm/sec (12h TNF-α).
8
Lastly, the ratio between the number of rolling and adhering cells was plotted for two different flow
conditions (Figure.3c and d). For unstimulated HUVECs, most of the circulating tumor cells were
observed to steadily roll over the endothelium monolayer, whereas the ratio decreases as the TNF-α
stimulation increases. At low flow rates (Q = 50 nL/min), the ratio for the HCT-15 cells was 0.845 ±
0.084 (-TNF-α); 0.713 ± 0.122 (TNF-α 6h) and 0.553 ± 0.096 (TNF-α 12h). Very similar are the ratios
quantified for the MDA-MB-231, for which it resulted 0.828 ± 0.067 (- TNF-α), 0.669 ± 0.034 (TNF-α
6h) and 0.597 ± 0.030 (TNF-α 12h). At high flow rates (Q = 100 nL/min), the ratio for the HCT-15 cells
was 0.875 ± 0.020 (- TNF-α), 0.728 ± 0.038 (TNF-α 6h) and 0.591 ± 0.017 (TNF-α 12h). Similarly, for
the MDA-MB-231, the ratio was 0.850 ± 0.061 (- TNF-α), 0.715 ± 0.015 (TNF-α 6h) and 0.651 ± 0.063
(TNF-α 12h). As reported before for other physical quantities, also in this case, no statistically significant
difference was determined between the two cell lines.
Cancer cell adhesion on inflamed endothelial cells under whole blood flow. Leukocyte recruitment
at inflamed tissues has a number of similarities with the colonization at distant sites of CTCs. In
particular, just like for leukocytes, CTCs tend to transiently interact with the blood vessel walls engaging
specific receptor molecules, then adhere and spread over the endothelial cells and, eventually, cross the
vascular barrier relocating in the extravascular space. Adhesion molecules are over-expressed in
postcapillary venules during an inflammatory process.(Granger and Senchenkova 2010; Strell and
Entschladen 2008; McEver and Zhu 2010). Moreover, it is well recognized that leukocyte rolling and
adhesion on the inflamed vascular endothelium is modulated by the presence of red blood cells (RBCs).
Specifically, experimental observation and simulations have shown that the deformability and shape of
RBCs allow them to concentrate within the core of blood vessels leaving a so-called 'cell free layer' next
to the vessel walls.(Goldsmith, Cokelet, and Gaehtgens 1989; Pappu and Bagchi 2007; Fedosov,
Noguchi, and Gompper 2014; Firrell and Lipowsky 1989) Leukocytes, which are two-times larger and
9
far less abundant than RBCs, tend to be pushed laterally in the cell free layer by the fast moving RBCs.
This process, known as 'margination', should also affect the vascular behavior of CTCs.
In this section, cancer cell rolling and adhesion over a monolayer of HUVECs is analyzed in the presence
of whole blood. The single-channel microfluidic chip was again covered by a confluent monolayer of
HUVECs, which were unstimulated or stimulated with TNF-α (12h only), and cancer cells re-suspended
in whole blood were directly injected at two different flow rates (Q = 50 and 100 nL/min). Whole blood,
freshly drawn from rats, contained all the cell and molecular components of blood, including red blood
cells, platelets, leucocytes and plasma proteins which may all contribute, at different extents, to cancer
cell rolling and adhesion.(Gay and Felding-Habermann 2011; Borsig et al. 2002) A fixed hematocrit of
40% was considered. Results for eight different working conditions are provided in Figure.4b and d,
which are for Q = 50 and 100 nL/min, respectively. As previously, breast cancer cells are identified by
blue bars whereas colon cancer cells are associated with red bars. On the left hand side (Figure.4a and
c), representative fluorescent microscopy images are shown for unstimulated HUVECs (- TNF-α), and
12h-stimulated HUVECs (+TNF-α 12h). The results unequivocally showed that blood cells favor the
adhesion of circulating tumor cells to the vascular walls, especially in the case of unstimulated
endothelium. In Figure.4b and d, the normalized number of adhering cells is reported. At Q = 50 nL/min,
the normalized number of adhering HCT-15 and MDA-MB-231 cells was 25.33 ± 4.762 and 18.19 ±
1.269 #/m2 in control experiments, 35.68 ± 10.99 and 46.96 ± 13.18 #/m2 after 12 hours of TNF-a
stimulation, respectively. At Q =100 nL/min, the normalized number of adhering HCT-15 and MDA-
MB-231 cells was 26.04 ± 9.90 and 17.59 ± 6.129 #/m2 in control experiments, 30.12 ± 4.011 and 23.04
± 4.406 #/m2 after 12 hours of TNF-a stimulation, respectively. Notably, even under these conditions,
no statistically significant difference in cell adhesion was detected between breast and colon cancer cells.
Interestingly, a statically significant difference was measured only between untreated and TNF-a treated
10
endothelial cells at the lowest flow rate (Q = 50 nL/min, in Figure.4b). At highest flow rates, the absolute
number of adhering cells reduces and twelve hours TNF-a stimulation is insufficient to induce a
statistically significant increase in cell deposition.
A direct comparison in terms of CTC vascular adhesion between whole blood flow and physiological
solution is now needed. Figure.4e and f collect all the data required for this comparison. Within an
unstimulated microvascular network, the presence of blood cells does dramatically increase CTC
adhesion (Figure.4e). For Q = 50 nL/min, the density of firmly adhering CTCs grows from about 10 to
20 #/m2 moving from physiological solution to whole blood flow. A slightly larger increase is observed
for Q = 100 nL/min. Differently, within an inflamed microvascular network, the presence of blood cells
does not significantly affect CTC adhesion (Figure.4f). The density of firmly adhering CTCs is around
40 #/m2 at 50 nL/min and reduces to about 30 #/m2 at 100 nL/min, with and without RBCs. This could
be interpreted as, under the current conditions, the density of adhering CTCs on the inflamed endothelium
has reached saturation and the presence of RBCs cannot further foster cell deposition. Also, RBC-CTC
collisions could limit any further increase in cell deposition. Indeed, additional experiments would be
needed to support this hypothesis. Interestingly, a direct comparison of the data presented in Figure.4e
and f would lead one to infer that, at higher flow rates (100 nL/min), the density of firmly adhering CTCs
on the inflamed and normal vasculature is comparable when a whole blood flow is considered. Again,
this could be due to a balance between shear stresses and cell-cell collisions. Indeed, this is not observed
at low flow rates (50 nL/min), where adhesion is higher on inflamed endothelium.
This data confirms that blood cells facilitate the vascular adhesion of CTCs, just like for leukocytes, and
open up to the following two considerations: CTCs would tend to adhere throughout the
microvasculature, on both inflamed and not inflamed endothelial cells, thus increasing the likelihood of
11
finding proper conditions for colonization; in microfluidic experiments, neglecting the role of blood cells
could dramatically underestimate the adhesion propensity of CTCs.
Predicting cancer cell adhesion and rolling on inflamed endothelial cells. In order to predict CTC
vascular behavior under different flow and adhesion conditions, a computational model was employed
based on previous works by the authors.(Decuzzi and Ferrari 2006; Coclite et al. 2016; Coclite et al.
2017) In this model, cancer cells were considered as rigid and deformable circular objects exposed to a
Poiseuille flow and capable of interacting with vascular walls (endothelial cells) via specific ligand-
receptor bonds (Figure.5a). Simulations were performed in a rectangular computational domain, with
height H (= 42 µm) and length 10H, resembling the longitudinal cross section of the single channel in
the microfluidic chip. The diameter of cancer cells was fixed to d= 15 µm, as quantified via bright field
microscopy (Supporting Figure.3). The ratio between the number of ligands decorating the surface of
cancer cells and the number of receptors expressed on the endothelium is rl. Two different ratios ρl were
considered, namely 0.3 and 0.6. These assumed ligand densities return a good agreement between the
experimental and numerical predictions for the cell rolling velocity over three different flow rates.
At first, cancer cells were assumed to be rigid, which is indeed the simplest possible hypothesis. Then,
simulations were performed for estimating the rolling velocities of cancer cells over the vascular wall as
a function of four different flow rates, namely Q = 25, 50, 75, and 100 nL/min; and two 𝜌" ratios, namely
0.3 and 0.6. The resulting data are shown in Figure.5b (lines) where a direct comparison with the
corresponding experimental data is also included (blue dots for HCT-15 cells). From the simulations, the
cell rolling velocity was predicted to grow quasi-linearly with the flow rate Q (R2 = 0.998 and 0.994 for
𝜌" = 0.3 and 0.6, respectively) and slightly decrease with an increase in 𝜌" = 0.3. Overall, the predicted
returning a relative error smaller than 0.74% and 3.10% for 𝜌" = 0.3 and 0.6, respectively. A larger
rolling velocities were found to be in good agreement with the experimental data for Q = 100 nL/min,
12
difference was observed at low flow rates, Q = 50 nL/min, where the relative error increased to about
43.01% and 52.07% for 𝜌" = 0.3 and 0.6, respectively. This might be due to the fact that this flow rate is
very close to the lower limit for the syringe pump used in the experiments. Note that an increase in 𝜌"
from 0.3 to 0.6 was associated with only a 3.5% decrease in rolling velocity. This is also in agreement
with the experimental data of Figure.3a and b documenting a modest variation in uroll with vascular
inflammation.
Although the 'rigid cell' approximation quite accurately modeled the rolling behavior of cancer cells, it
could not predict their firm vascular adhesion. Therefore, in a second set of simulations, the cancer cell
was considered as a deformable capsule characterized by the dimensionless capillary number Ca = 10 -2.
These data are plotted in Figures.5c and f for four different flow rates (Q = 25, 50, 75, and 100 nL/min);
two ligand-receptor densities (𝜌"=0.3 and 0.6). Also, a direct comparison between rigid and soft cells is
provided. Soft cells exhibited more complex vascular adhesion patterns. For 𝜌"=0.3, soft cells were
observed to establish an initial adhesive contact with the endothelial surface resulting in partial cell
deformation and increase in the number of ligand-receptor bonds. However, after reaching a maximum,
the adhesive interactions were not sufficient to counteract the dislodging hydrodynamic forces and,
consequently, the number of close bonds reduced tending eventually to zero. For 𝜌"=0.6, a larger number
of ligand-receptor bonds could be formed leading to stronger adhesive interactions. This is indeed
observed in the plots of Figure.5c and f. Also, for sufficiently high flow rates (Q ³ 50 nL/min), partially
adhering soft cells were deformed and pushed down to the wall thus maximizing their adhesive surface
and interface forces and leading to a 2 to 3-times higher number of ligand-receptor bonds as compared
to the corresponding rigid cell cases (Figure.5e and f). Notably, simulations predicted that rigid cells
would roll over the endothelium with a rolling velocity decreasing with an increasing surface density of
ligands (black and blue lines in Figure.5c and f). Differently, deformable cells would, for low ligand
surface densities, transiently adhere, detach and move away from the wall pushed by hydrodynamic lift
13
forces (red lines in Figure. 5c and f); whereas, for high ligand surface densities, deformable cells would
firmly adhere, deform under flow and increase the surface of adhesion as documented by the growth of
the number of the engaged ligand-receptor bonds. (green lines in Figure. 5c and f and Supporting
Figure.4). Although the present simulations can quite accurately predict the rolling velocities of
circulating cancer cells, it should be emphasized that only a fully 3D model, including deformable RBCs
and CTCs, could realistically predict the vascular behavior of cancer cells.(Fedosov, Caswell, and
Karniadakis 2011; Muller, Fedosov, and Gompper 2014)
CONCLUSIONS
A microfluidic chip was used to analyze the vascular transport of circulating tumor cells under different
biophysical conditions. The surface density of adhering cells and the velocity of rolling cells were
quantitatively characterized over a confluent endothelial monolayer as a function of the level of
inflammation (no TNF-a; TNF-a stimulation for 6h; TNF-a stimulation for 12h); flow rate (50 and 100
nL/min); and working fluid (physiological solution and whole blood, at 40% hematocrit). Two different
types of cancer cells -- colorectal HCT-15 and breast cancer MDA-MB-231 cells -- were considered.
It was confirmed that vascular inflammation facilitates cell adhesion in a way proportional to TNF-a
stimulation, whereas high flow rates are associated with lower cell deposition. Rolling velocities are only
slightly affected by vascular inflammation and grow proportionally with the flow rate. As compared to a
physiological solution, flowing cancer cells in whole blood enhances their firm deposition on healthy
endothelium rather than on the inflamed vasculature, for all tested conditions. No statistically significant
difference is observed for adhesion and rolling between HCT-15 and MDA-MB-231 cells.
These results would imply that neglecting the contribution of whole blood in the analysis of cancer cell
dynamics can significantly underestimate their vascular deposition. Furthermore, it can be concluded that
14
whole blood flow supports cancer cell deposition and facilitates metastatization over the entire
microvasculature.
EXPERIMENTAL SECTION
Fabrication of a single channel microfluidic chip. The single channel microfluidic chip was fabricated
following protocols previously demonstrated by the authors.(Manneschi et al. 2016) Briefly, a SU8-50
master was used as a mold for PDMS replicas of the chip. First a 40 µm thick layer of SU8-50 photoresist
(MicroChem) was spin coated on a silicon wafer (4"- P doped - <100> - 10 ÷ 20 W/cm2 -- 525 µm thick,
from Si-Mat) at 2000 rpm for 30 s. Then, the negative SU8-50 template was pre-and soft baked for
solvent evaporation; exposed to UV light and baked again for epoxy crosslinking; and finally developed.
This template was replicated using a mixture of PDMS and curing agent Sylgard 182 (Dow Corning
Corporation), with a ratio (w:w) 10:1. Specifically, the mixture was poured on the SU8-50 template,
cured in an oven at 60°C for 15 h, and moved at -20°C for 1 h. After peeling off from the template, the
channel extremities of the PDMS replica were punched via a biopsy puncher (OD = 1 mm, Miltex) to
form inlet and outlet ports. Finally, upon oxygen plasma treatment (Pressure = 0.5 mBar, Power = 15 w,
Time = 15 s; Plasma System Tucano, Gambetti), PDMS replica was sealed with a glass slide (20 x 60 x
0.17 mm) (No. 1.5H, Deckaläser). The resulting microfluidic chip has a rectangular cross section with a
width w = 210 µm, height h = 42 µm, and a port-to port length l = 2.7 cm.
Seeding of endothelial cells into the microfluidic chip. Chips were sterilized by autoclave, dried and
covered with 20 µg/mL of fibronectin to allow cell adhesion. HUVECs were introduced in the channel
from the inlet port at a density of 3×106 cells/mL by using a pipette tip. Then, chips were placed in an
incubator, to allow cell attachment and growth, and continuously perfused with Endothelial cell growth
medium supplement-mix (PROMOCELL) until cell confluency was achieved. HUVEC monolayers were
15
inflamed, at the occurrence, with 25 ng/mL of TNF-α for 6 or 12 hours. Each experiment was compared
to untreated HUVEC monolayer (-TNF-α).
Cancer cell adhesion and rolling under dynamic conditions. The microfluidic chip was placed on the
stage of an epi-fluorescence inverted microscope (Leica 6000). The working fluid was injected into the
chip using a syringe pump 33 Dual (Harvard apparatus). After the tripsinization, the cancer cells were
incubated for 30 minutes with CM-DIL, at 37 C (0,5%, Thermofisher) according to the manufacture's
protocol. Then, the cells were washed 3 times with PBS 1x (GIBCO) to remove the excess dye. Finally,
the cells were re-suspended in the RPMI medium (HCT-15) or EMEM medium (MDA-MB-231), without
FBS, that could interfere with the cell adhesion parameters, at 1x106 cells/mL. After each rolling
experiment, a washing with PBS was performed to remove the non-adherent cancer cells from the
endothelium. Tumor cells were introduced via a syringe pump on the HUVEC monolayer inside the
single channel chip. The inlet port of the chip was connected to the syringe pump through a polyethylene
tube (BTPE-60, Instech Laboratories), while the tube of the outlet port was in PBS, to ensure flow
equilibrium. After 1 minute of flow, the interaction of tumor cells with HUVECs was recorded for 15
consecutive minutes for each experiment. Two flow rates Q were imposed via the syringe pump, namely
50 and 100 nL/min. The resulting rolling velocity of tumor cells was calculated offline by post processing
the videos, using the distance traveled by the cell and divided by the time, within a region of interest
(ROI) (magnification 10 X, A = 1.22 X 10-6 m2). At least 15 cancer cells per experiment were monitored.
Each experiment was repeated three times for each different conditions and flow rates. For the study of
cell adhesion under whole blood flow, the working fluid was obtained by combining a cancer cell
suspension (density of 106 cells/mL) with whole blood from rat, collected in a standard blood test tubing
containing 3.2% of buffered citrate to prevent clotting. The working hematocrit was fixed to 40%.
Experiments and image acquisition were performed as described above.
16
SUPPLEMENTARY MATERIALS
See supplementary material for the complete description of the materials and methods used in the cell
culturing; CTCs adhesion measurement in static and dynamic conditions; and computational modelling.
Supplementary material include also supporting figures and movies about CTCs adhesion and rolling on
inflamed endothelium, the computed adhesion mechanics of deformable cells, and confocal microscopy
images of the adhesion molecules in the microfluidic chip.
ACKNOWLEDGEMENTS
This project was partially supported by the European Research Council, under the European Union's
Seventh Framework Programme (FP7/2007-2013)/ERC grant agreement no. 616695 and the AIRC
(Italian Association for Cancer Research) under the individual investigator grant no. 17664. The authors
acknowledge the precious support provided by the Nikon Center at the Italian Institute of Technology
for microscopy acquisitions and analyses. The authors acknowledge the help of Dr. Federica Piccardi
with the whole blood experiments. LP acknowledges affiliation with A.Li.Sa., Public Health Agency
(Liguria Region). ADC acknowledges affiliation with the Wolfson Institute of Preventive Medicine of
the Queen Mary University in London (UK) and the Division of Cancer Prevention and Genetics at the
European Institute of Oncology in Milan (Italy).
COMPETING INTERESTS: The authors declare that they have no competing interests.
REFERENCES
Barthel, S. R., J. D. Gavino, L. Descheny, and C. J. Dimitroff. 2007. 'Targeting selectins and selectin
ligands in inflammation and cancer', Expert Opin Ther Targets, 11: 1473-91.
17
Bersini, S., J. S. Jeon, G. Dubini, C. Arrigoni, S. Chung, J. L. Charest, M. Moretti, and R. D. Kamm.
2014. 'A microfluidic 3D in vitro model for specificity of breast cancer metastasis to bone',
Biomaterials, 35: 2454-61.
Borsig, L., R. Wong, R. O. Hynes, N. M. Varki, and A. Varki. 2002. 'Synergistic effects of L- and P-
selectin in facilitating tumor metastasis can involve non-mucin ligands and implicate leukocytes as
enhancers of metastasis', Proc Natl Acad Sci U S A, 99: 2193-8.
Burdick, M. M., J. M. McCaffery, Y. S. Kim, B. S. Bochner, and K. Konstantopoulos. 2003. 'Colon
carcinoma cell glycolipids, integrins, and other glycoproteins mediate adhesion to HUVECs under
flow', Am J Physiol Cell Physiol, 284: C977-87.
Chaffer, C. L., and R. A. Weinberg. 2011. 'A perspective on cancer cell metastasis', Science, 331:
1559-64.
Coclite, A., M. D. de Tullio, G. Pascazio, and P. Decuzzi. 2016. 'A combined Lattice Boltzmann and
Immersed boundary approach for predicting the vascular transport of differently shaped particles',
Computers & Fluids, 136: 260-71.
Coclite, A., H. Mollica, S. Ranaldo, G. Pascazio, M. D. de Tullio, and P. Decuzzi. 2017. 'Predicting
different adhesive regimens of circulating particles at blood capillary walls', Microfluidics and
Nanofluidics, 21: 168.
Decuzzi, P., and M. Ferrari. 2006. 'The adhesive strength of non-spherical particles mediated by
specific interactions', Biomaterials, 27: 5307-14.
Fedosov, D. A., H. Noguchi, and G. Gompper. 2014. 'Multiscale modeling of blood flow: from single
cells to blood rheology', Biomech Model Mechanobiol, 13: 239-58.
Fedosov, Dmitry A, Bruce Caswell, and George Em Karniadakis. 2011. 'Wall shear stress-based model
for adhesive dynamics of red blood cells in malaria', Biophysical Journal, 100: 2084-93.
18
Fidler, I. J. 1970. 'Metastasis: quantitative analysis of distribution and fate of tumor emboli labeled with
125 I-5-iodo-2'-deoxyuridine', J Natl Cancer Inst, 45: 773-82.
Firrell, J. C., and H. H. Lipowsky. 1989. 'Leukocyte margination and deformation in mesenteric
venules of rat', Am J Physiol, 256: H1667-74.
Flatmark, K., G. M. Maelandsmo, M. Martinsen, H. Rasmussen, and O. Fodstad. 2004. 'Twelve
colorectal cancer cell lines exhibit highly variable growth and metastatic capacities in an orthotopic
model in nude mice', Eur J Cancer, 40: 1593-8.
Gay, L. J., and B. Felding-Habermann. 2011. 'Contribution of platelets to tumour metastasis', Nature
Reviews Cancer, 11: 123-34.
Goldsmith, H. L., G. R. Cokelet, and P. Gaehtgens. 1989. 'Robin Fahraeus: evolution of his concepts in
cardiovascular physiology', Am J Physiol, 257: H1005-15.
Granger, D Neil, and Elena Senchenkova. 2010. "Inflammation and the Microcirculation." In
Colloquium Series on Integrated Systems Physiology: From Molecule to Function, 1-87. Morgan &
Claypool Life Sciences.
Holliday, D. L., and V. Speirs. 2011. 'Choosing the right cell line for breast cancer research', Breast
Cancer Res, 13: 215.
Hood, J. L., R. S. San, and S. A. Wickline. 2011. 'Exosomes released by melanoma cells prepare
sentinel lymph nodes for tumor metastasis', Cancer Res, 71: 3792-801.
Hoshino, A., B. Costa-Silva, T. L. Shen, G. Rodrigues, A. Hashimoto, M. Tesic Mark, H. Molina, S.
Kohsaka, A. Di Giannatale, S. Ceder, S. Singh, C. Williams, N. Soplop, K. Uryu, L. Pharmer, T.
King, L. Bojmar, A. E. Davies, Y. Ararso, T. Zhang, H. Zhang, J. Hernandez, J. M. Weiss, V. D.
Dumont-Cole, K. Kramer, L. H. Wexler, A. Narendran, G. K. Schwartz, J. H. Healey, P. Sandstrom,
K. J. Labori, E. H. Kure, P. M. Grandgenett, M. A. Hollingsworth, M. de Sousa, S. Kaur, M. Jain,
K. Mallya, S. K. Batra, W. R. Jarnagin, M. S. Brady, O. Fodstad, V. Muller, K. Pantel, A. J. Minn,
19
M. J. Bissell, B. A. Garcia, Y. Kang, V. K. Rajasekhar, C. M. Ghajar, I. Matei, H. Peinado, J.
Bromberg, and D. Lyden. 2015. 'Tumour exosome integrins determine organotropic metastasis',
Nature, 527: 329-35.
Huang, R., W. Zheng, W. Liu, W. Zhang, Y. Long, and X. Jiang. 2015. 'Investigation of Tumor Cell
Behaviors on a Vascular Microenvironment-Mimicking Microfluidic Chip', Sci Rep, 5: 17768.
Jeon, J. S., S. Bersini, M. Gilardi, G. Dubini, J. L. Charest, M. Moretti, and R. D. Kamm. 2015.
'Human 3D vascularized organotypic microfluidic assays to study breast cancer cell extravasation',
Proc Natl Acad Sci U S A, 112: 214-9.
Joyce, J. A., and J. W. Pollard. 2009. 'Microenvironmental regulation of metastasis', Nature Reviews
Cancer, 9: 239-52.
Kaplan, R. N., R. D. Riba, S. Zacharoulis, A. H. Bramley, L. Vincent, C. Costa, D. D. MacDonald, D.
K. Jin, K. Shido, S. A. Kerns, Z. Zhu, D. Hicklin, Y. Wu, J. L. Port, N. Altorki, E. R. Port, D.
Ruggero, S. V. Shmelkov, K. K. Jensen, S. Rafii, and D. Lyden. 2005. 'VEGFR1-positive
haematopoietic bone marrow progenitors initiate the pre-metastatic niche', Nature, 438: 820-7.
Kienast, Y., L. von Baumgarten, M. Fuhrmann, W. E. Klinkert, R. Goldbrunner, J. Herms, and F.
Winkler. 2010. 'Real-time imaging reveals the single steps of brain metastasis formation', Nat Med,
16: 116-22.
Kim, S., H. Takahashi, W. W. Lin, P. Descargues, S. Grivennikov, Y. Kim, J. L. Luo, and M. Karin.
2009. 'Carcinoma-produced factors activate myeloid cells through TLR2 to stimulate metastasis',
Nature, 457: 102-6.
Manneschi, C., R. C. Pereira, G. Marinaro, A. Bosca, M. Francardi, and P. Decuzzi. 2016. 'A
microfluidic platform with permeable walls for the analysis of vascular and extravascular mass
transport', Microfluidics and Nanofluidics, 20.
20
McEver, Rodger P, and Cheng Zhu. 2010. 'Rolling cell adhesion', Annual review of cell and
developmental biology, 26: 363-96.
Muller, K., D. A. Fedosov, and G. Gompper. 2014. 'Margination of micro- and nano-particles in blood
flow and its effect on drug delivery', Sci Rep, 4: 4871.
Myung, J. H., K. A. Gajjar, R. M. Pearson, C. A. Launiere, D. T. Eddington, and S. Hong. 2011. 'Direct
measurements on CD24-mediated rolling of human breast cancer MCF-7 cells on E-selectin', Anal
Chem, 83: 1078-83.
Nguyen, D. X., P. D. Bos, and J. Massague. 2009. 'Metastasis: from dissemination to organ-specific
colonization', Nature Reviews Cancer, 9: 274-84.
Niu, Y., J. Bai, R. D. Kamm, Y. Wang, and C. Wang. 2014. 'Validating antimetastatic effects of natural
products in an engineered microfluidic platform mimicking tumor microenvironment', Mol Pharm,
11: 2022-9.
Pappu, V., and P. Bagchi. 2007. 'Hydrodynamic interaction between erythrocytes and leukocytes
affects rheology of blood in microvessels', Biorheology, 44: 191-215.
Pisano, M., V. Triacca, K. A. Barbee, and M. A. Swartz. 2015. 'An in vitro model of the tumor-
lymphatic microenvironment with simultaneous transendothelial and luminal flows reveals
mechanisms of flow enhanced invasion', Integr Biol (Camb), 7: 525-33.
Popel, A. S., and P. C. Johnson. 2005. 'Microcirculation and Hemorheology', Annual Review of Fluid
Mechanics, 37: 43-69.
Provenzano, P. P., K. W. Eliceiri, and P. J. Keely. 2009. 'Multiphoton microscopy and fluorescence
lifetime imaging microscopy (FLIM) to monitor metastasis and the tumor microenvironment', Clin
Exp Metastasis, 26: 357-70.
Ríos-Navarro, Cesar, Carmen de Pablo, Víctor Collado-Diaz, Samuel Orden, Ana Blas-Garcia, María
Ángeles Martínez-Cuesta, Juan V Esplugues, and Angeles Alvarez. 2015. 'Differential effects of
21
anti-TNF-α and anti-IL-12/23 agents on human leukocyte -- endothelial cell interactions', European
journal of pharmacology, 765: 355-65.
Roberts, S. A., A. E. Waziri, and N. Agrawal. 2016. 'Development of a Single-Cell Migration and
Extravasation Platform through Selective Surface Modification', Anal Chem, 88: 2770-6.
Schluter, K., P. Gassmann, A. Enns, T. Korb, A. Hemping-Bovenkerk, J. Holzen, and J. Haier. 2006.
'Organ-specific metastatic tumor cell adhesion and extravasation of colon carcinoma cells with
different metastatic potential', Am J Pathol, 169: 1064-73.
Shiozawa, Y., E. A. Pedersen, A. M. Havens, Y. Jung, A. Mishra, J. Joseph, J. K. Kim, L. R. Patel, C.
Ying, A. M. Ziegler, M. J. Pienta, J. Song, J. Wang, R. D. Loberg, P. H. Krebsbach, K. J. Pienta,
and R. S. Taichman. 2011. 'Human prostate cancer metastases target the hematopoietic stem cell
niche to establish footholds in mouse bone marrow', J Clin Invest, 121: 1298-312.
Solinas, G., F. Marchesi, C. Garlanda, A. Mantovani, and P. Allavena. 2010. 'Inflammation-mediated
promotion of invasion and metastasis', Cancer Metastasis Rev, 29: 243-8.
Song, J. W., S. P. Cavnar, A. C. Walker, K. E. Luker, M. Gupta, Y. C. Tung, G. D. Luker, and S.
Takayama. 2009. 'Microfluidic endothelium for studying the intravascular adhesion of metastatic
breast cancer cells', PLoS One, 4: e5756.
Strell, Carina, and Frank Entschladen. 2008. 'Extravasation of leukocytes in comparison to tumor cells',
Cell Communication and Signaling, 6: 10.
Talmadge, J. E., and I. J. Fidler. 2010. 'AACR centennial series: the biology of cancer metastasis:
historical perspective', Cancer Res, 70: 5649-69.
Wang, X. Y., Y. Pei, M. Xie, Z. H. Jin, Y. S. Xiao, Y. Wang, L. N. Zhang, Y. Li, and W. H. Huang.
2015. 'An artificial blood vessel implanted three-dimensional microsystem for modeling
transvascular migration of tumor cells', Lab Chip, 15: 1178-87.
22
Wirtz, D., K. Konstantopoulos, and P. C. Searson. 2011. 'The physics of cancer: the role of physical
interactions and mechanical forces in metastasis', Nature Reviews Cancer, 11: 512-22.
Witz, I. P. 2008. 'The selectin-selectin ligand axis in tumor progression', Cancer Metastasis Rev, 27:
19-30.
Zervantonakis, I. K., S. K. Hughes-Alford, J. L. Charest, J. S. Condeelis, F. B. Gertler, and R. D.
Kamm. 2012. 'Three-dimensional microfluidic model for tumor cell intravasation and endothelial
barrier function', Proc Natl Acad Sci U S A, 109: 13515-20.
Zhang, Q., T. Liu, and J. Qin. 2012. 'A microfluidic-based device for study of transendothelial invasion
of tumor aggregates in realtime', Lab Chip, 12: 2837-42.
23
FIGURES AND GRAPHICS
Figure 1. Single-channel microfluidic chip. a. On the left, schematic representation of a single channel
microfluidic chip with length l = 2.7 cm, width w = 210 µm; height h = 42 µm. On the right, a single
channel microfluidic chip, with connecting inlet and outlet tubing, filled with a blue ink and placed on
the stage of a fluorescent inverted microscope. b. Representative confocal fluorescent microscopy images
of HCT-15 cells (membrane labeled in red with CM-DIL) flowing in the chip and interacting with a
confluent layer of HUVECs (nuclei stained in blue with DAPI). VE-cadherin adhesion molecules, arising
at boundaries of the endothelial cells, are stained in green. (Images are provided for unstimulated (-TNF-
α) and TNF-α stimulated HUVECs for 6 (+ TNF-α 6h) and 12 hours (+ TNF-α 12h). TNF-α
concentration: 25 ng/mL. Scale bar: 50 µm).
24
Figure 2. Cancer cell adhesion on inflamed endothelial cells under dynamic conditions. a.
Representative fluorescence microscopy images of breast cancer cells MDA-MB-231 (cell membrane
labeled in red with CM-DIL) flowing and interacting, in a single-channel microfluidic chip, with a
confluent monolayer of HUVECs (cell nuclei stained in blue with DAPI). b. Normalized number of
adhering cancer cells on a HUVEC monolayer at a flow rate of 50 nL/min, with and without stimulation
with TNF-α (25ng/mL), for 6 and 12 hours. c. Representative fluorescence microscopy images of colon
cancer cells HCT-15 (cell membrane labeled in red with CM-DIL) flowing and interacting, in a single-
channel microfluidic chip, with a confluent monolayer of HUVECs (cell nuclei stained in blue with
DAPI). d. Normalized number of adhering cancer cells on a HUVEC monolayer at a flow rate of 100
25
nL/min, with and without stimulation with TNF-α (25ng/mL), for 6 and 12 hours. (Data are plotted as
mean ± SD. n = 3. Statistical analysis ANOVA: *** symbol denotes statistically significant difference
p < 0.0001; ** symbol denotes statistically significant difference p < 0.001. (ninj=106 cells and A =
1.22´10-6 m2). HUVECs are not stimulated with TNF-α (-TNF-α) or stimulated with 25 ng/mL TNF-α
for 6h (+TNF-α 6h) or 12h (+TNF-α 12h)).
26
Figure 3. Cancer cell rolling on inflamed endothelial cells under dynamic conditions. (.a,.b) Rolling
velocity of colon cancer HCT-15 (red column) and breast cancer MDA-MB-231 (blue column) at 50
nL/min and 100 nL/min on a confluent layer of HUVEC. (.c,.d) Ratio between the number of rolling and
adhering cancer cells (HCT-15 - red column; breast cancer MDA-MB-231 - blue column) on a confluent
layer of HUVEC at 50 nL/min and 100 nL/min. HUVECs are not stimulated with TNF-α (-TNF-α) or
stimulated with 25 ng/mL of TNF-α for 6h (+TNF-α 6h) or 12h (+TNF-α 12h). (Data are plotted as mean
± SD. n = 3. Statistical analysis ANOVA. * symbol denotes statistically significant difference p<0.01;
** symbol denotes statistically significant difference p<0.001; *** symbol denotes statistically
significant difference p<0.0001).
27
Figure 4. Cancer cell adhesion on inflamed endothelial cells under whole blood flow. a.
Representative fluorescence microscopy images of breast cancer cells MDA-MB-231 (cell membrane
labeled in red with CM-DIL) flowing in whole blood and interacting, in a single-channel microfluidic
28
chip, with a confluent monolayer of HUVECs (cell nuclei stained in blue with DAPI). b. Normalized
number of adhering cancer cells on a HUVEC monolayer of colon cancer HCT-15 (red column) and
breast cancer MDA-MB-231 (blue column) at a flow rate (50 nL/min), with and without stimulation with
TNF-α (25ng/mL) in the presence of whole blood (hematocrit: 40%). c. Representative fluorescence
microscopy images of colon cancer cells HCT-15 (cell membrane labeled in red with CM-DIL) flowing
in whole blood interacting, in a single-channel microfluidic chip, with a confluent monolayer of
HUVECs (cell nuclei stained in blue with DAPI). d. Normalized number of adhering cancer cells on a
HUVEC monolayer of colon cancer HCT-15 (red column) and breast cancer MDA-MB-231 (blue
column) at high flow rate (100 nL/min), with and without stimulation with TNF-α (25ng/mL) in the
presence of whole blood (hematocrit: 40%). (e.,.f.) Normalized number of adhering colon cancer HCT-
15 (red column) and breast cancer MDA-MB-231 (blue column) on a HUVEC monolayer, without (f.)
and with (e.) stimulation of TNF-α (25ng/mL) for 12h at a flow rate of 50 nL/min and 100 nL/min, with
and without whole blood (hematocrit: 40%). (Data are plotted as mean ± SD. n = 3. Statistical analysis
ANOVA: * symbol denotes statistically significant difference p < 0.05; ** symbol denotes statistically
significant difference p < 0.01 (ninj = 106 cells and A = 1.22´10-6 m2)).
29
Figure 5. Predicting cancer cell adhesion and rolling on inflamed endothelial cells. a. Schematic
diagram presenting the computational problem with a close-up depicting ligand-receptor interactions at
the interface between cancer (up) and endothelial (lower) cells. b. Rolling velocities of cancer cells under
four different flow rates (Q = 25, 50, 75, and 100 nL/min) and two ligand-receptor bond concentrations
(𝜌" = 0.3 and 0.6). (Solid lines are simulated values; Dots are experimental values; dashed lines are
four different flow rates (Q = 25, 50, 75, and 100 nL/min), two ligand-receptor bond concentrations (𝜌"
theoretical values). c, d, e, f. Variation of the number of active ligand-receptor bonds over time, under
30
= 0.3 and 0.6), and for soft and rigid cancer cells. Nl is the number of closed bonds in each time step.
This number is computed as the ratio between the current number of closed bonds over the number of
closed bonds in the initial configuration."
31
|
1909.05493 | 1 | 1909 | 2019-09-12T07:54:05 | Tip-Enhanced Infrared Difference-Nanospectroscopy of the Proton Pump Activity of Bacteriorhodopsin in Single Purple Membrane Patches | [
"physics.bio-ph",
"cond-mat.mes-hall",
"physics.optics"
] | Photosensitive proteins embedded in the cell membrane (about 5 nm thickness) act as photoactivated proton pumps, ion gates, enzymes, or more generally, as initiators of stimuli for the cell activity. They are composed of a protein backbone and a covalently bound cofactor (e.g. the retinal chromophore in bacteriorhodopsin (BR), channel rhodopsin, and other opsins). The light-induced conformational changes of both the cofactor and the protein are at the basis of the physiological functions of photosensitive proteins. Despite the dramatic development of microscopy techniques, investigating conformational changes of proteins at the membrane monolayer level is still a big challenge. Techniques based on atomic force microscopy (AFM) can detect electric currents through protein monolayers and even molecular binding forces in single-protein molecules but not the conformational changes. For the latter, Fourier-transform infrared spectroscopy (FTIR) using difference-spectroscopy mode is typically employed, but it is performed on macroscopic liquid suspensions or thick films containing large amounts of purified photosensitive proteins. In this work, we develop AFM-assisted, tip-enhanced infrared difference-nanospectroscopy to investigate light-induced conformational changes of the bacteriorhodopsin mutant D96N in single submicrometric native purple membrane patches. We obtain a significant improvement compared with the signal-to-noise ratio of standard IR nanospectroscopy techniques by exploiting the field enhancement in the plasmonic nanogap that forms between a gold-coated AFM probe tip and an ultraflat gold surface, as further supported by electromagnetic and thermal simulations. IR difference-spectra in the 1450-1800 cm^{-1} range are recorded from individual patches as thin as 10 nm, with a diameter of less than 500 nm, well beyond the diffraction limit for FTIR microspectroscopy... | physics.bio-ph | physics | This is an open access article published under an ACS AuthorChoice License, which permits
copying and redistribution of the article or any adaptations for non-commercial purposes.
Cite This: Nano Lett. 2019, 19, 3104−3114
Letter
pubs.acs.org/NanoLett
°
Eglof Ritter,
⊥
⊥
Stefano Corni,
°
Matthias Broser,
Tip-Enhanced Infrared Difference-Nanospectroscopy of the Proton
Pump Activity of Bacteriorhodopsin in Single Purple Membrane
Patches
‡
Valeria Giliberti,*,†
Raffaella Polito,
Laura Zanetti-Polzi,
Ulrich Schade,
±
Paolo Biagioni,
†
Istituto Italiano di Tecnologia, Center for Life NanoScience, Viale Regina Elena 291, I-00161 Roma, Italy
‡
Department of Physics, Sapienza University of Rome, Piazzale Aldo Moro 2, I-00185 Roma, Italy
°Humboldt-Universitat zu Berlin, Institut fur Biologie, Invalidenstrasse 42, D-10115 Berlin, Germany
Helmholtz-Zentrum Berlin fur Materialien und Energie GmbH, Albert-Einstein-Str. 15, 12489 Berlin, Germany
⊥
Department of Physical and Chemical Sciences, University of L'Aquila, Via Vetoio, I-67010 L'Aquila, Italy
#Department of Chemical Sciences, University of Padova, Via Marzolo 1, I-35131 Padova, Italy
$CNR Institute of Nanoscience, Via Campi 213/A, I-41125 Modena, Italy
±Dipartimento di Fisica, Politecnico di Milano, Piazza Leonardo da Vinci 32, I-20133 Milano, Italy
and Michele Ortolani*,†,‡
#,$ Francesco Rusconi,
±
Isabella Daidone,
‡
Peter Hegemann,
°
Ljiljana Puskar,
Leonetta Baldassarre,
*S Supporting Information
ABSTRACT: Photosensitive proteins embedded in the cell
membrane (about 5 nm thickness) act as photoactivated proton
pumps, ion gates, enzymes, or more generally, as initiators of
stimuli for the cell activity. They are composed of a protein
backbone and a covalently bound cofactor (e.g. the retinal
chromophore in bacteriorhodopsin (BR), channelrhodopsin,
and other opsins). The light-induced conformational changes of
both the cofactor and the protein are at the basis of the
physiological functions of photosensitive proteins. Despite the
dramatic development of microscopy techniques, investigating
conformational changes of proteins at the membrane monolayer
level is still a big challenge. Techniques based on atomic force
microscopy (AFM) can detect electric currents through protein monolayers and even molecular binding forces in single-protein
molecules but not the conformational changes. For the latter, Fourier-transform infrared spectroscopy (FTIR) using difference-
spectroscopy mode is typically employed, but it is performed on macroscopic liquid suspensions or thick films containing large
amounts of purified photosensitive proteins. In this work, we develop AFM-assisted, tip-enhanced infrared difference-
nanospectroscopy to investigate light-induced conformational changes of the bacteriorhodopsin mutant D96N in single
submicrometric native purple membrane patches. We obtain a significant improvement compared with the signal-to-noise ratio
of standard IR nanospectroscopy techniques by exploiting the field enhancement in the plasmonic nanogap that forms between
a gold-coated AFM probe tip and an ultraflat gold surface, as further supported by electromagnetic and thermal simulations. IR
difference-spectra in the 1450−1800 cm−1 range are recorded from individual patches as thin as 10 nm, with a diameter of less
than 500 nm, well beyond the diffraction limit for FTIR microspectroscopy. We find clear spectroscopic evidence of a branching
of the photocycle for BR molecules in direct contact with the gold surfaces, with equal amounts of proteins either following the
standard proton-pump photocycle or being trapped in an intermediate state not directly contributing to light-induced proton
transport. Our results are particularly relevant for BR-based optoelectronic and energy-harvesting devices, where BR molecular
monolayers are put in contact with metal surfaces, and, more generally,
for AFM-based IR spectroscopy studies of
conformational changes of proteins embedded in intrinsically heterogeneous native cell membranes.
KEYWORDS: Infrared spectroscopy, bacteriorhodopsin, transmembrane proteins, protein conformational changes, AFM-IR,
plasmonic nanogap
R hodopsins as photosensitive transmembrane proteins
(TMPs)1 are fundamental biological macromolecules
that can act as light-gated ion pumps, channels, and enzymes
Received: February 4, 2019
Revised:
April 2, 2019
Published: April 5, 2019
© 2019 American Chemical Society
3104
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Downloaded via UNIV OF PADOVA on September 12, 2019 at 06:14:21 (UTC).See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles. Nano Letters
that transport ions or information across the cell membrane
(about 5 nm thickness), or as light
receptors initiating
enzymatic processes as phototaxis or vision.2,3 They generally
consist of seven transmembrane alpha-helices (opsin)1 and a
covalently bound photosensitive retinal cofactor that crucially
absorbs light. Light absorption is followed by a trans to cis
photoisomerization of the retinal.4 The visible-light-absorption
properties of the rhodopsins are mostly determined by specific
interaction of retinal with residues of the retinal binding pocket
forming the effective rhodopsin chromophore. Conformational
changes of this chromophore and of the protein backbone are
the key determinants of
the photosensitive rhodopsin
functions. In the relevant case of photoactivated pumps and
channels,
the protein
backbone lead to an opening of the conducting pore through
which ions are then transported along the electrochemical
gradient5,6 or to pK-changes of amino acid residues that result
in an uphill proton transport against
the electrochemical
gradient.7,8
these conformational changes of
Infrared (IR) spectroscopy is one of the main tools used for
the investigation of the conformational changes of rhodopsins
due to the high sensitivity of the IR vibrational absorption
frequencies and lineshapes to structural features.9,10 Moreover,
IR spectroscopy can be applied to rhodopsins without
interfering with the visible light employed for the optical
activation of the membrane functionalities.11 Because of the
extremely small variations in the protein structure deriving
from light-induced conformational changes of TMPs, FTIR
spectrometers are usually employed in difference-spectroscopy
mode, where IR vibrational absorption spectra are acquired
with and without visible-light illumination, and often only the
difference-spectrum is analyzed.12 FTIR difference-spectrosco-
py is widely applied to the study of, for example, the light-gated
proton pump bacteriorhodopsin (BR)13 or the light-gated
channelrhodopsin.10,14 FTIR spectroscopy, however, lacks the
spatial
resolution and the sensitivity required to address
individual parts of membranes and is therefore limited to
highly concentrated suspensions or thick films containing
about 109 cell membrane patches with a high density of light-
sensitive TMPs, preferentially of one single species.12 Efforts
have been undertaken to increase the sensitivity of FTIR-based
approaches by exploiting the attenuated total
reflectance
(ATR) effect15 or, to reach down to the membrane monolayer
level, by applying the surface-enhanced IR (SEIRA)
absorption16 by metal nanostructures.17,18 The typical
diameter of a native cell membrane patch is indeed of the
order of 1 μm only and is not even addressable by FTIR
microscopy (micro-FTIR).19 Large-area TMP monolayers
suitable for ATR-FTIR and SEIRA-FTIR difference-spectros-
copy have been obtained by reforming self-assembled thin
films where TMPs are embedded into an artificially
reconstructed cell membrane.18 However,
this approach
requires complex protein purification processes20 and prevents
the study of the TMPs in their native membrane environ-
ment.21 More recently, the requirement of large sample areas
of many squared millimeters covered by membrane mono-
layers, which represents a challenge for present TMP
biotechnologies,18,20 has been relaxed with the introduction
of SEIRA structures in micro-FTIR, but signal-to-noise ratios
are still to be improved.22
The atomic force microscope (AFM) is the instrument of
choice for studying single native membrane patches and
different AFM-based approaches have been applied to the
Letter
study of photosensitive TMPs.21,23−27 High-speed AFM24 and
conductive AFM25,26 have been applied to probe the light-
activated functions of BR embedded in 5 nm-thick individual
patches of its native cell membrane, usually referred to as the
purple membrane. Conductive AFM may,
indeed, provide
insights into local charge transport through BR molecules,25,26
while high-speed AFM has been applied to shed light onto the
main morphological changes of the BR backbone through fine
topography variations.24 These efforts are motivated by the
potential use of BR as biomaterial
for optoelectronic
applications including organic photovoltaic cells,28,29 where
purple-membrane patches are deposited on flat metal electro-
des. In particular, it has been found30 that the photoelectrical
activity of BR in the case of membrane monolayers on metal
surfaces is much less efficient than the ideal mechanism in
liquid suspensions or multilayers. However, a fundamental
understanding of such decreased activity, and therefore of the
underlying modifications of the BR photocycle imposed by the
combination of several effects such as direct physisorption,
chemisorption, or adhesion of BR on metal
is still
the reasons is that a purely AFM-based
lacking. One of
technique cannot detect
subtle conformational changes
connected to proton transport in the inner protein structure.
It
is therefore desirable to combine the IR spectroscopy
capabilities with AFM-based techniques as already proposed by
many authors;31−34 however, it has not yet been applied to IR
difference-spectroscopy of TMPs.
layers,
In this work, we present
the first IR difference-nano-
spectroscopy measurement of conformational changes of a
photosensitive TMP using the bacteriorhodopsin mutant BR
D96N,35 one of the simplest and best-characterized protein
systems. Samples consist of individual patches of two native
purple-membrane monolayers (from here on referred as
2NPM), that is, 10 nm-thick stacks of two purple membranes,
deposited on ultraflat gold surfaces. We focused on 2NPM,
instead of monolayers, because we experimentally observed
that prolonged contact-mode AFM measurements of 5 nm-
thick membrane-monolayer patches yields to a degradation of
our sample, which was evident from topography maps. IR
difference-spectroscopy data are obtained with an AFM-based
nanoscale IR technique that exploits the sample photothermal
expansion due to IR absorption (ref 36, here abbreviated as
AFM-IR). We achieve significant improvement in the signal-to-
noise ratio (SNR) over existing IR nanospectroscopy by
carefully choosing the power and the frequency tuning range of
the IR laser, so as to obtain maximum photothermal expansion
for each wavelength and by exploiting the field enhancement in
the plasmonic nanogap that forms between the apex of a gold-
coated AFM probe-tip and an ultraflat gold surface.36 We
demonstrate that when the nanogap is filled with 10 nm-thick
2NPM patches, AFM-IR can measure the IR difference-
spectrum of BR embedded in native purple membranes, with a
dramatic increase in the sensitivity per unit molecule compared
with FTIR-based approaches. Analyzing our AFM-IR differ-
ence-spectra and comparing them to more conventional FTIR
difference-spectra acquired on the same samples, we could
identify two distinct photocycles that we attribute to TMPs
that either adhere or not to the metal surfaces.
Bacteriorhodopsin Photocycle. BR absorbs photon
energy via the retinal chromophore that is bound to a lysine
of one of the helices via a protonated Schiff base, and acts as a
proton pump moving protons from the cytoplasm to the
extracellular space.8 In Figure 1a we provide a simplified sketch
3105
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
Figure 1. (a) Sketch of the BR photocycle. BR*: dark state; letters
from J to O: intermediate states; H+ and gray dashed arrow: proton
release or capture. Green light initiates the photocycle (green arrow);
in the D96N mutant, blue light brings proteins that accumulate in M
back to BR* (blue arrow). (b) FTIR absorption spectrum of a thick
film of native purple membranes in dark condition Adark (gray curve,
left axis); corresponding difference-spectrum ΔA = Agreen − Ablue
(orange curve, right axis). (c) Time-dependent spectral plot of [A(t)
− Adark] acquired while alternately turning on and off the green and
blue LEDs after a 2 min dark period, showing reproducible protein
response. (d) Time-cuts taken from (c) at the negative peak of CC
retinal stretching (1525 cm−1) and at the positive peak of CO
stretching of the COOH group of the proton acceptor Asp-85 (1760
cm−1). Rise/decay times are a few seconds. A linear baseline has been
subtracted from time-cuts in panel (d).
transitions,
of
the BR photocycle. Photon absorption triggers a fast
(∼100−200 fs) retinal isomerization from an all-trans to a 13-
cis configuration, resulting in the intermediate state labeled J,
which is converted in K and then in L.37 After this sequence of
fast
the photocycle of wild-type BR can be
described with the following important steps:37 (i) proton
transfer from the Schiff base to the amino acid Asp-85 acting as
proton-acceptor (L to M transition); (ii) proton release toward
the extracellular side in M (gray dashed arrow in Figure 1a);
(iii) reprotonation of the Schiff base from the amino acid Asp-
96 acting as proton-donor during M to N transition; (iv)
reprotonation of Asp-96 from the cytoplasmatic surface in N
(gray dashed arrow between the N and O intermediate states
in Figure 1a); (v) retinal reisomerization (N to O transition),
restoration of the initial photon absorption properties and
deprotonation of Asp-85.
The capability of the AFM-IR technique to register the light-
induced conformational changes of TMPs has been tested on
the BR mutant D96N. In BR D96N, the replacement of the
proton-donor Asp-96 with Asn-96, which cannot donate
protons, prolongs the lifetime of the M intermediate state up
to several seconds.35 Since photoexcitation of M state with
blue light leads back to a dark-state-like intermediate BR*,8,38
one can then control the photocycle by alternating green/
yellow and blue illuminations to induce the conversion BR* →
Letter
M and M → BR*. In our case we alternate illumination with
green (λ = 565 nm) and blue (λ = 420 nm) light provided by
loosely focused LED sources in order to control the photocycle
of BR D96N and average the AFM-IR difference-spectra over
many cycles to improve the SNR (see Methods).
Fourier-Transform Infrared Difference-Spectroscopy.
In order to obtain a reference absorption spectrum Adark(ω) of
BR D96N membrane patches in the same form as those used
for AFM-IR spectroscopy (poorly hydrated films on solid
substrates), we performed FTIR transmission spectroscopy of
thick films of purple membranes deposited by controlled
solvent evaporation from membrane-patch suspensions on an
IR-transparent CaF2 window. In Figure 1b, the absorbance
Adark(ω) in dark condition (no illumination with visible light)
is reported (gray curve, left axis). The spectrum clearly shows
two amide bands associated with the helical structure of BR,
that is, the amide-I band at ∼1660 cm−1 and the amide-II band
at ∼1540 cm−1 predominantly assigned to the CO
stretching and N−H bending of
the peptide bond,
respectively.9 Less intense bands associated with the lipid
environment and with other amino-acid absorption peaks are
seen at lower frequencies. The ratio between amide-I and
amide-II intensities allows us to assume a partial stacking of the
purple membranes, whose membrane plane is oriented mostly
parallel to the surface of the CaF2 substrate (see Supporting
Information), although orientation of
the cytoplasmatic/
extracellular side cannot be assessed from IR data in general.
The normal-incidence ΔA spectrum in Figure 1b then refers
mostly to the contribution parallel to the membrane plane.
The FTIR difference-spectrum ΔA(ω) = Agreen(ω) −
Ablue(ω) is obtained by consecutively illuminating the sample
with green light and then with blue light, directly inside the
FTIR spectrometer. In ΔA(ω) (orange curve in Figure 1b,
right axis), the pattern of negative and positive peaks between
1100 and 1800 cm−1 is indicative of conformational changes of
both the retinal and the protein backbone. In particular, the
peak pattern between 1580 and 1700 cm−1 is indicative of
conformational changes of
the protein backbone and the
ethylenic modes of the retinal. The peaks between 1150 and
1300 cm−1, instead, are assigned to the C−C retinal stretching,
while the strong negative peak at 1525 cm−1 and the less
intense positive peak around 1560 cm−1 are assigned to the
CC retinal stretching mode under blue and green light,
respectively. The positive peak at 1760 cm−1 in ΔA(ω) of
Figure 1b originates from the CO stretching mode of the
COOH functional group of
the residue Asp-85, also an
established spectroscopic marker of the proton transfer from
the Schiff base to the proton acceptor Asp-85.13,39,40 The
spectral features in the ΔA(ω) curve are clearly present also in
the time-dependent difference spectral plots of Figure 1c,d, and
they can be interpreted in terms of accumulation of a M-like
intermediate state at green illumination,39−41 even if in poor
hydration conditions.29,42,43
Infrared Difference-Spectra Acquired with AFM Tip-
Enhanced Nanospectroscopy. The AFM-IR platform,
based on the coupling of an AFM and a tunable mid-IR
quantum cascade laser (QCL), exploits the photothermal-
induced mechanical resonance of the AFM cantilever, which,
for soft samples and mid-IR laser excitation, has been
definitively attributed to thermal expansion of the sample
after IR radiation absorption.36,44,45 The capability of AFM-IR
to probe light-induced conformational changes of TMPs is first
demonstrated with the setup in Figure 2a on the same purple
3106
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
Letter
Figure 2. (a) Schematic of experimental setup for difference IR nanospectroscopy. QCL: quantum cascade laser, AFM: gold-coated atomic force
microscope probe. (b) Gray curve: AFM-IR spectrum of a thick-film region (d = 1 μm) located in the purple-membrane assembly analyzed in
Figure 1; violet curve: corresponding AFM-IR difference-spectrum ΔA; orange curve: ΔA calculated for 70° incidence by combining the normal
incidence data in Figure 1b with oblique incidence FTIR data.
Figure 3. (a) Sketch of a 2NPM (d = 10 nm) located in the plasmonic nanogap between Au surface and Au-coated AFM tip; zoom of a BR
molecule67 (PDB ID:1FBB); color arrows indicate the direction of the main IR dipoles: CO stretching, amide-I band;51 N−H bending, amide-II
band;51 CC stretching of retinal.52 (b) AFM topography map of one 2NPM patch with superimposed topography profile. (c) Ratio between two
subsequent AFM-IR spectra of a 2NPM in the full QCL tuning range (gray) and in two low-sample-absorption subranges (light and dark violet)
changing the metal-mesh filters in front of the QCL so as to obtain maximum photothermal expansion in each range regardless the value of the
sample absorption coefficient. (d) AFM-IR spectra acquired on purple-membrane films of different thickness d. An offset is subtracted so as to
make each spectrum null at 1800 cm−1 and the relative intensities are then normalized at 1660 cm−1. E-field orientations are sketched for the p-
polarized incident QCL beam (Ep‑pol) and for the plasmonic nanogap (E2NPM). (e−g) Simulated maps of: (e) E-field modulus; (f) E-field
component normal to the membrane plane; and (g) temperature increase when the heat source is the absorbed radiation at 1540 cm−1.
membrane sample and visible illumination conditions as used
for FTIR. A sample area of thickness d ∼ 1 μm was located by
AFM topography, and AFM-IR spectra were acquired under
repeated 70 visible-light illumination cycles paying attention to
keep the AFM tip located at the same sample position (see
Methods). The gray curve in Figure 2b is the average of all
AFM-IR spectra, and it represents the best possible estimate of
the AFM-IR membrane absorption spectrum A(ω). AFM-IR
difference spectra determined per each green/blue illumination
cycle have been averaged to obtain ΔA shown in Figure 2b
(purple curve, right axis). In Figure 2b, we also plot the ΔA
spectrum for 70° incidence,
for an electric field
that
is,
3107
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
orientation as in the AFM-IR experiment on thick samples,
calculated by combining the normal incidence data in Figure
1b with oblique incidence FTIR data (see Supporting
Information). One finds a satisfactory agreement in the entire
frequency range, demonstrating that AFM-IR can indeed be
used to probe the conformational changes of photosensitive
proteins. In two frequency windows, the AFM-IR data are not
reported due to the finite tuning range of the available QCL
modules (see Methods).
Infrared Difference-Nanospectroscopy of a Single
Purple-Membrane Patch. AFM-IR difference-nanospectro-
scopy experiments were then performed on samples of
decreasing thickness down to 2NPM patches, that is, 10 nm-
thick stacks of two purple membranes. Note that, contrary to
the techniques where an electrical signal is probed,30 AFM-IR
is not sensitive to the difference between cytoplasmatic and
extracellular sides of the membrane. To obtain IR difference-
spectra of isolated patches instead of thick films, the SNR of
the AFM-IR technique needs to be increased by plasmonic
field enhancement. To this aim, BR D96N purple membranes
were drop-casted onto ultraflat template-stripped gold surfaces
(Figure 3a). It has been demonstrated that, by depositing thin
films on metallic surfaces, the gold-coated AFM tip provides a
strong enhancement of
the IR radiation intensity in the
nanogap between the metallic tip apex and the surface,36
together with a fully surface-normal orientation of the IR
electric field below the tip apex. IR absorption spectra of
molecular monolayers36 and single purple membranes19 have
already been reported in the literature with the same
resonantly enhanced AFM-IR technique46 employed here,
which makes use of QCLs pulsed at high repetition rate (∼200
kHz), also limiting the transient sample heating to a few
degrees.36,47,46 A representative AFM topography map of a
2NPM patch is shown in Figure 3b. It reveals a thickness of d =
10 nm corresponding to two overlapping purple membranes
and a flat top surface within the AFM sensitivity (RMS of 0.5
nm). In Figure 3d the AFM-IR spectra in the amide-I and
amide-II regions are shown for samples of decreasing thickness
and on different surfaces. One can appreciate that the SNR
observed for 2NPMs is not significantly different from that
observed for thick films, because the volume reduction of the
probed material is compensated by the increased field intensity
in the narrower plasmonic nanogap.48,49 The electromagnetic
and thermal simulations presented in Figure 3e−g quantita-
tively explain the origin of
the AFM-IR photothermal
expansion signal
from a 10 nm-thick film with dielectric
function equal to that of BR proteins. In Figure 3e,f, the field-
enhancement for the total electric field Etot and for the surface-
normal component Ez are compared using the same color
scale, and one can appreciate that a negligible difference exists
in the nanoscale sample volume precisely located under the
apex tip (approximately, a cylinder of 20 nm radius in which
the field is oriented normal to the surface). This is also the
volume of highest photoinduced temperature increase ΔT (see
Figure 3g). Simulations were performed with IR radiation at
1540 cm−1, but no major qualitative differences are expected
over the entire 1500−1800 cm−1 range, apart from the value of
ΔT which depends on both the values of the frequency-
dependent absorption coefficient and of the emitted laser
power. For focused laser power of 10 mW, one obtains from
simulations ΔT ∼ 2 K at 1540 cm−1 at the end of each 260 ns-
long QCL pulse. The plot in Figure 3g suggests that, in this
specific thin-film-on-metal-surface configuration and at wave-
Letter
lengths where the sample absorption coefficient is high, the
sample volume expected to generate the AFM-IR signal
corresponds to the cylinder of 20 nm radius and height 10 nm,
containing approximately 150 BR trimers, which represent the
unit cell of the native assembling of BR molecules in purple
membranes.50
The role of dipole selection rules is evident in Figure 3d.
The dipole of the amide-II and of the CC retinal stretching
are mostly parallel to the membrane plane (red and yellow
arrows in Figure 3a indicating ∼70° and ∼73° with the surface
for N−H bending51 and CC stretching,52
normal
respectively) and therefore the signal intensity is suppressed
in the 1500−1600 cm−1 range with decreasing thickness (red
shaded area in Figure 3d) due to increasingly surface-normal
direction of the radiation electric field. Vice versa, the amide-I
dipole is mostly oriented normal to the membrane plane51
(∼40° with the surface normal, green arrow in Figure 3a), so
the intensity of the 1660 cm−1 peak related to α helix increases
by 15% in the 2NPM (surface-normal electric field) if
compared to the p-polarized incidence on thick films, where
the electric field is oriented at 20° with the surface normal
(note that in Figure 3d, all spectra have been normalized to the
amide-I peak). The modifications of the amide-I line shape
with varying electric field orientation seen in Figure 3d are in
agreement with previous IR linear dichroism studies of BR49,52
and are discussed in the Supporting Information.
In the CO stretching region around 1760 cm−1, the
absolute value of the AFM-IR signal is very low, but the 45°
orientation of the CO bond of Asp-8553 makes it anyway
possible to observe relevant difference signals from 2NPMs.
Normally the laser power in AFM-IR is set to a low value to
avoid saturation at the frequency of peak absorption (here
1660 cm−1). In order to further boost the SNR beyond the use
of plasmonic field enhancement,36 we optimized the spectral
acquisition procedure by adjusting the IR laser power, up to 50
mW peak power, according to the value of the absorption and
to the frequency-dependent laser emission power in each
range. The result of this procedure is evident in Figure 3c
where we compare the AFM-IR reproducibility line obtained
for two spectra acquired in the full 1480−1800 cm−1 range
with a metal-mesh filter with transmittance of 0.015, with that
obtained in the frequency intervals 1480−1640 cm−1 and
1690−1800 cm−1, where the sample absorption is low, with a
second and third metal-mesh filters with transmittance of 0.045
and 0.11,
the
reproducibility lines passes from ∼3% at low power, to ∼1%
for high power, enabling us to reach a SNR of ∼103 by
averaging over 70 difference-spectra acquisitions (see Meth-
ods).
Discussion of Difference-Spectra. In Figure 4a,b the
AFM-IR difference-spectra ΔA are shown for different film
thickness d, from 1 μm down to the 2NPM (d = 10 nm). Apart
from the d = 1 μm spectrum taken from Figure 2b and
measured on a CaF2 window, all other spectra were taken on
ultraflat gold surfaces. Since AFM-IR technique does not
provide an absolute value of the absorbance, the ΔA spectra in
Figure 4 are normalized using an arbitrary normalization
coefficient so as to equal the difference spectra at 1590 cm−1, a
wavenumber at which no dichroic behavior
is usually
observed.53,54 Turning our attention to the AFM-IR spectra
for the 2NPM (d = 10 nm) in Figure 4a,b, we clearly observe
three main variations arising in ΔA: (1) the intensity of the
negative peak of CC of retinal at ∼1525 cm−1 is reduced in
respectively. The standard deviation of
3108
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
Figure 4. (a,b) AFM-IR difference-spectra of purple-membrane films
of thickness d deposited either on CaF2 or Au. Of the full QCL tuning
range 1480−1800 cm−1, two subranges with high SNR are shown in
(a) and (b) respectively. The worse SNR for d = 50 nm is due to poor
tip indentation in that specific measurement. On top of panels (a) and
(b) the ΔA for the 2NPM (d = 10 nm, red curve) and for the thick
film (d = 1 μm, violet curve) are superimposed to highlight the
changes arising in the 2NPM spectrum at 1525, 1545 and 1760 cm−1,
respectively labeled by the numbers 1, 2, and 3 used in the text to
identify them.
Letter
the 2NPM down to 50% compared to the thick film (d = 1
μm) intensity; (2) a new negative peak appears at ∼1545
cm−1; (3) the intensity of the positive peak of CO of Asp-85
at 1760 cm−1 is reduced in the 2NPM down to 50% of the d =
1 μm intensity (see Figure 4a,b where the same numbering has
been used to highlight the three variations). The spectrum for
d = 50 nm presents the same features as the 2NPM spectrum,
but less intense, suggesting that the observed changes are
indeed related to the film thickness and possibly to the nature
of the substrate. The d = 1 μm and d = 10 nm ΔA are expected
to slightly differ in terms of relative intensity of positive and
negative peaks due to the selection rule effect. Variations
between the difference-spectra acquired under different
anisotropic polarization conditions can also be due to dipole
tilts during the BR photocycle. In the Supporting Information,
these effects are discussed in detail: in particular, the effect of
the anisotropic field orientation is discussed by comparing
micro-FTIR difference spectra and calculations performed with
a hybrid quantum/classical approach.55 The variations (1),
(2), and (3) observed between the d = 1 μm and the d = 10
nm sample in Figure 4 are simply too large to be interpreted as
a selection rule effect or as a dipole tilt, or as a combination of
the two. Note that the electric field orientation changes only by
20° degrees between the d = 1 μm and the d = 10 nm sample,
and that both the CC bond of retinal (1525 cm−1 negative
peak) and the CO bond of Asp-85 (1760 cm−1 peak) do not
significantly change their orientation during the photo-
cycle.53,54
The suppression of the Asp-85 proton acceptor peak at 1760
cm−1 seen in the case of 2NPMs, that is, variation (3) in Figure
4, may indicate that light-activated proton transport is partly
Figure 5. (a) Comparison of the AFM-IR ΔA data for d = 1 μm (thick film) and d = 10 nm (2NPM patch) with the calculated ΔA from a Gaussian
line shape model of the CC absorption of retinal. The model absorption spectra are reported in panels (b) and (d) for thick film and 2NPM,
respectively. The simplified model photocycles for thick films and 2NPMs, as derived from the line shape analysis, are shown in panels (c) and (e),
respectively. In the bottom part of panel (a), two relevant difference-spectra calculated from literature FTIR data59 (see text) are also shown for
comparison (dotted curves).
3109
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
rather,
illumination.
prevented in proteins embedded in the 2NPM.17 This
interpretation would also be in agreement with the reduced
photoelectrical activity often observed for BR molecules
adhering to solid surfaces by AFM-based and electrochemistry
techniques.30,56,57 In particular, a clear distinction between the
photoelectrical activity of BR monolayers and multilayers has
been verified in ref 30, pointing to an alteration of BR
photocycle in the case of monolayers adhering on a solid
surface compared to multilayers and suspensions. The
impossibility for net proton transport through the membrane
within a photocycle is typically accompanied by modifications
of the retinal and protein conformational changes if compared
to those of the standard BR photocycle.8,58 We therefore
suggest that the variations (1) and (2) in Figure 4 can be
interpreted as the spectral markers of a modified conformation
of the retinal and/or retinal Schiff base and/or its environment,
occurring under blue light
In the standard
photocycle of BR, the retinal isomerization induced by photon
the CC vibrational
absorption results in the shift of
frequency from its value around 1525 cm−1 in the dark state
BR* to a new value around 1560 cm−1 for the M intermediate
state59 (see Figure 4a). In the 2NPM spectrum, we instead
observe a branching of the negative peak at ∼1525 cm−1 into a
second negative peak at ∼1545 cm−1, that is, variation (2).
This feature can be assigned to a modified frequency for the
retinal CC absorption in a fraction of proteins that under
blue illumination are not brought back to the dark state BR*
(CC absorption at ∼1525 cm−1), but
they
correspond to a nonstandard intermediate state characterized
by a CC absorption at ∼1545 cm−1. One can then deduce a
branching of the photocycle: only a fraction of proteins follows
the standard photocycle contributing to the net proton
transport through the membrane, while the remaining proteins
undergo a different photocycle branch with zero net proton
transport through the membrane, which is characterized by
retinal CC vibration frequency at ∼1560 cm−1 and ∼1545
cm−1 under green and blue light illumination, respectively. The
the Asp-85 peak at 1760 cm−1, which is
intensity of
proportional
to the net proton transport
through the
membrane, is found to be suppressed to 50% in the 2NPM
with respect to its thick-film value, therefore we can estimate
that 50% of proteins participate to the standard photocycle and
50% to the photocycle branch.
In Figure 5, we present a model of
the difference-
spectroscopy line shape of the CC retinal through the
branching of the photocycle for 2NPMs. For thick films, the
CC retinal absorption band is modeled as a Gaussian line
shape rigidly shifting from 1522 cm−1 under blue light to 1563
cm−1 under green light in the standard BR photocycle (Figure
5b). For 2NPMs under blue light illumination, the CC
retinal absorption band splits into two sub-bands of equal
intensity centered at 1522 and 1546 cm−1 (Figure 5d). The
peak at 1546 cm−1 represents the retinal CC absorption of
the protein fraction that are not brought back to the dark state
BR* but are instead trapped in a different intermediate state
that we label X*. The simplified model photocycles for the
thick film and the 2NPM are sketched in Figure 5c,e. For
2NPMs,
into two
subpopulations that we label M and M*. The CC retinal
absorption band of M and M* is identical, but upon absorption
of blue light, M follows the standard proton-pump photocycle,
while M* is transformed into the X* intermediate state. Our
photocycle branching model also assumes that proteins falling
the M intermediate population is split
Letter
in the X* intermediate state thermally decay into BR*, insofar
reproducing the same final state as the standard BR D96N
photocycle under green/blue illumination cycles.
In the
bottom part of Figure 5a, we also present a comparison of
our data and our model with the difference-spectra calculated
from the extensive FTIR data of ref 59 characterizing the IR
spectra of all BR intermediate states. Here we employ the
spectrum of the L intermediate as a model for the X* spectrum
due to the similarity of the CC absorption frequency with
the X* intermediate state,59 and we calculate the difference-
spectrum of the mixture of BR* and L states (dotted pink
curve in Figure 5a), to be compared with the experimental
spectrum of the mixture of BR* and X* states (red curve in
Figure 5a). Summarizing, IR nanospectroscopy has enabled us
to characterize at
the nanoscale a protein conformational
change, related to the inhibition of the proton pump activity at
the membrane monolayer level when in contact with metal
surfaces. The alteration of the photocycle for proteins located
in the tip−surface nanogap may be ascribed to several
mechanisms, such as the inhibition of the standard protein
conformational changes and reprotonation process due to less
flexibility of those proteins directly adhering to the surface, or
alteration of membrane surface potential due to the presence
of the gold surface and the metallic AFM tip. However,
extended experimental validations would be necessary in order
to achieve a complete interpretation of the results, and this
would be beyond the scope of this paper.
In conclusion, we have demonstrated that atomic-force-
microscope-assisted infrared difference-spectroscopy with
plasmonic field enhancement at the probe tip can be fruitfully
employed to identify subtle conformational changes of
transmembrane proteins in nanoscale samples. The bacterio-
rhodopsin photocycle is studied in 2NPM cell membrane
patches, whose dimensions are well below the diffraction limit
for conventional infrared difference-spectroscopy. The number
of protein molecules probed in the mid-infrared, of the order
of ∼500, is comparable to that reached by spectroscopies based
on visible light lasers such as Forster resonance energy transfer
and micro-Raman spectroscopy, also capable of measuring
conformational changes. However, in the case of light-sensitive
proteins, those methods, which are based on visible-light lasers,
cannot be directly employed to study the native membrane
photocycle. Here instead we could observe a branching of the
photocycle, accounting for less than 1% relative infrared
absorption change at the retinal stretching mode frequency of
1525 cm−1, when the bacteriorhopsin molecules are in close
contact to metal surfaces. The effect could be attributed either
to mechanical inhibition of the conformational change or to
electrostatic potential arising from physisorption. The obtained
results are relevant for bioelectronic and biophotonic devices,
in which rhodopsin molecules are put in contact with metal
surfaces (e.g., electrodes).
In perspective, the presence of an atomic force microscopy
probe in emerging nanoinfrared techniques could be exploited
well beyond its plasmonic field-enhancement capabilities. For
example, the metal-coated tip could be used to introduce an
electric potential with respect to the solid surface, so as to
provide electrophysiology techniques with nanoscale resolu-
tion and simultaneous spectroscopic information on trans-
membrane protein conformational changes. Also, the future
implementation of liquid environments34,60 and/or controlled
atmospheres in nanoinfrared techniques could enable the
nanospectroscopy study of
relevant biological processes
3110
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
involving transmembrane gate proteins, such as optogenetics
and photoinduced drug delivery.
■ METHODS
Sample Preparation. Patches of purple membranes
densely filled with BR D96N were purified as reported in ref
61 and stored in buffer solution (20 × 10−3 M Bis-Tris
propane, 100 × 10−3 M NaCl, 1 × 10−3 M MgCl2) at −80 °C.
Droplets of the suspension containing purple membranes were
either cast onto IR-transparent CaF2 windows or onto 1 cm ×
1 cm template-stripped gold chips (Platypus Technologies, 0.3
nm rms roughness) and let dry in air. After 10 min, the samples
deposited on ultraflat gold surfaces were rinsed with the buffer
solution (phosphate buffered saline at pH = 7.8) and
subsequently dried for 1 h in an atmosphere with humidity
below 10%.
Visible Illumination. Light to control the BR photocycle
was provided by two LEDs (Thorlabs M565L3, center
wavelength at 565 nm (green) and M420L3, 420 nm
(blue)). The LED output beams were made collinear using
lenses and a dichroic filter (Thorlabs - MD480) and then
focused on the sample. The power density for both LEDs was
∼50 mW/cm2. Sample was light-adapted before all experi-
ments, and therefore, the photocycle starts from homogeneous
isomeric dark state (all-trans).
FTIR Spectroscopy. FTIR measurements were performed
in transmission mode with either a Bruker Vertex 70v (Figure
1b) or a Bruker Vertex 80v (Figure 1c) equipped with
blackbody sources and HgCdTe detectors and in vacuum
condition. The transmitted intensity was collected both in dark
condition (Tdark) or under visible illumination (Tblue or Tgreen)
by averaging 512 interferometer scans at 8 cm−1 spectral
resolution. The Adark curve of Figure 1b was then calculated as
Adark = −log(Tdark/T0) using pristine CaF2 as the reference
channel (T0). The ΔA curve of Figure 1b was calculated as ΔA
− Ablue = log(Tblue/Tgreen), as there is no dependence of
= Agreen
the reference T0 on the visible-light illumination. The A(t) −
Adark spectra of Figure 1c were calculated as A(t) − Adark =
log(Tdark/T(t)) where Tdark
is the transmission spectrum
obtained by averaging over the spectra acquired during the
first 2 min in dark condition.
Oblique incidence micro-FTIR spectra were obtained using
a gas-purged Nicolet Continuum Infrared Microscope with a
HgCdTe detector, a 15× Cassegrain reflective objective with
spherical aberration compensation (numerical aperture NA =
0.58) and a knife-edge aperture of 100 μm × 100 μm.
Reflection of the gold surface covered with the membrane films
in dark condition Rdark was acquired first, then Rblue and Rgreen
were acquired. A total of 1024 interferometer scans at 8 cm−1
spectral
resolution were accumulated for each spectrum.
Reflection from the pristine gold surface R0 was used as
reference to calculate the double-transmission spectrum of the
membrane film Rdark/R0 and the absorption Adark = −log(Rdark/
R0). The micro-FTIR difference-spectra ΔA were calculated as
ΔA = Agreen
− Ablue = log(Rblue/Rgreen).
AFM-IR Setup. AFM-IR measurements were performed
using a platform of tip-enhanced IR nanospectroscopy based
on the photothermal expansion effect (NanoIR2, Anasys
Instruments). AFM-IR measurements were performed by
purging with dry air both the optics and the sample
compartment for many hours. The p-polarized beam from a
broadly tunable mid-IR QCL (MIRCATxB, Daylight Sol-
utions, with a spectral range 900−1800 cm−1) is tightly
Letter
focused onto the gold-coated tip of an AFM probe. The IR
beam impinges from the side at an angle of 70° to the surface
normal, that is, with the electric filed oriented at 20° with
surface normal. The laser provided 260 ns long light pulses at a
repetition frequency that was chosen as to be in resonance with
the second mechanical bending mode of the cantilever at ∼200
kHz (resonantly enhanced infrared nanospectroscopy,
REINS36). The laser power was adjusted using transmission
metal-mesh filters in front of the QCL output and it was in the
range 5−50 mW depending on frequency range and sample
absorption.
AFM-IR Difference-Spectroscopy: Spectral Acquisi-
tion Protocol. ΔA were obtained with AFM-IR by averaging
over 70 difference-spectra Agreen(ω) − Ablue(ω). The AFM-IR
spectra were acquired with 4 cm−1 steps, ∼ 0.3 s per step
subdivided in integration time and tuning time. The ΔA
spectra of Figure 2b and Figure 4 were obtained by stitching
the data acquired in different subranges where the laser power
to have comparable AFM-IR signal
was tuned in order
intensity regardless of
the sample
absorption that depends on electric field orientation. After each
5 green/blue acquisitions, a topography map of the sample
area was acquired, in order to place the AFM tip exactly at the
same position for all acquisitions.
The AFM-IR ΔA could not be acquired in two narrow
frequency ranges around 1430 and 1680 cm−1 corresponding
to the boundaries of the tuning ranges of the individual laser
chips in the MIRcat system, in which fluctuations of the laser
emission power are highest.
the absolute value of
Electromagnetic and Thermal Simulations. The finite-
difference time-domain method (FDTD Solutions, Lumerical
Inc., Canada) has been used to simulate a sample consisting of
a 10 nm-thick dielectric film (representing the 2NPM) on top
of a 100 nm-thick Au layer, on top of 2 μm-thick epoxy layer.
The AFM tip was modeled as a smoothed-tip cone with half
apical angle of 12° and apex curvature radius of 25 nm,
following scanning electron microscope images of the probe.
The tip is made of Si with a 25 nm-thick Au coating. The IR
beam is modeled as a weakly focused Gaussian beam incident
at 70° to the surface normal with peak power density of 8 ×
106 W/m2. The 10 nm-thick film is assumed to have complex
refractive index n = 1.7 + 0.45i, taken at the frequency of the
amide-II band (∼1540 cm−1) from the out-of-plane
component of the refractive index (i.e., orthogonal to the
membrane plane) reported in ref 62. Thermal simulations are
performed with a commercial heat transport solver (HEAT,
Lumerical Inc., Canada) by imposing a fixed temperature
(room temperature) boundary condition at both the lower
surface of the substrate and the upper surface of the AFM tip.
For the 2NPM, we used a thermal conductivity of 0.25 W
m−1K−1 (ref 63), a mass density of 1250 kg m−3 (calculated
assuming a protein mass density of 1400 kg m−3 (ref 64), a
lipid mass density of 800 kg m−3 (ref 65), a protein:lipid ratio
of 75:25), and a heat capacity of 2.4 × 103 J kg−1 K−1 (ref 66).
In order to qualitatively mimic the unavoidable presence of a
thermal
interface resistance between the tip and the
membrane, we introduce a thin (0.1 nm) film with a low
thermal conductivity of about 0.0025 W m−1 K−1. The
extension of the nanoscale colder spot located below the tip
apex (an effect of the thermal reservoir represented by the tip)
depends on the poorly characterized properties of such a
surface resistance, while the maximum photoinduced temper-
3111
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
ature increase reached in the membrane changes by less than
20% after the introduction of the surface insulating layer.
IR Data Analysis. A smoothing spline algorithm and a
baseline correction have been applied to obtain the final AFM-
IR ΔA curves reported in Figure 2 and Figure 4 (see
Supporting Infromation).
■ ASSOCIATED CONTENT
*S Supporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: 10.1021/acs.nano-
lett.9b00512.
Native purple-membrane orientation in our samples;
effect of dipole selection rules on the difference-spectra;
comparison between micro-FTIR data and spectra
calculation in the amide-I band frequency interval;
model for photocycle branching; smoothing and baseline
correction of AFM-IR data; O−H stretching region
(PDF)
■ AUTHOR INFORMATION
Corresponding Authors
*E-mail: [email protected].
*E-mail: [email protected].
ORCID
Peter Hegemann: 0000-0003-3589-6452
Laura Zanetti-Polzi: 0000-0002-2550-4796
Isabella Daidone: 0000-0001-8970-8408
Stefano Corni: 0000-0001-6707-108X
Paolo Biagioni: 0000-0003-4272-7040
Leonetta Baldassarre: 0000-0003-2217-0564
Michele Ortolani: 0000-0002-7203-5355
Author Contributions
M.B. and P.H. prepared the purple membrane samples. V.G.
and R.P. constructed the illumination setup and performed the
AFM-IR experiments. V.G., M.O., E.R., L.P., and U.S.
performed the FTIR experiments. I.D., L.Z.-P., and S.C.
calculated the protein vibrational spectra. F.R. and P.B.
performed electromagnetic and thermal simulations. M.O.
and L.B. designed the experiment and coordinated the work.
V.G. and M.O. wrote the manuscript. All authors contributed
extensively to the analysis and discussion of the results.
Funding
M.O. and L.B. acknowledge funding by Sapienza University of
Rome through the program Ricerca d'Ateneo 2017 (Grant No.
PH11715C7E435F41). S.C. acknowledges funding by the
European Research Council, under
the grant ERC-CoG-
681285 (TAME-Plasmons). The research leading to this result
has been supported by the project CALIPSOplus under the
Grant Agreement 730872 from the EU Framework Programme
for Research and Innovation HORIZON 2020. E.R. and P.H.
acknowledge support from the German Federal Ministry of
Education and Research (BMBF) Grant No. 05K16KH1. M.B.
and P.H. were supported by the German Research Society
(DFG, SFB1078) and P.H. was supported by the Hertie
foundation.
Notes
The authors declare no competing financial interest.
Letter
■ ACKNOWLEDGMENTS
Measurements were also carried out at the IRIS beamline at
Helmholtz-Zentrum Berlin fur Materialien und Energie.
■ REFERENCES
(1) Ernst, O. P.; Lodowski, D. T.; Elstner, M.; Hegemann, P.;
Brown, L. S.; Kandori, H. Microbial and Animal Rhodopsins:
Structures, Functions, and Molecular Mechanisms. Chem. Rev. 2014,
114 (1), 126−163.
(2) Wald, G. The molecular basis of visual excitation. Nature 1968,
219 (5156), 800.
(3) Scheib, U.; Broser, M.; Constantin, O. M.; Yang, S.; Gao, S.;
Mukherjee, S.; Stehfest, K.; Nagel, G.; Gee, C. E.; Hegemann, P.
Rhodopsin-cyclases for photocontrol of cGMP/cAMP and 2.3 Å
structure of the adenylyl cyclase domain. Nat. Commun. 2018, 9 (1),
2046.
(4) Oesterhelt, D.; Meentzen, M.; Schuhmann, L. Reversible
Dissociation of
the Purple Complex in Bacteriorhodopsin and
Identification of 13-Cis and All-Trans-Retinal as Its Chromophores.
Eur. J. Biochem. 1973, 40 (2), 453−463.
(5) Jiang, Y.; Lee, A.; Chen,
J.; Cadene, M.; Chait, B. T.;
MacKinnon, R. The Open Pore Conformation of Potassium
Channels. Nature 2002, 417 (6888), 523−526.
(6) Nagel, G.; Szellas, T.; Huhn, W.; Kateriya, S.; Adeishvili, N.;
Berthold, P.; Ollig, D.; Hegemann, P.; Bamberg, E. Channelrhodop-
sin-2, a Directly Light-Gated Cation-Selective Membrane Channel.
Proc. Natl. Acad. Sci. U. S. A. 2003, 100 (24), 13940−13945.
(7) Lozier, R. H.; Bogomolni, R. A.; Stoeckenius, W. Bacterio-
rhodopsin: A Light-Driven Proton Pump in Halobacterium
Halobium. Biophys. J. 1975, 15 (9), 955−962.
(8) Haupts, U.; Tittor,
J.; Oesterhelt, D. Closing in on
Bacteriorhodopsin: Progress in Understanding the Molecule. Annu.
Rev. Biophys. Biomol. Struct. 1999, 28 (1), 367−399.
(9) Barth, A. Infrared Spectroscopy of Proteins. Biochim. Biophys.
Acta, Bioenerg. 2007, 1767 (9), 1073−1101.
(10) Ritter, E.; Puskar, L.; Bartl, F. J.; Aziz, E. F.; Hegemann, P.;
Schade, U. Time-Resolved Infrared Spectroscopic Techniques as
Applied to Channelrhodopsin. Front. Mol. Biosci. 2015, 2, 38.
(11) Alexiev, U.; Farrens, D. L. Fluorescence Spectroscopy of
Insights and Approaches. Biochim. Biophys. Acta,
Rhodopsins:
Bioenerg. 2014, 1837 (5), 694−709.
(12) Kottke, T.; Lorenz-Fonfría, V. A.; Heberle, J. The Grateful
Infrared: Sequential Protein Structural Changes Resolved by Infrared
Difference Spectroscopy. J. Phys. Chem. B 2017, 121 (2), 335−350.
(13) Lorenz-Fonfría, V. A.; Saita, M.; Lazarova, T.; Schlesinger, R.;
Heberle, J. PH-Sensitive Vibrational Probe Reveals a Cytoplasmic
Protonated Cluster in Bacteriorhodopsin. Proc. Natl. Acad. Sci. U. S. A.
2017, 114 (51), E10909−E10918.
(14) Lorenz-Fonfría, V. A.; Schultz, B.-J.; Resler, T.; Schlesinger, R.;
Bamann, C.; Bamberg, E.; Heberle, J. Pre-Gating Conformational
Changes in the ChETA Variant of Channelrhodopsin-2 Monitored by
Nanosecond IR Spectroscopy. J. Am. Chem. Soc. 2015, 137 (5),
1850−1861.
(15) Nyquist, R. M.; Ataka, K.; Heberle,
J. The Molecular
Mechanism of Membrane Proteins Probed by Evanescent Infrared
Waves. ChemBioChem 2004, 5 (4), 431−436.
(16) Neubrech, F.; Huck, C.; Weber, K.; Pucci, A.; Giessen, H.
Surface-Enhanced Infrared Spectroscopy Using Resonant Nano-
antennas. Chem. Rev. 2017, 117 (7), 5110−5145.
(17) Jiang, X.; Zaitseva, E.; Schmidt, M.; Siebert, F.; Engelhard, M.;
Schlesinger, R.; Ataka, K.; Vogel, R.; Heberle, J. Resolving Voltage-
Dependent Structural Changes of a Membrane Photoreceptor by
Surface-Enhanced IR Difference Spectroscopy. Proc. Natl. Acad. Sci.
U. S. A. 2008, 105 (34), 12113−12117.
(18) Ataka, K.; Stripp, S. T.; Heberle, J. Surface-Enhanced Infrared
Absorption Spectroscopy (SEIRAS) to Probe Monolayers of
Membrane Proteins. Biochim. Biophys. Acta, Biomembr. 2013, 1828
(10), 2283−2293.
3112
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Nano Letters
J. AFM: A Nanotool
(19) Giliberti, V.; Badioli, M.; Nucara, A.; Calvani, P.; Ritter, E.;
Puskar, L.; Aziz, E. F.; Hegemann, P.; Schade, U.; Ortolani, M.;
Baldassarre, L. Heterogeneity of
the Transmembrane Protein
Conformation in Purple Membranes Identified by Infrared Nano-
spectroscopy. Small 2017, 13 (44), 1701181.
(20) Carpenter, E. P.; Beis, K.; Cameron, A. D.;
Iwata, S.
Overcoming the Challenges of Membrane Protein Crystallography.
Curr. Opin. Struct. Biol. 2008, 18 (5), 581−586.
(21) Pfreundschuh, M.; Harder, D.; Ucurum, Z.; Fotiadis, D.;
Muller, D. J. Detecting Ligand-Binding Events and Free Energy
Landscape While Imaging Membrane Receptors at Subnanometer
Resolution. Nano Lett. 2017, 17 (5), 3261−3269.
(22) Pfitzner, E.; Seki, H.; Schlesinger, R.; Ataka, K.; Heberle, J. Disc
Antenna Enhanced Infrared Spectroscopy: From Self-Assembled
Monolayers to Membrane Proteins. ACS Sensors 2018, 3 (5), 984−
991.
(23) Dufrene, Y. F.; Ando, T.; Garcia, R.; Alsteens, D.; Martinez-
Martin, D.; Engel, A.; Gerber, C.; Muller, D. J. Imaging modes of
atomic force microscopy for application in molecular and cell biology.
Nat. Nanotechnol. 2017, 12 (4), 295.
(24) Shibata, M.; Yamashita, H.; Uchihashi, T.; Kandori, H.; Ando,
T. High-Speed Atomic Force Microscopy Shows Dynamic Molecular
Processes in Photoactivated Bacteriorhodopsin. Nat. Nanotechnol.
2010, 5 (3), 208−212.
(25) Casuso, I.; Fumagalli, L.; Samitier, J.; Padros, E.; Reggiani, L.;
Akimov, V.; Gomila, G. Nanoscale Electrical Conductivity of the
Purple Membrane Monolayer. Phys. Rev. E 2007, 76 (4), 041919.
(26) Berthoumieu, O.; Patil, A. V.; Xi, W.; Aslimovska, L.; Davis, J.
J.; Watts, A. Molecular Scale Conductance Photoswitching in
Engineered Bacteriorhodopsin. Nano Lett. 2012, 12 (2), 899−903.
in Membrane Biology.
(27) Muller, D.
Biochemistry 2008, 47 (31), 7986−7998.
(28) Renugopalakrishnan, V.; Barbiellini, B.; King, C.; Molinari, M.;
Mochalov, K.; Sukhanova, A.; Nabiev, I.; Fojan, P.; Tuller, H. L.;
Chin, M.; Somasundaran, P.; Padros, E.; Ramakrishna, S. Engineering
a Robust Photovoltaic Device with Quantum Dots and Bacterio-
rhodopsin. J. Phys. Chem. C 2014, 118 (30), 16710−16717.
Janfaza, S. Efficient Nanostructured
(29) Mohammadpour, R.;
Biophotovoltaic Cell Based on Bacteriorhodopsin as Biophotosensi-
tizer. ACS Sustainable Chem. Eng. 2015, 3 (5), 809−813.
(30) He, T.; Friedman, N.; Cahen, D.; Sheves, M. Bacteriorhodopsin
Monolayers for Optoelectronics: Orientation and Photoelectric
Response on Solid Supports. Adv. Mater. 2005, 17 (8), 1023−1027.
(31) Berweger, S.; Nguyen, D. M.; Muller, E. A.; Bechtel, H. A.;
Perkins, T. T.; Raschke, M. B. Nano-Chemical Infrared Imaging of
Membrane Proteins in Lipid Bilayers. J. Am. Chem. Soc. 2013, 135
(49), 18292−18295.
(32) Amenabar,
I.; Poly, S.; Nuansing, W.; Hubrich, E. H.;
Govyadinov, A. A.; Huth, F.; Krutokhvostov, R.; Zhang, L.; Knez,
M.; Heberle, J.; Bittner, A. M.; Hillenbrand, R. Structural analysis and
mapping of individual protein complexes by infrared nanospectro-
scopy. Nat. Commun. 2013, 4, 2890.
(33) Ruggeri, F. S.; Longo, G.; Faggiano, S.; Lipiec, E.; Pastore, A.;
Dietler, G. Infrared nanospectroscopy characterization of oligomeric
and fibrillar aggregates during amyloid formation. Nat. Commun.
2015, 6, 7831.
(34) Ramer, G.; Ruggeri, F. S.; Levin, A.; Knowles, T. P.; Centrone,
A. Determination of Polypeptide Conformation with Nanoscale
Resolution in Water. ACS Nano 2018, 12 (7), 6612−6619.
(35) Holz, M.; Drachev, L. A.; Mogi, T.; Otto, H.; Kaulen, A. D.;
Heyn, M. P.; Skulachev, V. P.; Khorana, H. G. Replacement of
Aspartic Acid-96 by Asparagine in Bacteriorhodopsin Slows Both the
Decay of the M Intermediate and the Associated Proton Movement.
Proc. Natl. Acad. Sci. U. S. A. 1989, 86 (7), 2167−2171.
(36) Lu, F.;
Jin, M.; Belkin, M. A. Tip-Enhanced Infrared
Nanospectroscopy via Molecular Expansion Force Detection. Nat.
Photonics 2014, 8 (4), 307−312.
Letter
J. K. Proton transfers
(37) Lanyi,
in the bacteriorhodopsin
photocycle. Biochim. Biophys. Acta, Bioenerg. 2006, 1757 (8), 1012−
1018.
(38) Ludmann, K.; Ganea, C.; Varo, G. Back Photoreaction from
J. Photochem.
Intermediate M of Bacteriorhodopsin Photocycle.
Photobiol., B 1999, 49 (1), 23−28.
(39) Hendler, R. W.; Meuse, C. W.; Braiman, M. S.; Smith, P. D.;
Kakareka, J. W. Infrared and Visible Absolute and Difference Spectra
of Bacteriorhodopsin Photocycle Intermediates. Appl. Spectrosc. 2011,
65 (9), 1029−1045.
(40) Rodig, C.; Chizhov, I.; Weidlich, O.; Siebert, F. Time-Resolved
Step-Scan Fourier Transform Infrared Spectroscopy Reveals Differ-
ences between Early and Late M Intermediates of Bacteriorhodopsin.
Biophys. J. 1999, 76 (5), 2687−2701.
(41) Rothschild, K. J. FTIR difference spectroscopy of bacterio-
rhodopsin: toward a molecular model. J. Bioenerg. Biomembr. 1992, 24
(2), 147−167.
(42) Korenstein, R.; Hess, B. Hydration Effects on the Photocycle of
Bacteriorhodopsin in Thin Layers of Purple Membrane. Nature 1977,
270 (5633), 184−186.
(43) Lorenz-Fonfría, V. A.; Furutani, Y.; Kandori, H. Active Internal
Waters in the Bacteriorhodopsin Photocycle. A Comparative Study of
the L and M Intermediates at Room and Cryogenic Temperatures by
Infrared Spectroscopy. Biochemistry 2008, 47 (13), 4071−4081.
(44) Dazzi, A.; Glotin, F.; Carminati, R. Theory of
Infrared
Nanospectroscopy by Photothermal Induced Resonance. J. Appl. Phys.
2010, 107 (12), 124519.
(45) O'Callahan, B. T.; Yan, J.; Menges, F.; Muller, E. A.; Raschke,
M. B. Photoinduced Tip−Sample Forces for Chemical Nanoimaging
and Spectroscopy. Nano Lett. 2018, 18 (9), 5499−5505.
(46) Lu, F.; Belkin, M. A. Infrared Absorption Nano-Spectroscopy
Using Sample Photoexpansion Induced by Tunable Quantum
Cascade Lasers. Opt. Express 2011, 19 (21), 19942−19947.
(47) Mancini, A.; Giliberti, V.; Alabastri, A.; Calandrini, E.; De
Angelis, F.; Garoli, D.; Ortolani, M. Thermoplasmonic Effect of
Surface-Enhanced Infrared Absorption in Vertical Nanoantenna
Arrays. J. Phys. Chem. C 2018, 122 (24), 13072−13081.
(48) Baldassarre, L.; Giliberti, V.; Rosa, A.; Ortolani, M.; Bonamore,
A.; Baiocco, P.; Kjoller, K.; Calvani, P.; Nucara, A. Mapping the
Amide I Absorption in Single Bacteria and Mammalian Cells with
Resonant Infrared Nanospectroscopy. Nanotechnology 2016, 27 (7),
075101.
(49) Giliberti, V.; Badioli, M.; Baldassarre, L.; Nucara, A.; Calvani,
P.; Ritter, E.; Puskar, L.; Hegemann, P.; Schade, U.; Ortolani, M.
Vibrational Contrast Imaging and Nanospectroscopy of Single Cell
Membranes by Mid-IR Resonantly-Enhanced Mechanical Photo-
expansion. 2016 41st International Conference on Infrared, Millimeter,
and Terahertz waves (IRMMW-THz), Copenhagen, Denmark, Sept.
25−30, 2016; pp 1−3.
(50) Muller, D. J.; Heymann, J. B.; Oesterhelt, F.; Moller, C.; Gaub,
H.; Buldt, G.; Engel, A. Atomic force microscopy of native purple
membrane. Biochim. Biophys. Acta, Bioenerg. 2000, 1460 (1), 27−38.
(51) Marsh, D.; Pali, T. Infrared dichroism from the x-ray structure
of bacteriorhodopsin. Biophys. J. 2001, 80 (1), 305−312.
(52) Earnest, T. N.; Herzfeld, J.; Rothschild, K. J. Polarized Fourier
Transform Infrared Spectroscopy of Bacteriorhodopsin. Transmem-
brane Alpha Helices Are Resistant
to Hydrogen/Deuterium
Exchange. Biophys. J. 1990, 58 (6), 1539−1546.
(53) Earnest, T. N.; Roepe, P.; Braiman, M. S.; Gillespie, J.;
Rothschild, K. J. Orientation of the Bacteriorhodopsin Chromophore
Probed by Polarized Fourier Transform Infrared Difference Spec-
troscopy. Biochemistry 1986, 25 (24), 7793−7798.
(54) Nabedryk, E.; Breton, J. Polarized Fourier Transform Infrared
(FTIR) Difference Spectroscopy of the M 412 Intermediate in the
Bacteriorhodopsin Photocycle. FEBS Lett. 1986, 202 (2), 356−360.
(55) Amadei, A.; Daidone, I.; Di Nola, A.; Aschi, M. Theoretical-
computational modelling of infrared spectra in peptides and proteins:
a new frontier for combined theoretical-experimental investigations.
Curr. Opin. Struct. Biol. 2010, 20 (2), 155−61.
3113
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
Letter
Nano Letters
J. Enhanced photocurrent
(56) Mukhopadhyay, S.; Cohen, S. R.; Marchak, D.; Friedman, N.;
Pecht, I.; Sheves, M.; Cahen, D. Nanoscale electron transport and
photodynamics enhancement
in lipid-depleted bacteriorhodopsin
monomers. ACS Nano 2014, 8 (8), 7714−7722.
(57) Patil, A. V.; Premaruban, T.; Berthoumieu, O.; Watts, A.; Davis,
J.
in engineered bacteriorhodopsin
monolayer. J. Phys. Chem. B 2012, 116 (1), 683−689.
(58) Sasaki, J.; Shichidas, Y.; Lanyi, J. K.; Maeda, A. Protein Changes
Associated with Reprotonation of the Schiff Base in the Photocycle of
Asp96 -- >Asn Bacteriorhodopsin. The MN intermediate with un-
protonated Schiff base but N-like protein structure. J. Biol. Chem.
1992, 267 (29), 20782−20786.
(59) Zscherp, C.; Heberle, J. Infrared Difference Spectra of the
Intermediates L, M, N, and O of the Bacteriorhodopsin Photoreaction
Obtained by Time-Resolved Attenuated Total Reflection Spectros-
copy. J. Phys. Chem. B 1997, 101 (49), 10542−10547.
(60) Khatib, O.; Wood, J. D.; McLeod, A. S.; Goldflam, M. D.;
Wagner, M.; Damhorst, G. L.; Koepke,
J. C.; Doidge, G. P.;
Rangarajan, A.; Bashir, R.; et al. Graphene-based platform for infrared
near-field nanospectroscopy of water and biological materials in an
aqueous environment. ACS Nano 2015, 9 (8), 7968−7975.
(61) Oesterhelt, D.; Stoeckenius, W. Isolation of the cell membrane
of Halobacterium halobium and its fractionation into red and purple
membrane. Methods Enzymol. 1974, 31, 667−678.
(62) Blaudez, D.; Boucher, F.; Buffeteau, T.; Desbat, B.; Grandbois,
M.; Salesse, C. Anisotropic optical constants of bacteriorhodopsin in
the mid-infrared: consequence on the determination of alpha-helix
orientation. Appl. Spectrosc. 1999, 53 (10), 1299−1304.
(63) Nakano, T.; Kikugawa, G.; Ohara, T. A molecular dynamics
study on heat conduction characteristics in DPPC lipid bilayer. J.
Chem. Phys. 2010, 133 (15), 154705.
(64) Fischer, H.; Polikarpov, I.; Craievich, A. F. Average protein
density is a molecular-weight-dependent function. Protein Sci. 2004,
13 (10), 2825−2828.
(65) Muddana, H. S.; Gullapalli, R. R.; Manias, E.; Butler, P. J.
Atomistic simulation of lipid and DiI dynamics in membrane bilayers
under tension. Phys. Chem. Chem. Phys. 2011, 13 (4), 1368−1378.
(66) Morad, N. A.; Idrees, M.; Hasan, A. A. Specific heat capacities
of pure triglycerides by heat-flux differential scanning calorimetry. J.
Therm. Anal. 1995, 45 (6), 1449−1461.
(67) Zander, U.; Bourenkov, G.; Popov, A. N.; De Sanctis, D.;
Svensson, O.; McCarthy, A. A.; Round, E.; Gordeliy, V.; Mueller-
Dieckmann, C.; Leonard, G. A. MeshAndCollect: an automated
multi-crystal data-collection workflow for synchrotron macromolecu-
lar crystallography beamlines. Acta Crystallogr., Sect. D: Biol.
Crystallogr. 2015, 71 (11), 2328−2343.
3114
DOI: 10.1021/acs.nanolett.9b00512
Nano Lett. 2019, 19, 3104−3114
|
1509.04031 | 2 | 1509 | 2017-05-05T14:40:02 | Simulated single molecule microscopy with SMeagol | [
"physics.bio-ph"
] | SMeagol is a software tool to simulate highly realistic microscopy data based on spatial systems biology models, in order to facilitate development, validation, and optimization of advanced analysis methods for live cell single molecule microscopy data. Availability and Implementation: SMeagol runs on Matlab R2014 and later, and uses compiled binaries in C for reaction-diffusion simulations. Documentation, source code, and binaries for recent versions of Mac OS, Windows, and Ubuntu Linux can be downloaded from http://smeagol.sourceforge.net. | physics.bio-ph | physics |
Simulated single molecule microscopy with SMeagol
Martin Lind´en, Vladimir ´Curi´c, Alexis Boucharin, David Fange, & Johan Elf†
Department of Cell and Molecular Biology, Uppsala University, Sweden. †[email protected]
Summary: SMeagol is a software tool to simulate highly realistic microscopy data based on spa-
tial systems biology models, in order to facilitate development, validation, and optimization of
advanced analysis methods for live cell single molecule microscopy data.
Availability and Implementation: SMeagol runs on Matlab R2014 and later, and uses compiled
binaries in C for reaction-diffusion simulations. Documentation, source code, and binaries for re-
cent versions of Mac OS, Windows, and Ubuntu Linux can be downloaded from
http://smeagol.sourceforge.net.
Supplementary information: Supplementary data are available at Bioinformatics online.
Recent advances in single particle tracking (SPT) microscopy 1 make it possible to obtain tens
of thousands macromolecular trajectories from within a living cell in just a few minutes. Since
molecules typically change their movement properties upon interactions, these trajectories contain
information about both locations and rates of intracellular reactions. This information is unfortu-
nately obscured by physical limitations of the optical microscope and noise in detection systems,
making statistical methods development for SPT analysis a very active research field. Unbiased
testing and comparison of such methods are however difficult given the absence of in vivo data of
intracellular dynamics where the true states of interaction are known, a.k.a. the ground truth. A
common resort is to instead use simulated, synthetic, data. However, tests using such data give
unrealistically optimistic results if the simplifying assumptions underlying the analysis method are
exactly satisfied. The need for realistic simulations is long recognized in microscopy and systems
biology 2–12, but systematic combinations of the two are only currently emerging 13,14.
We present the SMeagol package, that has been developed to generate highly realistic single
molecule microscopy time-lapse image series aimed primarily at single particle tracking applica-
tions. The purpose of SMeagol is to enable realistic comparisons between the output of advanced
analysis methods and known ground truth. SMeagol includes an extended MesoRD 11 version for
simulation of 3D diffusion in cellular compartments, diffusion limited reaction kinetics, surface
adsorption, reactions in membranes, and other complex aspects of reaction diffusion kinetics that
do occur in cells, but are not considered in SPT analysis algorithms. In addition to the molecules'
trajectories, SMeagol integrates the 3D point spread function of the microscope, the kinetics of
photo-activation, blinking and bleaching of the simulated fluorophores, background noise, and
camera specific parameters (Figure 1, movie S1,S2). Great flexibility is allowed by the possibil-
ity to supply these characteristic parameters either as tabulated experimental data for a particular
optical setup, or as theoretical models. The combination of using reaction diffusion kinetics in
cellular geometries and physics-based simulations of the emission and detection processes makes
the images more realistic than the synthetic data used for example by 5.
SMeagol can be used to optimize imaging conditions for specific systems in silico and to
1
benchmark methods for SPT analysis in analogy with the methods that has been developed to
benchmark localization methods for non-moving single particles 7. In the supplementary material,
we explore the robustness against localization errors and motional blur of the vbSPT software,
which extracts multi-state diffusive models from SPT data 15, and find that these effects can induce
overfitting under certain conditions. In addition, we provide a number of examples highlighting
possibilities, limitation and computational requirements of the SMeagol simulation engine.
When combined with increasingly refined simulations of intracellular processes, photo-physics
and optics; live-cell microscopy is moving closer to methods in fundamental physics, where com-
bined simulation of physical processes and detection systems have guided experimental design and
data analysis for a long time.
Figure 1: Simulated microscopy with SMeagol. (a) Workflow from stochastic reaction-diffusion
simulations to images. (b) The microscopy simulation starts from trajectories generated by stochas-
tic reaction-diffusion simulations, fills in stochastic motion and photon emission events between
the trajectory points, and finally combines PSF and camera noise models to simulate realistic im-
ages. (c) Simulated microscopy of fluorescently labeled MinE proteins in the Min oscillatory sys-
tem. Left: Stochastic reaction-diffusion simulation. Mid columns: Simulated SPT microscopy us-
ing an actual experimental background noise movie with continuous illumination and 4 ms/frame.
Right: A simulation of continuous illumination and 1 s/frame renders a conventional (non-single
molecule) fluorescence microscopy time-lapse movie. See also Supplementary movies S1,S2, and
the Supplementary material for further details.
2
timebrightnessspace-discretized particle trajectorytimepositionilluminationilluminationdiffusive path(a)(b)5 s2.7 μmMin oscillationsexperimentalbackgroundsimulatedfluorophores+single particle tracking4 ms exposure4 ms2.7 μmtime-lapse1 s exposure5 sSimulated microscopy- stochastic reaction/diffusion simulations- diffusion-limited mesoscopic kinetics - arbitrary geometry, multiple compartments- stochastic fluorophore photophysics- background & camera noise models - point-spread function of the optical system(c)Browian bridgeinterpolationPSFzrzrzrPSFPSFSimulated reaction-diffusion kinetics2.7 μmilluminationtimeposition Acknowledgements
We thank Fredrik Persson and Elias Amselem for helpful discussions about microscopy.
Funding
This work was supported by the European Research Council, Vetenskapsrdet, the Knut and Alice
Wallenberg Foundation, the Foundation for Strategic Research, and the Swedish strategic research
programme eSSENCE.
References
1. Suliana Manley, Jennifer M. Gillette, George H. Patterson, Hari Shroff, Harald F. Hess, Eric Betzig,
and Jennifer Lippincott-Schwartz. High-density mapping of single-molecule trajectories with photoac-
tivated localization microscopy. Nat. Meth., 5(2):155–157, 2008. doi:10.1038/nmeth.1176.
2. Stephanie Fullerton, Keith Bennett, Eiji Toda, and Teruo Takahashi. Camera simulation engine enables
efficient system optimization for super-resolution imaging. volume 8228 of Proc. SPIE, pages 822811–
7, 2012. doi:10.1117/12.906346.
3. Tristan S. Ursell, Eliane H. Trepagnier, Kerwyn Casey Huang, and Julie A. Theriot. Analysis of surface
protein expression reveals the growth pattern of the gram-negative outer membrane. PLoS Comput.
Biol., 8(9):e1002680, 2012. doi:10.1371/journal.pcbi.1002680.
4. Cox, S. et. al. Bayesian localization microscopy reveals nanoscale podosome dynamics. Nat. Meth., 9
(2):195–200, 2012. doi:10.1038/nmeth.1812.
5. Chenouard, N. et. al. Objective comparison of particle tracking methods. Nat. Meth., 11(3):281–289,
2014. doi:10.1038/nmeth.2808.
6. Jzsef Sink, Rbert Kkonyi, Eric Rees, Daniel Metcalf, Alex E. Knight, Clemens F. Kaminski, Gbor
Szab, and Mikls Erdlyi. TestSTORM: Simulator for optimizing sample labeling and image acquisi-
tion in localization based super-resolution microscopy. Biomed. Opt. Express, 5(3):778–787, 2014.
doi:10.1364/BOE.5.000778.
7. Daniel Sage, Hagai Kirshner, Thomas Pengo, Nico Stuurman, Junhong Min, Suliana Manley, and
Michael Unser. Quantitative evaluation of software packages for single-molecule localization mi-
croscopy. Nat. Meth., 12(8):717–724, 2015. doi:10.1038/nmeth.3442.
8. Boris M. Slepchenko, James C. Schaff, John H. Carson, and Leslie M. Loew. Computational cell
biology: spatiotemporal simulation of cellular events. Annu. Rev. Biophys. Biomol. Struct., 31:423–
441, 2002. doi:10.1146/annurev.biophys.31.101101.140930.
9. Rex A. Kerr, Thomas M. Bartol, Boris Kaminsky, Markus Dittrich, Jen-Chien Jack Chang, Scott B.
Baden, Terrence J. Sejnowski, and Joel R. Stiles. Fast Monte Carlo simulation methods for biological
reaction-diffusion systems in solution and on surfaces. SIAM J. Sci. Comput., 30(6):3126–3149, 2008.
doi:10.1137/070692017.
3
10. K. Takahashi, S. Tanase-Nicola, and P. R. ten Wolde. Spatio-temporal correlations can drastically
change the response of a MAPK pathway. Proc. Natl. Acad. Sci. U.S.A., 107(6):2473–2478, 2010.
doi:10.1073/pnas.0906885107.
11. D. Fange, A. Mahmutovic, and J. Elf. MesoRD 1.0: Stochastic reaction-diffusion simulations in the
microscopic limit. Bioinformatics, 28(23):3155–3157, 2012. doi:10.1093/bioinformatics/bts584.
12. Steven S. Andrews. Spatial and stochastic cellular modeling with the Smoldyn simulator. In Jacques
van Helden, Ariane Toussaint, and Denis Thieffry, editors, Bacterial Molecular Networks, volume 804,
pages 519–542. Springer, New York, NY, 2012.
13. Juan Angiolini, Nicolas Plachta, Esteban Mocskos, and Valeria Levi. Exploring the dynamics of cell
processes through simulations of fluorescence microscopy experiments. Biophys. J., 108(11):2613–
2618, 2015. doi:10.1016/j.bpj.2015.04.014.
14. Masaki Watabe, Satya N. V. Arjunan, Seiya Fukushima, Kazunari Iwamoto, Jun Kozuka, Satomi Mat-
suoka, Yuki Shindo, Masahiro Ueda, and Koichi Takahashi. A computational framework for bioimaging
simulation. PLoS ONE, 10(7):e0130089, 2015. doi:10.1371/journal.pone.0130089.
15. Fredrik Persson, Martin Lindn, Cecilia Unoson, and Johan Elf. Extracting intracellular diffusive
states and transition rates from single-molecule tracking data. Nat. Meth., 10(3):265–269, 2013.
doi:10.1038/nmeth.2367. Software: http://sourceforge.net/projects/vbspt/.
16. Winston C. Chow. Brownian bridge. WIREs Comp. Stat., 1(3):325–332, 2009. doi:10.1002/wics.38.
17. Maximilian H Ulbrich and Ehud Y Isacoff. Subunit counting in membrane-bound proteins. Nat. Meth.,
4(4):319–321, 2007. doi:10.1038/NMETH1024.
18. A. Finney and M. Hucka. Systems biology markup language: Level 2 and beyond. Biochem. Soc. T.,
31(6):1472–1473, 2003. doi:10.1042/bst0311472.
19. S F Gibson and F Lanni. Experimental test of an analytical model of aberration in an oil-immersion
objective lens used in three-dimensional light microscopy. J. Opt. Soc. Am. A, 9(1):154–166, 1992.
doi:10.1364/JOSAA.9.000154.
20. H. Kirshner, F. Aguet, D. Sage, and M. Unser. 3-D PSF fitting for fluorescence microscopy:
im-
J. Microsc., 249(1):13–25, 2013. doi:10.1111/j.1365-
plementation and localization application.
2818.2012.03675.x.
21. William H. Press, Brian P. Flannery, Saul A. Teukolsky, and William T. Vetterling. Numerical Recipes
in C: The Art of Scientific Computing. Cambridge University Press, Cambridge ; New York, 2 edition,
1992.
22. K. C. Huang, Y. Meir, and N. S. Wingreen. Dynamic structures in Escherichia coli: Spontaneous
formation of MinE rings and MinD polar zones. Proc. Natl. Acad. Sci. U.S.A., 100(22):12724–12728,
2003. doi:10.1073/pnas.2135445100.
23. David Fange and Johan Elf. Noise-induced Min phenotypes in E. coli. PLoS Comput. Biol., 2(6):e80,
2006. doi:10.1371/journal.pcbi.0020080.
4
24. David Fange, Otto G. Berg, Paul Sjberg, and Johan Elf.
Stochastic reaction-diffusion ki-
Proc. Natl. Acad. Sci. U.S.A., 107(46):19820–19825, 2010.
netics in the microscopic limit.
doi:10.1073/pnas.1006565107.
25. Sang-Hyuk Lee, Jae Yen Shin, Antony Lee, and Carlos Bustamante. Counting single photoactivat-
able fluorescent molecules by photoactivated localization microscopy (PALM). Proc. Natl. Acad. Sci.
U.S.A., 109(43):17436–17441, 2012. doi:10.1073/pnas.1215175109.
26. Andrew J. Berglund. Statistics of camera-based single-particle tracking. Phys. Rev. E, 82(1):011917,
2010. doi:10.1103/PhysRevE.82.011917.
27. Hendrik Deschout, Kristiaan Neyts, and Kevin Braeckmans. The influence of movement on the local-
ization precision of sub-resolution particles in fluorescence microscopy. J. Biophotonics, 5(1):97–109,
2012. doi:10.1002/jbio.201100078.
28. Johan Elf, Gene-Wei Li, and X Sunney Xie. Probing transcription factor dynamics at the single-
molecule level in a living cell. Science, 316(5828):1191–1194, 2007. doi:10.1126/science.1141967.
29. Kim I. Mortensen, L. Stirling Churchman, James A. Spudich, and Henrik Flyvbjerg. Optimized lo-
calization analysis for single-molecule tracking and super-resolution microscopy. Nat. Meth., 7(5):
377–381, 2010. doi:10.1038/nmeth.1447.
30. David MacKay. Information theory, inference, and learning algorithms. Cambridge University Press,
2003.
31. Alexandre Lazarescu and Kirone Mallick. An exact formula for the statistics of the current in the
TASEP with open boundaries. J. Phys. A: Math. Theor., 44(31):315001, 2011. doi:10.1088/1751-
8113/44/31/315001.
5
Supplementary material
S1 Simulated microscopy with SMeagol
SMeagol is a Matlab software suite that simulates microscopy images of randomly moving par-
ticles using two main ingredients: diffusive motion and stochastic photon emission events.
In
addition, noise from various sources (camera, background, optics, blinking and photobleaching,
etc.) can be included in a modular and flexible way. This makes it possible to evaluate how dif-
ferent aspects of the biological-, reporter- and detection- system influence the overall result of the
experiment.
Reactions and random motion in arbitrary geometries The microscopy simulation part of
SMeagol uses input trajectories in the form of a list of times, positions, particle id numbers, and
diffusive states, and can also include the time for creation and destruction of particles. The input
trajectories are interpolated using Brownian bridges 16 to generate individual emission positions of
every simulated photon. Brownian bridges simulate free diffusion, but the input data need not.
Thus, with a fine time step one can use SMeagol to simulate general types of motion. For an
example, see Sec. S5b and Fig. S4.
We have extended the reaction-diffusion simulation software MesoRD 11 to keep track of
individual molecules and write trajectories in the appropriate input format, and incorporated it in
SMeagol, but it is also possible to use indata from other sources. SMeagol's trajectory data format
is described in the software manual.
Tunable photophysics In parallel with the diffusion process, each particle in the simulation goes
through a simulated stochastic photophysical process which includes activation, Markovian transi-
tions between multiple photophysical states with different photon emission intensities, and eventu-
ally irreversible photobleaching. Short exposures can be simulated by setting an exposure time tE
shorter than the frame time ∆t, and photophysical effects of excitation, by specifying different pho-
tophysical transition rates during the illuminated (0 ≤ t < tE) and non-illuminated (tE ≤ t < ∆t)
phases. Separating photophysics and molecular diffusion makes it possible to simulate the same
reaction-motion trajectory under a wide range of experimental conditions.
EMCCD noise and background The emitted photons are mapped to the camera chip using a
point-spread function (PSF) model, combined with simulated EMCCD 17 and background noise.
The microscopy image, which is written to tif-stacks for further analysis.
Flexible, modular and user-friendly SMeagol is designed to allow easy incorporation of exper-
imental data and theoretical parameters at many levels. Thus, the user can specify arbitrary fluo-
rophore activation and photophysical kinetics, and also incorporate custom-written Matlab routines
for PSF and background models, by extending existing template files. It is also possible to use the
6
independently measured PSF for a specific optical set-up, or background movies from a specific
sample. Stochastic reaction-diffusion models are described using the systems biology markup lan-
guage (SBML) 18 with extensions to spatial models 11. The trajectories and the different building
blocks of the microscopy simulation are then combined and parameterized using either a graphical
user interface, parameter text files (runinput files), or Matlab structs.
S2 Point-spread function (PSF) model
For all microscopy simulations described here, we used a rotationally symmetric PSF model con-
structed from the Gibson-Lanni model 19, as implemented in PSFgenerator 20.
We simulated the Gibson-Lanni PSF model with high resolution for 584 nm light, NA=1.4,
and otherwise default settings PSFgenerator, computed the cumulative radial distribution function
(CRDF) for different focal planes and radii up to 5 µm, and constructed a Matlab look-up table for
the inverse CRDF. An individual photon emitted at xem., yem., zem. were then simulated as detected
at position
xdet.=xem. + r cos ν,
ydet.=yem. + r sin ν,
(S1)
(S2)
where the angle ν is uniformly distributed on [0, 2π], and r is sampled using the inverse transform
method 21, i.e.,
(S3)
with u uniformly distributed on (0, 1). The total intensity of the spot did not vary significantly in
the region (zem. − zfocus < 400 nm) relevant for our simulations.
r = CRDF −1(u; zem.),
In focus, the above PSF model has a standard deviation of about 335 nm. This is largely due
to large shoulders of the PSF, and the width of the central peak is about 95 nm.
S3 Simulated experiments with Min oscillations
To generate Fig. 1c, we simulated a minimal stochastic model of the Min oscillation cycle 22,23,
implemented using scale-dependent mesoscopic reaction rate constants 24.
Reaction-diffusion simulation To match experimental microscopy data, we choose an E. coli-
like geometry consisting of a cylinder with spherical end caps, with outer diameter 1 µm and total
length 2.7 µm. The outer 35 nm of the cell was modeled as the membrane region. In this geometry,
the oscillations have a period of 25-30 s.
We used a 10 nm spatial discretization, and initial conditions placing 1716 MinD and 483
7
MinE randomly in the membrane region at one half of the cell. After a 300 s simulation to reach
steady state, we ran a tracking production run, collecting snapshots (used for the left column of
Fig 1c) every 0.25 s, and tracking positions of all species involving MinE every 5 ms.
MinE SPT simulation (Fig 1c, mid columns, and supplementary movie S1). We simulated a sin-
gle particle tracking experiment with 4 ms frame rate and continuous illumination. The coordinate
system of the input trajectories where rotated and translated to fit a brightfield image of an E coli
bacterium expressing no fluorophores (Fig. 1c), and a fluorescence time-lapse movie of the same
cell was used as a background.
The Gibson-Lanni PSF model was used as described above, and an experimentally param-
eterized blink/bleach model for mEos2 25, with a bright state intensity of 125000 photons/s that
yields on average 500 photons per spot and 4 ms frame if no blinking occurs. Photoactivation
events where simulated every 10 s, with an activation probability of 20% per unconverted molecule.
We simulated an EMCCD gain of 40 (SMeagol inverse gain camera.alpha=1/40), i.e.,
every photon generates an exponentially distributed number of image counts with mean value 40.
Offset and readout noise are in this case included in the experimental background.
MinE long-exposure simulation (Fig 1c, right column, and supplementary movie S2). For the
long-exposure simulations, we randomly activated 50% of the MinE-containing molecules with a
constant emission intensity of 60 photons/s and no blinking or bleaching. For background, we set
a uniform intensity of 1 photon/pixel. We used continuous illumination and a frame rate of 1 Hz,
the Gibson-Lanni PSF model described above, an EM gain of 40, offset 100, and readout noise
with std. 4.
The reaction-diffusion model and trajectory output files, and the SMeagol runinput files for
these simulations, are included in Supplementary dataset S1.
S4 Analysis of diffusion with blur and localization errors
To illustrate how SMeagol could be used to evaluate analysis methods for live cell single particle
tracking experiments, we explore the ability of the vbSPT software 15 to correctly identify the
number of diffusive states in SPT data. We simulated an SPT experiment with normal diffusion
at a rate of D = 1 µm2 s−1 sampled every 3 ms in an E. coli geometry with varying fluorophore
brightness and exposure time (see Supplementary movie S3). vbSPT assumes that data come
from a Markov model with state-dependent diffusion constants, i.e., a model that neglects, e.g.,
z-dependent localization errors 27, motional blur 26,27, and confining effects of the cell boundaries.
The simulation thus contains many real features not included in the analysis model, which could
lead vbSPT to overfit the data by incorrectly identifying more than one diffusive state. The question
8
Figure S2: Sensitivity of vbSPT to imaging artifacts.
(a) Depending on the experimental pa-
rameters, vbSPT correctly finds one diffusive state (pink) or incorrectly finds two diffusive states
(black). The experimental parameters are spot intensity, exposure times tE, and localization error
threshold. (b) Spot selection criteria illustrated on a distribution of estimated pointwise localization
errors, with σmax/σmode being the error threshold in (a). (c) Comparison of the diffusion constants
found by vbSPT with an effective diffusion constant defined via the theoretical step length vari-
ance 26.
(d) Comparison of the true and estimated (as in (b)) root-mean-square error for every
trajectory in (a).
9
true err. [nm]102030est. err. [nm]102030(d)52030017010075426420tE[ms](a)(c)Deff.[µm2/s]DvbSPT[µm2/s]0.511.520.60.811.21.4photons/spoterror threshold(b)1 state2 states050100log(freq.)10-410-310-210-1σmodeest. error [nm] Table S1: Settings for the microscopy simulations of simple diffusion (Fig. S2).
sample time : 3 ms per frame
photophysics : constant emission intensity, no bleaching
ROI
: 80 nm pixels, 55 × 20 pixel ROI, focal plane in the mid-
: plane of the bacteria
: offset=100, readout noise std.=4,
: EM gain=20 counts/photon
camera
background : constant, on average 1 photon/pixel per frame
is under which experimental conditions this is likely to happen.
Reaction-diffusion simulation We simulated simple diffusion of a single fluorescent particle in
an E coli-like geometry, built as a cylinder with length 3 µm and diameter of 0.8 µm, plus spherical
end caps. We used 10 nm voxels, and wrote particle positions every 7 ms.
SMeagol simulation We generated SPT movies with a frame duration of 3 ms in a range of
imaging conditions from the above diffusive trajectory, by varying the exposure time tE and the
average number of photons per spot (Nphot. = tE × emission intensity). In particular, we used all
combinations of tE = 0.5 ms, 1 ms, 2 ms, 3 ms, and Nphot. = 75, 100, 300, 520. Other microscopy
parameters used in all cases are summarized in table S1.
Estimated number of states Fig. S2a shows number of states learned by vbSPT as a function of
three tuning parameters: the exposure time tE, the average number of photons per spot, and the
maximally allowed pointwise localization error (Fig. S2b). The correct and overfitting conditions
are indicated in purple and black, respectively. In general, all three parameters influence the over-
fitting tendency in a non-trivial way. Continuous illumination (exposure time=frame time) leads
to overfitting in almost all conditions, but a modest decrease in exposure time using, e.g., strobo-
scopic illumination 28, leads to significant improvement due to decreased motional blur. We also
note that if the number of photons per localized molecule is limited, it is advantageous to include
only positions with high localization accuracy.
Estimating the diffusion constant As the analysis model of vbSPT neglects both localization
errors and motional blur, one should not take the numerical estimates of the diffusion constants at
face value. However, the estimates can be interpreted using a theory of motional blur for diffusing
particles 26. A closer inspection of the analysis algorithm 15 shows that vbSPT effectively looks at
the step length variance, which in the absence of localization error and blur is simply
(cid:10)∆x2(cid:11) = 2DvbSPT∆t.
10
(S4)
(cid:10)∆x2(cid:11) = 2D∆t(1 − 2R) + 2(cid:10)σ2
(cid:11) ,
A more detailed model that includes motional blur and localization errors 26 instead predicts
where R = tE
Eliminating the step length variance from the above equations, we find that
3∆t is the motional blur coefficient, and (cid:104)σ2
x
(S5)
x(cid:105) is the mean-square localization error.
DvbSPT = Deff. ≡ D(1 − 2R) +
(cid:104)σ2
x(cid:105)
∆t
.
(S6)
Fig. S2c plots the effective diffusion constant Deff. vs. the posterior mean of DvbSPT for the different
data sets (using our estimated average localization errors), and we see that the prediction of Eq. (S6)
is reproduced well when a single diffusive state is correctly identified.
Point localization We localized the spots using a maximum-likelihood fit of a symmetric Gaus-
sian plus constant background to a 7-by-7 fit region, using the EMCCD likelihood function of Ref.
29, with the offset and gain settings of table S1. Each spot is thus described by 5 fit parameters:
background b, spot amplitude N, spot standard deviation s, and spot position µx, µy. We used
Matlab's built-in function fminunc for numerical optimization of the log likelihood, which was
parameterized to allow only positive values of b, N, and s2, and used the true spot positions, and
the average PSF width and amplitude to construct an initial guess for each fit. To minimize con-
finement artifacts from the cell walls, we analyzed motion and uncertainties along the long cell
axis (x coordinate) only.
Estimating point-wise localization uncertainty Due to fluorophore motion during exposure, ran-
dom photon emission, z-dependence of the PSF, etc., the quality of the fit varies from spot to spot.
We used a Laplace approximation 30 (also known as the saddle point approximation in statistical
physics) of the likelihood function to estimate the localization uncertainty of individual spots, as
follows: Let IM denote the fit region of the image used for localization, and θ = (µx, µy, . . .)
the fit parameters. We approximate the likelihood function p(IMθ) by a Gaussian centered at the
maximum likelihood estimate θ∗ using a Taylor expansion in (θ − θ∗),
p(IMθ) ≈ exp
ln p(IMθ∗) + ∇θ ln P (IMθ)(cid:12)(cid:12)θ∗
(cid:125)
(θ − θ∗)
(cid:123)(cid:122)
(cid:16)
(cid:124)
=0
(S7)
where the first order term disappears since θ∗ is a local maximum, and the second order term is
given by the inverse covariance matrix,
,
(θ − θ∗)T Σ−1(θ − θ∗) + . . .
(cid:17)
− 1
2
(cid:12)(cid:12)(cid:12)(cid:12)θ∗
Σ−1 = −∂2 ln p(IMθ)
∂θ2
.
(S8)
This can be interpreted as the Bayesian posterior distribution (with a flat prior). The uncertainty of
the parameters are then characterized by their posterior covariances 30. In particular, the posterior
variance of µx is approximately given by Σµx,µx.
11
As a simple test of this estimator, we compare the true and estimated average root-mean-
square (RMS) error for all points in every trajectory (Fig. S2d). We find it to be correct on average,
i.e.,
(S9)
for true RMS errors (cid:46) 20 nm, but biased downwards for larger errors, probably because the Gaus-
sian approximation of the posterior density (Eq. (S7)) is inaccurate in those cases.
x − µx,true)2(cid:11) ≈ (cid:104)Σµx,µx(cid:105) ,
(cid:10)(µ∗
Selection criteria To build diffusion trajectories for vbSPT analysis, we first discarded spots
where the numerical optimization failed. We then built a histogram of estimated standard errors
σx =(cid:112)Σµx,µx, and identified the most likely estimated error σmode. One such histogram is shown
in Fig. S2b. Finally, we discarded spots that had either σx < σmode/2 as being unrealistically pre-
cise, or σx > σmode × (error threshold), using error thresholds in the range 1.3 − 6 as our third
control variable in Fig. S2a. The highest threshold of 6 included practically all spots.
The fraction of retained spots as well as average trajectory length vary with both simulation
parameters and error threshold, but for this experiment we generated enough images to construct
data sets with 80 000 diffusive steps for all conditions.
vbSPT settings For the trajectory analysis, we used vbSPT 1.1.2 15. For each data set, we ran 20
independent runs of the greedy model search algorithm with up to 15 hidden states. We used an
inverse gamma prior with mean value 1 µm2 s−1 and strength 5 (std. ≈ 0.6 µm2 s−1) for the diffu-
sion constants (with 80 000 diffusive steps in the trajectory, this prior is completely overwhelmed
by the data), flat Dirichlet priors for the initial state and state-change probability distributions, and
a Beta distribution with mean 0.02 s and std. 2 s for the mean dwell time of the hidden states. (For
detailed definitions, we refer to the vbSPT manual).
S5 Misc. model examples
Here, we briefly describe some additional model examples, to illustrate the capability and limita-
tions of SMeagol in various settings. Source files for these examples are included in Supplementary
data S5.
S5a. Affinity to cell poles To illustrate the geometric modeling capabilities of MesoRD and
SMeagol, we construct a simplified model of a diffusing cytoplasmic protein P with specific affinity
to binding partners localized at the cell poles.
We use an E coli-like rod shape, where cytoplasm is modeled by a union of a cylinder and
two spheres with radii 465 nm.
In the cytoplasm, the proteins diffuse with diffusion constant
2.5 µm2 s−1. The membrane region is modeled as an additional 35 nm outer shell, from which
polar regions in the form of small spherical caps are created, as shown in Fig. S3a.
In these
12
Figure S3: Simulating a diffusing protein with polar binding regions. a) Model geometry, with the
cytosol (black) inside a thin membrane region (blue) that also contains polar caps (red,magenta)
where binding receptors are localized. b) Snapshot from a simulated SPT experiment, with both
polar bound (red arrows) and freely diffusing molecules visible.
Figure S4: Simulating non-diffusive transport. (a) Snapshot from TASEP simulation (black, parti-
cles) on a helical path (red). (b) Simulated images with focal plane placed near the upper, middle,
and lower part of the helical curve emphasize different parts of the path.
caps, we assume a constant concentration [R] of receptors. Then, the binding rate can be written
r = ka[R][P ], where ka is the association rate constant, and the unbinding rate constant is kd. We
choose ka[R] = 20 s−1, kd = 0.01 s−1, and diffusion constant 0.01 µm2 s−1 for the bound complex
(confined to the membrane caps).
We ran 6 s of stochastic reaction-diffusion simulations starting with 125 P molecules uni-
formly distributed outside the polar caps, and after a burn-in of 2 s, wrote positions to file every
1 ms. For the microscopy simulation, we randomly activated 5% of the molecules, and used the
same settings as in the simple diffusion experiment above (table S1), except slightly larger re-
gion of interest, a fluorophore intensity corresponding to giving on average 270 photons/frame and
fluorophore, and photobleaching with a mean lifetime of 1 s. A snapshot is shown in Fig. S3b.
13
a)b)a)b)-400-2000400200z400200y0x-20015001000-400500focus belowfocus midcellfocus above Figure S5: Simulating a diffusing transcription factor in a yeast cell. (a) The model geometry
includes a cytoplasm compartment with two buds (red) and a spherical nucleus (blue). Also shown
is the bacterial model of Fig. S3 (black). (b,c) A snapshot and 500-frame average, respectively,
from a 5 ms/frame SPT simulation. (d) Wall time versus voxel size for a 5 s stochastic reaction
diffusion simulation tracking all proteins (RD all), and microscopy image simulation. The former
closely follows the expected 1/∆x2 scaling, while the latter is essentially constant. (e) Wall times
for the ∆x = 30 nm case from (d), but with varying number of involved proteins. "RD none" and
"RD 25" refers to stochastic reaction diffusion simulations with tracking deactivated and tracking
only 25 proteins irrespective of total copy number.
S5b. Non-diffusive transport along helical membrane filaments To illustrate the possibility of
simulating more complex motion than diffusion using SMeagol, we constructed an active transport
model of particles moving on a helical path (Fig. S4). More specifically, we wrote a Matlab script
to simulate a totally asymmetric exclusion process (TASEP 31) with open boundaries: Particles
(black) are created at the left end of the helix in Fig. S4a, walk forwards in 36 nm steps along the
helical path (red) with a rate of 10 s−1 (under a site exclusion constraint), until they fall off at the
right end. The insertion rate was 2 s−1, putting the TASEP in the low density phase 31. We then
wrote particle coordinates to a trajectory text file at regular intervals (all in the same chemical state),
and fall-off events as particle destructions in the reactions text file. For microscopy simulations,
we set D = 0, which disables the Brownian bridges and leads to linear interpolation between the
trajectory coordinates. Simulation scripts and runinput files are included in data set S5.
14
a)c) 500 frame averageb) snapshot153060120240∆x [nm]10-1100101102wall time [min]t∝∆x-2RD allimaged)e)102103N100101102wall time [min]RD allRD noneRD 25imagequadr.linear S5c. A large cell To illustrate the computational requirements of SMeagol, we use a simple
model of transcription factor (TF) motion in a budding yeast cell. The model has a cytoplasmic
compartment, made of two spheres joined to an ellipse with a spherical nucleus compartment
inside. As seen in Fig. S5a, it is several times larger than the bacterial-like examples described so
far.
The main computational load for fine-scale MesoRD simulations consists of diffusive motion
(total diffusive hopping rate is 6D/∆x2), so we use a simple kinetic model where the TF has a
'free' state that diffuses with diffusion constant 2.5 µm2 s−1 in the entire cell, and can interconvert
to a 'bound' state (0.01 µm2 s−1) inside the nucleus. In addition, we simulate active transport into
the nucleus by setting the diffusion rate from the cytoplasm to the nucleus 20 times larger than
that in the reverse direction. Protein accumulation in the nucleus is clearly visible in the simulated
images (Fig. S5b,c).
Fig. S5d shows wall time1 for a 5 s stochastic reaction-diffusion simulation, starting with
about 375 uniformly distributed TFs and various voxel sizes, which fits very well with the expected
inverse quadratic scaling ∝ ∆x−2. Second, we generated a 2.5 s simulated microscopy movie with
5 ms frame time, 100 80 nm pixels, and 10% of the proteins activated and emitting on average
270 photons/frame. This simulation is limited by evaluating Brownian bridges and parsing the
trajectory file, and thus independent of the discretization of the trajectories.
Secondly, we varied the number of proteins, while keeping the discretization constant at
∆x = 30 nm. As seen in Fig. S5d, the image simulation part scales linearly as expected, while the
RD simulation scales quadratically if all proteins are tracked, but linearly with tracking deactivated
or if tracking only a fixed number of proteins. Thus, we see that while fairly large cells can be sim-
ulated, the RD simulations can be computationally demanding especially for models that include
large cell size, tracking of many molecules, and small details that require fine discretizations.
Finally, we note that the total size in itself does not influence computing time significantly,
as long as the number of subvolumes fit in the RAM memory of the computer. As an illustration,
we rescaled all length in the ∆x = 30 nm case of Fig. S5d by a factor 0.3 (thus decreasing the
volume by 97%) but kept the discretization and number of molecules constant. However, the RD
simulation of the smaller model only took 25% less time.
S5d. Complex shapes To highlight SMeagol's abilities to model geometries beyond the tractable
possibilities given by transformation and combinations of simple geometric primitives, we also
model particle diffusion in an erythrocyte shaped geometry, shown in Fig. S6a. The starting point
is here a freely available 3D model2, which we imported to and exported from Blender3 to make
1All computing times were measured on a 2.40GHz Intel Xeon CPU with 24 GB RAM.
2 http://www.turbosquid.com/FullPreview/Index.cfm/ID/509576, accessed 2016-02-04)
3www.blender.org
15
Figure S6: Modeling an erythrocyte. a) Rendering of the SBML model with 50 nm subvolumes.
b) Alignment of the model (particle positions indicated by black dots) relative to the simulated
region of interest (ROI) and focal plane of the simulation. c) Snapshot from a simulated movie
with 25 ms exposure time.
it a triangular mesh. The mesh was then converted to SBML compatible format using a custom
Python script, and incorporated into an SBML model, where about 70 particles diffuse freely (D=2
µm2 s−1) within the erythrocyte.
For the microscopy simulation, we rotated the model to align it with the focal plane as shown
in Fig. S6b and used similar settings as for the yeast example to produce the snapshot in Fig. S6c.
S6 Misc. supplementary material
• Supplementary movie S1 illustrates the simulated MinE single particle tracking experiment
(Fig. 1c, mid columns), showing both true particle positions, the experimental background,
and the simulated result in three separate panels.
• Supplementary movie S2 illustrates the simulated MinE fluorescence microscopy time-
lapse movie (Fig. 1c, right column), as well as the true particle positions.
• Supplementary movie S3 shows simulated single particle tracking experiments that are
analyzed in Fig. S2.
• Supplementary dataset S4 contains the SBML files, reaction-diffusion trajectory output,
and SMeagol runinput files for the simulated Min oscillation experiments (Fig. 1c).
• Supplementary dataset S5 contains model and runinput files, scripts, and source files for
the example models described in Sec. S5.
16
a)b)c)xzyx150 x 150 px ROI |
1707.07621 | 1 | 1707 | 2017-06-28T02:36:17 | Coarse-Grained Molecular Dynamics Modeling of Defective Erythrocyte Membrane and Sickle Hemoglobin Fibers | [
"physics.bio-ph",
"physics.med-ph"
] | In this work, we develop a two-component coarse-grained molecular dynamics (CGMD) model for simulating the erythrocyte membrane. This proposed model possesses the key feature of combing the lipid bilayer and the erythrocyte cytoskeleton, thus showing both the fluidic behavior of the lipid bilayer and elastic properties of the erythrocyte cytoskeleton. The proposed model facilitates simulations that span large length-scales (~ micrometer) and time-scales (~ ms). By tuning the interaction potential parameters, diffusivity and bending rigidity of the membrane model can be controlled. In the study of the membrane under shearing, we find that in low shear strain rate, the developed shear stress is mainly due to the spectrin network, while the viscosity of the lipid bilayer contributes to the resulting shear stress at higher strain rates. In addition, the effects of the reduction of the spectrin network connectivity on the shear modulus of the membrane are investigated. | physics.bio-ph | physics | Coarse-Grained Molecular Dynamics Modeling of Defective Erythrocyte
Membrane and Sickle Hemoglobin Fibers
He Li
University of Connecticut, 2014
Understanding the complex behavior of the normal and defective red blood cell (RBC)
membrane requires the development of detailed computational models. In this work, we show
that implementation of coarse-grained molecular dynamics (CGMD) methods can answer
fundamental questions related to protein diffusion and membrane loss in erythrocytes. In
particular, we first developed a two-component CGMD erythrocyte membrane model comprising
the lipid bilayer and an implicitly-represented cytoskeleton. The model reproduced the
mechanical properties of the RBC membrane illustrating both the fluidic behavior of the lipid
bilayer and the elastic properties of the erythrocyte cytoskeleton. By applying shear deformation,
we found that the shear stress was mainly due to the cytoskeleton at low shear strain rates while
the viscosity of the lipid bilayer contributed to the resulting shear stress at higher strain rates.
Building up on the experience acquired from the model above, we then created a RBC membrane
model with explicit representation of the cytoskeleton. In addition to the normal RBC membrane,
the model can describe the membrane of defective RBCs from patients with hereditary
spherocytosis (HS) and hereditary elliptocytosis (HE) by introducing defects that reduce the
connectivity between the lipid bilayer and the cytoskeleton (vertical connectivity) or the
connectivity of the cytoskeleton (horizontal connectivity), respectively. The model was then
applied in the study of band-3 protein diffusion in the normal RBCs and in the RBCs with
membrane protein defects. We showed that the diffusion of band-3 protein is more pronounced
in the defective RBCs than normal RBC, particularly in HE RBCs, which is in agreement with
experimental observations. Further, the explicit two-component CGMD RBC membrane model
was used to simulate vesiculation in the normal and defective RBC membrane induced by the
spontaneous curvature of the membrane domain or by compression on the lipid bilayer. We
found that the vesicle size depended on the membrane connectivity. Lower vertical or horizontal
connectivity caused the generation of larger vesicles than the high connectivity. Finally, we
introduced a CGMD model to simulate single hemoglobin S (HbS) fibers as a chain of CG
particles. We showed that the proposed model was able to efficiently simulate the mechanical
behavior of single and interacting HbS fibers. Simulations of the zippering process between two
HbS fibers illustrated that the depletion forces induced by HbS monomers were comparable to
direct fiber-fiber interaction via the Van der Waals forces.
ii
Coarse-Grained Molecular Dynamics Modeling of Defective
Erythrocyte Membrane and Sickle Hemoglobin Fibers
He Li
B.S., Beijing University of Technology, 2005
M.S., University of Saskatchewan, 2008
A Dissertation
Submitted in Partial Fulfillment of the
Requirements for the Degree of
Doctor of Philosophy
at the
University of Connecticut
2014
iii
Copyright by
He Li
2014
iv
APPROVAL PAGE
Doctor of Philosophy Dissertation
Coarse-Grained Molecular Dynamics Modeling of Defective Erythrocyte Membrane and Sickle
Hemoglobin Fibers
Presented by
He Li, B.S., M.S.
Major Advisor ______________________________________________________________
George Lykotrafitis
Associate Advisor ___________________________________________________________
Tai-Hsi Fan
Associate Advisor ___________________________________________________________
Greg Huber
Associate Advisor ___________________________________________________________
Horea Ilies
Associate Advisor ___________________________________________________________
David Pierce
University of Connecticut
2014
v
Acknowledgment
I am very honored to have Professor George Lykotrafitis as my research advisor. During my
Ph.D. study, Professor George Lykotrafitis introduced me to the interesting and exciting field of
cellular mechanics. Meanwhile, Professor George Lykotrafitis gave me unlimited guidance,
encouragement and patience. He helped me to improve my presentation skills and supported me
to attend research conferences where I can exchange and discuss ideas with other researchers. He
also gave me important suggestions on the developing my future career.
Second, I would like to thank my committee members Professor Tai-Hsi Fan, Professor Greg
Huber, Professor Horea Ilies and Professor David Pierce for taking time to review my research
work and offering their valuable advices.
Third, I am thankful for my colleagues working in the cellular mechanics lab who create a
enjoyable research environment.
Last but not least, I would like to send my most sincere and deepest appreciation to my parents
and my wife who are being supportive and taking care of me in my personal life so that I can
concentrate on my research work. In addition, I would like to thank ME Entertainment Frontier
Group, especially all the "ME Big4" members and their families, QiangGe and Xinxuan for
giving me so much joy and happiness during the past six years.
vi
Table of Contents
Title Page ..................................................................................................................................... iii
Approval Page .............................................................................................................................. iv
Acknowledgment ......................................................................................................................... vi
Table of Contents ....................................................................................................................... vii
List of Figures ............................................................................................................................. xii
Abstract of the Dissertation ................................................................................................... xxvi
Chapter 1. Introduction
1.1. Human Erythrocytes ................................................................................................................1
1.2. Hereditary Spherocytosis and Elliptocytosis ............................................................................5
1.3. Sickle Cell Disease and Sickle Hemoglobin Fiber ...................................................................6
1.4. Existing RBC Membrane Models ............................................................................................9
1.5. Research Motivations..............................................................................................................10
1.6. Dissertation Outline ............................................................................................................... 13
Chapter 2. Two-Component Coarse-Grain Molecular Dynamics Model for the Human
Erythrocyte Membrane
2.1. Introduction .............................................................................................................................14
2.2. Simulation method
2.2.1. Model ...........................................................................................................................19
2.2.2. Pair potential ................................................................................................................21
2.2.3. Simulation details .........................................................................................................24
2.3 Results
vii
2.3.1. Self-Diffusion ..............................................................................................................26
2.3.2. Membrane elasticity .....................................................................................................28
2.3.3. Phase diagram and the effects of the interaction potential parameters on membrane
properties......................................................................................................................31
2.4 Applications .............................................................................................................................33
2.5 Summary ..................................................................................................................................37
Chapter 3. Erythrocyte Membrane Model with Explicit Description of the Lipid Bilayer
and the Spectrin Network
3.1. Introduction .............................................................................................................................40
3.2. Model and simulation method ................................................................................................45
3.3 Membrane properties
3.3.1. Membrane fluidity ........................................................................................................52
3.3.2. Measurements of membrane bending rigidity, shear modulus and viscosity ...............53
3.3.3. Measurements of fluctuation frequencies of the spectrin filaments and the lipid
bilayer ..........................................................................................................................57
3.3.4. Timescales of the simulations ......................................................................................60
3.4 Mechanical interaction between the spectrin network and the lipid bilayer ............................61
3.5 Summary ..................................................................................................................................67
Chapter 4. Modeling of Band-3 Proteins Diffusion in the Normal and Defective Red Blood
Cell Membrane
4.1. Introduction .............................................................................................................................71
4.2. Model and simulation methods ...............................................................................................76
4.3 Results and discussion
viii
4.3.1. Band-3 diffusion in RBC membrane with perfect cytoskeleton, in the normal RBC
membrane and in the lipid membrane ..........................................................................78
4.3.2. Band-3 diffusion in RBC membrane with protein defects in the vertical interactions 82
4.3.3. Band-3 diffusion in RBC membrane with protein defects in the horizontal
interactions ...................................................................................................................84
4.3.4. Anomalous diffusion of the band-3 in the membrane with protein defects in the
horizontal interactions ..................................................................................................86
4.3.5. Effects of attractive forces between the spectrin filaments and lipid bilayer on the
band-3 diffusion ...........................................................................................................89
4.4 Summary ..................................................................................................................................94
Chapter 5. Modeling of Vesiculation in Healthy and Defective Human Erythrocyte
Membrane
5.1. Introduction .............................................................................................................................97
5.2. Model and simulation method ..............................................................................................104
5.3 Results and discussion
5.3.1. Vesiculation in health RBC membrane .....................................................................107
5.3.1.1 Vesiculation due to spontaneous curvature of membrane domain ................107
5.3.1.2 Vesiculation due to the compression on the lipid bilayer ..............................109
5.3.1.3 Vesiculation due to the combined effects ......................................................113
5.3.2. Vesiculation in defective RBC membrane .................................................................115
5.3.2.1 Vesiculation due to the spontaneous curvature of the membrane domain in HS
and HE cell membrane ................................................................................115
5.3.2.2 Vesiculation due to the compression in the HS RBC membrane ..................116
5.3.2.3 Vesiculation due to the compression in the HE RBC membrane ..................118
ix
5.4 Summary ................................................................................................................................120
Chapter 6. Modeling Sickle Hemoglobin Fibers as Four Chain of Coarse-Grained
Particles
6.1. Introduction ...........................................................................................................................122
6.2. Model and method ................................................................................................................127
6.3 Results and discussion
6.3.1. Measurements of material properties of HbS fiber ....................................................134
6.3.1.1. Bending rigidity ............................................................................................135
6.3.1.2. Torsional rigidity ..........................................................................................138
6.3.1.3. Effect of the bending and torsional potentials on the HbS fiber model ........139
6.3.2. Modeling the zippering of two HbS fibers .................................................................142
6.3.3. Bending of two zippered fibers .................................................................................150
6.4 Summary ................................................................................................................................152
6.5 Appendix 6.A .........................................................................................................................153
Chapter 7. Modeling Sickle Hemoglobin Fibers as One Chain of Coarse-Grained
Particles
7.1. Introduction ...........................................................................................................................158
7.2. Model and method ................................................................................................................160
7.3 Results and discussion
7.3.1. Measurements of material properties of HbS fiber ....................................................165
7.3.1.1. Bending rigidity ............................................................................................166
7.3.1.2. Torsional rigidity ..........................................................................................168
x
7.3.2. Modeling the zippering of two HbS fibers .................................................................169
Chapter 8. Epilogue ...................................................................................................................174
Chapter 9. References ................................................................................................................182
xi
List of Figures
Figure 1.1. Schematic model of the RBC membrane. The vertical and horizontal interactions
between its components are indicated. Disruption of vertical interactions causes spherocytosis
via loss of the unsupported PB, while disruption of actin-spectrin junctions causes elliptocytosis
because the network cannot recover its initial shape after undergoing large deformations in the
circulation (1). .................................................................................................................................2
Figure 1.2. (A) Spherocytes. (B) Elliptocytes. ................................................................................5
Figure 1.3. (A) The HbS fiber is composed of 7 double strands, shown as like-colored pairs (after
(2)). (B) Homogeneous nucleation. (C) Heterogeneous nucleation (2). (D) 2D fiber domain, after
(3). ...................................................................................................................................................6
Figure 1.4. (A–F) Evolution of HbS fiber polymerization (4). Image in F corresponds to
equilibrium between polymerized fiber and solution after 1 min of polymerization. Elongated
spherulites in (B–E) evolve into isometric spherulites in (F). ........................................................8
Figure 2.1. (A) Schematic of the membrane of human RBC. The blue circle represents the lipid
particle and the red circle signifies the actin junction. The yellow and green circles correspond to
a band-3 complex connected to the spectrin network and a free band-3 complex respectively. (B)
Two-component human RBC membrane model. "A" type particles represent actin junctions. "B"
type particles represent band-3 complex that are connected to the spectrin network. "C" type
xii
particles represent band-3 complex that are not connected to the network. "D" type particles
represent lipid particles. Red dash line highlights the structure of 2D six-fold spectrin network....15
Figure 2.2. The pair-wise interaction potential between lipid and lipid particles. (A) Comparison
with LJ6-12 potential. (B) The interaction potential profile as a function of . (C) Increasing the
parameter Rcut widens the potential well, giving rise to a softer core of the inter-particle
interaction . The soft core is essential for the stabilization of membrane at the fluid phase. .......23
Figure 2.3. (A) Thermally equilibrated fluid membrane of N = 35500 particles at a reference
time, representing a membrane with dimension of approximately 1 μm × 1 μm The tiles with
different colors tiles are used to differentiate the positions of the particles at a reference time. (B)
After 5×105 time steps, the particles are seen to be mixed due to diffusion, demonstrating the
fluidic behavior of the model membrane. (C) Linear time dependence of MSD of the two-
component membrane model. (D) Radial distribution function in the 2D fluid membrane
embedded in 3D .............................................................................................................................28
Figure 2.4. Vertical displacement fluctuation spectrum of two-component membrane as a
function of the wave number q. ....................................................................................................30
Figure 2.5. (A) Phase diagram. The hand-drawn lines show the phase boundaries. (B)
Dependence of the diffusion coefficient D on the temperature T for Rcut= 2.6d and 2.8d,
respectively. (C) Dependence of the bending rigidity κ on the parameter α at temperature T =
0.175, 0.22 and 0.275, respectively. .............................................................................................33
xiii
Figure 2.6. (A) Thermally equilibrated fluid membrane at a temperature of. (B) Sheared
membrane at a temperature of at engineering shear strain of 1. (C) Shear stress-strain response
of the membrane at two strain rates. Red curve represents the response of the "bare" spectrin
network. The blue and black curves signify the shear stress obtained for the strain rates of
0.001d/ts and 0.01d/ts, respectively. (D) Shear moduli of the membrane versus spectrin network
connectivity at the engineering shear strains of 0.1 and 0.9, respectively. ...................................35
Figure 3.1. (A) Schematic of the human RBC membrane. The blue sphere represents a lipid
particle and the red sphere signifies an actin junctional complex. The grey sphere represents a
spectrin particle and the black sphere represents a glycophorin particle. The yellow and green
circles correspond to a band-3 complex connected to the spectrin network and a mobile band-3
complex respectively. A mesoscale detailed membrane model. (B) top view of the initial
configuration, (C) side view of the initial configuration, and (D) a side view of the equilibrium
configuration. "A" type particles represent actin junctional complexes, "B" type particles
represent spectrin proteins, "C" type particles represent glycophorin proteins, "D" type particles
represent a band-3 complex that are connected to the spectrin network ("immobile" band-3), "E"
type particles represent band-3 complex that are not connected to the network ("mobile" band-3),
and "F" type particles represent lipid particles. ............................................................................43
Figure 3.2. The interaction potentials employed in the membrane model. The blue curve
represents the pair-wise potential between lipid particles. The green curve represents the spring
potential between spectrin particles. The red curve represents the spring potential between actin
xiv
and spectrin particles. The black curve represents the repulsive L-J potential between the lipid
and spectrin particles. ....................................................................................................................48
Figure 3.3. (A) Thermally equilibrated fluid membrane of N = 29567 particles at a reference
time, representing a membrane with dimension of approximately 0.8 μm × 0.8 μm The tiles with
different colors are used to differentiate the positions of the particles at a reference time. (B)
After 1×106 time steps, the particles are mixed due to diffusion, demonstrating the fluidic
behavior of the membrane model. (C) Linear time dependence of the mean square displacement
(MSD) of the two-component membrane model. (D) Radial distribution function of the 2D fluid
membrane embedded in 3D. .........................................................................................................53
Figure 3.4. Vertical displacement fluctuation spectrum of membrane model as a function of the
dimensionless quantity, where q is the wave number and σ is the unit length corresponding to
~4.45 nm. .....................................................................................................................................55
Figure 3.5. (A) Thermally equilibrated fluid membrane at a temperature of kBT/ε = 0.22. (B)
Sheared thermally equilibrated membrane at engineering shear strain of 1. (C) Shear stress-strain
response of the membrane at two different strain rates. The red curve represents the response of
the "bare" spectrin network. The blue and green curves signify the shear stress obtained at the
strain rates of 0.001σ/ts and 0.01σ/ts, respectively. The purple curve represents the shear stress
measured from the "bare" spectrin network plus the 8μN/m attributed to viscosity. ...................57
xv
Figure 3.6. (A) Power spectrum density (PSD) corresponding to thermal fluctuations of the
membrane model. The data marked as red circles are generated from the average displacement of
the particles belonging to a triangular compartment of the lipid bilayer. The data marked as blue
squares are generated from a particle positioned in the middle of a spectrin filament. fc represents
the thermal fluctuation frequency of the cytoskeleton and fl represents the thermal fluctuation
frequency of the lipid bilayer. Inset: equilibrium state of the proposed RBC membrane model. (B)
PSD corresponding to thermal fluctuations of a particle positioned in the middle of a spectrin
filament in the "bare" cytoskeleton. Inset: equilibrium state of "bare" cytoskeleton. ..................59
Figure 3.7. The pressure distribution exerted by the spectrin filaments on the lipid bilayer with
attraction parameters (A) n = 0 (B) n = 0.05 (C) n= 0.1 (D) n = 0.2. dee is the end to end distance
of the spectrin filaments. The blue curve represents the pressure distribution measured from the
membrane model. The purple curve represents the pressure distribution obtained from the
analytical estimation for a normal membrane (5). The red curve represents the pressure
distribution applied on the membrane with ankyrin protein defects. The black curve represents
the pressure distribution obtained analytically for a membrane with ankyrin protein defects (5).
The green curve represents the pressure distribution measured from the membrane with spectrin
protein defects. ..............................................................................................................................63
Figure 4.1. (A) Schematic of the human RBC membrane. Two-component human RBC
membrane model (B) top view at equilibrium configuration, (C) side view at initial configuration
(D) side view at equilibrium configuration. "A" type particles represent actin junctions. "B" type
particles represent spectrin particles. "C" type particles represent glycophorin proteins. "D" type
xvi
particles represent band-3 complexes that are connected to the spectrin network (immobile band-
3). "E" type particles represent band-3 complexes that are not connected to the network (mobile
band-3). "F" type particles represent lipid particles. ....................................................................72
Figure 4.2. (A) Schematic of the HS RBC membrane. (B) Schematic of the HE RBC
membrane. .....................................................................................................................................74
Figure 4.3. Trajectories of a mobile band-3 particle (A) hopping to a neighboring compartment in
the normal RBC membrane and (B) diffusing in a lipid bilayer without cytoskeleton. The
trajectories are extracted from simulation with duration of 2×106 time steps and the positions of
mobile band-3 particles were recorded every 2000 time steps. (C) MSDs against time and
corresponding diffusion coefficients of the mobile band-3 diffusing in the RBC membrane with
perfect cytoskeleton, in the normal RBC membrane and in membrane without cytoskeleton. ....82
Figure 4.4. MSDs against time and corresponding diffusion coefficients of the mobile band-3
diffusing in the RBC membrane with different vertical connectivities. .......................................84
Figure 4.5. (A) Comparison between the trajectory of a band-3 particle diffusing within the
compartment of an RBC membrane without defects (blue line) and the trajectories of two mobile
band-3 particle diffusing in RBC membrane with defects in the cytoskeleton (green and red
lines). The trajectories are extracted from simulation with duration of 2×106 time steps and the
positions of mobile band-3 particles were recorded every 2000 time steps. (B) MSDs against
xvii
time and corresponding diffusion coefficients of the mobile band-3 diffusing in the RBC
membrane with different cytoskeleton connectivities. .................................................................86
Figure 4.6. (A) log-log plot of the MSDs against time for the RBC membrane with cytoskeleton
connectivities Chorizontal = 0%, 30%, 50%, 70%, 90% and 100%. (B) The corresponding
anomalous exponent α for the previously mentioned horizontal cytoskeleton connectivities. ....88
Figure 4.7. Ratios of the band-3 diffusion coefficients D measured in the RBC membrane with
protein defects to Dperfect at a variety of attractive forces between the spectrin filaments and the
lipid bilayer (n = 0, 0.05, 0.1 and 0.15). We test membrane models with (A) vertical
connectivities of Cvertical = 100%, 90%, 70%, 50%, 30%, and 0%, and with (B) horizontal
cytoskeleton connectivities of Chorizontal = 100%, 90%, 70%, 50%, 30%, and 0%. The black dot
line represents the results predicted by the percolation theory in (6). ..........................................90
Figure 5.1. (A) Schematic of the human RBC membrane. The blue and yellow spheres represent
two types of lipid particles corresponding to membrane areas with different spontaneous
curvatures. The red sphere signifies an actin junctional complex. The white sphere represents a
spectrin particle and the black sphere represents a glycophorin particle. The light blue circle
corresponds to a band-3 complex that is connected to the spectrin network while green circle
represents a band-3 complex that does not connect to the lipid bilayer. (B) Top view of the
membrane model with two types of lipid CG particles completely mixed. (C) Top view of the
membrane model after the G type particles aggregate into membrane domains. "A" type particles
represent actin junctional complexes, "B" type particles represent spectrin proteins, "C" type
xviii
particles represent glycophorin proteins, "D" type particles represent a band-3 complex that are
connected to the spectrin network ("immobile" band-3), "E" type particles represent band-3
complex that are not connected to the network ("mobile" band-3), "F" type particles represent
majority of lipid particles and "G" type particles represent minority of the lipid particles. . .....103
Figure 5.2. Vesiculation induced by the spontaneous curvature of the membrane domains. (A)
When the spontaneous curvature is small (β = 0.1), the membrane domains only bud off from the
membrane and no vesicle is formed. (B) When β is increased to 0.18, vesiculation is observed
and the released vesicles are made of the G type particles only. (C) When β is increased to 0.24,
more vesicles are formed from the membrane. . .........................................................................109
Figure 5.3. Vesiculation induced by compression on the lipid bilayer. (A) When the compression
ratio Rcompression = 2%, only a small bud is created on the membrane and the cytoskeleton
conforms to the bud. (B) When Rcompression is increased to 5%, the bud grows up and the
cytoskeleton retracts from the budding area. (C) When Rcompression is increased to 15%, two
vesicles are released from the membrane and they consist of both G and F lipid particles. . ....111
Figure 5.4. Vesiculation induced by the combined effects of spontaneous curvature and
compression on the lipid bilayer. (A) When the spontaneous curvature is very small (β = 0.05),
the vesiculation processes at compression ratio of Rcompression = 5% and 10%, only produce a big
bud. (B) When Rcompression = 5% and β = 0.1, one vesicle is released. The vesicle has a size of ~60
nm and consists of only G lipid particles. (C) When Rcompression = 10%, three vesicles are released
from the membrane with sizes ranging from 40-60 nm. . ...........................................................114
xix
Figure 5.5. Vesiculation in the membrane with protein defects in the vertical interactions. (A)
When the compression ratio Rcompression is low (Rcompression = 5%), only one bud is created from the
membrane for all the selected Cvertical. (B) When Rcompression = 15% and vertical connectivity
Cvertical = 60%, one vesicle is created. (C) Summary of dependence of membrane vesiculation on
vertical connectivity Cvertical and compression ration Rcompression. In the figure, × indicates no
vesiculation occurs. ① indicates one vesicle is observed. ② indicates two vesicles are observed.
③ indicates three vesicles are observed. . ..................................................................................118
Figure 5.6. Vesiculation in the membrane with protein defects in the horizontal interactions. (A)
When the compression ratio Rcompression = 5% and horizontal connectivity Chorizontal = 20%, only
one bud is formed from the membrane. (B) When Rcompression = 10% and Chorizontal = 40%, one
vesicle with cytoskeleton fragment inside is created (C) Summary of dependence of membrane
vesiculation on horizontal connectivity Cvertical and compression ratio Rcompression. ① indicates one
vesicle is observed. ② indicates two vesicles are observed. In the figure, × indicates no
vesiculation occurs. . ...................................................................................................................120
Figure 6.1. (A) A schematic of HbS fibers in a sickle cell (B) A single HbS fiber comprises 7
pairs of double strands of hemoglobin tetramers. The pairs are shown as same-color agents. It is
drawn after (2). (C) Top view of the HbS fiber model. (D) Side view of the HbS fiber
model. ..........................................................................................................................................124
xx
Figure 6.2. The finitely extendable nonlinear elastic (FENE) potential, the corresponding
harmonic potential, the Lennard-Jones potential and the truncated Lennard-Jones potential
employed in the HbS fiber model are shown. .............................................................................129
Figure 6.3. (A) Bending and (B) twist of the HbS fiber model. ................................................131
Figure 6.4. Characteristic configuration of the HbS fiber model. .............................................136
Figure 6.5. Variation with time of the bending rigidity of the HbS fiber model. ......................137
Figure 6.6. The distribution of HbS fiber midpoint displacements and the associated normalized
Gaussian probability distribution. ...............................................................................................138
Figure 6.7. Variation with time of the torsional rigidity of the HbS fiber model. .....................139
Figure 6.8. Example of the behavior of the HbS fiber model without applying the bending and
the torsional potentials. ...............................................................................................................140
Figure 6.9. Variation with time of (A) the bending rigidity and (B) the torsional rigidity of the
HbS fiber model without applying the bending and the torsional potentials. .............................141
xxi
Figure 6.10. (A) Initial states of two parallel HbS fibers before zippering. (B) Intermediate state
of two initially parallel HbS fibers zippering. (C) Equilibrium configuration of two initially
parallel HbS fibers. The value of the L-J parameter k is 200. ....................................................142
Figure 6.11. (A) Position and (B) speed of the fiber junction during the zippering process of
two parallel HbS fibers. ..............................................................................................................144
Figure 6.12. Equilibrium states of two parallel HbS fibers zippered under different interfiber
forces. The L-J parameter k was selected to be (A) k = 100, (B) k = 50, (C) k = 10,
(D) k = 5. .....................................................................................................................................144
Figure 6.13. (A) Initial state of two HbS fibers before zippering. (B) Intermediate state of the
two HbS fibers during the "Y" shape zippering. (C) Equilibrium configuration of the two HbS
fibers. ..........................................................................................................................................145
Figure 6.14. (A) Position and (B) speed of the fiber junction during the "Y" shape zippering
process. ........................................................................................................................................147
Figure 6.15. Equilibrium states of two HbS fibers zippered as "Y" shapes under different
interfiber forces. The L-J parameter k is selected to be (A) k = 100, (B) k = 200, (C) k = 300, (D)
k = 400. (E) The final position of the fiber junction measured from point D for different
interfiber forces. ..........................................................................................................................148
xxii
Figure 6.16. Equilibrium states of two HbS fibers zippered in a "Y" shape as a result of a third
fiber crossing the other two fibers near to their ends for different interfiber forces. The L-J
parameter k is selected to be (A) k = 200, (B) k = 300, (C) k = 400, (D) k = 500. (E) The final
position of the fiber junction measured from point D for different interfiber forces. ...............149
Figure 6.17. Equilibrium states of two zippered HbS fibers under compression for the same final
deformation ratio e = 0.2 and at different interfiber forces. The L-J parameter k is selected to be
(A) k = 400, (B) k = 300, (C) k = 200, (D) k = 100. (E) Separations of the two fibers measured
from point C, at different interfiber forces. .................................................................................151
Figure 6.18. Equilibrium states of two zippered HbS fibers under compression at different final
deformation ratios and for the same interfiber force . The deformation ratio e is selected to be (A)
e = 0.02, (B) e = 0.05, (C) e = 0.15, (D) e = 0.2. (E) Separations of the two fibers measured from
point C at different deformation ratios. .......................................................................................152
Figure 7.1. (A) Schematic of HbS fibers in a sickle cell. (B) A single HbS fiber comprises 7
pairs of double strands of hemoglobin tetramers. The pairs are shown as same-color agents. It is
drawn after (2). (C) Each bead represents the entire cross-section of a single HbS fiber. (D) Side
view of the HbS fiber model. ......................................................................................................158
Figure 7.2. (A) Bending angle θ and (B) twist angle ΔΦ implemented in the HbS fiber
model............................................................................................................................................161
xxiii
Figure 7.3. The finitely extendable nonlinear elastic (FENE) potential applied between
consecutive fiber particles, the corresponding harmonic potential, and the Lennard-Jones
potential applied between particles belonging to different HbS fibers. The additional bending and
torsional FENE potentials employed in the simulations have similar behavior with the FENE
potential shown here. ..................................................................................................................163
Figure 7.4. The distribution of HbS fiber end displacements and the associated normalized
Gaussian probability distribution. ...............................................................................................167
Figure 7.5. Characteristic configuration of the HbS fiber model. The colors of the particles
represent the value of the rotation angle Φ, the blue color corresponds to Φ = 0 and the red color
corresponds to Φ = 0.72. The color bar is shown at the top of the figure. ................................168
Figure 7.6. The distribution of the accumulation of the twist angle at half of the pitch length (7th
chain particle) Ψ and the associated normalized Gaussian probability distribution. ..................169
Figure 7.7. (A) Initial 3D configuration of two HbS fibers (red particles) arranged in "Y" shape.
(B) Equilibrium 3D configuration of the two HbS fibers after zippering. The value of the L-J
parameter is k = 30. (C) Initial 3D configuration of two HbS fibers arranged in "Y" shape
surrounded by HbS monomer particles (white particles). (D) Equilibrium 3D configuration of the
two HbS fibers surrounded by HbS monomer particles after zippering. The value of the L-J
parameter is k = 30 and the density of the HbS monomers is ρ = 0.23 particles/σ3. It is observed
xxiv
that the zippered length of two fibers is increased by 7σ compared to the (B) due to the presence
of the HbS monomer particles. ...................................................................................................170
Figure 7.8. Zippered length ratios of two fibers (Lzip/L) for different attractive forces between the
HbS fibers. The solid lines describe the ratios of the zippered length to the fiber length for
different values of the parameter k. The solid lines in different colors represent different densities
of HbS monomers. ......................................................................................................................172
xxv
Abstract of the Dissertation
In this work, we develop a two-component coarse-grained molecular dynamics (CGMD) model
for simulating the erythrocyte membrane. This proposed model possesses the key feature of
combing the lipid bilayer and the erythrocyte cytoskeleton, thus showing both the fluidic
behavior of the lipid bilayer and elastic properties of the erythrocyte cytoskeleton. The proposed
model facilitates simulations that span large length-scales (~ µm) and time-scales (~ ms). By
tuning the interaction potential parameters, diffusivity and bending rigidity of the membrane
model can be controlled. In the study of the membrane under shearing, we find that in low shear
strain rate, the developed shear stress is mainly due to the spectrin network, while the viscosity
of the lipid bilayer contributes to the resulting shear stress at higher strain rates. In addition, the
effects of the reduction of the spectrin network connectivity on the shear modulus of the
membrane are investigated.
We extend the previously developed two-component red blood cell (RBC) membrane model and
develop a mesoscale implicit-solvent CGMD model of the erythrocyte membrane which
explicitly describes the phospholipid bilayer and the cytoskeleton in order to study the
cytoskeleton and lipid bilayer interactions, protein diffusion in the lipid bilayer and the
membrane vesiculation in the normal and defective RBC membrane. We show that the proposed
model represents RBC membrane with the appropriate bending stiffness and shear modulus. The
timescale and self-consistency of the model are established by comparing our results to
experimentally measured viscosity and thermal fluctuations of the RBC membrane. Furthermore,
we measure the pressure exerted by the cytoskeleton on the lipid bilayer. We find that defects at
the anchoring points of the cytoskeleton to the lipid bilayer (as in spherocytes) cause a reduction
xxvi
in the pressure compared to an intact membrane, while defects in the dimer-dimer association of
a spectrin filament (as in elliptocytes) cause an even larger decrease in the pressure. We
conjecture that this finding may explain why the experimentally measured diffusion coefficients
of band-3 proteins are higher in elliptocytes than in spherocytes, and higher than in normal RBCs.
We employ the RBC membrane model which explicitly describes the phospholipid bilayer and
the cytoskeleton, to study the band-3 protein diffusion in blood disorders such as hereditary
spherocytosis (HS) and hereditary elliptocytosis (HE). Furthermore, we investigate the effect of
the attraction between the spectrin filaments and lipid bilayer on the band-3 diffusion. First, we
measure the band-3 protein diffusion coefficients from healthy RBC membrane and from the
membrane after the cytoskeleton is removed. Comparison of these two coefficients clearly
illustrates the steric hindrance of the cytoskeleton on the band-3 mobility. Second, we measure
the band-3 diffusion from defective RBC membrane and quantify the relation between the band-
3 diffusion coefficients and degree of protein defects, via which we interpret the experimentally
measured band-3 diffusion coefficients in HS and HE. The diffusion coefficients measured from
our simulation are consistent with the experimentally measured diffusion coefficients as well as
the diffusion coefficients calculated based on the percolation analysis. By comparing the
diffusion coefficients measured from different types of protein defects, we find that the band-3
mobility is primarily controlled by the connectivity of the spectrin network, in agreement with
previous experimental results. Meanwhile, we study the dependence of band-3 anomalous
diffusion exponent on protein defects in the cytoskeleton and we find that the effect of
cytoskeleton on the anomalous diffusion exponent is small. At last, we applied the attractive
forces between spectrin filaments and lipid bilayer, and find that the attractive forces have the
xxvii
functions of decelerating the band-3 diffusion. The band-3 diffusion measurements conducted
when the spectrin filaments are fully attached to the lipid bilayer generally agree with the
calculation based on the percolation theory. The simulation results and the comparisons with the
existing experimental measurements are also used to predict the scale of the undetermined
attractive forces between the spectrin filament and lipid bilayer.
We employ a two-component RBC membrane model to simulate the diffusion of band-3 proteins
in the normal RBC and in the RBCs with protein defects. We introduce protein defects which
reduce the connectivity between the lipid bilayer and the cytoskeleton or reduce the connectivity
of the cytoskeleton and these defects are associated with the blood disorders of HS and HE,
respectively. We first measure the band-3 diffusion coefficients in the normal RBC membrane
and in the RBC membrane without cytoskeleton. Comparison of these two coefficients
demonstrates that the cytoskeleton limits the band-3 lateral mobility. Second, we study band-3
diffusion in defective RBC membranes and quantify the relation between the band-3 diffusion
coefficients and the percentage of protein defects in HS and HE RBCs. By comparing the
diffusion coefficients measured in these two cases, we conclude that the band-3 mobility is
primarily controlled by the cytoskeleton connectivity. Third, we study how the band-3
anomalous diffusion exponent depends on the percentage of protein defects in the membrane
cytoskeleton. Our measurements show that the effect of the cytoskeleton connectivity on the
anomalous diffusion exponent is small. Finally, we show that introduction of attraction between
the lipid bilayer and the spectrin network can reduce band-3 diffusion. By comparing our
measurements to experimental data, we predict that the attractive force between the spectrin
xxviii
filament and the lipid bilayer is at least 20 times smaller than the binding forces at the two
membrane major binding sites at band-3 and glycophorin.
.
Normal and defective RBCs release microvesicles of different compositions and sizes. Recent
advances have revealed that RBC-derived microvesicles play essential roles in coagulation and
in immune response. In blood disorders induced by membrane protein defects, microvesiculation
is a result of deteriorated stability of the lipid bilayer. Despite extended experimental and
theoretical work conducted on the vesiculation, the effects of the RBC cytoskeleton and its
connectivity on vesicle formation are not clear. In this work, we employ a recently developed
RBC membrane model, which explicitly describes the lipid bilayer and the spectrin network, to
study mechanisms of vesicle formation in the normal RBC membrane and in membranes with
defects related to HS and HE. We specifically correlate the size of the vesicle to
microvesiculation procedure and to membrane defects. We also determine conditions under
which the released microvesicles contain cytoskeleton fragments. Our simulations show that
lateral compression on the membrane can cause formation of vesicles with size similar to the size
of the basic cytoskeleton corral. When we consider a lipid bilayer model with different areas of
spontaneous curvature, corresponding to different phospholipid composition, then the produced
vesicles have a homogeneous composition. If lateral compression is applied on the previous
system, then the formation of vesicles originated from the curved membrane domain is facilitated.
In HS, where the vertical connectivity between the lipid bilayer and the spectrin network is
reduced, and in HE, where the lateral network connectivity is reduced, we find that the sizes of
microvesicles is diverse compared to the sizes of the microvesicles released from the healthy
RBC. This is due to the reduced confinement of the lipid bilayer by the RBC cortex. An
xxix
increased vertical connectivity between the lipid bilayer and the cytoskeleton causes the
generation of multiple vesicles with sizes similar to the cytoskeleton corral dimension. In the
case of a low vertical connectivity, the membrane tends to release larger vesicles under the same
compression ratio as above. It is noted that vesicles released from the HE RBCs may contain
cytoskeleton components due to the fragmentation of the cytoskeleton while vesicles released
from the HS RBCs are depleted of cytoskeleton components.
We develop a solvent-free CGMD HbS fiber model which represents a single hemoglobin fiber
as four tightly bonded chains. Because the intracellular polymerization of deoxy-sickle cell
hemoglobin has been identified as the main cause of sickle cell disease, the material properties
and biomechanical behavior of polymerized HbS fibers becomes important in the studies of the
sickle cell disease. In this developed fiber model, a finitely extensible nonlinear elastic (FENE)
potential, a bending potential, a torsional potential, a truncated Lennard-Jones potential and a
Lennard-Jones potential are implemented along with the Langevin thermostat to simulate the
behavior of a polymerized HbS fiber in the cytoplasm. The parameters of the potentials are
identified via comparison of the simulation results to the experimentally measured values of
bending and torsional rigidity of single HbS fibers. After it is shown that the developped model
is able to very efficiently simulate the mechanical behavior of single HbS fibers, it is employed
in the study of the interaction between HbS fibers. It is illustrated that frustrated fibers and fibers
under compression require a much larger interaction force to zipper than free fibers resulting to
partial unzippering of these fibers. Continuous polymerization of the unzippered fibers via
heterogeneous nucleation and additional unzippering under compression can explain the
xxx
formation of HbS fiber networks and consequently the wide variety of shapes of deoxygenated
sickle cells.
Finally, we simplified the four-chain fiber model and simulate single HbS fibers as a chain of
particles. The overall strategy for this one-chain model is similar to the one applied in the four-
chain fiber model. However, the representation of the cross-section by one CG particle discards
the need for calculating the interactions between neighboring chains. More importantly, the one-
chain fiber model can be easily used to simulate the formation of fiber bundles and fiber domains.
For example, the zippering process between two HbS fibers is studied and the effect of depletion
forces is investigated. Simulation results illustrate that depletion forces play a role comparable to
direct fiber-fiber interaction via Van der Waals forces.
xxxi
Chapter 1.
Introduction
In this chapter, I present a brief review on the RBC and the structure of healthy RBC membrane.
Following that, HS and HE, which are caused by the proteins defects in the RBC membrane, will
be introduced. Afterwards, structure of the HbS fibers and sickle cell disease (SCD) are
introduced and described. This section ends with a review on the existing models for RBC
Membrane.
1.1 Human Erythrocytes
A healthy quiescent RBC is biconcave in shape. Its diameter is approximately 8 μm and its
thickness is approximately 2.5 μm. Its main function is to deliver oxygen to the body tissues via
the blood flow through the blood circulations. An erythrocyte completes almost 250,000 cycles
during its 100 to 120 days life span and often passes through areas that induce high shear stress
such as the aortic valve, and endures multiple cycles of swelling and shrinkage at lungs and
kidneys. A RBC not only has to be robust but it must be very compliant too, as it undergoes very
large deformations when it passes through narrow capillaries of diameters as low as 3 - 4 μm in
the brain (1) and when it squeezes through spleen slits, which are roughly 1 μm ×2 μm openings
and about 2 μm deep formed by adjacent endothelial cells (2).
Figure 1.1. Schematic model of the RBC membrane. The vertical and horizontal interactions
between its components are indicated. Disruption of vertical interactions causes spherocytosis
via loss of the unsupported PB, while disruption of actin-spectrin junctions causes elliptocytosis
because the network cannot recover its initial shape after undergoing large deformations in the
circulation (3)
The erythrocyte's remarkable mechanical properties originate from the unique architecture of its
cell membrane (4), which is the main load bearing component, as there are no stress fibers inside
a normal RBC. A schematic representation of a part of the membrane is shown in Fig. 1.1(5, 6).
The cell membrane consists of spectrin tetramers that form a 2D six-fold network tethered to a
phospholipid bilayer (PB) (width: approximately 5 nm) comprising various types of
phospholipids, sphingolipids, cholesterol, and intramembrane proteins. The spectrin tetramers
are formed of spectrin heterodimers consisting of intertwined α-spectrin and β-spectrin filaments
running antiparallel to each other (Fig. 1.1). The intramembrane proteins band 3 and glycophorin
tether the spectrin network to the bilayer via additional binding proteins (e.g., ankyrin and 4.1).
2
The spectrin tetramers are connected via actin filaments and additional proteins forming
junctional complexes (5-8). The PB behaves as a 2D fluid and it resists bending but it cannot
resist shear loading that is sustained by the spectrin network. The mechanical properties and
shape of the RBC rely on the competition between the elastic energy of the lipid membrane,
including membrane tension and bending stiffness, and the shear and dilation elasticity of the
cytoskeleton. Within the framework of continuum mechanics, the free energy of a lipid
membrane can be described by
the classical Canham-Helfrich
theory
(9), as
E
Lipid Bilayer
2
C C C
1
2
2
0
C C dA
1
2
,
(1.1)
where
1C and
2C are the principal curvatures,
0C is the spontaneous curvature , γ is the surface
tension, κ and κ are the bending and Gaussian rigidities, respectively. The first term on the
right-hand side of Eq. (1.1) is the area-expansion energy, while the second and third terms
represent the normal and Gaussian bending energies, respectively. Since the topology of the
membrane remains unchanged the Gaussian bending energy results in a constant contribution,
and this energy term can be dropped. Surface tension is defined as
γ
σ
11
ZLσ
22
2
, where
11σ and
22σ are the components of the in-plane Virial stress (10) in a 3D periodic supercell
calculation, and Lz is height of the supercell, so that γ has the unit of N/m. Note that the so-called
nonlocal bending energy (11-13) which is induced by the lipid inner and outer layer area
difference is not included. This term appears in the area-difference elasticity model of bending
energy, but not in the classical spontaneous curvature model (9).
The cytoskeleton shrinks to a 3 to 5-fold smaller area after the entire membrane of the RBC is
removed (14), indicating that the cytoskeleton is stretched by its attachment to the lipid bilayer
3
(13, 15), and thus exerts a compression force on the lipid bilayer. The compression force is
balanced by the membrane bending stress. The elastic energy of the cytoskeleton is expressed as
E
Cytoskeleton
K
2
2
dA
dA
(1.2)
where α = λ1λ2 -1 and β = (λ1 - λ2)2/2λ1λ2 are the local area and shear strain invariants,
respectively. λ1 and λ2 are the local principal stretches. Kα and μ are the linear elastic moduli for
stretching and shear, respectively. We add the elastic energy of the cytoskeleton to Eq. (1.1) and
therefore obtain the total free energy for the RBC membrane
E
Total
E
Lipid Bilayer
E
Cytoskeleton
2
2
dA
dA
K
2
C C C
1
2
0
2
C C dA
1
2
(1.3)
In addition, the RBC membrane is composed of a host of different lipids and membrane proteins,
forming a heterogeneous bilayer. The steric interactions mismatch between different types of
lipids or between lipids and proteins results in a line tension which induces lateral phase
separation. The phase boundary line tension prefers to reduce the phase domain boundary length.
The line tension could promote the membrane vesiculation and the energy induced by the line
tension is described by
Eline
dl
4
(1.4)
where l is boundary of the phase separation and σ is the energy per unit length at the phase
boundary.
1.2 Hereditary Spherocytosis and Elliptocytosis
Because of the importance of the membrane skeleton for the stability of a RBC, defects in any of
its components lead to blood disorders (3, 5), such as HS and HE, which will be briefly
introduced as follows.
The hemolytic disorders of hereditary spherocytosis, and hereditary elliptocytosis affect the lives
of millions of individuals worldwide. HS is by far the most common congenital hemolytic
anemia in northern European descendants with a frequency of at least 1:5,000 (16). HE has a
worldwide distribution but is more common in people with African and Mediterranean ancestry
(17). Erythrocytes in patients with HS are characterized by a spherical shape of a smaller
diameter than healthy RBCs (Fig. 1.2A). In HE and its variants erythrocytes are elongated into a
cigar or oval shape (Fig. 1.2B).
A
B
Figure 1.2. (A)Spherocytes.(B) Elliptocytes.
(18)
In HS
and HE,
connections
between
components of
the RBC membrane break
resulting to loss of structural and functional
integrity of the membrane. Two types of
interactions
are
identified
as:
vertical
interactions between the spectrin network and
the PB and horizontal interactions at the actin-
5
spectrin junctions of the cytoskeleton (Fig. 1.1) (3). The prevailing hypothesis is that defects in
vertical interactions lead to spherocytosis via loss of unsupported membrane, while defects in
horizontal interactions lead to elliptocytosis because the network loses the ability to recover its
initial shape after undergoing large deformations in circulation (3, 16, 19-24).
The typical clinical picture of HS and HE combines evidences of hemolysis with spherocytosis.
Patients with severe HS, by definition, have life-threatening hemolytic anemia that requires
blood transfusion and they show an incomplete response to splenectomy. In addition to the risks
of recurrent transfusions, these patients are particularly prone to aplastic crises (bone marrow
failure), which occasionally develop retardation and delayed sexual maturation. Most cases of
HS are heterozygous since homozygosity is lethal (25, 26).
1.3 Sickle Cell Disease and Sickle Hemoglobin Fiber
A
B
C
D
Figure 1.3. (A) The HbS fiber is composed of 7 double strands, shown as like-colored pairs (after
(27)). (B) Homogeneous nucleation. (C) Heterogeneous nucleation (27). (D) 2D fiber domain,
after (28).
SCD affects approximately 1:400 African-American individuals in the United States (29-37). It
is an inherited blood disorder caused by a single point mutation in one of the genes encoding
6
hemoglobin which results in polymerization of deoxygenated abnormal sickle hemoglobin (HbS)
and the formation of long stiff rodlike fibers causing sickling of RBCs (27, 32, 38-40). Electron
microscopy has revealed that the HbS fiber consists of 7 double strands in a hexagonally shaped
cross-section twisted about a common axis in a rope-like fashion retaining their atomic contacts
(Fig. 1.3A) (41-43). The fibers have an average radius of approximately 11 nm and a mean
helical path of approximately 270 nm, defined as the length over which the fibers twist through
180o (41, 44). In vivo, HbS fibers are formed in an unusual fashion by two types of nucleation
processes. Homogeneous nucleation entails the nucleation of a new fiber from an aggregate of
monomers (see Fig. 1.3B) (27, 45). In heterogeneous nucleation, the surface of a formed polymer
assists the nucleation of a new polymer (46, 47) (see Fig. 1.3C). Since the homogeneous
nucleation is very slow, there are only a few homogeneous nuclei in the deoxygenated state
leading to a very low number of polymer domains via the dominant process of heterogeneous
nucleation, which generates aligned polymers. The process of fiber domain formation is shown
in Fig.1.4. First, the fiber polymerization initiates from in the HbS solution and form few
nucleuses, which develop to HbS fibers through homogeneous nucleation. Then, the HbS fibers
elongate with new fibers growing on the surface of the existing fibers, which eventually
generates fiber arrays and form fiber domains (See Fig.1.4B-F).
In fact, most RBCs gel with only a single polymer domain (48, 49). Another important
observation is that the sickle RBC membrane considerably enhances the nucleation by a factor of
6 (50) and that HbS fibers are directly attached to the cytoplasmic tail of the band 3 protein(51).
However, because the bulk majority of band 3 proteins are not attached to the cytoskeleton, HbS
fibers can create long protrusions consisting of PB without support from the spectrin network (6,
7
35). Due to the presence of the fiber domains inside the cell, sickled RBC shows increased cell
rigidity and adhesion, along with morphologies, such as sickle, holly leaf, granular, which result
from the interactions between the RBC membrane and intracellular HbS fiber domains (52). The
increased cell rigidity and adhesion raise the blood flow resistance and potentially trigger vaso-
occlusion in the microcirculation.
Figure 1.4. (A–F) Evolution of HbS
fiber polymerization (53). Image in f
corresponds
to equilibrium between
polymerized fiber and solution after 1
min of polymerization. Elongated
spherulites
in
(B–E)
evolve
into
isometric spherulites in (F).
Pathology in SS-SCD begins with the increased rigidity and cell adherence of the sickled RBCs
leading to vasoocclusion (36, 50, 54-57), which often causes significant tissue damage all over
the body and severe complications including acute and chronic pulmonary dysfunction, aseptic
necrosis of the hip, sickle cell retinopathy, and severe and chronic pain (31, 58-60). Overt stroke
caused by the occlusion of large blood vessels is the most deadly complication of SCD (61-64).
Silent stroke due to cerebral microinfarcts is the most common neurological complication in
children with the SCD resulting to both motor and neurocognitive impairment and to progressive
cerebral injury (65-68).
8
1.4 Existing RBC Membrane Models
Continuum membrane models based on membrane elasticity (9, 13, 69-73) have been adopted to
elucidate vesicle shape transitions and estimate thermal fluctuations of fluid membranes at length
scales much larger than the bilayer thickness (74, 75). At the opposite end of the length scale
spectrum, atomistic simulations have been extensively used to rationalize the molecular
mechanisms of various functions of bilayer membranes (76-79). However, owing to the
prohibitive computational cost, atomistic models are inadequate for direct comparisons with the
length and time scales of typical laboratory experiments.
The limitations of atomistic simulations and continuum approaches have motivated a continual
search for CGMD methods that bridge atomistic and continuum models (80-101). The complete
picture of CGMD and relevant references can be found in recent reviews (82, 96, 100). CGMD
models can be generally categorized into explicit solvent and implicit solvent (solvent-free)
schemes. Explicit solvent schemes employ the hydrophobic interactions between membrane and
solvent particles (a water molecule or a group of several water molecules) to stabilize the 2D
membrane (84, 87, 94). Explicit solvent models frequently employ dissipative particle dynamics
(DPD), a very efficient method that represents a large volume of the solvent with a soft bead,
thus significantly accelerating the computations (102-105). The technique has been extended to
lipid bilayers by introducing spring forces between representative particles in the polymer chains
(87, 106). In the case of implicit solvent schemes, the solvent particles are not directly
represented in the simulation and their effect is taken into account by employing effective
multibody interaction potentials, based on the local particle density (83, 86, 95), or by
9
implementing different pair-potentials between particles representing the hydrophobic tail and
those representing the hydrophilic head of the lipids (88, 91, 92).
In order to largely extend the accessible length and time scale of membrane model, an early
CGMD lipid membrane model featuring orientation-dependent interactions was developed by
Drouffe et al. (83), where an anisotropic attractive interaction between spherical particles, along
with a hard core repulsive interaction and a density-dependent "hydrophobic" multibody
potential, was used to describe the lipid interactions. Branningan and Brown (107) developed a
solvent-free model in which lipids are represented as rigid, asymmetric spherocylinders
interacting through orientation-dependent attractions. Kohyama (101) extended the Drouffe's
model by introducing an extra degree of freedom that corresponds to the effective curvature
caused by thermal fluctuations. Yuan et al. (108) simulated the biological fluid membranes by
introducing a one-particle-thick, solvent-free, coarse-grained model, in which the interparticle
interaction is described by a soft-core pairwise potential. The model essentially combines
Drouffe's (83) and Cooke's approaches (92). The interaction strength is also dependent on the
relative orientations of the particles, but there is no membrane cytoskeleton involved. In all the
models, the orientation-dependent interactions are utilized to describe the hydrophobicity of the
tail groups of the lipids, and are essential to the self-assembly of lipid bilayer in aqueous
environment.
1.5 Research Motivation
Previously developed CGMD models of the RBC membrane either only simulated the RBC
cytoskeleton with application of a bending potential to represent the bending stiffness induced by
10
the lipid bilayer (80, 81, 91, 109, 110) or only modeled the lipid bilayer (84, 86, 91, 92, 95, 108)
with implicit spectrin network (111). Therefore, we develop a two-component CGMD model for
simulating the erythrocyte membrane. The developed model possesses the key feature of
combing the lipid bilayer and the erythrocyte cytoskeleton, thus showing both the fluidic
behavior of the lipid bilayer and elastic properties of the erythrocyte cytoskeleton. Then, by
extending this two-component RBC membrane model, we developed a mesoscale implicit-
solvent CGMD model of the erythrocyte membrane which explicitly describes the phospholipid
bilayer and the cytoskeleton.
We employed the developed model to study:
(1) Diffusion of the mobile band 3 particles in the normal RBC membrane and in the membrane
with defective membrane proteins. Recent experimental studies have demonstrated that the
properties of band 3 change significantly in HS and HE (112). In HS cells, the fast population of
band 3 (free band 3 proteins) takes the prominence while the slow population (fixed band 3)
dominates the behavior in healthy cells (73% of the total band 3). Similarly, a substantial fraction
of band 3 was found to redistribute from the slower to faster diffusing population in HE samples.
In addition, the diffusion coefficient for the HE samples were approximately 10 times faster than
in normal cells (112). With the proposed model, we can quantitatively study the dependence of
the band 3 diffusion coefficient on the proteins defects. First, all the linkages between the
spectrin proteins and the immobile band 3 proteins will be broken (vertical interactions), which
commonly happens in the HS where defective band 3, protein 4.2 and ankyrin proteins are found
in the RBC membrane. Then, in the next case, the spectrin chains are dissociated in the middle or
from the actin junctions to represent the proteins defects in horizontal interactions, which happen
11
in HE. At last, the comparison of the band 3 diffusion coefficients measured from the reduced
vertical connections and horizontal connections is performed to find out which is the major
determinant of the lateral diffusion rate of band 3. Moreover, comparison is made between band
3 diffusion of the proposed model and the previous two-component WLC RBC membrane model
(111) and demonstrate the hindering effect of the spectrin filaments on the band 3 mobility.
(2) Vesicluation in the healthy and defective erythrocyte membrane. The aggregation of membrane
proteins of specific types that have mutual affinity, is the dominate process driving the formation of the
nanovesicles (diameter of 60 ~ 100 nm)(113), while the compression forces induced by cytoskeleton
stiffening lead to buckling and formation of the microvesicles (diameter of 60 ~ 300 nm (15, 113).
The proposed model will be used to simulate the membrane vesiculation process and explore the
dependence of vesiculation on the compression ratio and cytoskeleton connectivity. Moreover,
different types of the lipids or proteins can be introduced into the model. The aggregation of the
same types of the particles will form membrane domains. The domain boundary line tension
prefers to reduce the domain boundary length and is expected to contribute to the membrane
vesiculation. In addition, we will use our model to test the two predominant mechanisms that
have been hypothesized to explain how defects in vertical interactions result to HS (25). One
suggests that membrane loss is induced by defective interactions between the lipid bilayer and
the cytoskeleton. The unsupported lipid bilayer in the cytoskeleton deficient area will bud off
and form vesicles. The other hypothesis assumes that the membrane is stabilized by the
interactions between band 3 proteins and their neighboring lipids. Defects in membrane proteins
enhance the lateral mobility of band 3 proteins (112, 114-118), and the band 3, therefore, may
aggregate and form clusters in the defective RBC membrane, which lead to loss of the
unsupported lipids in band 3 deficient areas (25).
12
1.6 Dissertation Outline
The rest of this dissertation is organized as follows. A two-component CGMD model for the
erythrocyte membrane is introduced in Chapter 2, which possesses the key feature of combing
the lipid bilayer and the erythrocyte cytoskeleton, thus showing both the fluidic behavior of the
lipid bilayer and elastic properties of the erythrocyte cytoskeleton. In Chapter 3, we extend the
two-component RBC membrane model introduced in the Chapter 2 and develop a mesoscale
implicit-solvent CGMD model of the erythrocyte membrane which explicitly describes the
phospholipid bilayer and the cytoskeleton in order to study the cytoskeleton and lipid bilayer
interactions, protein diffusion in the lipid bilayer and the membrane vesiculation in the normal
and defective RBC membrane. In Chapter 4, we applied the RBC membrane model which
explicitly describes the phospholipid bilayer and the cytoskeleton, to study the band-3 protein
diffusion in blood disorders such as hereditary spherocytosis (HS) and hereditary elliptocytosis
(HE). In Chapter 5, we employ the RBC membrane model which explicitly describes the
phospholipid bilayer and the cytoskeleton, to study vesiculation in the healthy and Defective
Human Erythrocyte Membrane. In Chapter 6, we develop a solvent-free CGMD HbS fiber model
which represents a single hemoglobin fiber as four tightly bonded chains to simulate the material
properties and biomechanical behavior of polymerized HbS fibers. In Chapter 7, we simplified
the four-chain fiber model introduced in the Chapter 6 and simulate single HbS fibers as a chain
of particles, which can be easily used to simulate the formation of fiber bundles and fiber
domains. Epilogue is provided in Chapter 8.
13
Chapter 2.
Two-Component Coarse-Grain Molecular Dynamics Model for the
Human Erythrocyte Membrane
Abstract
We present a two-component coarse-grain molecular dynamics model for simulating the
erythrocyte membrane. The proposed model possesses the key feature of combing the lipid
bilayer and the erythrocyte cytoskeleton, thus showing both the fluidic behavior of the lipid
bilayer and elastic properties of the erythrocyte cytoskeleton. In this model, three types of
coarse-grained particles are introduced to represent clusters of lipid molecules, actin junctions
and band-3 complexes, respectively. The proposed model facilitates simulations that span large
length-scales (~ µm) and time-scales (~ ms). By tuning the interaction potential parameters,
diffusivity and bending rigidity of the membrane model can be controlled. In the study of the
membrane under shearing, we find that in low shear strain rate, the developed shear stress is
mainly due to the spectrin network, while the viscosity of the lipid bilayer contributes to the
resulting shear stress at higher strain rates. In addition, the effects of the reduction of the spectrin
network connectivity on the shear modulus of the membrane are investigated.
2.1 Introduction
Mature human erythrocytes do not have an internal structure, meaning that their mechanical
properties originate from the cell membrane which is the only structural element of the cell (6).
14
The red blood cell (RBC) membrane is composed of a two-dimensional (2D) six-fold spectrin
network tethered to a lipid bilayer comprising various types of phospholipids, sphingolipids,
cholesterol and integral membrane proteins (see Fig. 2.1). The spectrin network consists of
spectrin tetramers that are connected via actin filaments and additional proteins forming
junctional complexes. Each spectrin tetramer comprises two heterodimers, which consist of
intertwined and antiparallel α-spectrin and β-spectrin filaments (6). The integral membrane
proteins band-3 and glycophorin tether the cytoskeleton network to the phospholipid bilayer via
additional peripheral proteins (e.g., Ankyrin and 4.1) (3). The lipid bilayer is essentially a 2D
fluid-like structure embedded in a three-dimensional (3D) space. It resists bending but cannot
sustain in-plane shear stress because the lipids and most of the proteins can diffuse freely within
the membrane to relax the shear stress. The stiffness of RBCs arises primarily from the spectrin
network. It is also noted that the material properties of the lipid bilayers play important roles in
the function and distribution of the membrane proteins (119).
15
Figure 2.1. (A) Schematic of the membrane of human RBC. The blue circle represents the lipid
particle and the red circle signifies the actin junction. The yellow and green circles correspond to
a band-3 complex connected to the spectrin network and a free band-3 complex respectively. (B)
Two-component human RBC membrane model. "A" type particles represent actin junctions. "B"
type particles represent band-3 complex that are connected to the spectrin network. "C" type
particles represent band-3 complex that are not connected to the network. "D" type particles
represent lipid particles. Red dash line highlights the structure of 2D six-fold spectrin network.
Continuum models based on membrane elasticity (9, 13, 69-73) have been adopted to elucidate
vesicle shape transitions and estimate thermal fluctuations of fluid membranes at length scales
much larger than the bilayer thickness (74, 75). At the opposite end of the length scale spectrum,
atomistic simulations have been extensively used to rationalize the molecular mechanisms of
various functions of bilayer membranes (76-79). However, owing
to
the prohibitive
16
computational cost, atomistic models are inadequate for direct comparisons with the length and
time scales of typical laboratory experiments.
The limitations of atomistic simulations and continuum approaches, along with the practical need
to treat the heterogeneous nature of RBC membrane have motivated a continual search for
Coarse-Grained Molecular Dynamics (CGMD) methods that bridge atomistic and continuum
models (80-101, 120, 121). The complete picture of CGMD and relevant references can be found
in recent reviews (82, 96, 100). CGMD models can be generally categorized into explicit solvent
and implicit solvent (solvent-free) schemes. Explicit solvent schemes employ the hydrophobic
interactions between membrane and solvent particles (a water molecule or a group of several
water molecules) to stabilize the 2D membrane (84, 87, 94). Explicit solvent models frequently
employ dissipative particle dynamics (DPD), a very efficient method that represents a large
volume of the solvent with a soft bead, thus significantly accelerating the computations (102-
105). The technique has been extended to lipid bilayers by introducing spring forces between
representative particles in the polymer chains (87, 106). In the case of implicit solvent schemes,
the solvent particles are not directly represented in the simulation and their effect is taken into
account by employing effective multibody interaction potentials, based on the local particle
density (83, 86, 95), or by implementing different pair-potentials between particles representing
the hydrophobic tail and those representing the hydrophilic head of the lipids (88, 91, 92).
In order to largely extend the accessible length and time scale of membrane model, an early
CGMD lipid membrane model featuring orientation-dependent interactions was developed by
Drouffe et al. (83), where an anisotropic attractive interaction between spherical particles, along
17
with a hard core repulsive interaction and a density-dependent "hydrophobic" multibody
potential, was used to describe the lipid interactions. Branningan and Brown (107) developed a
solvent-free model in which lipids are represented as rigid, asymmetric spherocylinders
interacting through orientation-dependent attractions. Kohyama (101) extended the Drouffe's
model by introducing an extra degree of freedom that corresponds to the effective curvature
caused by thermal fluctuations. Yuan et al. (108) simulated the biological fluid membranes by
introducing a one-particle-thick, solvent-free, coarse-grained model, in which the interparticle
interaction is described by a soft-core pairwise potential. The model essentially combines
Drouffe's (83) and Cooke's approaches (92). The interaction strength is also dependent on the
relative orientations of the particles, but there is no membrane cytoskeleton involved. In all the
models, the orientation-dependent interactions are utilized to describe the hydrophobicity of the
tail groups of the lipids, and are essential to the self-assembly of lipid bilayer in aqueous
environment.
Here, we extend the coarse-grained solvent-free approach followed in the simulation of lipid
bilayers (108), to model the entire RBC membrane by introducing the two-component membrane
model. Three types of particles are used to represent lipid molecules, actin junctions and band-3
complexes. The actin junction particles along with a number of band-3 particles are connected
via the worm-like chain (WLC) potential to form a hexagonal network linked to the lipid bilayer.
Additional band-3 particles that do not sense the WLC potential are employed to simulate the
band-3 protein complexes that can diffuse freely in the lipid bilayer. An orientation-dependent
potential, similar to the potentials introduced in (108), is employed between all particles to
simulate the two-dimensional fluidic nature of the membrane.
18
2.2 Simulation method
2.2 .1Model
The model describes the RBC membrane as a two-component system, including the cytoskeleton
and the lipid bilayer. Three types of particles are introduced (see Fig. 2.1B). The blue color
particles represent a cluster of lipid molecules, which have a diameter of 5 nm, a similar value to
the thickness of the lipid bilayer. The red particles signify actin junctions and they form a
canonical hexagonal network representing the spectrin network. An actin junction has a diameter
of approximately 35 nm and resides in the cytoplasm (122, 123). It is connected to the lipid
bilayer via glycophorin which has a size comparable to thickness of the lipid bilayer. Thus, the
fluidic behavior of the membrane is not affected by the size of the actin junctions but only by the
size of the glycophorin. Therefore, the particles that represent the actin junctions in the model
have a diameter of 5 nm, similar to the size of glycophorin. The yellow particles denote band-3
complexes with diameter of 7.5 nm (6) and they are connected to the spectrin network, while the
green particles represent the free band-3 complexes that are not connected to the spectrin
network. The masses of the particles are given in the section 2.3.
In the model, each particles carry both translational and rotational degrees of freedom
i
i nx ,
,
where
ix and
in are the position and the orientation (director vector) of particle i, respectively.
The rotational degrees of freedom obey the normality condition
1in
. Thus, each particle
effectively carries 5 degrees of freedom. We define
x
ij
x
j
x
i
as the distance vector between
19
particles i and j . Correspondingly,
ijr
x
ij
is the distance, and
x
ij
rx
ij
ij
is a unit vector. The
particles interact with one another via a pair-wise additive potential
xnn
i
ij
,
,
j
ru
ij
R
u
ij
xnn
aαA
i
ij
,
,
,
j
ij
ru
A
,
with
,
A a
n n x
i
ij
,
,
j
1
a
n n x
i
ij
,
,
j
1
,
(2.1)
(2.2)
where
ijR ru
and
ij
A ru
are the repulsive and attractive components of the pair potential,
respectively. is a tunable linear amplification factor. The energy well of this potential is
essential to regulate the fluid-like behavior of the membrane, therefore we define a function
aαA ,
to enable tuning the energy well. The effects of function
in the potential will be
aαA ,
discussed in the section 2.2. The interaction between two particles depends not only on their
distance but also on their relative orientation via the function
ij
xnn
ia
,
,
j
which varies from -1
to +1 and adjusts the attractive part of the potential. We specify that
ijx
n
jn and both are normal to vector
in is parallel to
case when
1a
corresponds to the
n
i
j
x
ij
, and the
value
n
i
1a
n
j
to the case when
ij
x
in is anti-parallel to
jn and both are perpendicular to vector
ijx
. The former instance is energetically favored due to the maximum attractive
interaction between particles i and j, while the latter is energetically disfavored due to the
20
maximum repulsive interaction. In essence, the energy difference between these two cases acts
as the thermodynamic driving force responsible for the self-assembly of lipid bilayer. One simple
ij
xnn
ia
,
,
j
form of
that captures these characteristics is
,
xnn
ia
ij
,
j
xn
ij
i
n
j
x
ij
nn
i
j
ij
xnxn
i
ij
j
.
(2.3)
The equations of motion include the translational and rotational components
m
i
x
i
( )
V
x
i
,
m
i
n
i
)
(
V
n
i
)
(
V
n
i
n n
i
,
i
(2.4)
(2.5)
i
m
i
n n n
i
i
where
V
N
j
1
u
i j
,
im~ is a pseudo-mass with units of energy × time2, and the right-hand side
(RHS) of Eq. (2.5) conforms to the normality constraint
1in
.
2.2.2 Pair potential
To stabilize a fluid membrane, the lipid particles should be allowed to move past each other with
relatively low resistance, while the overall cohesive integrity is preserved. Cooke et al. (92)
discovered that the classical LJ6-12 potential,
ru
4
rdε
12
rd
6
, where ε has the unit of
energy and d has the unit of length, possesses very strong repulsive core below the equilibrium
distance
1 62r
d
. This suppresses position exchange of particles and makes it difficult to form
a liquid phase. As temperature T is raised from absolute zero, the initial crystal phase sublimes to
the vapor phase without going through the liquid phase. Compared to traditional particle systems,
21
the vapor phase in the model with orientation disorder is thermodynamically more favorable
because of the entropic contributions from its rotational degrees of freedom. To describe a fluid
membrane by self-assembled moving particles, we employ the following repulsive and attractive
potentials:
ij
R
u r
u r
u r
A
R
ij
ij
R
2
u r
A
ij
r
cut
cut
R
0,
cut
ij
r
ij
R
r
min
8
for
r R
<
ij
cut
R
cut
r
min
4
for
r
ij
R
cut
for
r
ij
R
cut
(2.6)
We chose
l
lr
min
1/ 6
2
d
for the pair potentials between lipid and lipid particles, so that the
potential described in (2.6) has the same minimum energy and equilibrium distance as the LJ6-
l ar
12 potential. The association energy between lipid and lipid particles is ε. While min
a ar
and min
in
the pair potentials between lipid and actin particles, and between actin and actin particles are also
d6/12
l br
, min
a br
, min
b br
and min
in the pair potentials between lipid and band-3 particles, between
actin and band-3 particles, between band-3 and band-3 particles are
1.25 2 d
1/6
1.25 2 d
1/6
,
and
1.5 2 d
1/6
, respectively, as diameter of band-3 particles is 1.5 of diameter for the lipid and actin
particles. The association energy between lipid and band-3 particles, between lipid and actin
particles, as well as between actin and band-3 particles is chosen to be 1.4ε. The Figs. 2.2A and
2.2B only show the potential between lipid particles. Another appealing characteristic of the
iju r
above potential (hereon denoted by "Poly4-8"), is that the 1st, 2nd and 3rd derivatives of
with respect to rij are all continuous at
ijr
R
cut
.
22
Figure 2.2 The pair-wise interaction potential between lipid and lipid particles. (A) Comparison
with LJ6-12 potential. (B) The interaction potential profile as a function of
aαA ,
. (C)
Increasing the parameter Rcut widens the potential well, giving rise to a softer core of the inter-
1
aαA
,
particle interaction
the fluid phase.
. The soft core is essential for the stabilization of membrane at
Fig. 2.2A shows that the effective potential in the case of
aαA
1
has a wider energy well
,
than the LJ6-12 potential, which gives rise to a larger "maneuverability" near the equilibrium
distance than that of the classical LJ6-12 potential. This characteristic facilitates particles
squeezing past each other and stabilizes the liquid phase of the particle ensemble against the gas
phase. The effect of the function
aαA ,
on the potential is illustrated in Fig. 2.2B. As the
function
aαA ,
changes from +1 to negative values, the total potential shifts from attractive to
repulsive, penalizing the non-parallel arrangement of neighboring particles. This property of the
23
potential ensures the formation of a 2D membrane sheet rather than a dense 3D structure with
higher coordination numbers. We emphasize that because the parameter α regulates the energy
penalty paid by membrane particles when their relative orientation changes, it plays an important
role in adjusting the bending rigidity of the membrane.
The restoring force near the equilibrium distance between same types or different types of
CGMD particles can be tuned by adjusting the parameter Rcut. Fig. 2.2C shows that a decreasing
Rcut narrows the potential well between the lipid and lipid particles, which results in a rapid
increase of the repulsive force. This increase in repulsive force, in turn, prevents the particles
from passing each other and stabilizes them into a solid-like crystal phase. The influence of Rcut
on the stabilization of fluid membranes is a general feature of soft-core potentials independent of
the precise functional form of the potentials (92).
A WLC potential (124, 125) is employed to simulate the spectrin network. The WLC forces
between the connected actin and band-3 particles is described by
f
WLC
( )
L
k T
B
p
1
4(1
2
x
)
1
4
x
,
(2.7)
where
x L L
max
0,1
, Lmax = 100 nm is maximum or the contour length of the spectrin chain
between an actin particle and a band-3 particle. L is the instantaneous chain length or the
instantaneous distance between an actin and the corresponding band-3 particles. p = 10 nm is the
persistent length of the spectrin filaments (126). kB is the Boltzmann constant and T is the
24
temperature. We used the WLC model to represent the RBC cytoskeleton without introducing
bending rigidity to the spectrin network because the RBC membrane cytoskeleton is a highly
flexible structure with bending rigidity κ ~ 0.024 – 0.24kBT (127, 128), which is about two orders
of magnitude smaller than the bending rigidity of the lipid bilayer κ ~ 10 - 20 kBT(126 , 129).
2.2.3 Simulation details
We perform (CGMD) simulations of a planar membrane to extract the elastic properties of the
membrane. Periodic boundary conditions (PBCs) were imposed in the plane of the membrane (x-
and y-directions), while the z-direction (perpendicular to the membrane surface) is completely
free without any constraints. The system consists of N = 35500 particles, including 34600 lipid
particles, 90 actin junctions particles and 810 band-3 complex particles out of which, 540 band-
3 particles are not connected to the spectrin network. The dimension of the membrane is
approximately 1 μm × 1 μm. The numerical integration of the equations of motion (Eqs. (2.4)
and (2.5)) was performed using the Beeman algorithm (10, 130). The Nose-Hoover thermostat
was employed to control the temperature. The model was implemented in the NAT ensemble (95,
131, 132). Since the model is solvent free and the membrane is a two-dimensional structure, we
controlled the projected area instead of the volume. The projected area was adjusted to result in
zero tension for both the lipid bilayer and the spectrin network at the equilibrium state. The time
scale that guides the choice of the time step in the CGMD simulations is
t
s
i
m d
2
1 2
. The
chosen time step is 0.01ts. In addition, to match the time steps used for translational and
rotational degrees of freedom, the pseudomass
im~ is chosen to be
m m d
~ l
i
i
2
, where
l
im is the
mass of the lipid particles. In section 3.1, we determined an appropriate timescale by comparing
the in-plane diffusion coefficient obtained from the simulations with experimental values (133).
25
We use ε as the energy unit, and d as the unit length to quantify the geometrical and mechanical
properties of lipid bilayer and membrane proteins. Given the diameter of the lipid particles
l
lr
min
1/6
2
d
5nm
, one finds
nm45.4d
. Each coarse-grained lipid particle carries a mass of
lm
95.6kDa
, which was obtained by averaging over those of various lipids that compose an
area of approximately 20 nm2 of the erythrocytic membrane (6). The band-3 complex consists of
band-3 protein and glycophorin A, with molecular weights of 102 kDa and 14 kDa, respectively.
It also includes Ankyrin, which connects the band-3 protein with the spectrin cytoskeleton, and
protein 4.2 which binds with the band-3 and Ankyrin. Ankyrin and 4.2 proteins have molecular
weights of 210 kDa and 72 kDa respectively (3). Therefore, the total mass of the band-3 complex
is approximately 398 kDa, corresponding to 4 lm . As mentioned above, the band-3 complex is
approximated as a sphere with diameter of 7.5 nm, equivalent to
1.5 l lr . An actin junction is
min
composed of 14 kDa glycophorin C, short actin filaments (with a weight of 546 kDa), and 80
kDa protein 4.1, 200 kDa adducin, 41 kDa tropomodulin and a 28 kDa tropomyosin (3). Thus,
the actin junction is approximated as a sphere with total weight of 1000 kDa and a diameter of 5
lr . The maximum or the contour length of the spectrin chain
l
nm, corresponding to 10 lm and min
between actin particles and band-3 particles is equivalent to max
L
r
l
20 l
min
. As it will be shown
below, the system is found to be a fluid membrane with bending rigidity well within the
experimentally established range at B
k T
0.22
and Rcut = 2.6d. The amplification factor α in
Eq. (2.2) was chosen to be 1.55.
2.3. Results
26
2.3.1. Self-Diffusion
In this section, we show that the proposed model simulates a membrane with the appropriate
mechanical and physical properties. Simulations are initiated from a flat triangular crystalline
lattice that occupies the entire mid-plane of the super-cell. At zero temperature, the particles
form a 2D solid phase with a long-range order. As the temperature increases, the membrane
starts to fluctuate and is gradually equilibrated at a fluid phase at the temperature of B
k T
0.22
(see Fig. 2.3A). To ensure that the model represents a fluid membrane, it is necessary to examine
whether the lipid and band-3 particles can diffuse in a fluidic manner such that the particle's
positions are exchangeable. Fig. 2.3A and 2.3B show the positions of the particles of a thermally
equilibrated fluid membrane at two separate times. The tiles with different colors are used to
differentiate the positions of the particles at the initial moment. After 5×105 time steps the
colored particles are mixed due to diffusion, demonstrating the fluidic nature of the membrane.
We further verify the fluidic characteristics of the membrane with two other methods described
below. First, we show in Fig. 2.3C that the mean square displacement (MSD) of lipid particles,
defined by
1
N
j N
x
j
1
t
j
x
j
2
0
increases linearly with time, indicating that self-diffusion
and the changes of the particle connectivity take place easily and that the membrane is
essentially a fluid within the simulation time. Using the expression of the diffusion coefficient
D
lim
t
t
41
MSD
, we obtained from Fig. 2.3C that D = 2.7×10-2 d2/ts, where ts is the time scale.
A typical value of the diffusion coefficient for the lipid of the phospholipid bilayer membranes is
D ≈ 10-7cm2/s (133). By comparing with the real value, we can determine the time scale
st
0.07 μs
. Second, we found by examining the radial distribution function at Rcut = 2.6d and
27
k T
B
0.22
(see Fig. 2.3D) that the membrane does not possess a crystalline order at a distance
larger than the equilibrium distance, further demonstrating the characteristic of the fluid
membrane.
Figure 2.3. (A) Thermally equilibrated fluid membrane of N = 35500 particles at a reference
time, representing a membrane with dimension of approximately 1 μm × 1 μm The tiles with
different colors tiles are used to differentiate the positions of the particles at a reference time. (B)
After 5×105 time steps, the particles are seen to be mixed due to diffusion, demonstrating the
fluidic behavior of the model membrane. (C) Linear time dependence of MSD of the two-
component membrane model. (D) Radial distribution function in the 2D fluid membrane
embedded in 3D
2.3.2. Membrane elasticity
Within the framework of continuum mechanics, the free energy of a fluid membrane can be
described by the classical Canham-Helfrich theory (69), as
28
κ
CCC
12
0
2
2
dACCκ
21
,
(2.8)
F
γ
where
1C and
2C are the principal curvatures,
0C is the spontaneous curvature , γ is the surface
tension, κ and κ are the bending and Gaussian rigidities (134), respectively. The first term on
the right-hand side of Eq. (2.8) is the area-expansion energy, while the second and third terms
represent the normal and Gaussian bending energies, respectively. Since the topology of the
membrane remains unchanged, the Gaussian bending energy results in a constant contribution,
and thus this energy term can be dropped. Surface tension is defined as
σ
11
ZLσ
22
γ
2
,
where
11σ and
22σ are the components of the in-plane Virial stress (10) in a 3D periodic
supercell calculation, and Lz is height of the supercell, so that has the unit of N/m. Since the
spectrin network is described by the WLC model, which does not directly introduce resistance to
bending, we expect that the Canham-Helfrich theory for lipid bilayers is still valid in the
equilibrium. In what follows, we extract the bending rigidity from the power spectrum of the
height thermal fluctuations.
Applying the equipartition theorem on the Helfrich free energy in the Monge representation (9,
69, 74, 126), one can express the power spectrum as
2
~
qh
TK
B
2
q
q
4
,
2
l
(2.9)
29
where
~
qh
is the discrete Fourier transform of the out-of-plane displacement
rh
of the
membrane, defined as
~
qh
l
L
ieh
rqr
n
,
(2.10)
with L being the lateral size of the supercell,
l
l
l
r
min
1 6
2
smallest particles, and q is the norm of the wave vector
q
d
y
x qq ,
sets the mesh size equal to size of
Lnn
,
, i.e.,
2
qq
x
,
y
x
y
.
Figure 2.4. Vertical displacement fluctuation spectrum of two-component membrane as a
function of the wave number q.
To compute the bending rigidity, we first constructed an equilibrated membrane at kBT/ε = 0.22
and with zero tension. This effectively excludes the membrane tension effect, and consequently,
one expects that the power spectrum exhibits only a
4q dependence on the wave vector. Then,
30
we calculated
~
2
qh
from the raw data of each recorded configuration and averaged them over
all the available configurations to evaluate
~
2
qh
. Finally, by fitting the numerical data to
Eq.(2.9) for γ = 0, via a minimization procedure based on a nonlinear regression scheme, one
obtains the bending rigidity κ = 11.3kBT of the membrane model (see Fig. 2.4). The numerical
result lies within the experimental range of
10
k T
B
20
k T
B
for lipid bilayer (126, 129).
2.3.3 Phase diagram and the effects of the interaction potential parameters on membrane
properties
The phase diagram of the membrane model based on the potential parameters Rcut and T is
plotted. The dependence of the diffusion coefficient D on the temperature T and Rcut as well as
the dependence of the bending rigidity on the parameter α are investigated to study the effects of
the parameters of the interaction potential on the behavior of the membrane model.
First, we determined the conditions under which the model behaves as a stable fluid membrane.
The main parameters which govern the behavior of the model are Rcut and T. We constructed a
two-dimensional phase-diagram, as shown in Fig. 2.5A, where simulations are performed to
determine the diffusivity of the membrane model. A similar analysis has been conducted by
Cooke et al.(135). The phase diagram is based on a system consisted of N = 35500 coarse-
grained particles. At low temperature and for small Rcut, the model behaves as a gel with very
low diffusivity. At large values of Rcut, the model is unstable independently of the temperature.
As the temperature increases, the range of Rcut/d within which the membrane behaves as stable
fluid widens. The gel-fluid boundary is identified by a sudden increase in the diffusion
31
coefficient and the disappearance of the crystalline order at a distance larger than the equilibrium
distance. The transition from fluid to unstable state (gas phase) is characterized by the
observation that the tensionless membrane starts to break apart (135) .
Second, the dependence of the diffusion coefficient D on the temperature T for Rcut= 2.6d and
2.8d, and α = 1.55 is studied. Measurements of D are limited in the fluid phase of the membrane.
As shown in Fig.2.5B, D increases monotonically with T for both cases of Rcut= 2.6d and 2.8d, as
higher temperature leads to larger kinetic energy. On the other hand, an increase in Rcut from 2.6d
to 2.8d broadens the interaction potential well between the particles, as shown in Fig. 2.2C,
resulting in a decrease in the repulsive force. This decrease in repulsive force makes the particles
pass each other with lower resistance and thus contribute to an increase in the diffusion
coefficient. In addition, we also studied the dependence of the bending rigidity on the parameter
α at three temperatures. Rcut is chosen to be 2.6d. The main role of α in the interaction potential is
to regulate the energy well when the two interacting particles are disoriented from their favorable
relative orientation (see Fig. 2.2B) and therefore is closely related to the bending rigidity of the
membrane model. As shown in the Fig. 2.5C, the bending rigidities κ increases monotonically
with α for three different temperatures. We note that κ increases faster when the temperature is
lower. As α increases, the function
aαA ,
decreases and the total potential shifts from attractive
to repulsive, imposing greater penalty for the non-parallel arrangement of neighboring particles
and forcing the membrane to become a flat 2D sheet. It is also illustrated in the Fig.2.5C that the
temperature affects the bending rigidity of the membrane. The bending rigidity drops
significantly when the temperature increases because a higher temperature increases the
equilibrium distance between the membrane particles.
32
Figure 2.5. (A) Phase diagram. The hand-drawn lines show the phase boundaries. (B)
Dependence of the diffusion coefficient D on the temperature T for Rcut= 2.6d and 2.8d,
respectively. (C) Dependence of the bending rigidity κ on the parameter α at temperature T =
0.175, 0.22 and 0.275, respectively.
2.4. Application
In this section, the developed two-component membrane model is applied in shearing
experiments where the shear moduli of the membrane are measured. The numerical results are
compared to experimentally obtained values. We also examine the effect of the reduction of the
spectrin network connectivity on the shear modulus.
As discussed above, phospholipid bilayers behave as 2D fluid-like structures and thus they resist
bending but cannot sustain static in-plane shear stress because the lipids and the proteins can
diffuse freely within the membrane. The resistance of the membrane under low shearing strain
33
rate is due only to the cytoskeleton. At the high shear strain rate, however, the viscosity of the
lipid bilayer also contributes to the total resistance to the deformation. The membrane is sheared
up to a shear strain of 1 at a temperature of kBT/ε = 0.22 (see Fig. 2.6 A and B) at the two strain
rates of 0.001d/ts and 0.01d/ts, respectively. The response of the membrane to shearing at the low
strain rate of 0.001d/ts is illustrated by the blue line in Fig. 2.6C. At this strain rate, the shear
viscosity of the lipid bilayer is not detectable and the total resistance is due only to the spectrin
network. This becomes apparent from the fact that the red line in Fig. 2.6C, which describes the
shear stress – shear strain response of the corresponding WLC network, follows the pattern of the
blue line, which describes the shear stress-shear strain response of the entire membrane. It is
worth noting that the WLC model captures the experimentally identified stiffening behavior of
the RBC membrane (136) and the behavior of a detailed spring-based model of cytoskeleton
network (109). When the strain rate is increased to 0.01d/ts, the viscosity of the lipid bilayer
contributes to the total shear resistance (black line in Fig. 2.6C). As a result the initial value of
the shear stress is not zero but 8 μN/m, which is equal to the shear stress measured during the
shear deformation of a one-component lipid bilayer model.
34
Figure 2.6. (A) Thermally equilibrated fluid membrane at a temperature of B
k T
0.22
. (B)
Sheared membrane at a temperature of B
k T
0.22
at engineering shear strain of 1. (C) Shear
stress-strain response of the membrane at two strain rates. Red curve represents the response of
the "bare" spectrin network. The blue and black curves signify the shear stress obtained for the
strain rates of 0.001d/ts and 0.01d/ts, respectively. (D) Shear moduli of the membrane versus
spectrin network connectivity at the engineering shear strains of 0.1 and 0.9, respectively.
Up to this point, the connectivity of the spectrin network is maintained at 100%, meaning that
each actin junction is connected with its 6 neighboring junctions through the WLC potential
forming a perfect 2D six-fold structure. However, there are both experimental evidence (137-140)
and sound theoretical studies (141, 142) showing that the connectivity of the spectrin network in
the RBC membrane is not 100% and that the network is not a perfect 2D six-fold structure.
Electron microscopy images demonstrate that while most of actin junctions are connected with
other 6 actin junctions, a small amount of actins junctions are connected with 5 or 7 actin
junctions (137, 138). Subsequent atomic force microscopy (AFM) results showed a lower
35
connectivity of cytoskeleton is about 3 and 4 links per actin junction, leading to square, pentagon
and hexagon-like structures in the network (139, 143, 144). A CGMD model of the RBC
cytoskeleton predicted that the biconcave shape of RBCs can be attained only if the network
undergoes constant remodeling to relax the in-plane shear elastic energy to zero during the
deformation meaning that the connectivity of the spectrin network is changeable (142). Recent
experimental results presented the evidence that the remodeling of the spectrin network plays a
vital role in determining the cell deformation and biconcave shape of the RBCs (140).
As shown in Fig.2.6C, when the spectrin network connectivity is 100%, the shear modulus of the
membrane at small deformations is approximately 10 μN/m while it is 28 μN/m at 90%
engineering shear strain. The linear elastic shear modulus of the model is larger than the
experimentally measured values of 7.3 μN/m (136) and 4 - 9 μN/m (126) as well as the
simulation results of 8.3 μN/m (142). This is probably because the connectivity of the employed
spectrin network is 100% while, as it was mentioned above, the actual spectrin network is not
perfect. Since the membrane resistance to the shearing results mainly from the spectrin network,
it is expected that the shear modulus of the membrane will decrease as the connectivity of the
spectrin network is reduced. To investigate the effect of the spectrin network connectivity on the
mechanical properties of the membrane model, the shear modulus of the membrane is measured
when the connectivity of the spectrin network is reduced from 100% to 50%. The resulting shear
moduli at small and large deformations are shown in Fig. 2.6D, which clearly demonstrated that
their values drop dramatically with decreased spectrin network connectivity especially at large
deformations. The shear moduli of the membrane fall within the experimentally measured value
range when the network connectivity is between 60% and 90%. RBC with lower spectrin
36
network connectivity is softer and it is easier to deform when passing through narrow blood
vessels. However, the low connectivity could lead to insufficient surface shear resistance to
maintain the cell's biconcave shape and integrity during the blood circulation. Our simulation
results show that when the connectivity drops below 50%, the membrane model does not resist to
shear deformation indicating that the RBCs completely lose the ability of recovering its
biconcave shape after deformation (109).
2.5. Summary
We have developed a two-component CGMD model for RBC membranes. In this model, the
lipid bilayer and the cytoskeleton are considered as an ensemble of discrete particles that interact
through a direction-dependent pair potential. Three types of coarse-grained particles are
introduced, representing a cluster of lipid molecules, actin junctions and band-3 complexes. The
large grain size extends the accessible length and time scales of the simulations to ~ µm and ~
ms. By tailoring only a few parameters of the inter-grain interaction potential, the model can be
stabilized into a fluid phase manifested by the free diffusion of the particles and the reproduction
of the essential thermodynamic properties of the RBC membrane. An important feature of the
proposed model is the combination of the spectrin network with the lipid bilayer, which
represents a more compete representation of the RBC membrane compared to one-component
CGMD model and to most of the continuum models. This model allows us to study the behavior
of the membrane under shearing. The behavior of the model under shearing at different strain
rates illustrates that at low strain rates up to 0.001d/ts, the developed shear stress is mainly due to
the spectrin network and it shows the characteristic non-linear behavior of entropic networks,
while the viscosity of the fluid-like lipid bilayer contributes to the resulted shear stress at higher
37
strain rates. Decrease of the spectrin network connectivity results to significant decrease of the
shear modulus of the membrane, which demonstrates that the cytoskeleton carries most of the
load applied on the cell while the lipid bilayer only functions as 2D fluid. The values of the shear
moduli measured from the membrane with reduced spectrin network connectivity are in a good
agreement with previous experimental, theoretical, and numerical results.
38
Chapter 3.
Erythrocyte Membrane Model with Explicit Description of the
Lipid Bilayer and the Spectrin Network
Abstract
The membrane of the red blood cell (RBC) consists of spectrin tetramers connected at actin
junctional complexes, forming a 2D six-fold triangular network anchored to the lipid bilayer.
Better understanding of the erythrocyte mechanics in hereditary blood disorders such as
spherocytosis, elliptocytosis, and especially, in sickle cell disease requires the development of a
detailed membrane model. Here, we introduce a mesoscale implicit-solvent coarse-grained
molecular dynamics (CGMD) model of the erythrocyte membrane which explicitly describes the
phospholipid bilayer and the cytoskeleton, by extending a previously developed two-component
RBC membrane model. We show that the proposed model represents RBC membrane with the
appropriate bending stiffness and shear modulus. The timescale and self-consistency of the
model are established by comparing our results to experimentally measured viscosity and
thermal fluctuations of the RBC membrane. Furthermore, we measure the pressure exerted by
the cytoskeleton on the lipid bilayer. We find that defects at the anchoring points of the
cytoskeleton to the lipid bilayer (as in spherocytes) cause a reduction in the pressure compared to
an intact membrane, while defects in the dimer-dimer association of a spectrin filament (as in
elliptocytes) cause an even larger decrease in the pressure. We conjecture that this finding may
explain why the experimentally measured diffusion coefficients of band-3 proteins are higher in
elliptocytes than in spherocytes, and higher than in normal RBCs. Finally, we study the effects
that possible attractive forces between the spectrin filaments and the lipid bilayer on the pressure
39
applied on the lipid bilayer by the filaments. We discover that the attractive forces cause an
increase in the pressure while they diminish the effect of membrane protein defects. As this
finding contradicts with experimental results, we conclude that the attractive forces are moderate
and do not impose a complete attachment of the filaments to the lipid bilayer.
3.1. Introduction
The human red blood cell (RBC) repeatedly undergoes large elastic deformations when passing
through narrow blood vessels. The large flexibility of the RBCs is primarily due to the cell
membrane, as there are no organelles and filaments inside the cell. The RBC membrane is
essentially a two-dimensional (2D) structure, comprised of a cytoskeleton and a lipid bilayer,
tethered together. The lipid bilayer includes various types of phospholipids, sphingolipids,
cholesterol, and integral membrane proteins, such as band-3 and glycophorin (see Fig. 3.1A). It
resists bending but cannot sustain in-plane static shear stress as the lipids and the proteins diffuse
within the lipid bilayer at equilibrium. The RBC membrane cytoskeleton is a 2D six-fold
structure consisting of spectrin tetramers, which are connected at the actin junctional complexes.
The cytoskeleton is tethered to the lipid bilayer via "immobile" band-3 proteins at the spectrin-
ankyrin binding sites and via glycophorin at the actin junctional complexes (see Fig. 3.1A).
Although the mechanical properties and biological functions of the RBC membrane have been
well studied in the past decades, the interactions between the lipid bilayer and cytoskeleton as
well as the interactions between the cytoskeleton and transmembrane proteins are not yet fully
understood. The cytoskeleton plays a major role in the integrity of the RBC membrane, as is
evident in blood disorders where defects in membrane proteins lead to membrane loss and
reduced mechanical robustness of the RBC (3, 25, 145). In hereditary spherocytosis (HS), the
40
tethering of the cytoskeleton to the lipid bilayer (vertical interaction in Fig. 3.1A) is partially
disrupted resulting in membrane loss and subsequently in the spherical shape of the RBCs. In
hereditary elliptocytosis (HE), the cytoskeleton is disrupted at α-β spectrin linkages or at
spectrin-actin-4.1R junctional complexes (horizontal interactions in Fig. 3.1A) (3, 25, 145, 146).
This partial disruption of the cortex diminishes the ability of the RBC to recover its biconcave
shape after undergoing large deformations. In addition, because the spectrin filaments act as
barriers, restricting the lateral diffusion of the "mobile" band-3 proteins, defects in the vertical
and horizontal interactions between the cytoskeleton and the lipid bilayer modify the regular
diffusion of band-3 proteins (115, 147-150).
Several approaches have been followed for the mathematical description and modeling of the
RBC membrane. At one end of the spectrum, there are the continuum membrane models based
on elasticity theory (9, 69-74, 151, 152). At the other end, atomistic simulations mainly study the
behavior of lipids in the lipid bilayer (76-79). However, it is very challenging for continuum
models to account for the detailed structure and defects in the RBC cytoskeleton, while it is not
feasible for atomistic methods to simulate a representative sample of the RBC membrane
including lipids, membrane proteins, and the cytoskeleton. Because of these limitations, particle-
based mesoscale models were introduced to study the biomechanical behavior of the RBC
membrane. These models fall into two main groups. In one group, the membrane is modeled as a
2D canonical hexagonal network of particles where the immediate neighbors are connected via a
worm-like chain (WLC) potential that represents an actin filament. The bending rigidity induced
by the lipid bilayer is represented by a bending potential applied between two triangles with a
common side (80, 81, 110, 141, 142). In the other category, the lipid bilayer is simulated by
41
coarse-grained methods where each lipid molecule is coarse-grained into several connected
beads (84, 86, 88, 91, 92, 95). The lipids are forced to assemble the bilayer by additional solvent
particles, in the case of explicit solvent models, or by an additional potential, in the case of
solvent-free models. In the explicit solvent models, hydrophobic interactions are employed
between the lipids and solvent particles to represent effects of the water molecules (84, 87, 94,
102-105). In the implicit solvent models, the effect of the solvent is taken into account by
employing orientation-dependent interaction potentials between the lipid particles (83, 91, 92, 96,
101, 107). At a higher level of coarse-graining, a group of lipid molecules are coarse-grained into
one bead (108, 111). Drouffe et al. (83) simulated biological membranes by introducing a one-
particle-thick, solvent-free, coarse-grained model, in which the inter-particle interaction is
described by a Lennard-Jones (LJ) type pair potential depending not only on the distance
between the particles but also on their directionality. Noguchi and Gompper (98) developed a
one-particle thick, solvent-free, lipid bilayer model by introducing a multibody potential that
eliminated the need of the rotational degree of freedom. Yuan et al. (108) introduced a similar
approach, but instead of the LJ potential, a soft-core potential was used to better represent the
particle self-diffusion. An overview of particle-based models for the RBC membrane can be
found in recent reviews (82, 96, 100, 153).
42
Figure 3.1 (A) Schematic of the human RBC membrane. The blue sphere represents a lipid
particle and the red sphere signifies an actin junctional complex. The grey sphere represents a
spectrin particle and the black sphere represents a glycophorin particle. The yellow and green
circles correspond to a band-3 complex connected to the spectrin network and a mobile band-3
complex respectively. A mesoscale detailed membrane model. (B) top view of the initial
configuration, (C) side view of the initial configuration, and (D) a side view of the equilibrium
configuration. "A" type particles represent actin junctional complexes, "B" type particles
43
represent spectrin proteins, "C" type particles represent glycophorin proteins, "D" type particles
represent a band-3 complex that are connected to the spectrin network ("immobile" band-3), "E"
type particles represent band-3 complex that are not connected to the network ("mobile" band-3),
and "F" type particles represent lipid particles.
Absence of explicit representation of either the lipid bilayer or the cytoskeleton in the
aforementioned particle-based models limits their applications in the study of the interactions
between the cytoskeleton and the lipid bilayer, and between the cytoskeleton and diffusing
membrane proteins. Recently, a model consisting of two layers of 2D triangulated networks,
where one layer represents the cytoskeleton while the other one represents the lipid bilayer, was
introduced (154). This approach is computationally very efficient and it can be used in
simulations of blood flow. However, it is not capable of modeling the interactions between the
spectrin filaments and the lipid bilayer, which is frequently essential, as exemplified in the study
of the diffusion of membrane proteins such as the band-3 (115, 147-150). Li and Lykotrafitis
(111) introduced a two-component RBC membrane model where the lipid bilayer consists of CG
particles of 5 nm size while the cytoskeleton is comprised of particles that represent actin
junctional complexes and form a canonical hexagonal network. These "actin junction" particles
are connected by the WLC potential representing a spectrin filament. Due to the implicit
representation of the spectrin filaments, this approach does not consider the interactions between
the filaments and the lipid bilayer as well as between the filaments and membrane protein during
diffusion. Auth et. al (155) developed an analytical model to study the cytoskeleton and lipid
bilayer interactions. This analytical model is able to estimate the pressure exerted on the lipid
bilayer by the cytoskeleton because of the thermal fluctuations of the spectrin filaments. The
44
computed pressure is then used to obtain the deformation of the lipid bilayer induced by the
spectrin filaments, which was calculated to be 0.1 nm ~ 1 nm. This deformation is small
compared to the local deformation (~15 nm) of the lipid bilayer caused by the lateral
compression applied by the spectrin cytoskeleton (151, 156).
In this chapter, we introduce a two-component implicit-solvent CGMD RBC membrane model
comprised of the lipid bilayer and the RBC cytoskeleton by extending a previously developed
RBC membrane model (111). The key feature of this new model is the explicit representation of
the cytoskeleton and the lipid bilayer by CG particles. We determine the parameters of the model
by matching the bending stiffness and shear modulus of the membrane model with existing
experimental results (129, 136, 157). Then, we extract the timescale of our simulations by
measuring the viscosity of the membrane model and comparing it with the experimentally
obtained values. The timescale is later confirmed by measuring the thermal fluctuation frequency
of the lipid bilayer and the spectrin filaments. In addition, we measure the pressure exerted by
the spectrin filaments on the lipid bilayer. These results are compared with analytically estimated
pressure values (118, 155). Finally, we investigate the effects of the attraction between spectrin
filaments and the lipid bilayer on the pressure applied by the cytoskeleton on the lipid bilayer in
the case of the normal RBC membrane and for membranes with defective proteins.
3.2. Model and simulation method
The proposed model describes the RBC membrane as a two-component system, comprised of the
cytoskeleton and the lipid bilayer. We first introduce the cytoskeleton, which consists of spectrin
filaments connected at the actin junctional complexes forming a hexagonal network. The actin
45
junctional complexes are represented by red particles (see Fig. 3.1B-D) that have a diameter of
approximately 15 nm and are connected to the lipid bilayer via glycophorin. Spectrin is a protein
tetramer formed by head-to-head association of two identical heterodimers. Each heterodimer
consists of an α-chain with 22 triple-helical segments and a β-chain with 17 triple-helical
segments (6). In the proposed model, the spectrin is represented by 39 spectrin particles (grey
particles
in Fig. 3.1B-D) connected by unbreakable springs. The spring potential,
u
s-s
cy
( )
r
(
k r
0
-
r
s-s
eq
2
) / 2
, is plotted as the green curve in Fig. 3.2, with equilibrium distance
between the spectrin particles
r
Ls-s
eq
max / 39
, where Lmax is the contour length of the spectrin
(~200 nm) and thus
r
s-s
eq
5
nm. The spectrin chain is linked to the band-3 particles (yellow
particles) at the area where the α-chain and the β-chain are connected. The two ends of the
spectrin chains are connected to the actin junctional complexes via the spring potential
u
a-s
cy
( )
r
k r r
(
-
0
a-s
eq
2
) / 2
, where the equilibrium distance between an actin and a spectrin particle is
r
a-s
eq
10
nm. The spring constant k0 will be determined below. Spectrin particles that are not
connected by the spring potential interact with each other via the repulsive part of the L-J
potential
4
12
r
ij
6
r
ij
0
r
ij
R
cut,LJ
s-s
r
eq
r
ij
R
cut,LJ
s-s
r
eq
u
rep
(
r
ij
)
(3.1)
where ε is the energy unit and σ is the length unit. rij is the distance between spectrin particles.
The cutoff distance of the potential Rcut,LJ is chosen to be the equilibrium distance r s-s
eq between
46
two spectrin particles. The potential is plotted as the black curve in Fig. 3.2. The spring constant
k0 = 57 ε/σ2 is chosen to be identical to the curvature of
u
LJ (
r
ij
) 4
/
12
/
r
ij
r
ij
6
at
the energy well bottom to reduce the number of free parameters. The cytoskeleton introduced
above is directly connected to the lipid bilayer via the band-3 particles D and the glycophorin
particles C (see Fig. 3.1C). In addition, we have investigated the effect of attractive interactions
of various strengths between spectrin filaments and the lipid bilayer on the diffusion of band-3
particles. Previous studies on lipid–spectrin filaments interactions have suggested that spectrin
binds to the negatively charged lipid surfaces with association constants of 2-10×106 M-1 (158-
163), while the association constants of spectrin-ankyrin, ankyrin-band-3, spectrin-protein 4.1-
actin are 2×107 M-1, 2×108 M-1 and 2×1012 M-2, respectively (164, 165). For simplicity, we
applied the attractive part LJ potential between spectrin and lipid particles
u
att
(
r
ij
)
4
n
12
r
ij
6
r
ij
0
n
r
ij
l-s
r
eq
,
r
ij
l-s
r
eq
(3.2)
where n is a parameter used to tune the attractive energy between the spectrin filaments and lipid
bilayer.
r
l-s
eq
5nm
is the equilibrium distance between spectrin particles and lipid particles. Our
simulation results show that the cytoskeleton is completely attached to the lipid bilayer when n ≥
0.2. Because the cytoskeleton of the RBC membrane is a highly flexible structure with a bending
rigidity of κ ~ 0.024 – 0.24kBT (127, 128), which is about two orders of magnitude smaller than
the bending rigidity of the lipid bilayer κ ~ 10 - 20 kBT (129, 157) as well as the bending rigidity
47
of the RBC membrane κ ~ 10 - 50 kBT (157), we did not consider bending rigidity for the
spectrin network in this model.
Figure 3.2 The interaction potentials employed in the membrane model. The blue curve
represents the pair-wise potential between lipid particles. The green curve represents the spring
potential between spectrin particles. The red curve represents the spring potential between actin
and spectrin particles. The black curve represents the repulsive L-J potential between the lipid
and spectrin particles.
Three types of CG particles are introduced to represent the lipid bilayer and band-3 proteins (see
Fig. 3.1B-D). The blue color particles denote a cluster of lipid molecules. Their diameter of 5 nm
is approximately equal to the thickness of the lipid bilayer. The black particles represent
glycophorin proteins with the same diameter as the lipid particles. The band-3 protein consists of
two domains: (i) the cytoplasmic domain of band-3 with a dimension of 7.5×5.5×4.5nm that
contains the binding sites for the cytoskeletal proteins, and (ii) the membrane domain, with a
dimension of 6×11×8 nm, whose main function is to mediate anion transport (166, 167). We
represent the membrane domain of band-3 by a spherical CG particle with a radius of 5 nm. The
48
volume of the particle is similar to the excluded volume of the membrane domain of a band-3.
However, when band-3 proteins interact with the cytoskeleton, the effect of the cytoplasmic
domain has to be taken into account and thus the effective radius is approximately 12.5 nm. One
third of band-3 particles, which are connected to the spectrin network, are depicted as yellow
particles (see Fig. 3.1B). The rest of the band-3 particles, which are free to diffuse in the lipid
bilayer, simulate the mobile band-3 proteins and they are shown as green particles (see Fig.
3.1B). The CG particles, which form the lipid bilayer and transmembrane proteins, carry both
translational and rotational degrees of freedom (xi, ni), where xi and ni are the position and the
orientation (direction vector) of particle i, respectively. The rotational degrees of freedom obey
the normality condition ni = 1. Thus, each particle effectively carries 5 degrees of freedom. xij
= xj - xi is defined as the distance vector between particles i and j. rij ≡ xij and ij
x
x r
ij
ij
are the
distance and the unit vector respectively. The particles, forming the lipid membrane and
membrane proteins, interact with one another via the pair-wise potential
n , n , x
i
j
ij
u
R
r
ij
A
,
a
n , n , x
i
j
ij
u
A
r
ij
,
(3.3)
u
m em
with
,
R
cut mem
r
ij
R
cut,mem
-
r
eq
8
r
eq
u r
R
ij
u r
A
ij
u r
R
ij
k
2
k
u
A
r
ij
R
cut mem
,
-
r
ij
R
cut,mem
0,
49
u
4 8
(
r
ij
)
-
4
k
for
r R
ij
<
cut,mem
k
r
ij
for
R
cut,mem
for r R
cut,mem
ij
(3.4)
where uR(rij) and uA(rij) are the repulsive and attractive components of the pair potential,
respectively. α is a tunable linear amplification factor. The function A(α,a(ni,nj,xij))=
1+α(a(ni,nj,xij) -1) tunes the energy well of the potential, through which the fluid-like behavior
of the membrane is regulated. In the simulations, α is chosen to be 1.55 and the cutoff distance of
the potential Rcut,mem is chosen to be 2.6σ. The parameters α and Rcut,mem are selected to maintain
the fluid phase of the lipid bilayer. Detailed information about the selection of the potential
parameters can be found from author's previous work (111). k is selected to be 1.2 for the
interactions among the lipid particles and k = 2.8 for interactions between the lipid and the
protein particles, such as glycophorin and band-3. Fig. 3.2 shows only the potential between lipid
particles (blue curve). The interactions between the cytoskeleton and the lipid bilayer are
represented by the repulsive part of the L-J potential as shown in Eq. (3.1). The cutoff distance of
the potential Rcut,LJ is adjusted to be the equilibrium distances between different pairs of CG
particles. The equilibrium distance r a-l
eq between the actin particles and the lipid particles is 10
nm while r a-b
eq
between the actin particles and the band-3 particles is 20 nm. The equilibrium
distance r l-s
eq
between the spectrin particles and the lipid particles is 5 nm, while the equilibrium
distance r b -s
eq
between the spectrin particles and the band-3 particles is 15 nm.
The equation of translational motion for all the CG particles is
m
i
x
i
( )
V
x
i
,
(3.5)
50
where for the CG particles forming the lipid membrane and the transmembrane proteins
V
N
j
1
(
u
mem ,ij
u
)
LJ,ij
, and for the CG particles comprising the cytoskeleton,
V
N
j
1
(
u
LJ,ij
)
u
.
cy, i
The equation of rotational motion for the CG particles forming the lipid bilayer and proteins in
the lipid bilayer is
(
m
i
n
i
N
j
1
u
mem,ij
)
n
i
u
mem,ij
j
N
1
(
n
i
)
n n
i
i
m
i
n n n
i
i
i
,
(3.6)
where
im~ is a pseudo-mass with units of energy × time2, and the right-hand side of Eq. (3.6)
obeys the normality constraint ni = 1. The pseudomass
im~ is chosen to be
l
im is the mass of the lipid particles.
m m
2
~ l
i
i
, where
The system consists of N = 29567 CG particles. The dimension of the membrane is
approximately 0.8 μm × 0.8 μm. The numerical integrations of the equations of motion (Eqs. (3.5)
and (3.6)) are performed using the Beeman algorithm (10, 130). The temperature of the system is
maintained at kBT/ε = 0.22 by employing the Nose-Hoover thermostat. The model is
implemented in the NAT ensemble (95, 131, 132). Periodic boundary conditions are applied in
all three directions. Since the model is solvent-free and the membrane is a two-dimensional
structure, we controlled the projected area instead of the volume. The projected area is adjusted
to result in zero tension for the entire system at the equilibrium state. Given the diameter of the
lipid particles l-l
r
eq
1/6
2
5nm
, the length unit σ is calculated to be σ = 4.45 nm. The timescale
that guides the choice of the timestep in the MD simulations is
t
s
51
m d
2
i
1 2
. The timestep of
the simulation is selected to be Δt = 0.01ts. Since the CG particles used in the simulations do not
correspond to real molecules, the employed timescale does not have an immediate correlation
with the real system. The timescale in our simulation is established by measuring the viscosity of
the membrane model and comparing it with the experimentally measured value for the RBC
membrane. For an independent confirmation, we also obtain the timescale by measuring the
thermal fluctuation frequencies of the spectrin filaments and the lipid membrane. The details of
the timescale will be discussed in the section 3.4. For simplicity, we assume that the cytoskeleton
possesses a perfect two-dimensional (2D) six-fold triangular structure with a fixed connectivity
and that each spectrin filament is anchored to the band-3 proteins at its mid-point. In reality, the
RBC cytoskeleton contains numerous defects while the band-3-ankyrin connections and the
cytoskeleton undergo dynamic remodeling (109, 140, 168).
3.3. Membrane properties
3.3.1 Membrane fluidity
In this section, we show that the proposed model of the RBC membrane reproduces the
appropriate mechanical properties. To ensure that our system represents the lipid bilayer, it is
necessary to examine whether the CG particles can diffuse in a fluidic manner. Figs. 3.3A and B
show the positions of the CG particles of a thermally equilibrated fluid membrane at two
separate times. The tiles with different colors are used to differentiate the positions of the
particles at the initial moment. After 1×106 time steps, the colored particles are mixed due to
diffusion, demonstrating the fluidic nature of the membrane. We further verify the fluidic
characteristics of the membrane by first showing that the mean squared displacement (MSD),
defined by
1
N
Nj
j
1
x
j t
x
j
20
, increases linearly with time (see Fig. 3.3C), and second by
52
showing that that the correlations between the CG lipid particles are lost beyond a few particle
diameters, suggesting a typical fluidic behavior of the membrane model at B
k T
0.22
and
Rcut,mem=2.6σ (see Fig. 3.3D).
Figure 3.3 (A) Thermally equilibrated fluid membrane of N = 29567 particles at a reference time,
representing a membrane with dimension of approximately 0.8 μm × 0.8 μm The tiles with
different colors are used to differentiate the positions of the particles at a reference time. (B)
After 1×106 time steps, the particles are mixed due to diffusion, demonstrating the fluidic
behavior of the membrane model. (C) Linear time dependence of the mean square displacement
(MSD) of the two-component membrane model. (D) Radial distribution function of the 2D fluid
membrane embedded in 3D.
3.3.2 Measurements of membrane bending rigidity, shear modulus and viscosity
Here, we compute the bending rigidity, shear modulus, and viscosity of the membrane model.
We obtain the bending rigidity of the membrane by measuring the fluctuation spectrum of the
53
membrane at zero tension and then fitting the data to the expression of the Helfrich free energy
in the Monge representation (9, 69, 74, 157),
2
~
qh
TK
B
2
q
q
4
2
l
(3.7)
where
~
qh
is the discrete Fourier transform of the out-of-plane displacement
rh of the
membrane, defined as
~
qh
l
L
rqr
ieh
n
,
where L is the lateral size of the supercell,
l
particles, and q is the wave vector
q
l-l
r
eq
x qq ,
(3.8)
1 6
2
y
, i.e.,
qq
x
,
sets the mesh size equal to size of lipid
y
nn
x
2
,
L
y
. The power
spectrum in the Fig. 3.4 exhibits a q−4 dependence on the wave vector. The deviation at large
wave vectors is due to limitation defined by the size of the CG particles, as at wave lengths
smaller than the particle size, the continuum approximation breaks down. The bending rigidity of
the membrane is found to be κ = 11.3kBT (see Fig. 3.4), which lies within the experimental range
of (10kBT – 20kBT) for lipid bilayer (129, 157).
54
Figure 3.4 Vertical displacement fluctuation spectrum of membrane model as a function of the
dimensionless quantity q, where q is the wave number and σ is the unit length corresponding
to ~4.45 nm.
For the measurements of the shear moduli, the membrane is sheared up to a shear strain of 1 with
the strain rate =0.001σ/ts, as shown in Figs. 3.5A and B. The response of the membrane to
shearing is illustrated by the blue line in Fig. 3.5C. The proposed model captures the experimentally
identified stiffening behavior of the RBC membrane, which is due to the spectrin network. The shear
modulus of the membrane at small deformations is approximately 12 μN/m while it is increased to 27
μN/m at engineering shear strain of 0.9. The initial elastic shear modulus of the model is larger than
the experimentally measured values of 4 - 9 μN/m (136, 157). This is most likely due to the
implementation of a perfect network in the present model, while the spectrin network of the RBC
membrane is not perfect (137, 138, 141, 142) probably because of ATP-induced dissociations (151,
155). Experimental measurements showed that ATP-induced dissociations played a crucial role
in forming defects in the cytoskeleton. In addition, a larger shear modulus was measured in ATP-
depleted RBCs compared to normal RBCs (169). At the strain rate of =0.001σ/ts, the shear
viscosity of the lipid bilayer is not detectable and the total resistance to the shearing is caused
55
only by the cytoskeleton. This is confirmed by the fact that the shear stress-strain response of the
cytoskeleton (red curve in Fig. 3.5C) follows the pattern of the shear stress-strain response of the
entire membrane (blue curve in Fig. 3.5C). This result is in agreement with our previous two-
component membrane model where the spectrin network is represented implicitly (111). At the
higher strain rates of 0.005σ/ts and to 0.01σ/ts, the viscosity of the lipid bilayer contributes to the total
resistance during shearing. For example, the total value of the measured shear stress corresponding to
0.01σ/ts strain rate (green curve in Fig. 3.5C) is the sum of the shear stress due to the cytoskeleton
(purple curve in Fig. 3.5C) and of the shear stress due to the viscosity of the lipid bilayer. By
subtracting the shear stress due to the cytoskeleton from the total shear stress, we obtain the viscous
shear stress to be τ = 0.008 ε/σ2. At the strain rate of = 0.005σ/ts, we computed the viscous shear
stress to be τ = 0.004 ε/σ2. We assume that during the shearing process, the homogeneous lipid
bilayer behaves as a simple two-dimensional viscous Newtonian fluid. Then, by applying the
shear stress - strain rate relation
, we calculate the shear viscosity of the lipid bilayer to
be
0.8
st L
/
3
, where L is the length of the membrane model. The shear viscosity of the lipid
bilayer can also be measured from shearing the membrane without the cytoskeleton. At the same
shear strain rate, the measured shear stress remains constant value with respect to different shear
strains. At the strain rate of =0.001σ/ts, no shear stress is measured from the lipid bilayer. At
the strain rate of = 0.005σ/ts, the shear stress is measured to be τ = 0.004 ε/σ2. When the strain
rate is further increased to = 0.01σ/ts, the shear stress is measured to be τ = 0.008 ε/σ2. Therefore,
the obtained shear viscosity of the lipid bilayer is consistent with the value we measured from
shearing the lipid bilayer with the cytoskeleton. The comparison between the numerical value
predicted by the model and the experimental value, is performed in the section 3.4 where we
determine the timescale
st for our simulations.
56
Figure 3.5 (A) Thermally equilibrated fluid membrane at a temperature of kBT/ε = 0.22. (B)
Sheared thermally equilibrated membrane at engineering shear strain of 1. (C) Shear stress-strain
response of the membrane at two different strain rates. The red curve represents the response of
the "bare" spectrin network. The blue and green curves signify the shear stress obtained at the
strain rates of 0.001σ/ts and 0.01σ/ts, respectively. The purple curve represents the shear stress
measured from the "bare" spectrin network plus the 8μN/m attributed to viscosity.
3.3.3 Measurements of fluctuation frequencies of the spectrin filaments and the lipid bilayer
The fluctuation frequency of the spectrin filaments in the cytoskeleton is measured by tracking
the normal displacement with respect to the lipid bilayer of a single spectrin particle in the
middle of the two binding sites. Then, we apply Fast Fourier Transform (FFT) on the
displacements and compute the power spectral density (PSD) of the vibrations. The
57
corresponding results in Fig. 3.6A show that the dominant fluctuation frequency of the filaments
is measured to be approximately fc = 0.009/ts. Regarding the fluctuation of the lipid bilayer, we
measure the average fluctuation displacement of the lipid bilayer patch in one triangular
compartment formed by the spectrin filaments. The frequency is then obtained by using the PSD
plot and it is found to be fl = 0.001/ts, which is approximately one order of magnitude smaller
than the fluctuation frequency of the filaments. Next, in a different simulation, we measure the
fluctuation frequencies of the filaments in the "bare" cytoskeleton, meaning that the lipid bilayer
is completely removed from the RBC membrane leaving only the cytoskeleton. Since the
cytoskeleton applies compression on lipid bilayer, the lipid bilayer is crumpled at equilibrium
and it exerts extension force to the cytoskeleton. Therefore, when the lipid bilayer is removed, an
effective repulsive potential is applied between two neighboring actin junctions in the
cytoskeleton to represent the effect of the lipid bilayer. The effective repulsive potential
A Au
( )
r
is described by
u
A A
( )
r
4
lipid
3
r
)
flat
H R
(
flat
r
)
(3.9)
R
(
R
flat
where Rflat is the distance between the two neighboring action junctions when the membrane
becomes flat. We take Rflat = 1.2 ReqA-A , where ReqA-A is the equilibrium distance between the
action junctions. ReqA-A is approximately 90 nm in our simulations. κlipid is the bending stiffness
of the lipid bilayer. H(x) is the Heaviside step function. α is chosen to be 0.36 so that the
pressure of the "bare" cytoskeleton is nearly zero at equilibrium temperature. More details about
the "bare" cytoskeleton can be found in a previous work by one of the authors (109). We found
that the frequency of the "bare" spectrin network is fc = 0.001/ts (see Fig. 3.6B), which is one
58
order of magnitude smaller than the fluctuation frequency of a spectrin filament in the
cytoskeleton that is connected to the lipid bilayer. The reason for this difference is that in the
"bare" spectrin network, the spectrin filaments are connected only to the actin junctional
complexes, while in the case of the cytoskeleton attached to the lipid bilayer, the spectrin
filaments are also anchored to the band-3 proteins. This means that in the "bare" network each
filament is connected only at its two ends while in the proposed membrane model each filament
has an additional binding site in the middle causing faster vibrations. The result is in agreement
with the theoretical prediction that the conformation time tc of a spectrin filament depends on the
number of parts per filament divided by the band-3 binding sites, Nb,
3
N (170). In the
b
t
c ~1 /
proposed membrane model, each filament comprises two parts (Nb = 2), as there is one band-3
binding site per filament. Therefore, the conformation time for the filaments in the RBC
membrane is about 8 times smaller than the conformation time for the "bare" network in which
each filament has only one part (Nb = 1).
Figure 3.6 (A) Power spectrum density (PSD) corresponding to thermal fluctuations of the
membrane model. The data marked as red circles are generated from the average displacement of
the particles belonging to a triangular compartment of the lipid bilayer. The data marked as blue
squares are generated from a particle positioned in the middle of a spectrin filament. fc represents
the thermal fluctuation frequency of the cytoskeleton and fl represents the thermal fluctuation
59
frequency of the lipid bilayer. Inset: equilibrium state of the proposed RBC membrane model. (B)
PSD corresponding to thermal fluctuations of a particle positioned in the middle of a spectrin
filament in the "bare" cytoskeleton. Inset: equilibrium state of "bare" cytoskeleton.
3.3.4 Timescales of the simulations
As we described previously, since the particles introduced in CGMD simulations do not
correspond to real atoms or molecules, the timescale
t
s
m
21
2 i
defined in the MD
simulation is not directly related to the evolution of a real physical system. Only through
comparison with a physical process can such a correspondence be established. Here, we compare
the shear viscosity obtained in the section 3.2, with the experimentally measured value of
3.6×10-7 N·s/m (171), via which we conclude that the timescale of our simulation is
approximately
ts ~ 3×10-6s. The corresponding
timestep of
the simulations
is
then
t
0.01
t
s
3 10
s
8
. After determining the timescale by employing the viscosity of the RBC
membrane, we need to test the self-consistency of the model by confirming that its predictions
for the fluctuation frequencies of the spectrin filaments and of the lipid bilayer are in agreement
with the expected physical values. The thermal fluctuation frequency of the lipid membrane
obtained from our simulation is fl = 0.001/ts ~ 333Hz, which is comparable with the theoretical
and experimentally measured frequency ~1000Hz (151, 169). Analytical estimation shows that
the conformation
time
conform
for
the
spectrin
filaments
in
the cytoskeleton
is
conform
3
cytoplasm ee
d
/ (
N k T
3
b
B
)~100 s
, where dee is the end to end distance of the spectrin filaments,
Nb is the number of parts per filament divided by the band-3 binding sites and ηcytoplasm is the
viscosity of the cytoplasm (170). In our model, the thermal fluctuation frequency of the spectrin
filament is measured to be fc = 0.009/ts. By using the timescale obtained above, we determine
60
that the model predicts the conformation time for the spectrin filaments to be
conform
,
~330 s
which is comparable with the analytically estimated conformation time ~100µs (170).
3.4. Mechanical interaction between the spectrin network and the lipid bilayer.
The lipid bilayer and the spectrin network along with their interactions play an essential role in
the structure and biological functions of the RBC. Experimental measurement of the pressure
applied on the lipid bilayer by spectrin filaments is challenging, as the length scale of the
interactions between the cytoskeleton and lipid is not easily accessed by dynamic cell mechanics
experiments. In this section, we compute the pressure exerted on the lipid bilayer by the spectrin
filaments and investigate how it is affected by the attraction between the spectrin filament and
the lipid bilayer. The pressure obtained from the simulations is compared with analytically
estimated values for the case of flexible linear polymer attached to the lipid bilayer at its two
ends (118, 155). Lastly, we examine how defects in the anchoring of the spectrin network to the
lipid bilayer and defects in the structure of the spectrin network influence the pressure applied on
the lipid bilayer.
In the proposed model, the interaction between the spectrin filaments and the lipid bilayer is
described by the repulsive part of the LJ-12 potential, as shown in Eq. (3.1). The filament-
induced local pressure is calculated by using the expression proposed by Cheung and Yip (172),
who consider the pressure as summation of the momentum flux and force across a planar unit
area in a time interval. Because in our model the spectrin filaments do not penetrate the lipid
bilayer and the local deformation of the lipid bilayer caused by the spectrin filaments is ~1 nm
(155), the pressure is measured by considering only the forces between the lipid bilayer and the
61
spectrin filaments. In the RBC membrane without defects, the two ends of the spectrin filaments
are connected at the actin junctional complexes, which are anchored to the lipid bilayer by
glycophorin proteins. The spectrin filaments are additionally anchored to the lipid bilayer via
band-3 proteins at the spectrin-ankyrin binding sites. The blue curve in Fig. 3.7A shows that the
pressure increases at the area close to band-3 binding site, where the spectrin particles are
connected to the lipid bilayer. The pressure is large there because of the continuous interactions
between the spectrin particles and the lipid bilayer. When moving away from the band-3 binding
site, the pressure drops rapidly as the frequency of interactions between the spectrin particles and
the lipid bilayer reduces. The pressure measured from the simulation is close to the analytical
pressure distribution introduced in (155),
P
(
)
,
,
1
2
k T
B
4
R
2
g
((
[
)
1
2
2
1
e
(
1
2
2
) / ( 4
R
2
g
)
(
e
1
2
2
) / ( 4
R
2
g
)
2
1
3
) 2
R
3
(
1
2
2
g
2
1
2
) ]
2
6
R
2
g
(3.10)
where Rg is the radius of gyration of the spectrin filaments, ρ is the pressure measurement
location, ρ1 and ρ2 are the anchoring locations of the filament to the lipid bilayer. Assuming that
the measurement points always remain on the line connecting the two anchoring locations, the
pressure profile based on Eq. (3.10) is plotted as the purple curve in Fig. 3.7A. It is noted that the
pressures in the analytical estimation are significantly higher than the simulation measurements
at the two ends of the filaments. The reason is that in the analytical calculations, the filaments are
assumed to be directly attached to the lipid bilayer, while in our numerical simulation and in the
actual RBC membrane, they are connected to the actin junctional complexes which then bind to
the lipid bilayer through the glycophorin proteins.
62
Figure 3.7 The pressure distribution exerted by the spectrin filaments on the lipid bilayer with
attraction parameters (A) n = 0 (B) n = 0.05 (C) n= 0.1 (D) n = 0.2. dee is the end to end distance
of the spectrin filaments. The blue curve represents the pressure distribution measured from the
membrane model. The purple curve represents the pressure distribution obtained from the
analytical estimation for a normal membrane (118). The red curve represents the pressure
distribution applied on the membrane with ankyrin protein defects. The black curve represents
the pressure distribution obtained analytically for a membrane with ankyrin protein defects (118).
The green curve represents the pressure distribution measured from the membrane with spectrin
protein defects.
63
Next, we investigate the effect that the introduced attractive forces between the spectrin filament
and the lipid bilayer have on the pressure applied to the lipid bilayer by the spectrin filaments.
The parameter n in the Eq. (3.2) is selected to be 0.05, 0.1 and 0.2, as the spectrin-lipid
interaction is relatively weak compared to the two primary connections between the cytoskeleton
and lipid bilayer (158-165). We chose 0.2 as the maximum value of n because at this value the
spectrin filaments are completely attached to the lipid bilayer. The pressure plots corresponding
to n = 0.05 and 0.1 (blue curves in the Fig. 3.7B and C) show that the pressure profiles are
similar to the profile obtained when n = 0, but the magnitudes of the pressure increase as n
increases. The increased attractive forces reduce the average distances between the filaments and
the lipid bilayer, resulting in more frequent interactions between the spectrin particles and the
lipid bilayer, and subsequently in a larger pressure on the lipid bilayer. When n = 0.2, the
attractive forces are large enough to cause attachment of the entire filament to the lipid bilayer,
and thus generate a dramatic increase in the pressure (Fig. 3.7D). We note that the pressure is
lower at the two ends of the filament because these points are connected to actin junctional
complexes and not directly to the membrane. Previous experimental studies (115, 147-150, 173)
and analytical modeling (116-118, 174-177) have shown that the pressure fence induced by the
cytoskeleton hinders the lateral diffusion of mobile band-3 proteins and lipid molecules.
Therefore, it is reasonable to predict that an increase in the attractive forces between the filament
and lipid bilayer causes a reduction in the lateral diffusion of the band-3 proteins and lipid
molecules.
64
Another question is how disruption of the connections between spectrin filaments and immobile
band-3 proteins (vertical interactions in Fig. 3.1A) modifies the pressure field applied on the
lipid bilayer by the spectrin network. This is relevant in HS where defects in proteins such as
ankyrin and band-3 proteins, which play an important role in the coupling between the lipid
bilayer and the cytoskeleton, result in a partial detachment between the network and the lipid
bilayer. The red curve in the Fig. 3.7A shows that when only repulsive interaction between the
lipids and the spectrin filaments is assumed (n = 0), the pressure is higher at the two ends of the
filaments, compared to the middle portion of the filaments. This reduction is due to a decrease in
the frequency of interactions between the middle sections of the filaments and the lipid bilayer.
In particular, the pressure profile is consistent with the analytically estimated pressure
distribution defined by Eq. (3.10) (see black curve in Fig. 3.7A) with the exception that the
analytical pressure is much higher at the two ends of the filament. This is due to the assumption
in the theoretical analysis that the two ends are directly anchored to the lipid bilayer while in our
simulations they are connected to the junctional complexes. Therefore, we expect that a
disruption of the connection between the filaments and the lipid bilayer at the band-3 anchoring
points weakens the pressure fence. This causes an increase in the probability for band-3 particles
to cross the boundaries of the filament-formed compartment, thus it enhances the diffusion of
band-3 proteins in HS (112, 115, 118, 178). The introduction of small attractive forces between
the lipid bilayer and the spectrin network (n = 0.05 and n = 0.1), when the connection between
the filament and the lipid at the band-3 site is severed, results in a constant pressure along the
filament. We note that the pressure is higher than in the case of only repulsive interactions
between the lipid bilayer and the network. When the attractive force is large (n = 0.2), and the
entire filament is in contact with the surface of the lipid bilayer, the pressure is similar with the
65
pressure field developed in the normal membrane with the exception of the middle segment of
the filament. In conclusion, the disruption of the anchoring of the spectrin filaments to the lipid
bilayer results in a lower and nearly constant pressure along the filament. Since the spectrin
filament-induced pressure acts as a pressure fence, hindering the lateral movement of mobile
band-3 proteins and lipid molecules, it is predicted that the lower pressure facilitates the
diffusion of lipids and band-3 proteins across the spectrin filaments. This results in a higher
mobile band-3 diffusion coefficient compared to the band-3 diffusion coefficient in the normal
RBC membrane.
Finally, we explore how the dissociation of a spectrin filament to two filaments affects the
pressure applied to the membrane. Here, we aim to simulate defective spectrin dimer-dimer
interactions that most commonly happen in HE (145). The pressure applied from both filaments
on the membrane is shown as green curves in Fig. 3.7A-D for different attractive forces between
the filaments and the membrane. We observe that when no attractive forces (n = 0) or small
attractive forces (n = 0.05) are assumed, the pressure is lower than in the normal membrane or in
a membrane with defects in the vertical interactions. It is reasonable to assume that a weaker
pressure fence on the lipid bilayer causes an increase in the probability of band-3 particles to
cross the spectrin filaments, justifying the observed higher diffusion coefficients of band-3
proteins in HE (112, 115, 118, 178). We also note that our simulations show that the pressure
applied on the lipid bilayer in the case of dimer-dimer disruption, which corresponds to HE, is
lower than the pressure measured in the case of disruption of vertical interactions, which
corresponds to HS. Based on this result, we conjecture that our model predicts greater band-3
diffusion in HE than in HS, in agreement with experimental observations (112). Of course direct
66
study of band-3 diffusion is required to validate these predictions. When the attractive forces are
large (n = 0.2), the difference between the pressure measured in the normal membrane and in the
membrane with protein defects becomes negligible with the exception of the area close to the
band-3 binding point in the middle where the pressure is large in the normal membrane. This
means that if the attractive force between the spectrin network and the lipid bilayer are large, the
diffusion coefficients of band-3 should have been similar in normal RBCs and in RBCs from
patients with HS or HE. However, since the band-3 diffusion coefficients measured in normal
RBCs are smaller than the ones measured in RBCs in HS and HE (112), we conjecture that the
attractive forces between the spectrin filaments and the lipid bilayer should be low and the
parameter n that defines the strength of the attractive forces cannot be larger than n = 0.1, which
justifies our selection of n in Section 2.
3.5. Summary
We introduce a particle-based model for the erythrocyte membrane that accounts for the most
important structural components of the membrane, including the lipid bilayer, the spectrin
network, and the proteins that play an important role in the anchoring of the spectrin cortex to the
lipid bilayer, as well as the band-3 proteins. In particular, five types of CG particles are used to
represent actin junctional complexes, spectrin, glycophorin, immobile band-3 protein, mobile
band-3 protein and an aggregation of lipids. We first demonstrate that the model captures the
fluidic behavior of the lipid bilayer and then that it reproduces the expected mechanical material
properties of bending rigidity and shear modulus of the RBC membrane. The timescale of our
simulations, which is found to be ts ~ 3×10-6s, is inferred by comparing the viscosity of the
membrane model to experimentally measured values. Then, the self-consistency of the model
67
with respect to the timescale is tested by comparing the computed vibration frequency of the
spectrin filaments and lipid membrane to analytically obtained values. We confirm that vibration
frequency of the spectrin filaments and lipid membrane measured from the proposed membrane
model, are also in agreement with experimental values. At last, we study the interactions
between the cytoskeleton and the lipid bilayer, and measure the pressure applied on the
membrane by the spectrin filaments. We also investigate how disruption of the connection
between the spectrin network and the lipid bilayers, which simulates defects in HS, and rupture
of the dimer-dimer association, which simulates defects in HE, affect the pressure exerted on the
lipid bilayer. We show that overall the introduction of defects in the spectrin network or in the
vertical connection between the lipid bilayer and the cytoskeleton results in lower pressure,
which is consistent with prediction in (155). In addition, we find that the defects related to HE
have a stronger effect than the defects related to HS. This result implies that diffusion of band-3
proteins in RBCs from patients with HS and HE is enhanced compared to the normal RBCs.
Moreover, elliptocytes exhibit more prominent diffusion of band-3 proteins than spherocytes.
Both conclusions are supported by experimental results. The level of attraction forces between
the lipid bilayer and the membrane cytoskeleton is another important parameter that regulates the
pressure applied by a spectrin filament to the lipid bilayer. We show that as the attractive force
increases, it causes an overall increase in the pressure and it diminishes the differences in the
pressure generated by membrane protein defects and, consequently, the differences in the
diffusion of band-3 proteins in normal and defective erythrocytes. Since this finding is not
supported by experimental results, we conjecture that the attractive force between the lipid
bilayer and the spectrin filaments should be low, resulting in a membrane model where the
filaments are not completely attached to the lipid bilayer. A detailed study of the band-3
68
diffusion in this model and direct comparison with experimental results is necessary in order to
form a more accurate picture of how the model regulates diffusion. Because of the explicit
representation of the lipid bilayer and the cytoskeleton, the proposed model can be potentially
used in the investigation of a variety of membrane related problems in RBCs in addition to
diffusion. For example, membrane loss through vesiculation and membrane fragility in
spherocytosis and elliptocytosis, interaction between hemoglobin fibers and RBC membrane in
sickle cell disease, and RBC adhesion are problems where the applications of the proposed
membrane model could be beneficial.
69
Chapter 4.
Modeling of the Band-3 Proteins Diffusion in the Normal and
Defective Red Blood Cell Membrane
Abstract
We employ a two-component red blood cell (RBC) membrane model to simulate the diffusion of
band-3 proteins in the normal RBC and in the RBCs with protein defects. We introduce protein
defects which reduce the connectivity between the lipid bilayer and the cytoskeleton or reduce
the connectivity of the cytoskeleton and these defects are associated with the blood disorders of
hereditary spherocytosis (HS) and elliptocytosis (HE), respectively. We first measure the band-3
diffusion coefficients in the normal RBC membrane and in the RBC membrane without
cytoskeleton. Comparison of these two coefficients demonstrates that the cytoskeleton limits the
band-3 lateral mobility. Second, we study band-3 diffusion in defective RBC membranes and
quantify the relation between the band-3 diffusion coefficients and the percentage of protein
defects in HS and HE RBCs. By comparing the diffusion coefficients measured in these two
cases, we conclude that the band-3 mobility is primarily controlled by the cytoskeleton
connectivity. Third, we study how the band-3 anomalous diffusion exponent depends on the
percentage of protein defects in the membrane cytoskeleton. Our measurements show that the
effect of the cytoskeleton connectivity on the anomalous diffusion exponent is small. Finally, we
show that introduction of attraction between the lipid bilayer and the spectrin network can reduce
band-3 diffusion. By comparing our measurements to experimental data, we predict that the
attractive force between the spectrin filament and the lipid bilayer is at least 20 times smaller
than the binding forces at the two membrane major binding sites at band-3 and glycophorin.
70
4.1. Introduction
Diffusion of membrane proteins has been studied for a long time because of its importance in
various cellular processes (179-182). The early fluid mosaic model developed for the plasma
membrane postulated that proteins were freely diffusing in the plane of the membrane and the
proteins followed a normal diffusion pattern, meaning that the mean square displacement (MSD)
of the diffusing protein is proportional to time, MSD ~ t (183). However, as the experimental
techniques advanced, the protein diffusion in the membrane was found to be more complicated
and it was observed that proteins undergo anomalous diffusion (subdiffusion) in biological cells
(184-190). The MSD of the diffusing protein is proportional to a fractional power of time, MSD
~ tα, where α < 1 is called anomalous diffusion exponent. Anomalous diffusion of proteins in the
cell membrane mainly originates from the interactions between the diffusing protein and the
immobile and mobile particles in the membrane (191, 192). In addition, the diffusing proteins
may stick to or be hindered by the cell cytoskeleton, causing the deviation from the normal
diffusion (116, 175, 193-195). RBC membrane cytoskeleton consists of spectrin tetramers
connected at actin junctional complexes, forming a two-dimensional (2D) six-fold triangular
network tethered to a lipid bilayer, as shown in Fig. 4.1A. While the main function of the
cytoskeleton is to maintain the mechanical properties and integrity of the RBC, experimental
studies have demonstrated the cytoskeleton sterically hinders the lateral diffusive motion of the
band-3 proteins (147, 149, 176, 178, 196). In the RBC membrane, approximately one third of the
band-3 proteins bind to the cytoskeleton and they are considered to be immobile band-3. The rest
of the band-3 proteins, which are not connected to the cytoskeleton, are mobile band-3 proteins.
Although mobile band-3 proteins do not directly bind to the cytoskeleton, their lateral diffusive
71
motion in the membrane is hindered by the presence of the cytoskeleton (115-118, 147-150, 174-
177, 197). In the short time range (~10 ms), the motions of the mobile band-3 proteins are
constrained within the compartments formed by the spectrin network underneath the lipid bilayer
(178). In the long time range, the band-3 proteins occasionally hop from one compartment to a
neighboring compartment due to the fluctuation of the spectrin network (117, 118, 177, 198) and
the fluctuation of lipid bilayer (199). Therefore, two types of diffusive motions are defined,
namely the microscopic diffusion, which describes the motion of band-3 protein within a
compartment, and the macroscopic diffusion, which describes the global motion of band-3
proteins over the surface of the cell.
Figure 4.1. (A) Schematic of the human RBC membrane. Two-component human RBC
membrane model (B) top view at equilibrium configuration, (C) side view at initial configuration
(D) side view at equilibrium configuration. "A" type particles represent actin junctions. "B" type
particles represent spectrin particles. "C" type particles represent glycophorin proteins. "D" type
72
particles represent band-3 complexes that are connected to the spectrin network (immobile band-
3). "E" type particles represent band-3 complexes that are not connected to the network (mobile
band-3). "F" type particles represent lipid particles.
Although experimental measurements clearly demonstrate that the band-3 proteins undergo hop
diffusion in the RBC membrane, the exact mechanisms of how the band-3 proteins escape from
one compartment to a neighboring one are still not well-understood because of the limited length
and time scales that can be explored by experiments. Therefore, a number of analytical models
were developed to investigate the mechanism of the hop diffusion (116, 118, 174, 175, 198-201).
In general, three mechanisms have been proposed. The first mechanism considers that band-3
molecules escape from one compartment and move to the neighboring compartment due to the
remodeling of the cytoskeleton (116, 174, 175, 198). According to the second and third
mechanism, the thermal fluctuation of the spectrin filaments and the thermal fluctuation of the
lipid bilayer respectively allow for the band-3 hop diffusion (118, 199, 201). In addition, the
binding forces between the spectrin and specific type of lipid molecules, which were reported in
a number of studies (158-163, 202-205), may enhance the confinement of the cytoskeleton on the
lateral motions of the band-3 proteins. However, since the measurement of the forces between
the spectrin filaments and lipid bilayer is not an easy task, studies on the effects of the binding
force between the spectrin filaments and the lipid bilayer on the band-3 diffusion are still
incomplete.
73
Figure 4.2. (A) Schematic of the HS RBC membrane. (B) Schematic of the HE RBC membrane.
Study of the diffusion of band-3 protein in the hemolytic blood disorders raises important
questions. Band-3 diffusion measurements in the HS and HE RBCs showed that the mobile
band-3 diffusion was accelerated and the number of mobile band-3 proteins was increased (112,
115, 197). More importantly, the abnormal band-3 diffusive motion may relate to the instability
of RBC membrane and membrane vesiculation in the blood disorders (16, 26, 145, 206). In HS,
there are protein defects in the vertical interactions (see Fig. 4.2A) between the cytoskeleton and
the lipid bilayer (3, 16, 19, 21-24, 207), whereas in HE, protein defects occur in the horizontal
interactions (see Fig. 4.2B) at the spectrin dimer-dimer connecting sites or at the actin-spectrin
junctions of the cytoskeleton (3). Disruptions of either the vertical interaction or horizontal
interaction change the lateral diffusion of the mobile band-3 proteins (112, 115, 116, 118, 197,
74
208, 209). While numerous experimental measurements and analytical modeling were conducted
on the band-3 diffusion in the RBC membrane, only a limited number of papers quantitatively
relate the band-3 diffusion coefficient to the percentage of the protein defects in the cells with
blood disorders. In these experimental measurements, protein defects can only be identified by
measuring the ratio between band-3 and spectrin content in HS and HE (115, 197). As for the
analytical modeling, Saxton (116) studied the dependence of the band-3 diffusion coefficients on
the cytoskeleton connectivity by performing Monte Carlo simulations. His simulation results
showed that the band-3 diffusion coefficients fall significantly as the cytoskeleton connectivity
was increased. However, the effects of the spectrin filaments and lipid bilayer fluctuations were
not considered in these Monte Carlo simulations. Auth and Gov (118) proposed an analytical
model, where the effect of the spectrin filaments is simulated by static pressure field. This model
was capable of showing the band-3 diffusion in the normal RBC membrane and in the membrane
with ankyrin protein defects, but protein defects in the cytoskeleton were not considered.
In this work (210), we simulate the band-3 hop diffusion in the RBC membrane due to the
thermal fluctuation of the spectrin filaments and the thermal fluctuation of the lipid bilayer.
Second, we quantify the relation between the percentage of membrane defects and band-3
diffusion coefficients in HS and HE. Third, we compute the band-3 anomalous diffusion
exponents α in HE RBCs, by fitting the MSD measured from our simulation to the MSD ~ tα
relation for the appropriate time scale. The calculated exponents are compared with exponents
obtained from the fixed obstacles model and picket fence model. Finally, we explore the effects
of the attractive forces between the spectrin filaments and the lipid bilayer on the band-3
diffusion. In order to achieve our goals, we apply a recently developed two-component coarse-
75
grained molecular dynamics (CGMD) model for the human RBC membrane which explicitly
represents the cytoskeleton and the lipid bilayer by coarse-grained (CG) particles (211).
Previously developed CGMD models explicitly simulate either only the RBC cytoskeleton with
implicit representation of the lipid bilayer (80, 81, 109, 110, 142) or only the lipid bilayer (84, 86,
91, 92, 95, 108) with implicit spectrin network (111). Absence of either the explicit lipid bilayer
or the cytoskeleton limits their application in studying the protein diffusion in the RBC
membrane. Recently, a model consisting of two layers of 2D triangulated networks, where one
layer represents the cytoskeleton while the other one represents the lipid bilayer, was introduced
(154). Although
this approach simulates
two-component RBC membrane with high
computational efficiency, it cannot be used to model the protein diffusion. The two-component
RBC membrane model applied in this work explicitly comprises both the lipid bilayer and the
cytoskeleton, and thus can be used to simulate interactions between the cytoskeleton and the
lipid bilayer as well as interactions between the cytoskeleton and diffusing membrane proteins.
In addition, the two-component model naturally simulates the band-3 hop diffusion due to the
combined effects of the thermal fluctuation of the spectrin filaments and of the lipid bilayer.
Furthermore, the explicit representation of the cytoskeleton allows us to easily incorporate a
variety of proteins defects into this model and introduce the attractive force between the spectrin
filaments and lipid bilayer.
4.2. Model and methods
In the applied RBC membrane model, three types of coarse-grained (CG) particles are introduced
to represent lipid bilayer and two types of transmembrane proteins (see Fig. 4.1A-D), including
the blue color particles signifying a cluster of lipid molecules, the black particles denoting
76
glycophorin proteins, the yellow particles representing the band-3 proteins that are connected to
the spectrin network through the ankyrin proteins (immobile band-3), and the green particles
representing the band-3 proteins that are not attached to the spectrin network (mobile band-3).
To simplify our model, the connections between spectrin and immobile band-3 are assumed to be
fixed. The cytoskeleton consists of spectrin filaments connected at actin junctions forming a
canonical hexagonal network. The actin junctions are represented by the red particles in Fig.4.1
A-D, and are connected to the lipid bilayer via glycophorin. The spectrin filament is represented
by 39 spectrin particles (white particles) connected by unbreakable springs. The details of the
potentials between the CG particles applied in the model, along with the mass and dimensions of
the CG particles can be found in the authors' previous work (211). The interactions between the
cytoskeleton and the lipid bilayer are represented by the repulsive part of the L-J potential
u
LJ
(
r
ij
)
4
12
r
ij
6
r
ij
0
r
ij
R
cut,LJ
r
ij
R
cut,LJ
(4.1)
where ε is the energy unit and σ is the length unit.
ijr is the distance between the CG particles.
r
Given the diameter of the lipid particles l-l
eq
1/62
5 nm
, one finds σ = 4.45 nm. The cutoff
distance of the potential is chosen to be
R
1/ 6
2
r
eq
, where eqr is the equilibrium distance
cut,LJ
between different types of CG particles. More information about the model can be found in
authors' previous work (211). In our simulations, the dimension of the membrane is
approximately ~ 0.8 × 0.8 μm2. The system comprises N = 36606 CG particles. By applying the
Nose-Hoover thermostat, the temperature of the system is maintained at kBT/ε = 0.22. The model
77
is implemented in the NAT ensemble (131). Because the model is solvent-free and the
membrane is a two-dimensional structure, we controlled the projected area instead of the volume.
The projected area is adjusted to result in zero tension at the equilibrium state. The time step is
selected to be
0.01t , where ts is the time scale. The time scale ts in our simulation will be
s
determined in the following section by comparing the band-3 diffusion coefficients measured in
our model with the former experimental results.
4.3. Results and discussion
4.3.1. Band-3 diffusion in RBC membrane with perfect cytoskeleton, in the normal RBC
membrane and in the lipid membrane
In this section, we measure the diffusion coefficients of the mobile band-3 particles in a RBC
membrane model with perfect cytoskeleton, in the normal RBC membrane and when only the
lipid bilayer is considered. The diffusion coefficient of the mobile band-3 is calculated by
measuring
the slope of
the MSD with
respect
to
time. MSD
is defined as
1
N
band
3
3
x
bandN
j
1
t
j
x
j
2
0
(212), where Nband-3 is the number of the mobile band-3 particles.
First, we simulated the diffusion of band-3 particles in the membrane with perfect cytoskeleton,
meaning that the cytoskeleton is considered as a perfect 2D six-fold triangular structure with a
fixed connectivity. Fig. 4.3A displays a representative trajectory of a mobile band-3 particle
hopping to the neighboring compartment during 2×106 time steps. The trajectory of the band-3
particle clearly illustrates the steric hindrance effect of the cytoskeleton on the lateral motion of
band-3 particles. Plot of the MSD of the mobile band-3 particles with respect to the time in Fig.
4.3C (black curve) shows that MSD rises rapidly for
t
t
micro
10000
t
s
, where tmicro is the
78
timescale for the microscopic diffusion, indicating that band-3 particles diffuse within the
compartments for
t
t
. For
t
t
micro
micro
, MSD increases linearly with time although the slope of
MSD against time becomes much smaller than the slope in the microscopic diffusion. This
means that the mobile band-3 particles explore areas larger than the triangular compartments for
t
t
micro
and the increase of MSD results from the intercompartment hop diffusion (macroscopic
diffusion). We extract the macroscopic diffusion coefficient of the mobile band-3 particles by
fitting a straight line to the MSD for large times and then applying the definition of the diffusion
coefficient
D
t
t
micro
lim 1 4
t MSD
(212). The band-3 diffusion coefficients mentioned in the rest
of the chapter all refer to the macroscopic diffusion coefficient. The diffusion coefficient for the
band-3 particles
in
the RBC membrane with perfect network
is
found
to be
D
perfect
1.23 10
4
t
/
2
. It is noted that the value of the MSD at
s
t
t
micro
10000
t
s
in our
simulation is about
55
2
2
( 1130nm )
, which corresponds to a circular area with diameter of
8.5σ ~ 38 nm. This estimation is in agreement with the numerical results obtained in (118) and 2
times smaller than the value observed in the single particle tracking experiment (178). The
discrepancy between our result and the experimental value is probably caused by the application
of perfect 2D hexagonal spectrin network in our model. For the normal RBC membrane, electron
microscopy images revealed that although most of actin junctions were linked with other 6 actin
junctions, a small amount of actins junctions were linked with 5 or 7 actin junctions (137, 138).
In addition, Li et.al (142) demonstrated that the biconcave shape of RBCs can be obtained only if
the spectrin network undergoes remodeling of the cytoskeleton to relax the in-plane elastic
energy. The metabolic remodeling of the spectrin network is affected by the adenosine 5′-
triphosphate (ATP) contents in the RBC(140, 151). The defects in the cytoskeleton of the normal
79
RBC membrane can effectively increase the area of the compartments. To obtain the
connectivity of the cytoskeleton in our membrane model that corresponds to the normal RBC
membrane, we match the numerical value of the shear modulus measured from our RBC
membrane model with the experimental values of ~8.5μN/m obtained in (141, 213). Based on
our previous work in (111, 211), the shear modulus of our membrane model falls within the
range of
the experimentally measured values when
the cytoskeleton connectivity
is
approximately 90%. Therefore, we select the membrane with cytoskeleton connectivity of 90%
to represent the normal RBC membrane. We reduce the cytoskeleton connectivity to 90% by
randomly disconnecting the spectrin filaments in the middle and repeat the diffusion
measurement performed for the perfect cytoskeleton. We find that the MSD (see Fig. 4.3C, blue
curve) goes through a fast increase until
t
t
micro
25000
t
s
, and then it increases slowly and
linearly with respect to time. Although the general behavior of the MSD is similar to the
behavior of the MSD measured in the perfect cytoskeleton case, the timescale for the
microscopic diffusion and the corresponding area explored by the band-3 particles during
microscopic diffusion are increased due to the reduced cytoskeleton connectivity. The
corresponding diffusion coefficient of the mobile band-3 particles in the membrane with
cytoskeleton connectivity of 90% is found to be
normalD
1.29 10
4
t
/
2
. The experimentally
s
measured band-3 diffusion coefficient in the RBC membrane is
D
4 10 cm /s
11
2
(214). By
comparing our numerical result to the experimental value, the timescale of our simulation is
determined to be s
t
0.7 μs
. This estimation is comparable to the timescale calculated using the
viscosity of the RBC membrane in the authors' previous work (211). Then, the timescale for the
microscopic diffusion of the mobile band-3 particles in the normal RBC membrane is computed
to be approximately micro
t
25000
t
s
17.5 ms
, which is consistent with the experimentally
80
measured timescale of ~ 10-20 ms for the microscopic diffusion in (178, 196). At last, we
remove the cytoskeleton from the membrane model and simulate the band-3 diffusion in the lipid
bilayer without the obstruction from the spectrin filaments. Fig. 4.3B shows that the trajectory of
the band-3 particle covers, in the same time as before, larger distance than in the cases where the
spectrin cytoskeleton is implemented. This result demonstrates that the cytoskeleton has a clear
confining effect in the diffusive motion of the band-3 particles. The MSD of the band-3 particles,
plotted in Fig. 4.3C, increases linearly with respect to time and the corresponding diffusion
coefficient is measured to be
D
52 10
no_cyto
4
t
/
2
, approximately 40 times larger than
s
D
normal
1.29 10
4
t
/
2
. This measurement is comparable with experimental results showing that
s
band-3 diffuses about 50 times faster in mouse erythrocytes that lack major components of the
cytoskeleton than in the healthy erythrocytes (147).
Figure 4.3. Trajectories of a mobile band-3 particle (A) hopping to a neighboring compartment in
the normal RBC membrane and (B) diffusing in a lipid bilayer without cytoskeleton. The
81
trajectories are extracted from simulation with duration of 2×106 time steps and the positions of
mobile band-3 particles were recorded every 2000 time steps. (C) MSDs against time and
corresponding diffusion coefficients of the mobile band-3 diffusing in the RBC membrane with
perfect cytoskeleton, in the normal RBC membrane and in membrane without cytoskeleton.
4.3.2. Band-3 diffusion in RBC membrane with protein defects in the vertical interactions
In this section, we reduce the connectivity between the spectrin filaments and immobile band-3
particles in our model to simulate band-3 diffusion in the RBC membrane with defective proteins
in the vertical interactions. This situation is encountered in the RBCs of patients suffering from
HS. In this case, defects in proteins that play a direct or indirect role in the binding of the
cytoskeleton to the lipid bilayer, such as band-3, 4.2, and ankyrin proteins, result in local
detachments of the lipid bilayer (3, 145). Experimental measurements showed that the diffusion
of the mobile band-3 was accelerated and mobile band-3 portion was increased in the HS cells
(112, 115, 197). Auth and Gov (118) demonstrated, by using a pressure fence analytical model,
that removal of all the connections between the spectrin filaments and immobile band-3 proteins
leads to a 20-fold increase of the mobile band-3 diffusion coefficient. To simulate the diffusion
of band-3 particles in the HS RBC membrane, we reduce the connectivity between the spectrin
filaments and the immobile band-3 particles. We then compute the band-3 diffusion coefficient
and validate it by comparing the numerical diffusion coefficients with experimentally measured
values. In addition, we quantified the relation between the percentage of the protein defects and
band-3 diffusion coefficients. To examine the effects of protein defects in the vertical
connections only, the cytoskeleton connectivity is conserved at 100% for all the measurements in
this section. In our simulations, the connectivities between the spectrin filament and band-3
82
particles are selected to be Cvertical = 70%, 50% and 0%, respectively. The MSD of band-3
particles with respect to time are plotted in Fig. 4.4. When Cvertical = 70%, the diffusion
coefficient is found to be
D
C
vertical
=70%
2.5 10
3
t
/
2
, approximately 2 times larger than
D
.
perfect
s
The increase in the band-3 diffusion coefficient is attributed to the increased average distance
between the spectrin filaments and the lipid bilayer as the connections between the filaments and
band-3 particles are removed. Larger distance between the filaments and lipid bilayer causes an
increase in the probability of band-3 particles passing to the neighboring compartments and thus
results in a larger band-3 diffusion coefficient. When Cvertical is further decreased to 0%, the
band-3 diffusion coefficient increases to
D
C
vertical
=0%
9.6 10
3
t
/
2
approximately 8 times larger
s
than
D
perfect
. By applying the timescale of our simulation s
t
0.7 μs
, we find that the mobile
band-3 particle diffusion coefficient in the membrane with protein defects in the vertical
interactions
is approximately
D
0.8 3 10 cm /s
10
2
. This result
is consistent with the
experimental values of the band-3 diffusion coefficients in HS RBCs (115, 197), suggesting that
the band-3 diffusion coefficient increases, but not significantly, in HS cell (112).
Figure 4.4. MSDs against time and corresponding diffusion coefficients of the mobile band-3
diffusing in the RBC membrane with different vertical connectivities.
83
4.3.3. Band-3 diffusion in RBC membrane with protein defects in the horizontal
interactions
In this section, we disrupt the spectrin filaments in our membrane model to simulate the band-3
diffusion in the membrane with defective proteins in the horizontal interactions. This situation is
encountered in the RBCs of patients suffering from HE where defects in α-spectrin, β-spectrin,
and/or protein 4.1 cause dissociation of the spectrin tetramer into dimers or disconnection of the
spectrin filaments from the actin junctions (3, 145). As a result, the diffusion coefficient of the
mobile band-3 is largely increased in the HE RBCs (112, 115). Saxton performed Monte Carlo
simulation based on the percolation analysis and showed that the mobile band-3 diffusion
coefficients strongly depends on the cytoskeleton connectivity (116). To simulate the HE RBC
membrane, we disconnect the spectrin chains in the middle. The cytoskeleton connectivities are
selected to be Chorizontal = 80%, 70%, 60%, 50% and 30%, respectively. Because of the ruptured
spectrin filaments, band-3 particles can travel easily between the compartments with disrupted
boundaries, resulting in an enhanced lateral mobility. The recorded band-3 trajectory when the
cytoskeleton connectivity is Chorizontal = 70% (see Fig. 4.5A) shows that band-3 particles
trajectories cover larger areas (red and green curves) comparing to band-3 particles in the normal
RBC membrane (blue curve). This means that the steric hindrance effects of the cytoskeleton are
reduced as the cytoskeleton connectivity decreases. The MSDs and corresponding diffusion
coefficients of band-3 particles measured with a variety of Chorizontal (see Fig. 4.5B) show that as
the Chorizontal is decreased from 80% to 30% the band-3 diffusion coefficients increase
significantly from
CD
horizontal
=80%
3.27 10
4
t
/
2
to
CD
s
horizontal
=30%
50 10
4
t
/
2
, indicating that the
s
cytoskeleton connectivity plays an important role in controlling band-3 diffusion. Particularly,
84
when Chorizontal = 30%, the band-3 diffusion coefficient increases to
CD
horizontal
=30%
50 10
4
t
/
2
,
s
which is comparable to the band-3 diffusion coefficients measured from the membrane model
without cytoskeleton
D
52 10
no_cyto
4
t
/
2
. This means that at low connectivity the
s
cytoskeleton loses the function of hindering the band-3 lateral diffusion. Comparison between
the band-3 diffusion coefficients measured in the HS and HE RBC membrane models shows that
the cytoskeleton connectivity is the major determinant of the lateral diffusivities of band-3. This
explains the experimental observation that band-3 diffusion coefficients measured in HE RBC
are larger than those measured in HS RBCs (112).
Figure 4.5. (A) Comparison between the trajectory of a band-3 particle diffusing within the
compartment of an RBC membrane without defects (blue line) and the trajectories of two mobile
85
band-3 particle diffusing in RBC membrane with defects in the cytoskeleton (green and red
lines). The trajectories are extracted from simulation with duration of 2×106 time steps and the
positions of mobile band-3 particles were recorded every 2000 time steps. (B) MSDs against
time and corresponding diffusion coefficients of the mobile band-3 diffusing in the RBC
membrane with different cytoskeleton connectivities.
4.3.4. Anomalous diffusion of the band-3 in the membrane with protein defects in the
horizontal interactions
The MSDs of the band-3 particles obtained in the previous section can be used to calculate the
anomalous diffusion exponent in the case of the RBC membrane with protein defects in the
horizontal interactions. Single particle tracking experiments have clearly demonstrated that
mobile band-3 proteins undergo subdiffusion in the RBC membrane (112, 176, 178, 190),
meaning that MSD of the band-3 follows the relation MSD ~ tα, where α < 1 is the anomalous
diffusion exponent. Several mechanisms have been hypothesized to explain protein subdiffusion
in the cell membrane, including confinement by the membrane skeleton (116, 148, 150, 174-176,
215), obstruction by the lipid rafts or protein domains (192, 216-218), and binding of the
diffusing proteins to the cytoskeleton (193, 219-221). Although the subdiffusion of the proteins
in the cell membrane has been studied for a long time, quantitative relations between the degree
of the obstruction and the anomalous diffusion exponents are difficult to be established due to
limitations of the experimental techniques. Only few analytical models have been developed to
study the dependence of the anomalous diffusion exponent on the membrane protein defects (192,
195). As seen in Fig. 4.5B, the MSDs of the mobile band-3 particles measured at high
cytoskeleton connectivities do not increase proportionally with respect to time, illustrating a
86
typical anomalous diffusion pattern. The anomalous diffusive behavior of the band-3 particles in
our model results from the steric hindrance effect of the cytoskeleton, therefore, as the
cytoskeleton connectivity decreases, the band-3 diffusion approaches to normal diffusion, as
shown in Fig. 4.5B. To determine the quantitative relation between the band-3 anomalous
diffusion exponents α and the cytoskeleton connectivity Chorizontal, we plot the MSDs of mobile
band-3 particles against time with a variety of cytoskeleton connectivities in log-log coordinates.
The log-log plots of the MSD against time exhibit a linear relationship with slight changes in
slopes from 1 (normal diffusion) to 0.905 when the Chorizontal is increased from 0 % to 70% (Fig.
4.6A). We observe that in the cases of Chorizontal = 90% and 100%, the dependence of the
log(
)MSD on log( )t deviates from linearity at large time scales. This is because at large time
scales the band-3 particles undergo macroscopic diffusion where the increase of the MSD is due
to the band-3 hop diffusion. The small diffusivities in the macroscopic diffusion cause the
deviation of the MSD curves from the MSD ~ tα relation. These observations suggest that the
anomalous diffusion relation MSD ~ tα does not hold for the long time when describing the
cytoskeleton-induced subdiffusion of band-3 proteins, which is in agreement with observations
from previous analytical simulations (192, 195). At time scales where the anomalous diffusion
relation holds, Fig. 4.6B shows that when the cytoskeleton connectivity (Chorizontal) increases
from 0% to 100%, the corresponding anomalous diffusion exponents α varies from 1 to 0.86.
Thus, anomalous diffusion exponents of band-3 particles induced by the cytoskeleton lie in the
range of 1-0.86, which is larger than the exponents range of 1–0.94 induced by picket fence but
smaller than the exponents range of 1–0.65 induced by lipid rafts or protein domains (195). The
cytoskeleton-induced obstruction only has a small effect on the band-3 anomalous diffusion
exponent because the anomalous diffusion relation MSD ~ tα holds true only during short time
87
scales (microscopic diffusion) and during the transition period from the microscopic diffusion to
the macroscopic diffusion. The decrease in the cytoskeleton connectivity barely influences band-
3 microscopic diffusion. In contrast, fixed obstacles and lipid rafts, which are not considered
here, can largely alter band-3 microscopic diffusion motion in the microscopic diffusion time
scale.
Figure 4.6. (A) log-log plot of the MSDs against time for the RBC membrane with cytoskeleton
connectivities Chorizontal = 0%, 30%, 50%, 70%, 90% and 100%. (B) The corresponding
anomalous exponent α for the previously mentioned horizontal cytoskeleton connectivities.
4.3.5. Effects of attractive forces between the spectrin filaments and lipid bilayer on the
band-3 diffusion
88
Here, we apply attractive force between the spectrin filaments and lipid bilayer to investigate its
effect on the mobile band-3 diffusion in the normal RBC membrane and in the membrane with
defective vertical and horizontal interactions. Previous studies of the lipid–spectrin interactions
showed that the spectrin binds to the negatively charged lipid surfaces with association constants
of 2-10×106 M-1 (158-163, 202-205), whereas the association constants at the lipid bilayer-
cytoskeleton major binding sites, such as spectrin-ankyrin, ankyrin-band-3, spectrin-protein 4.1-
actin were 2×107 M-1, 2×108 M-1 and 2×1012 M-2, respectively (164, 165). Comparison between
the association constants suggests that the spectrin-lipid affinity is not as strong as the primary
binding sites where the spectrin filaments are anchored to the lipid bilayer. But the binding force
between the spectrin and lipid molecules could reduce the average distance between the lipid
bilayer and cytoskeleton and thus enhance the steric hindrance effect on the band-3 diffusion.
Since the binding force between the spectrin filaments and lipid bilayer is difficult to measure,
we apply the attractive part of the L-J potential to represent the effective binding forces between
the spectrin particles and lipid particles,
u
att
(
r
ij
)
4
n
12
r
ij
6
r
ij
0
n
r
ij
l-s
r
eq
r
ij
l-s
r
eq
(4.2)
where n is a parameter to tune the attractive energy between the spectrin filaments and lipid
bilayer.
r
l-s
eq
5nm
is the equilibrium distance between the spectrin particles and lipid particles.
Our previous simulations (211) showed that the cytoskeleton was completely attached to the
lipid bilayer when n ≥ 0.2, which contradicts with experimental observations. Therefore, in the
89
current simulations, the parameters n applied in the Eq. (4.2) are selected to be n = 0, 0.05, 0.1
and 0.15, respectively.
Figure 4.7. Ratios of the band-3 diffusion coefficients D measured in the RBC membrane with
protein defects to Dperfect at a variety of attractive forces between the spectrin filaments and the
lipid bilayer (n = 0, 0.05, 0.1 and 0.15). We test membrane models with (A) vertical
connectivities of Cvertical = 100%, 90%, 70%, 50%, 30%, and 0%, and with (B) horizontal
cytoskeleton connectivities of Chorizontal = 100%, 90%, 70%, 50%, 30%, and 0%. The black dot
line represents the results predicted by the percolation theory in (222).
90
First, we examine the effects of the attractive forces between the spectrin filaments and lipid
bilayer on the band-3 diffusion in a RBC membrane with perfect cytoskeleton and in a
membrane with protein defects in the vertical interactions. The connectivities between the
spectrin filaments and immobile band-3 Cvertical are selected to be 100%, 90%, 70%, 50%, 30%
and 0%, respectively. The ratios of the measured band-3 diffusion coefficients D to the band-3
diffusion coefficients in a RBC membrane with perfect cytoskeleton Dperfect are plotted in the Fig.
4.7A. The black curve in the Fig. 4.7A shows that when no attractive force is applied (n = 0), the
ratios of diffusion coefficients gradually rise to 8 with decreasing Cvertical, while it only grows to
2 with application of a small attractive force n = 0.05 (blue curve in Fig. 4.7A), meaning that the
attractive force enhance the steric hindrance effect on the band-3 diffusion. This conclusion can
be further proven if we apply a larger attractive force n = 0.1 (green curve in Fig. 4.7A). With
this attractive force, no band-3 hop diffusion is observed when Cvertical = 100% and 90% because
the distance between the filaments and the lipid bilayer is reduced such that the band-3 particles
cannot pass the boundaries of the compartments. As Cvertical is decreased to 70%, band-3 hopping
diffusion is observed, but the band-3 macroscopic diffusion is still largely suppressed, resulting
in smaller diffusion coefficients than Dperfect. Even as Cvertical is reduced to 0%, the ratio of
diffusion coefficients is only 0.6. When the attractive parameter n is further increased to 0.15, the
spectrin filaments are found to be fully attached to the lipid bilayer, preventing the band-3
particles from hopping to neighboring compartments. Thus, the band-3 diffusion coefficients
become zero for all values of vertical connectivities (red curve in Fig. 4.7A). Based on the above
discussion, we conclude that the attractive forces between the spectrin filaments and the lipid
bilayer can reinforce the steric hindrance effects of the cytoskeleton on the band-3 motion and
thus reduce the band-3 diffusion coefficients. In particular, small attractive forces diminish the
91
effects of defective vertical interactions on increasing the band-3 diffusivity while large
attractive forces almost totally negate the effects of the defective vertical interactions between
the cytoskeleton and lipid bilayer on band-3 diffusivity.
Next, we study the effect of the attractive forces between the spectrin filaments and lipid bilayer
on the band-3 diffusion in membrane with protein defects in the horizontal interactions. The
connectivities of the cytoskeleton Chorizontal are selected to be 100 %, 90%, 70%, 50%, 30% and 0%
(lipid bilayer only), respectively. The corresponding ratio of the measured band-3 diffusion
coefficients D to Dperfect are plotted in the Fig. 4.7B. When no attractive force is applied (n = 0),
the band-3 diffusion coefficients are boost dramatically with decreasing Chorizontal (black curve in
the Fig. 4.7B), as Chorizontal plays the major role on regulating the band-3 diffusion motion. When
attractive forces are applied, lower band-3 diffusion coefficients are measured, as shown by the
blue, red and green curves in the Fig 4.7B. These observations suggest that the attractive forces
between the spectrin filaments and lipid bilayer can also diminish the enhancement of the band-3
diffusion due to the protein defects in the horizontal interactions. From the band-3 diffusion
measurements in membrane with defective vertical interaction, we discover that when n ≥ 0.1,
the band-3 can barely pass the boundaries of the compartments, due to the small distances
between the filaments and lipid bilayer, leading to small or zero band-3 diffusion coefficients
(see Fig. 4.7A). However, Fig. 4.7B shows that even when n is increased to 0.1 and 0.15,
significant band-3 diffusion coefficients are measured at Chorizontal ≤ 50%. This difference is
attributed to the fact that in the HE RBC membrane, the band-3 particles can travel to the
neighboring compartments by passing the broken boundaries, even though the spectrin filaments
92
are attached to the lipid bilayer. This also explains why the red and green curves measured at n =
0.1 and 0.15 overlap in the Fig. 4.7B and demonstrates that the band-3 diffusion coefficients do
not differ for attractive forces with n ≥0.1.
At last, we compare the band-3 diffusion coefficients measured at large attractive forces (n ≥0.1)
to the band-3 diffusion coefficients computed by using Monte Carlo simulation where the
authors studied the dependence of the protein diffusion coefficients on the cytoskeleton
connectivity (116, 222). Both of these two measurements ignore the effect of the thermal
fluctuation of the lipid bilayer and spectrin filaments on the band-3 diffusion. In these Monte
Carlo simulation, diffusing particles are assumed to travel between sites connected with the
unblocked bonds (corresponding to the broken spectrin filaments in our present simulations).
The percolation analysis suggested that if the connections between the travelling sites are static,
the diffusion coefficients dropped linearly down to zero as Chorizontal was increased from 0 to 0.35
(see the dot lines in Fig. 4.7B), which is the percolation threshold (116). When Chorizontal > 0.35,
the diffusion coefficients of the travelling particles become zero. The diffusion coefficients
measured in our simulations at n = 0.1 and 0.15 show that when Chorizontal = 100%, 90% and 70%,
the band-3 diffusion coefficients are approximately zero, in agreement with percolation analysis.
However, a notable band-3 diffusion coefficient of
/t
6 10
4
2
is measured at Chorizontal= 50%,
s
contradicting with the percolation analysis. This discrepancy is believed to result from the fact
that the band-3 particles in our simulation have not explored all the accessible areas in the
membrane (Can you explain this). The time for the band-3 particles to fully explore the RBC
membrane with Chorizontal = 50% cannot be accessed by our simulation. When Chorizontal = 30%,
93
which is below the percolation threshold, the diffusion coefficient measured from our simulation
is again consistent with the percolation analysis. But when Chorizontal is reduced down to 10%, the
diffusion coefficients measured from our simulation are lower than the ones predicted by
percolation analysis. This difference is probable caused by the fact that in the percolation
analysis the diffusing particles can pass the broken boundaries of the compartments without
interfering with the broken spectrin filaments whereas in our model, the broken spectrin
filaments can still interact with the band-3 particles and hinder their lateral motions. Although
the band-3 diffusion coefficients measured at large attractive forces (n ≥0.1) are generally
consistent with the percolation analysis, we note that our diffusion measurements only agree with
the experimental measurements on the HS and HE cells (112, 115, 197) for the cases of n = 0
and 0.05 where the band-3 diffusion coefficients increase, but not significantly in the HS, and
largely boosted in the HE (112). Therefore, it is reasonable to predict that the effective attractive
force between the spectrin filaments and lipid bilayer is at least 20 times smaller than the binding
forces at the two major binding sites between the cytoskeleton and the lipid bilayer. in agreement
with the experimental measurements(204, 205).
4.4. Summary
We apply a two-component RBC membrane model to study band-3 diffusion in the normal RBC
membrane and in the membranes with defective vertical and horizontal interactions. We measure
the band-3 diffusion coefficients and quantify the relation between the band-3 diffusion
coefficients and percentage of protein defects. Our measurements show that the band-3 diffusion
coefficients are increased by 8 times when connectivity between the spectrin filaments and
immobile band-3 particles Cvertical is decreased from 100% to 0%, whereas the band-3 diffusion
94
coefficients can be boosted by 20-40 times with significantly reduced cytoskeleton connectivity
Chorizontal. These results are in agreement with the experimental measurements of the band-3
diffusion in the HS and HE RBCs (112, 115, 147, 197). By comparing the measured band-3
diffusions coefficients, we demonstrate that cytoskeleton connectivity is the major determinant
of the lateral diffusivity of band-3. In addition, we quantify the relations between the anomalous
diffusion exponent and the percentage of protein defects in the horizontal interactions. We find
that the cytoskeleton has a small effect on the anomalous diffusion exponent, comparing to the
obstructions due to the lipid rafts or protein domains. At last, we introduce attractive forces
between the spectrin filaments and the lipid bilayer, and study the effects of the attractive forces
on the band-3 diffusion. We find that application of the attractive forces slows down the band-3
diffusion in the normal and defective RBC membrane. Especially, large attractive forces can
prevent the band-3 hop diffusion in the normal RBC membrane and in the membrane with
protein defects in the vertical interactions. But, in the cases of the membrane with protein defects
in the horizontal interactions, the band-3 diffusion coefficients become independent of the
attractive forces when attractive forces are large. This is because the band-3 particles travel to
different cytoskeleton compartments through the broken spectrin filaments. Moreover, we show
that the band-3 diffusion coefficients measured at the large attractive forces generally agree with
the percolation analysis in (116). By comparing the band-3 diffusion coefficients from our
simulation with the experimental measured band-3 diffusion coefficients in HS and HE (112, 115,
147, 197), we estimated the scale of the effective attractive force between the spectrin filaments
and lipid bilayer is at least 20 times smaller than the binding forces at the two major binding sites
at the immobile band-3 proteins and the glycophorin C. The approach used here for the
95
simulation of band-3 diffusion in the defective RBC membrane provides a basis for the future
study of the mechanism of the membrane vesiculation in HS and HE.
96
Chapter 5.
Modeling of Vesiculation in Healthy and Defective Human
Erythrocyte Membrane
Abstract
Red blood cells-derived microvesicles play important roles in regulating cell communications
and modulating immune response and other patho-physiological processes. In the blood
disorders, microvesiculation in the defective red blood cell (RBC) membrane becomes
complicated due to the various protein defects. In addition, membrane domains which change the
local curvature of the RBC membrane could also affect the microvesiculation. In this work, we
simulate the vesiculation in the RBC membrane by applying a two-component coarse-grained
molecular dynamics (CGMD) RBC membrane model. We study the effect of the spontaneous
curvature of the membrane domain on the microvesiculation induced by the compression on the
RBC membrane. Furthermore, we study the vesiculation in RBCs from patients suffering from
the blood disorders of hereditary spherocytosis (HS) and elliptocytosis (HE). Our simulation
results show that the large spontaneous curvature of a membrane domain can induce vesicles
made of homogeneous composition and the diameters of the vesicles are less than 50 nm. On the
other hand, the compression on the membrane can cause the formation of vesicles with
heterogeneous composition and the sizes of the obtained vesicles are similar to the size of the
cytoskeleton corral. When both effects are considered, the compression on the membrane tends
to facilitate the formation of vesicles originated from the curved membrane domain. In the HS
and HE, we find that the sizes of microvesicles become more diverse comparing to the sizes of
97
the microvesicles released from the healthy RBC, due to the reduced confinement from the
cytoskeleton on the lipid bilayer. When the vertical connectivity between the lipid bilayer and
the cytoskeleton is elevated, multiple vesicles, with sizes similar to the cytoskeleton corral
dimension, are shed from the compressed membrane, whereas membrane with low vertical
connectivity tends to release larger vesicles under the same compression ratio as above. It is
noted that vesicles released from the HE RBCs could contain cytoskeleton components due to
the fragmentation of the cytoskeleton while vesicles released from the HS RBCs are depleted of
cytoskeleton components.
5.1. Introduction
Human RBC circulates in human body for about half million times during its 120 days lifespan
to deliver oxygen from the lung to all tissues and transport carbon dioxide from tissues back to
the lung. During its circulation, RBC loses surface area and the lipid content by approximately
20%, mainly by shedding of hemoglobin-contained vesicles (223, 224). RBCs, which constitute
40% of the total blood volume, are one of the major vesicle-secreting cells in the circulating
blood (225), particularly in the course of some pathological conditions such as HS and sickle cell
disease (SCD)(226). Membrane released vesicles are commonly grouped into two categories:
namely nanovesicles and microvesicles, distinguished by the size of the vesicle. Nanovesicles
have a mean diameter of approximately 25 nm (15, 227, 228), whereas microvesicles have an
approximate diameter in a range of 60-300 nm (229). The formation of nanovesicles is caused by
the aggregation of specific types of the membrane proteins or lipids which have mutual affinity
and capability of forming membrane domains (229). Membrane domains with significant
spontaneous curvatures can bud off from the RBC membrane and form vesicles (229). The
98
compositions of the nanovesicles mainly contain the lipids or the proteins that constitute the
membrane domains. Due to the small sizes of the nanovesicles (~25 nm) comparing to the size of
the corrals formed by the cytoskeleton (~90 nm), the formation of the nanovesicles is not
affected by the integrity of cytoskeleton. On the other hand, the formation of microvesicles
originates from the lipid bilayer area that uncouples from the cytoskeleton during the aging of
the RBC (230). Previous studies have shown that the RBC cytoskeleton is under stretch when it
is tethered to lipid bilayer and therefore exerts a compression force on the lipid bilayer (13, 120).
This compression force is balanced by the bending force induced by the membrane curvature (13,
118). As the cell ages, the stiffness of the cytoskeleton is increased by 20% (231) and its density
is raised by 30-40% (30). As a result, the cytoskeleton imposes larger compressive forces on the
cell membrane. Thus, the curvature of the membrane has to increase to accommodate the
enhanced compression forces, leading to membrane buckling and buds formation at the scale of
the cytoskeleton corral-size. The buds could grow in size with further increased compression
forces, and subsequently detach from the membrane in order to release the bending stress (118,
151). The compositions of the microvesicles are more heterogeneous than those of the
nanovesicles. For example, the lipid composition of the microvesicles is similar to that of the
parent RBCs, but the components of the membrane cytoskeleton, such as the spectrin and actin,
are in general not detected (232-234). In addition, the microvesicles contain the major membrane
integral proteins in approximately the same relative amounts as the parent cell membrane (235).
Vesiculation from the RBCs of the patients suffering from blood disorders, such as HS and HE,
raises interesting questions as the erythrocyte-released vesicles were found to be increased in the
99
blood content of the patients with blood disorders (3, 4, 16, 20, 26, 145, 236, 237). After losing
vesicles, the RBCs transfer from discocytes to irregular-shaped RBC with the reduced cell
deformability, resulting in early removal from circulation by the spleen (3, 25, 145, 207, 236,
238). In HS, the defects in the proteins that tether the cytoskeleton to the lipid bilayer, such as
ankyrin, protein 4.2 and band-3 proteins, stiffen the cytoskeleton and produce an area in the RBC
membrane which is not supported by the cytoskeleton, facilitating the vesiculation. A common
feature of HS is loss of membrane surface area and resultant change in cell shape from
discocytes to spherocytes. In HE, the protein defects, such as α-spectrin, β-spectrin and protein
4.1 defects, occur in membrane cytoskeleton and disrupt the cytoskeleton. Since the cytoskeleton
is the main load carrier of the RBC (237, 239, 240), the HE RBCs become vulnerable in the high
shear-stress blood flow. As a result, they shed vesicles and undergo progressive shape
transformations from the biconcave to the elliptical shapes. Various protein defects in the
defective RBC membrane not only impact the RBC morphology, but also determine the contents
of released vesicles. For example, the microvesicles shed from the normal RBCs have similar
compositions as the parent RBCs (241), while microvesicles released from spectrin/ankyrin-
deficient RBCs are enriched in band-3 proteins (242). In addition, protein defects that cause the
disruption of the cytoskeleton or weakened connections between the lipid bilayer and
cytoskeleton could promote the protein diffusion in the membrane and potentially facilitate the
formation of the membrane domains and nanovesiculation (112, 115, 118, 197).
Numerous experimental studies have been conducted to understand the mechanism of RBC
membrane vesiculation and the change of the RBC morphology. For example, experimental
measurements showed that increasing the intracellular calcium concentration could promote
100
RBC vesiculation and transition of the cell shape from discocyte to echinocyte (243, 244).
Microvesicles were shed from the tip of echinocyte spicules (244). Calcium-induced vesicles are
characterized by being devoid of phosphatidylcholine (PC), spectrin, actin, and glycophorin, but
preserving the band-3 proteins. Depletion of adenosine triphosphate (ATP) in the RBC is found
to induce similar morphological changes as boosting calcium content, but the protein
compositions in the released vesicles are different (234, 244, 245). In addition, membrane
vesiculation is found to serve as a way of removing the nonfunctional proteins and auto-antigens
through releasing the damaged membrane patches (224). Recently, de Vooght et al (246)
proposed that mature RBCs are probably the only cells that do not secrete exosomes and shed
only microvesicles. Although the experimental measurements facilitate the studies of the
membrane vesiculation, they are not sufficient to fully understand the mechanism of the
vesiculation. Therefore, analytical studies were performed to explain the cell morphology
changes and vesicle formation (11, 13, 15, 74, 129, 151, 247-249). Sheetz and Singer first raised
a bilayer-couple hypothesis which assumed the sequential transition of stomatocyte–discocyte–
echinocyte was induced by the area difference between the two leaflets of the plasma membrane
alone (250). However, the area–difference–elasticity (ADE) model developed based on the
bilayer-couple hypothesis showed that as the area difference increased, the model tended to form
buds instead of the echinocytosis (251). This impasse was solved by introducing the elastic
energy of the cytoskeleton into the ADE model (252-254). Since the cytoskeleton is largely
stretched or sheared during the budding formation, the energy corresponds to the budding
configuration is significantly increased, leaving the echinocyte to be the preferred shape. Lim et
al (13) model the stomatocyte–discocyte–echinocyte sequence by varying only the area
difference between the two leaflets in the ADE model with the effect of the cytoskeleton. The
101
authors also noted that area difference had equivalent effects as the spontaneous curvature on
changing the cell shape because the spontaneous curvature naturally originated from the area
difference between the two leaflets. The spontaneous curvature of the RBC membrane could
result from the molecule asymmetry (9), area mismatch between two leaflets of the lipid bilayer
(13, 249), or protein-lipid hydrophobic mismatch, or protein-assisted curvatures (255-257). It can
be enhanced by incorporating the effects of increased anionic amphipaths, high salt, high pH,
ATP depletion and cholesterol enrichment. Although the ADE model with consideration of the
cytoskeleton successfully describes the cell shape transition, other mechanisms for the regulating
the shape of cells were proposed. Another popular mechanism states that the budding and
vesiculation in the membrane is provoked by compression on the lipid bilayer induced by the
stiffened cytoskeleton. The increased stiffness of the cytoskeleton may result from the depleted
ATP as RBC ages (151, 229), or cause by the protein defects which weaken the vertical
connections between the lipid bilayer and cytoskeleton, such as in the HS RBCs (3, 145, 236).
Spangler et al. (258) introduced a lipid membrane sphere model by combining particle-based
lipid bilayers to an elastic meshwork to study cytoskeleton-induced blebbing. Their simulation
results clearly demonstrate that membrane blebbing results from a localized disruption or a
uniform contraction of the cytoskeleton. The enhanced vesiculation observed in the HS RBC can
be explained by the Spangler et al. (258)'s simulation and thus favor the mechanism that the
budding and vesiculation in the membrane are induced by cytoskeleton-induced compression.
Microvesiculation in the RBC membrane is more likely to result from the spontaneous curvature
and cytoskeleton-induced compression simultaneously because the increased intracellular
calcium or depleted ATP not only promote asymmetry of leaflets, but also activate sproteolytic
enzymes like calpain which breaks down the tethering points between the membrane
102
cytoskeleton and the lipid bilayer (236). Therefore, the exact underlying mechanisms of RBC
vesiculation are not completely deciphered.
Although previously developed model succeeded in explaining the cell morphology change and
membrane budding and vesiculation, there are still questions that remain unanswered. For
example, the spontaneous curvatures and boundary line tension of the lipid rafts or protein-
enriched membrane domain, which play important roles in the RBC signaling, trafficking and
material transport (259-262), can results in a domain morphological transition from a flat to a
dimpled state (263). The effects of the elastic properties on formation of the membrane domain
as well as the interactions between the domains were studied in (263, 264). However, how does
the spontaneous curvature of the membrane domain affect the microvesiculation is not clear.
Also, the increase in raft lipids and sphingolipids could excites a lipid raft coalescence process
by clustering the raft components, which potentially promotes the vesiculation from the
membrane domain (265). In addition, mechanism of the vesiculation in RBCs in the pathological
conditions such as HS and HE is still not well understood due to the complication induced by the
different protein defects. In order to answer the above questions, we apply our previously
developed two-component CGMD RBC membrane model (see Fig. 5.1A-C) to simulate the RBC
membrane vesiculation in the normal RBC membrane and in the RBC membrane with protein
defects. An important feature of the applied membrane model lies in introducing the rotational
degrees of freedom into the interactions between the lipid CG particles, glycophorin and band-3
proteins, which allow us to introduce spontaneous curvatures into the lipid bilayer domains. In
addition, the applied model explicitly represent the lipid bilayer, band-3 proteins and the
cytoskeleton by CG particles which can be used to study the interactions between the lipid
103
bilayer and cytoskeleton and the function of the cytoskeleton on maintaining the integrity of the
RBC membrane. First, we simulation the vesiculation induced by the spontaneous curvature of
the membrane domain and vesiculation induced by the compression on the lipid bilayer,
respectively. Second, we simulate vesiculation in the RBC membrane due to the spontaneous
curvature of the membrane domain and due to the compression on the lipid bilayer. At last, we
study the vesiculation in the RBC membrane with protein defects, such as in HS and HE cells
and investigate the effects of protein defects on the vesiculation(266).
Figure 5.1. (A) Schematic of the human RBC membrane. The blue and yellow spheres represent
two types of lipid particles corresponding to membrane areas with different spontaneous
curvatures. The red sphere signifies an actin junctional complex. The white sphere represents a
spectrin particle and the black sphere represents a glycophorin particle. The light blue circle
corresponds to a band-3 complex that is connected to the spectrin network while green circle
104
represents a band-3 complex that does not connect to the lipid bilayer. (B) Top view of the
membrane model with two types of lipid CG particles completely mixed. (C) Top view of the
membrane model after the G type particles aggregate into membrane domains. "A" type particles
represent actin junctional complexes, "B" type particles represent spectrin proteins, "C" type
particles represent glycophorin proteins, "D" type particles represent a band-3 complex that are
connected to the spectrin network ("immobile" band-3), "E" type particles represent band-3
complex that are not connected to the network ("mobile" band-3), "F" type particles represent
majority of lipid particles and "G" type particles represent minority of the lipid particles.
5.2. Model and simulation method
The two-component CGMD human RBC membrane model applied in this work explicitly
comprises both the lipid bilayer and the cytoskeleton. Three types of particles are introduced to
represent lipid bilayer (see Fig. 5.1A). The blue and grey color particles signify a cluster of lipid
molecules, which have a diameter of 5 nm, a similar value to the thickness of the lipid bilayer.
The difference between the two types of lipid particles will be introduced later. The black
particles denote glycophorin proteins with the same diameter of the lipid particles. The third type
of particles denotes band-3 complexes. One third of band-3 complexes, represented by the light
bule particles, are connected to the spectrin filaments. The rest of the band-3 complexes can
freely diffuse in the lipid bilayer and they are represented by green particles. The cytoskeleton is
comprised of the hexagonal spectrin filaments connected at actin junctions. The actin junctions
are represented by the red particles in Fig.5.1A and it is connected to the lipid bilayer via
glycophorin. Spectrin is a protein tetramer formed by head-to-head association of two identical
heterodimers which consists of an α-chain with 22 triple-helical segments and a β-chain with 17
105
triple-helical segments. In the proposed model, therefore, the spectrin filament is represented by
39 spectrin particles (white particles) connected by unbreakable springs. The details of the
potentials applied in the cytoskeleton of the membrane model and the geometry of the CG
particles can be found in the authors' previous work (211). The lipid particles, the glycophorin
particles and band-3 particles carry both translational and rotational degrees of freedom (xi, ni),
where xi and ni are the position and the orientation (director vector) of particle i, respectively.
The rotational degrees of freedom obey the normality condition ni = 1. Thus, each particle
effectively carries 5 degrees of freedom. xij = xj - xi is defined as the distance vector between
particles i and j. Correspondingly, rij ≡ xij is the value of the distance, and ij
x
x r
ij
ij
is a unit
vector. The lipid particles, the glycophorin particles and the band-3 particles interact with one
another via a pair-wise additive potential
u
m e m
n n x
,
,
j
i
ij
u
R
r
ij
A
,
a
n n x
,
,
j
i
ij
u
A
r
ij
,
(5.1)
r
ij
cut ,mem
r
eq
cut ,mem
r
ij
R
R
cut ,mem
R
2
r
ij
u
A
R
cut ,mem
0,
u
R
u
A
u
R
r
ij
r
ij
r
ij
8
r
eq
4
for
r R
ij
<
cut ,mem
for
r
ij
R
cut ,mem
for
r
ij
R
cut ,mem
(5.2)
where uR(rij) and uA(rij) are the repulsive and attractive components of the pair potential,
respectively. α is a tunable linear amplification factor and it is chosen to be 1.55. The function
A(α,a(ni,nj,xij))= 1+α(a(ni,nj,xij) -1) which tunes the energy well of the potential, regulate the
fluid-like behavior of the membrane. The interaction between two CG particles depends not only
106
on their distance but also on their relative orientation via the function
ij
xnn
ia
,
,
j
which varies
from -1 to +1 and adjusts the attractive part of the potential. We specify that
the case when
in is parallel to
jn and both are normal to vector
ijx
n
i
1a
n
j
corresponds to
x
ij
, and the
value
n
i
1a
n
j
to the case when
ij
x
in is anti-parallel to
jn and both are perpendicular to vector
ijx
. The former instance is energetically favored due to the maximum attractive
interaction between particles i and j, while the latter is energetically disfavored due to the
maximum repulsive interaction. One simple form of
characteristics is
ij
xnn
ia
,
,
j
that captures these
n ,n ,x
ij
i
j
a
n n
i
j
n x
i
ij
n x
j
ij
.
(5.3)
where β is the parameter that adjusts the curvature of the lipid bilayer, which allow us to
introduce a variety of spontaneous curvatures into lipid bilayer. The effects of function
A(α,a(ni,nj,xij)) and the details of the applied potentials between the lipid CG particles were
discussed in the previous works (111, 211). The system consists of N = 32796 CG particles. The
dimension of the membrane is approximately 0.8 μm × 0.8 μm. The temperature of the system is
maintained at kBT/ε = 0.22 with the Nose-Hoover thermostat. The model was implemented in the
NAT ensemble (95, 131, 132). Since the model is solvent-free and the membrane is a two-
dimensional structure, we controlled the projected area instead of the volume. The projected area
was adjusted to result in zero tension for both the lipid bilayer and the cytoskeleton at the
equilibrium state. The details about the potentials applied in the membrane model, geometries of
107
the CG particles and the timescales of the model can be found in the authors' previous work (111,
211).
5.3. Results and discussion
5.3.1 Vesiculation in health RBC membrane
5.3.1.1 Vesiculation due to spontaneous curvature of membrane domain
First, we simulate the vesiculation due to spontaneous curvature of membrane domain.
Membrane domains, such as lipid rafts, have draw significant attentions as increasing evidences
have shown that they play an crucial role in regulating cellular processes including cell polarity,
protein trafficking and signal transduction (263, 267-272). Membrane domains consist of
specific type of lipids or proteins which are more ordered and tightly packed than the
surrounding bilayer. For example, lipid rafts in plasma membranes are enriched in sterol- and
sphingolipids with size ranging from 10 to 200 nm (259-261). In order to create phase separation
and form membrane domains in the applied model, we randomly assign the lipid CG particles as
type F and G. The number ratio between component G (blue) and F (gray) is 2:8 and it is
conserved throughout simulations. Type G particles are considered to be the minority and
clusters of G particles are treated as domains. The line tension between two types of lipid
particles F and G is simulated by assuming that the association energies between the F-F and G-
G particles are the same, but the association energy between the F-G particles is only 70% of the
association energy between the F-F and G-G. We start the simulation with two types of lipid
particles completely mixed, as shown in Fig. 5.1B. Driving by the different association energies,
the G particles are aggregating gradually to small islands and the islands can merge to become
the domains (see Fig. 5.1C). The domain size ranges from 20 to 100 nm, consistent with the
108
range of 10 to 200 nm observed in (259-261). Then, we introduce a variety of spontaneous
curvatures to the G particles to mimic different domain curvatures which caused by the mismatch
between two leaflets or protein-assisted curvatures. In our model, the spontaneous curvature can
be controlled by tuning the parameter β in the potential introduced in Eq.(5.3). Fig. 5.2A shows
that when β = 0.1, the domains only bud off from the membrane and no vesicles are formed,
meaning that the energy resulting from the line tension and spontaneous curvature is not large
enough to drive the vesiculation. As the spontaneous curvature increases, the buds become more
curved. When the β is increased to 0.18, vesiculation is observed and the released vesicles are
made of the G type particles only (see Fig. 5.2B). This means the spontaneous curvature of
membrane domain has to be large enough to drive the vesiculation and the vesicles induced by
the spontaneous curvature has the same compositions as the membrane domains. As the β
continues to increase, more vesicles are formed from the membrane (see Fig. 5.2C). Since the
size of the vesicle depends on the spontaneous curvature of domain of the G particles, the
maximum size of the vesicles induced by the spontaneous curvature is obtained at β = 0.18, and
the diameter of the obtained vesicles is about 40 nm, which is larger than the size of the
nanovesicles but smaller than the lower end of the 60-300 nm range for the size of the
microvesicles (118). With the spontaneous curvature increasing, the sizes of the vesicles can
drop down to 20 nm, similar to the size of the nanovesicles. Owing to the smaller vesicle size
comparing to the typical size of the cytoskeleton corral (~90 nm), the cytoskeleton probably does
not affect the vesiculation induced by the spontaneous curvature.
109
Figure 5.2. Vesiculation induced by the spontaneous curvature of the membrane domains. (A)
When the spontaneous curvature is small (β = 0.1), the membrane domains only bud off from the
membrane and no vesicle is formed. (B) When β is increased to 0.18, vesiculation is observed
and the released vesicles are made of the G type particles only. (C) When β is increased to 0.24,
more vesicles are formed from the membrane.
5.3.1.2 Vesiculation due to the compression on the lipid bilayer
In this section, we simulate vesiculation in the RBC membrane model driven by the compression
on the lipid bilayer. When the cytoskeleton is tethered to the lipid bilayer, it is under tension
because the cytoskeleton can shrink by 3-5 times after the lipid bilayer is removed (273). On the
other hand, the cytoskeleton applies a compressive force on the lipid bilayer. As the RBC ages,
110
the depleted ATP level in the cell leads to increased density and stiffness of the cytoskeleton,
which impose a larger compressive force on the lipid bilayer and thus facilitate the vesicle
formation (151). As a result, RBC constantly loses cell membrane through shedding vesicles
during its circulation in the human body (151, 274). The size of RBC-released vesicles is similar
to the size of the corral of the RBC cytoskeleton (275). In this section, we simulate the effects of
the stiffened cytoskeleton by applying compression on the applied RBC membrane model. In the
applied RBC membrane model, the soft cytoskeleton shrink after compression while the lipid
bilayer is incompressible. Thus the applied compression on the membrane increases the area
ratio between the lipid bilayer and cytoskeleton, similar to the method applied in a previous
study (258). The compression ratios, which is defined as the percentage of the area reduced from
initial projected area of the membrane, are selected to be Rcompression = 2%, 5%, 10%, 15% and
18%, respectively. In order to better understand the membrane vesiculation process, we explain
our simulation results based on Helfrich's membrane model (9). It is known that the free energy
function of non-sheared membrane is controlled by the membrane curvature induced by the
thermal fluctuation, spontaneous curvature, membrane domain line tension and cytoskeleton
stretching energy (ref.). The expression for the free energy is described by
F
b
2
C C C dA
2
0
1
2
l
dA K
l
/ 2
2
dA
c
(5.4)
where (C1+C2)/2 is the mean curvature originated from the thermal fluctuation of the membrane,
C0 is the spontaneous curvature, γ is the line tension at the boundary of the membrane domains,
Kα is linear elastic moduli of the cytoskeleton and α is the cytoskeleton local area invariants
(276). In addition, the lipid membrane and integral membrane proteins are constrained when they
111
are moving towards the negative z-direction by applying a harmonic potential to mimic the effect
of the cytoplasm inside the RBC. The potential is described as
V
confine
1
2
K
confine
2
z
(5.5)
where Kconfine is the confinement coefficient and z is the distance measured from the mid-plain of
the supercell. Kconfine is chosen to be ε/σ2.
Figure 5.3. Vesiculation induced by compression on the lipid bilayer. (A) When the compression
ratio Rcompression = 2%, only a small bud is created on the membrane and the cytoskeleton
conforms to the bud. (B) When Rcompression is increased to 5%, the bud grows up and the
cytoskeleton retracts from the budding area. (C) When Rcompression is increased to 15%, two
vesicles are released from the membrane and they consist of both G and F lipid particles.
112
Fig. 5.3A shows that when Rcompression = 2%, only buds are created on the membrane and the
cytoskeleton is observed to conforms to buds. Under this condition, the free energy of the
membrane is controlled by the curvature energy of the lipid bilayer and the elastic energy of the
cytoskeleton induced by the local stretch on the budding areas. As the compression ratio
increases, the buds grow up and the cytoskeleton maintains conforming to the growing buds until
the stretching force is strong enough to pull the cytoskeleton off from the budding areas. Then
the budding area is unsupported from the cytoskeleton when Rcompression = 5% and 10%, as shown
in the Fig. 5.3B and the cytoskeleton becomes stress-free. Therefore, the free energy of the
membrane is only controlled by the bending energy of the lipid bilayer. Further increase in the
compression ratio is balanced by the increased curvature in budding area. At Rcompression = 15%,
the neck of the bud close itself and pinch the vesicle off from the membrane to release the
bending energy, as shown in the Fig. 5.3C. It is noted that the released vesicles have a more
heterogeneous composition (consists of both G and F lipid CG particles) compared to the
vesicles that form due to domain spontaneous curvature. This observation indicates that the line
tension induced by the different association energy between the type A and B lipid particles do
not play an important role in the vesiculation caused by compression on the lipid bilayer. In
addition, the sizes of the vesicles are ~90 nm, similar to the size of the cytoskeleton corral, larger
than the vesicles induced by the membrane domain curvatures, which is consistent with the
experimental observations (16, 206, 236). A larger compression ratio (Rcompression = 18%) in our
simulation create multiple vesicles at the similar sizes, instead of a single larger size of vesicle.
The larger-size microvesicles may result from the breakage between the band-3 and spectrin
113
filaments due to the depletion of ATP in the cell-aging process, which will be discussed in the
section 3.2.2 below.
5.3.1.3 Vesiculation due to the combined effects
In the last two sections, we describe the membrane vesiculation induced by the spontaneous
curvature of membrane domain and by the compression on the lipid bilayer, respectively. In the
RBC vesiculation process, these two effects could co-exist. For example, the curved membrane
domains or lipid rafts happen to locate in the membrane areas which are not supported by the
cytoskeleton. Therefore, in this section, we simulate the membrane vesiculation under the
combined effects of spontaneous curvature of membrane domain and compression. First, we
simulate the conditions where the spontaneous curvature of the membrane domain is small so
that the domains only form buds but not vesicles. The Rcompression are selected to 5% and 10%,
respectively. We find that when the spontaneous curvature is very small (β = 0.05), the
vesiculation processes at Rcompression = 5% and 10%, only produce a big bud, as shown in Fig.
5.4A, are similar to the cases when membrane is under compression and no spontaneous
curvature is introduced. When Rcompression is increased to 15%, vesicles consisting of two types of
lipid particles are released. This means that small spontaneous curvatures do not affect the
membrane vesiculation due to the compression. However, when the spontaneous curvature is
increased to β = 0.1, vesicles made of only G lipid particles are observed at Rcompression = 5% and
10% (see Fig. 5.4B and 5.4C). As shown in Fig.5. 2A, the spontaneous curvature of β = 0.1 alone
cannot drive vesiculation. Therefore, the compression applied here promotes the vesiculation of
the membrane domain. It is noted that the compression on the lipid bilayer prefer to stimulate the
growth of the curved membrane domain, instead of promoting the buds from membrane patches
114
made of two types of lipid particles, like the case of β = 0.05 under the same compression. When
Rcompression = 5% and β = 0.1, two vesicles with size of ~60 nm are observed (see Fig. 5.4B).
When Rcompression = 10%, three vesicles are released from the membrane with sizes ranging from
40-60 nm, depending on the size of the originated membrane domains (see Fig. 5.4C). As β
increases, more vesicles made of only G lipid particles are released from the membrane at
Rcompression = 5% and 10%. For the cases of β >= 0.18, where the small-sized vesicles can be
formed spontaneously without compression, our simulations show that when the membrane is
compressed at Rcompression = 5% and 10%, the self-driven small-sized vesicles are first released
from the membrane and then larger size vesicles stimulated by the compression are observed.
Figure 5.4. Vesiculation induced by the combined effects of spontaneous curvature and
compression on the lipid bilayer. (A) When the spontaneous curvature is very small (β = 0.05),
the vesiculation processes at compression ratio of Rcompression = 5% and 10%, only produce a big
115
bud. (B) When Rcompression = 5% and β = 0.1, one vesicle is released. The vesicle has a size of ~60
nm and consists of only G lipid particles. (C) When Rcompression = 10%, three vesicles are released
from the membrane with sizes ranging from 40-60 nm.
5.3.2. Vesiculation in defective RBC membrane
In this section, we simulate the vesiculation in the RBC membrane with protein defects, such as
in the HS and HE RBCs, and study the effects of the percentage of protein defects on the
vesiculation. The defects in the cytoskeleton proteins or in the proteins that tether the
cytoskeleton to the lipid bilayer can create an area in the membrane which is not supported by
the cytoskeleton, and thus facilitate the RBC vesiculation. As a results, erythrocyte-released
vesicles were found to be increased in the blood content of the patients with blood disorders (3, 4,
16, 20, 26, 145, 237, 277), leading to the altered cellular morphology, reduced cell deformability
and early removal by the spleen (3, 25). Here, we introduce the protein defects into the applied
membrane model and simulate the vesiculation process in the HS and HE RBC membrane,
respectively.
5.3.2.1 Vesiculation due to the spontaneous curvature of the membrane domain in HS and HE
cell membrane
First, we study the effects of the protein defects in the vertical interaction between the lipid
bilayer and cytoskeleton, and the protein defects in the horizontal interaction within the
cytoskeleton on the vesiculation originated from the membrane domain. We repeat the
simulation conducted in the section 3.1.1, with the connectivity in the vertical interaction and
horizontal interaction selected to be Cvertical = 50% and Chorizontal = 50%, respectively. We find
116
similar results to the membrane vesiculation in the normal RBC membrane, meaning that the
protein defects in the membrane do not affect the vesiculation caused by the spontaneous
curvature. This probably results from the fact that the sizes of the vesicles induced by the
spontaneous curvature of the membrane domain are smaller than the size of the corral in the
cytoskeleton.
5.3.2.2 Vesiculation due to the compression in the HS cell membrane
Second, we simulate the vesiculation in the membrane with protein defects in the vertical
interactions. It could be caused by the defective band-3, protein 4.2 and ankyrin proteins in the
HS (3, 145, 236). In addition, the decreased Cvertical also can be induced by the ATP depletion as
RBC ages (151, 229). The effects of protein defects in the vertical interaction are represented by
breaking the connections between band-3 and spectrin particles. The Rcompression are selected to be
5%, 10%, 15% and 18% of the initial projected area of the membrane. The band-3 and spectrin
connectivity (vertical connectivity) Cvertical between the lipid membrane and cytoskeleton is
reduced from 100% to 0%. When the Rcompression is low (Rcompression = 5%), Fig. 5.5A shows that
only one bud is created from the membrane and no vesicle is released from the membrane,
similar to cases for the normal RBC membrane. When a higher Rcompression = 10% is applied, the
unsupportive budding area becomes larger to accommodate the increased compression. Under
this condition, vesiculation still does not occur for all Cvertical and only two buds are formed. At
large Rcompression (15% and 20%), we find vesiculation for all the Cvertical, as shown in the Fig 5.5C.
However, vesiculation process is different for the low Cvertical and high Cvertical. At high Cvertical, the
vesiculation starts from the area within the corral of the cytoskeleton and thus creates vesicles
with sizes similar to the corral (~90nm), while at the low Cvertical, buds can migrate and merge to
117
larger buds and then form vesicles with sizes of ~400 nm, consistent with (229). This means that
high vertical connectivity Cvertical in the normal RBC membrane can slow down or prevent the
migration of the buds, therefore favors the formation of vesicles with sizes similar to the size of
corral. This is clearly demonstrated in the Fig. 5.5C which shows that at high Rcompression = 15%
and 18%, the membrane tends to release multiple corral-sized vesicles at high Cvertical, while
when the Cvertical is low, the sizes of the released become more diverse (from 90 nm to 400 nm)
because there is less constrain from the cytoskeleton. These simulation results explain the
various sizes of vesicles shed by RBC (16, 229, 236). In addition, our simulations illustrate that
the cytoskeleton does not fragment during vesiculation, resulting in vesicles completely depleted
of cytoskeleton components, as shown in the Fig. 5.5B. This result is consistent with the
experimental observation on the development of the spherocytes from the RBCs with defective
band 3 or protein 4.2, but normal amounts of spectrin in (278). The vesicles released from the HS
RBC membrane all contain band-3 proteins (see Fig. 5.5B), which is consistent with the
hypothesis raised by Eber and Lux (16). Since we do not assign band-3 particles structural
functions on stabilizing the membrane in this model, the vesiculation in the HS RBC membrane
here mainly result from membrane area which is not supportive by the cytoskeleton, instead of
from band-3 deficient area.
118
Figure 5.5. Vesiculation in the membrane with protein defects in the vertical interactions. (A)
When the compression ratio Rcompression is low (Rcompression = 5%), only one bud is created from the
membrane for all the selected Cvertical. (B) When Rcompression = 15% and vertical connectivity
Cvertical = 60%, one vesicle is created. (C) Summary of dependence of membrane vesiculation on
vertical connectivity Cvertical and compression ration Rcompression. In the figure, × indicates no
vesiculation occurs. ① indicates one vesicle is observed. ② indicates two vesicles are observed.
③ indicates three vesicles are observed.
5.3.2.3 Vesiculation due to the compression in the HE RBC membrane
At last, we simulate the vesiculation in the membrane with protein defects in the horizontal
interactions, such as α-spectrin, β-spectrin and protein 4.1 defects in the HE cell membrane. The
protein defects in the horizontal interaction are represented by dissociations of spectrin chains at
the spectrin dimer-dimer interactions. The cytoskeleton connectivity (horizontal connectivity)
119
Chorizontal = 0%, 20%, 40%, 60%, 80% and 100% are tested, respectively. The compression ratios
Rcompression are selected to be 5%, 10% and 15%, respectively. When the Rcompression is low
(Rcompression = 5%), only one bud is created from the membrane and no vesicle is released from
the membrane for all the selected Chorizontal, similar to cases for the normal RBC membrane and
HS cell membrane. However, it is noted that when the Chorizontal is low, such as the case of
Chorizontal = 20% shown in the Fig. 5.6A, the fragments of the cytoskeleton are observed to go
with the bumped lipid membrane as the cytoskeleton is disrupted at this low Chorizontal. When a
higher Rcompression = 10% is applied, the unsupportive budding area grows up to accommodate the
increased compression. Vesiculation does not occur at Chorizontal ≥ 60% and only two buds are
formed. But as the Chorizontal decreases, vesicles are observed, as shown in the Fig. 5.6B, as more
membrane area becomes unsupported from the cytoskeleton. The vesiculation from the
membrane with reduced horizontal connectivity Chorizontal indicates that the cytoskeleton of the
RBC has the functions of preventing the membrane from losing vesicles. Moreover, it is noted
that at Rcompression = 10%, no vesiculation is observed from the HS cell membrane for all the
selected Cvertical. The difference of the vesiculation in the HS and HE cells indicates that the
cytoskeleton plays the major role in maintaining the stability and integrity of the RBCs. When
Rcompression is increased to 15%, Fig. 5.6C shows that vesiculation occurs for all the selected
Chorizontal. Similar to the results obtained from the HS cell membrane, two corral-sized vesicles
are created at high Chorizontal (Chorizontal ≥ 60%) while one larger vesicle is found at low Chorizontal
(Chorizontal ≤ 40%) due to the reduced confinement of the cytoskeleton on the lipid bilayer. It is
interesting to see that the vesicles resulting from the HE RBC membrane may contain
cytoskeleton fragments (see Fig. 5.6B), distinguished from the vesicles shed from the normal
RBC and HS RBCs which are depleted from the cytoskeleton components.
120
Figure 5.6. Vesiculation in the membrane with protein defects in the horizontal interactions. (A)
When the compression ratio Rcompression = 5% and horizontal connectivity Chorizontal = 20%, only
one bud is formed from the membrane. (B) When Rcompression = 10% and Chorizontal = 40%, one
vesicle with cytoskeleton fragment inside is created (C) Summary of dependence of membrane
vesiculation on horizontal connectivity Cvertical and compression ratio Rcompression. ① indicates
one vesicle is observed. ② indicates two vesicles are observed. In the figure, × indicates no
vesiculation occurs.
5.4 Summary
We applied a two-component CGMD RBC membrane model to simulate vesicle formation in the
normal RBC membrane and in the HS and HE RCB membrane. We assumed vesiculation
induced by differences in the spontaneous curvatures of membrane domains and vesiculation
induced by compression on the lipid bilayer. Our results show that the difference in the
121
spontaneous curvatures of the RBC membrane domains induce nanovesiculation or the formation
of small-sized microvesicles that consist of same type of particles. The compression on the
membrane caused the generation of cytoskeleton corral-sized vesicles with heterogeneous
composition. When both effects were considered, the compression on the membrane tends to
facilitate the formation of vesicles that were made of the same type of particles. Furthermore, we
modeled the vesiculation in HS and HE RBC membrane by reducing the vertical and horizontal
connectivities in our membrane model. We found that the sizes of the vesicles released from the
HS and HE cell membrane are more diverse than the sizes of the vesicles released from the
normal RBC membrane. This is due to the higher constrains induced by the cytoskeleton on the
lipid bilayer in the normal RBC membrane compared to the defective RBC membranes. When
the vertical or horizontal connectivity is high, multiple vesicles with sizes similar to the
cytoskeleton corral size are created from the membrane under large compression ratios. In
contrast, membrane with low vertical or horizontal connectivity tends to produce larger vesicles.
Moreover, we found that under the same compression ratio, membrane with protein defects in
the horizontal interactions are more likely to lose vesicles than in membranes with defects in the
vertical interactions. This means that the cytoskeleton plays a major role in maintaining the
stability and integrity of the RBCs. In addition, vesicles released from HE RBC membrane may
contain fragments of cytoskeleton while the vesicles shed from HS RBC membrane are devoid of
cytoskeleton components as the cytoskeleton is still intact.
122
Chapter 6.
Modeling Sickle Hemoglobin Fibers as Four Chain of Coarse-
Grained Particles
Abstract
The intracellular polymerization of deoxy-sickle cell hemoglobin (HbS) has been identified as
the main cause of sickle cell disease. Therefore, the material properties and biomechanical
behavior of polymerized HbS fibers is a topic of intense research interest. A solvent-free coarse-
grain molecular dynamics (CGMD) model is developed to represent a single hemoglobin fiber as
four tightly bonded chains. A finitely extensible nonlinear elastic (FENE) potential, a bending
potential, a torsional potential, a truncated Lennard-Jones potential and a Lennard-Jones
potential are implemented along with the Langevin thermostat to simulate the behavior of a
polymerized HbS fiber in the cytoplasm. The parameters of the potentials are identified via
comparison of the simulation results to the experimentally measured values of bending and
torsional rigidity of single HbS fibers. After it is shown that the proposed model is able to very
efficiently simulate the mechanical behavior of single HbS fibers, it is employed in the study of
the interaction between HbS fibers. It is illustrated that frustrated fibers and fibers under
compression require a much larger interaction force to zipper than free fibers resulting to partial
unzippering of
these fibers. Continuous polymerization of
the unzippered fibers via
heterogeneous nucleation and additional unzippering under compression can explain the
formation of HbS fiber networks and consequently the wide variety of shapes of deoxygenated
sickle cells.
123
6.1. Introduction
A red blood cell (RBC) contains 25-30% hemoglobin whose main function is to carry oxygen
from the lungs to tissues. Normal RBCs contain hemoglobin A (HbA) that has 2 subunits
denoted α and 2 denoted β. However, RBCs of people suffering from sickle cell disease contain
HbS in which a charged surface group glu at β6 is replaced by a hydrophobic group val (27).
This replacement promotes polymerization of deoxygenated hemoglobin at high enough
concentrations resulting to abnormal sickle-shaped RBCs (see Fig. 6.1A) which are less
compliant and more adherent than normal RBCs. Because of increased stiffness and cell
adherence to the endothelium, the circulation of sickle cells through the body's narrow blood
vessels, such as arterioles, venules, and capillaries, is often obstructed resulting in infarctions and
organ damage (23, 50, 279). In addition, overt stroke caused by occlusion of large cerebral
arteries is one of the main complications of sickle-cell disease (62, 280-282). Since polymerized
HbS fibers are stiff, they create protrusions and cause vesiculation of the RBC membrane (35).
The increased stiffness of HbS fibers is considered to be the main reason for the wide variety of
shapes that deoxygenated RBCs from patients with SCD acquire (27, 35, 283). Since HbS fibers
play a very important role in causing sickle cell disease, the material properties and thermal
behavior of HbS during the polymerization process have been widely studied for an extended
period of time.
124
Figure 6.1. (A) A schematic of HbS fibers in a sickle cell (B) A single HbS fiber comprises 7
pairs of double strands of hemoglobin tetramers. The pairs are shown as same-color agents. It is
drawn after (27). (C) Top view of the HbS fiber model. (D) Side view of the HbS fiber model.
Electron microscopy has revealed that a single HbS fiber consists of 7 double strands in a
hexagonally shaped cross-section twisted about a common axis in a rope-like fashion retaining
their atomic contacts (42, 43, 284-286). Single HbS fibers have an average radius of
approximately 11 nm and a mean helical path of approximately 270 nm, defined as the length
over which the fiber twists through 180(cid:2925) (39, 284, 285, 287). Fig. 6.1B, drawn after Fig. 1 in
(27), illustrates the structure of a single HbS fiber. In vivo, HbS fibers are formed in an unusual
fashion by two types of nucleation processes. Homogeneous nucleation entails the nucleation of
a new fiber from an aggregate of monomers. In heterogeneous nucleation, the surface of a
formed polymer assists the nucleation of a new polymer (46, 288). Because the homogeneous
nucleation is very slow, there are only a few homogeneous nuclei in the deoxygenated state
125
leading to a very low number of polymer domains via the dominant process of heterogeneous
nucleation, which generates aligned polymers. In fact, most red blood cells gel in a single
polymer domain (49, 289). Another important observation is that the red blood cell membrane
considerably enhances the nucleation by a factor of 6 (50), and that HbS is associated with the
red blood cell membrane and specifically with the cytoplasmic tail of the band-3 protein.
The properties of the HbS fiber have been extensively studied experimentally in the past two
decades. Bending rigidities and persistence lengths of a group of HbS fibers with different
thickness were measured by differential interference contrast (DIC) microscopy in (290). All the
tested fibers were approximately 10 μm long while the variation in the thickness of HbS fibers
was due to the bundling of single fibers. The persistence length of a single HbS fiber was found
to be (cid:1864)(cid:3043) = 0.24 mm and the associated bending rigidity was measured (cid:2018) = 8.3 × 10(cid:2879)(cid:2870)(cid:2873) N m(cid:2870). In
another approach, the bending and torsional rigidities of approximately 1 μm long single HbS
fibers were obtained by cryo-electron microscopy (291, 292). The bending of each fiber was
quantified by measuring the normal deviation of its middle point from the straight line joining its
ends. In these experiments, frozen hydrated HbS fibers were used. The measurements showed
that the bending rigidity of a single HbS fiber is approximately (cid:2018) = 5.2 × 10(cid:2879)(cid:2870)(cid:2873) N m(cid:2870), which is
consistent with the value measured by (290), while the torsional rigidity is approximately
6 × 10(cid:2879)(cid:2870)(cid:2875) J m (292). While in homogeneous materials the bending and torsional rigidities are on
the same order of magnitude, in single HbS fibers the bending rigidity is two orders larger than
the torsional rigidity meaning that the material is anisotropic and the axial shear response is
much softer than the extension (292). It is likely that the difference between the various types of
contacts of the double strands is the source of the anisotropy. HbS fibers are among the stiffest
126
filaments in a mammalian cell. For comparison, the persistence length of microtubules is
approximately 5 mm, of intermediate filaments it is in the range of 200 nm to 1.3 μm, of F-actin
it is approximately 10 μm, and of genomic DNA it is 53 ± 2 nm (126). Amyloid fibrils, which
may be involved in chronic neurodegenerative diseases such as Alzheimer's disease, have
persistence lengths that vary approximately from 5 μm to 0.3 mm, depending on their structure
(293).
The interactions between sickle hemoglobin fibers were studied in (294). The authors found that
HbS gels and fibers were fragile and easily broken by mechanical perturbation. They also
observed different types of fiber cross-links in gel and the process of fibers zippering. In (44), the
focus of the studies was on the interactions between two HbS fibers. They estimated that the
attractive binding energy between two zippered HbS fibers is (cid:1873) = 7.2 (cid:1863)(cid:3003)(cid:1846) μm(cid:2879)(cid:2869) with a
corresponding characteristic length (cid:1864)(cid:3030) = (cid:1863)(cid:3003)(cid:1846) (cid:1873)⁄ ≅ 140 nm, which is significantly smaller than
the persistence length of a single fiber. The fact that (cid:1864)(cid:3030) ≪ (cid:1864)(cid:3043) means that single HbS fibers are
very unlikely to spontaneously separate because of thermal vibrations. However, this is true for
non-stressed quiescent fibers. It will be shown that HbS fibers under bending can separate
without additional external force other than the ones that cause bending.
So far, most of the studies on the HbS fibers were based on experimental observations. There is
only limited molecular dynamics literature on HbS, which investigated the contact points of the
double strands for the formation of the HbS fiber and in general the crystal structure of HbS (295,
296). Here, a coarse-grain (CG) model of a single HbS fiber is introduced. The parameterization
scheme employed fits the CG potential to the experimentally measured material properties of
127
single fibers, namely the bending and torsional rigidities (290, 292). It ignores the detailed
structure of the fibers and the dynamics of the polymerization in order to reach a very large time
scale, which is necessary in the study of the mechanical behavior of the fibers. A similar
approach has been successfully used in the representation of the spectrin cytoskeleton of
erythrocyte membrane (109), and in the study of cross-linked actin-like networks (297, 298).
Essentially, the proposed model follows a top-down approach and the parameters are validated
on the basis of experimental measurements (299). The approach is much simpler than multiscale
coarse-graining strategies where molecular dynamics simulations considering atomistic details
are performed and the CG force field is parametrized to match the atomistic forces and to predict
the thermodynamic behavior (299-303) and references therein. Finally, the interactions between
two HbS fibers and the zippering process are investigated.
6.2. Model and method
The single hemoglobin fiber studied here is modeled as four tightly bonded chains, each of
which is composed of 100 soft particles. The cross-section of the HbS fiber model is shown in
Fig. 6.1C. The distance between the centers of neighboring particles is 10 nm. The total length
of the simulated HbS fiber is 1 μm and the radius of the fiber is approximately 10 nm (see Fig.
6.1D), which is consistent with the geometry of single hemoglobin fibers described in
experimental studies (292). The displacements of all particles are governed by the Langevin
equation (304)
(cid:1865)(cid:3036)
(cid:3031)(cid:3118)(cid:2200)(cid:3284)
(cid:3031)(cid:3047)(cid:3118) = (cid:2162)(cid:3036) − (cid:1858) (cid:3031)(cid:2200)(cid:3284)
(cid:3031)(cid:3047)
+ (cid:2162)(cid:2191)
(cid:3003),
(6.1)
128
where (cid:1865)(cid:3036) is the mass of the ith particle, (cid:1858) is the friction coefficient and it is identified to be
20 (cid:1865)(cid:3036) (cid:1872)(cid:3046)⁄ , where (cid:1872)(cid:3046) is the time unit. (cid:2200)(cid:3036) is the position vector of the ith particle, and (cid:1872) is time.
(cid:2162)(cid:3036) = − (cid:2034)(cid:1847) (cid:2034)(cid:2200)(cid:3036)
⁄
is the deterministic force produced by the implemented total potential (cid:1847), (cid:2162)(cid:3036)
(cid:3003) is
Gaussian white noise that obeys the fluctuation-dissipation theorem (98, 305)
〈(cid:2162)(cid:3036)
(cid:3003)〉 = 0 ,
〈(cid:2162)(cid:3036)
(cid:3003)(cid:2162)(cid:3037)
(cid:3003) 〉 = (cid:2870)(cid:3038)(cid:3251)(cid:3021)(cid:3033)(cid:3083)(cid:3284)(cid:3285)
∆(cid:3047)
,
(6.2)
(6.3)
where (cid:1863)(cid:3003) is the Boltzmann's constant, (cid:1846) is the absolute environmental temperature, (cid:2012)(cid:3036)(cid:3037) is the
Kronecker delta, and (cid:1986)(cid:1872) is the time-step. The friction coefficient (cid:1858) is selected to maintain the
environmental temperature at 300 K . The Langevin thermostat is similar to Berendsen's
thermostat but it introduces a damping effect that slows down the dynamics of the system (299,
306).
In the simulations, the energy unit is (cid:2013) (cid:1863)(cid:3003)(cid:1846)⁄
= 1, and the length unit is (cid:2026) = 2(cid:2879)(cid:2869)/(cid:2874) (cid:1870)(cid:2868), where
(cid:1870)(cid:2868) = 10 nm is the diameter of the particles. Since each hemoglobin tetramer has a molecular
weight of approximately 68,000 Da, we assign to each coarse-grained particle a mass of about
(cid:1865)(cid:3036) ≅ 3.95 × 10(cid:2879)(cid:2870) kg. The time scale (cid:1872)(cid:3046) = ((cid:1865)(cid:3036)(cid:2026)(cid:2870) (cid:2013)⁄ )(cid:2869) (cid:2870)⁄ ≅ 2.7 × 10(cid:2879)(cid:2877) s of the simulation is
set by the motion due to the deterministic force (130). The time step for the numerical solution of
the Langevin equation is chosen to be (cid:1986)(cid:1872) = 0.001 (cid:1872)(cid:3046). However, the time scale (cid:1872)(cid:3046) in CGMD does
not correspond to a real time since the particles do not represent real atoms. Only via comparison
with a physical process, the correspondence between the simulation time and the "real" time can
be established.
129
Figure 6.2. The finitely extendable nonlinear elastic (FENE) potential, the corresponding
harmonic potential, the Lennard-Jones potential and the truncated Lennard-Jones potential
employed in the HbS fiber model are shown.
A finitely extendable nonlinear elastic (FENE) potential (see Fig. 6.2) is introduced between
adjacent particles that belong to the same chain (e.g., D and E in Fig. 6.1D) and between
particles that belong to different chains that are initially in the same cross-sectional plane (A-B,
A-C, A-D, B-C, B-D and C-D in Fig. 6.1C). The FENE potential is described by the equation
(cid:1847)(cid:3046) = − (cid:2869)
(cid:2870)
(cid:1837)(cid:3046) (cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) ln [1 − ((cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)
∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
)(cid:2870)],
(6.4a)
where (cid:1837)(cid:3046) is a parameter related to the constant (cid:1837)(cid:3046)(cid:3043) = (cid:1837)(cid:3046)/(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) of the harmonic potential that
has the same stiffness with the above potential at equilibrium, (cid:1870)(cid:3036)(cid:3037) is the distance between
particles (cid:1861) and (cid:1862), and (cid:1870)(cid:2868) is the equilibrium distance between the particles. The FENE potential is
130
harmonic at its minimum but the bonds cannot be stretched more than ∆(cid:1870)(cid:3040)(cid:3028)(cid:3051). The force is given
by the expression
(cid:1832) = −(cid:1837)(cid:3046)
⁄
(cid:3435)(cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)(cid:3439) ∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
(cid:2869)(cid:2879)(cid:3427)(cid:3435)(cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)(cid:3439) ∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
⁄
(cid:3118) .
(cid:3431)
(6.4b)
In these simulations, the maximum extension allowed between two particles is set to be (cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) =
0.3 (cid:1870)(cid:2868). The main role of the FENE potential is to bond the four chains, maintain the geometry of
the model, and ensure that interacting fibers cannot move through one another. Since the
proposed model ignores the detailed structure of the HbS fibers, only one spring constant is
considered. The value of (cid:1837)(cid:3046) is determined based on the following argument: the Young's
modulus of an elastic beam is related to its bending rigidity via the expression (cid:2018) = (cid:1831)(cid:1835), where (cid:1831)
is the Young's modulus of the beam and (cid:1835) is the moment of inertia about the axis along the beam.
However, in particle models, the spring potential does not generate the expected bending rigidity
because of the free rotation of the neighboring points. Nevertheless, the generated overall
Young's modulus has to be consistent with the bending rigidity. In the appendix A, we show that
the value of (cid:1837)(cid:3046) that generates the Young's modulus (cid:1831) which is consistent with the bending
rigidity of the HbS fiber is approximately (cid:1837)(cid:3046) = (cid:1837)(cid:3046)(cid:3043)(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) = (cid:2024) (cid:1831)(cid:1844)(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) 4⁄ = 3600 (cid:2013) (cid:1870)(cid:2868)⁄ .
A truncated Lennard-Jones (L-J) potential (see Fig. 6.2) is applied between two particles in the
same fiber, but in different cross-sectional planes (A-E, A-F, B-E and B-F in Fig. 6.1D), to
provide repulsive force when the two particles are in close range and thus prevent particle
overlap. The expression of the truncated L-J potential is given by
131
(cid:1847)(cid:3013)(cid:3011)(cid:3435)(cid:1870)(cid:3036)(cid:3037)(cid:3439) = (cid:3422)
4(cid:2013)(cid:4593) (cid:4680)(cid:3436) (cid:3097)
(cid:3045)(cid:3284)(cid:3285)
(cid:2869)(cid:2870)
(cid:3440)
(cid:2874)
(cid:3440)
− (cid:3436) (cid:3097)
(cid:3045)(cid:3284)(cid:3285)
(cid:3047)(cid:3045)
0 (cid:1870)(cid:3036)(cid:3037) ≥ (cid:1870)(cid:3030)(cid:3048)(cid:3047)
(cid:4681) + (cid:2013)(cid:4593) (cid:1870)(cid:3036)(cid:3037) < (cid:1870)(cid:3030)(cid:3048)(cid:3047)
(cid:3047)(cid:3045) = 2(cid:2869) (cid:2874)⁄ (cid:2026)
(6.5)
where (cid:1870)(cid:3036)(cid:3037) is the distance between particles (cid:1861) and (cid:1862). The cut-off distance (cid:1870)(cid:3030)(cid:3048)(cid:3047)
(cid:3047)(cid:3045) for truncated L-J
potential is chosen to be (cid:1870)(cid:2868). In order to limit the number of the independent model parameters,
we determined (cid:2013)(cid:4593) = (cid:1837)(cid:3046)(cid:2026) (36 2(cid:2870)/(cid:2871))
⁄
by requiring that the truncated L-J potential has the same
(cid:3047)(cid:3045) .
curvature as the FENE potential at (cid:1870)(cid:3030)(cid:3048)(cid:3047)
Figure 6.3. (A) Bending and (B) twist of the HbS fiber model.
A bending FENE potential (cid:1847)(cid:3029) is employed to describe the bending rigidity of the HbS fiber
(cid:1847)(cid:3029) = − (cid:2869)
(cid:2870)
(cid:1837)(cid:3029)(cid:1986)(cid:2016)(cid:3040)(cid:3028)(cid:3051) ln [1 − ( (cid:3087)(cid:2879)(cid:3087)(cid:3116)
∆(cid:3087)(cid:3288)(cid:3276)(cid:3299)
)(cid:2870)],
(6.6)
132
where (cid:1837)(cid:3029) is the parameter that directly regulates the bending stiffness of the hemoglobin fiber,
and (cid:2016) is the angle formed by three consecutive particles in the same chain, as illustrated in Fig.
6.3A. (cid:2016)(cid:2868) is the equilibrium angle and it is chosen to be (cid:2024), meaning that the three consecutive
particles are initially located in-line. ∆(cid:2016)(cid:3040)(cid:3028)(cid:3051) is the maximum allowed bending angle between two
particles and is set to be 0.3 (cid:2016)(cid:2868). (cid:1837)(cid:3029) is selected via a trial and error process in order to produce a
bending rigidity identical to the experimental value and it is determined to be (cid:1837)(cid:3029) = 5 × 10(cid:2871)(cid:2013).
The torsional rigidity of the proposed model is introduced via a torsional FENE potential
between particles belonging to consecutive cross-sectional planes defined by four particles. The
potential is expressed as
(cid:1847)(cid:3047) = − (cid:2869)
(cid:2870)
(cid:1837)(cid:3047)∆(cid:2004)(cid:3040)(cid:3028)(cid:3051) ln [1 − ( (cid:3075)
∆(cid:3075)(cid:3288)(cid:3276)(cid:3299)
)(cid:2870)],
(6.7)
where (cid:1837)(cid:3047) is the parameter that regulates the torsional stiffness of the HbS fiber, and (cid:2004) is the
angle between two directional vectors defined in two consecutive cross-sectional planes, as it is
illustrated in Fig. 6.3B. ∆(cid:2004)(cid:2923)(cid:2911)(cid:2934) is the maximum allowed twisting angle between two consecutive
cross-sectional planes and is set to be ≅ 0.3 (cid:2024) . (cid:1837)(cid:3047) is tuned to match the model against
experimentally obtained values of the torsional rigidity of single hemoglobin fibers (44) and it is
chosen to be (cid:1837)(cid:3047) = 500(cid:2013).
The L-J potential, as plotted in Fig. 6.2, is employed between particles that belong to different
fibers to produce attractive interfiber forces when two fibers are close. The expression of L-J
potential is given by
133
(cid:1847)(cid:3013)(cid:3011)(cid:3435)(cid:1870)(cid:3036)(cid:3037)(cid:3439) = (cid:3422)
k ∗ 4(cid:2013) (cid:4680)(cid:3436) (cid:2978)
(cid:3045)(cid:3284)(cid:3285)
(cid:2869)(cid:2870)
(cid:3440)
(cid:2874)
(cid:3440)
− (cid:3436) (cid:2978)
(cid:3045)(cid:3284)(cid:3285)
(cid:4681) + k(cid:2013) (cid:1870)(cid:3036)(cid:3037) < (cid:1870)(cid:3030)(cid:3048)(cid:3047)
,
(6.8)
0 (cid:1870)(cid:3036)(cid:3037) ≥ (cid:1870)(cid:3030)(cid:3048)(cid:3047)
where (cid:1870)(cid:3030)(cid:3048)(cid:3047) = 2.5(cid:2026) and k is a parameter used to adjust the interfiber forces between two fibers.
The value of k that is used for the plot in Fig. 6. 2 is 500.
As it was mentioned earlier, the spring constant (cid:1837)(cid:3046)(cid:3043) is related to the Young's modulus of the
fiber through the expression (cid:1837)(cid:3046)(cid:3043) = (cid:2024) (cid:1831) (cid:1844)/4. Βy approximating the fiber as a circular cylindrical
rod of radius (cid:1844) and cross-sectional moment of inertia (cid:1835) = (cid:2024)(cid:1844)(cid:2872)/4, the relationship between the
bending rigidity and the spring constant can be established as (cid:1837)(cid:3046)(cid:3043) = (cid:2018)/(cid:1844)(cid:2871). Then, a relevant
time-scale for the vibration of the HbS fibers can be estimated as (cid:1872)(cid:3030) = 2(cid:2024) (cid:3493)(cid:1865) (cid:1837)(cid:3046)(cid:3043)⁄
=
2(cid:2024)(cid:3493)(cid:1865)(cid:1844)(cid:2871) (cid:2018)⁄ ≅ 176 ps, which is approximately 15 times smaller than (cid:1872)(cid:3046) and this justified why
the time step (cid:1986)(cid:1872) is 10 times smaller than the value usually employed in MD simulations.
Because the Langevin equation acts as a thermostat, the noise and friction are designed to
maintain a given temperature. Deviations from that temperature are corrected with a decay time
of (cid:1872)(cid:3031) = (cid:1865)(cid:3036)/(cid:1858). By choosing (cid:1858) = 20 (cid:1865)(cid:3036)/(cid:1872)(cid:3046), the decay time becomes (cid:1872)(cid:3031) = (cid:1872)(cid:3046) 20⁄ = 5 (cid:1986)(cid:1872). This
means that the system corrects the temperature variations in about 5 time steps. Another typical
relevant time for the fluctuations of the HbS fibers is the duration of such fluctuations. This time
can be estimated as (cid:1872)(cid:3033) = (cid:1858)/(cid:1837)(cid:3046)(cid:3043) (307). By using that (cid:1858) = 3(cid:2024)(cid:2015) (cid:2868) , where (cid:2015) = 8.9 × 10(cid:2879)(cid:2872) Pa s
approximates the dynamic viscosity of the solvent, and the relation between the single spring
134
constant and the overall bending rigidity of the fiber, the relaxation time can be estimated as
(cid:1872)(cid:3033) = 3(cid:2024)(cid:2015)(cid:1870)(cid:2868)
(cid:2872) (cid:2018)⁄ ≅ 162 ps. The fact that (cid:1872)(cid:3030) and (cid:1872)(cid:3033) have a ratio almost one means that the motion
of the fibers is overdamped.
For the integration of Eq. (6.1), a modified version of the leapfrog algorithm is used:
(cid:2204)(cid:2191)((cid:1872) + (cid:1986)(cid:1872) 2⁄ ) = (cid:2204)(cid:3036)((cid:1872)) + (cid:1986)(cid:1872) 2⁄ (cid:2183)(cid:3036)((cid:1872)),
6.9(a)
(cid:2200)(cid:2191)((cid:1872) + (cid:1986)(cid:1872)) = (cid:2200)(cid:3036)((cid:1872)) + (cid:1986)(cid:1872) (cid:2204)(cid:2191)((cid:1872) + (cid:1986)(cid:1872) 2⁄ ),
(cid:2204)(cid:3557)(cid:2191)((cid:1872) + (cid:1986)(cid:1872)) = (cid:2204)(cid:3036)((cid:1872) + (cid:1986)(cid:1872) 2⁄ ) + (cid:1986)(cid:1872) 2⁄ (cid:2183)(cid:3036)((cid:1872)),
(cid:2183)(cid:2191)((cid:1872) + (cid:1986)(cid:1872)) = (cid:2183)(cid:2191)(cid:3435)(cid:2200)(cid:2191)((cid:1872) + (cid:1986)(cid:1872)), (cid:2204)(cid:3557)(cid:2191)((cid:1872) + (cid:1986)(cid:1872))(cid:3439),
(cid:2204)(cid:2191)((cid:1872) + (cid:1986)(cid:1872)) = (cid:2204)(cid:2191)((cid:1872) + (cid:1986)(cid:1872) 2⁄ ) + (cid:1986)(cid:1872) 2⁄ (cid:2183)(cid:3036)((cid:1872) + (cid:1986)(cid:1872)).
6.9(b)
6.9(c)
6.9(d)
6.9(e)
Because the total force applied on a particle depends on the velocity of the particle, a prediction
is made for the new velocity, which is denoted as (cid:2204)(cid:3557), and it is corrected in the last step. A similar
approach was employed for the modification of the velocity-Verlet algorithm used in (308).
6. 3. Results and discussion
6. 3.1 Measurements of material properties of HbS fiber
135
We show that the proposed model is able to simulate a single HbS fiber with the appropriate
mechanical properties. In particular, by tuning the parameters (cid:1837)(cid:3029), and (cid:1837)(cid:3047), the model reproduces
the experimental values for the bending rigidity (cid:2018), and the torsional rigidity (cid:1855), of a HbS fiber
(292). During the simulations, one end of the HbS fiber is fixed and it is used as the reference
point for the central axis of the fiber.
6. 3.1.1 Bending rigidity
Thermally driven fluctuations of semi-flexible fibers can be used to obtain the bending moduli of
the fibers (126). One method is to measure the mean-squared amplitude of each dynamical mode
of vibrations and by using the principle of equipartition of energy to extract the bending rigidity
for each mode independently (309). However, because HbS fibers are stiff with a large
persistence length of approximately 120 μm and the simulated fibers have a total length of 1μm,
they bend a little resulting to a large uncertainty when the motion is decomposed into Fourier
modes. Here, a method introduced by (290) is followed. The bending rigidity of the HbS fiber is
calculated from the equation
(cid:2018) = (cid:1863)(cid:3003)(cid:1846)(cid:1864)(cid:3043) =
(cid:3038)(cid:3251)(cid:3021)(cid:3013)(cid:3119)
(cid:2872)(cid:2876)〈((cid:3048)(cid:3299)((cid:3013) (cid:2870)⁄ ))(cid:3118)〉 ,
(6.10)
where (cid:1864)(cid:3043) is the persistence length of the HbS fiber, (cid:1838) is the length of the fiber, and (cid:1873)(cid:3051)((cid:1838) 2⁄ ) is
the spatial displacement of the projected fiber midpoint from the central axis. The central axis,
from which the displacement of the fiber's middle point is measured, is defined as a line that is
perpendicular to the cross-section of the fiber at the fixed point and it passes through the center
136
of the cross-sectional area (dashed line in Fig. 6.4). Once the bending rigidity is obtained, the
persistence length can be easily calculated by the expression
(cid:1864)(cid:3043) = (cid:2018) (cid:1863)(cid:3003)(cid:1846)⁄
.
(6.11)
The measured bending rigidity of HbS fiber with respect to time is plotted in Fig. 6.5. The Initial
fiber configuration at 0°K was a straight line. The free end of the HbS fiber model first began to
fluctuate and then the kinetic energy was transferred in the fiber from the free end towards the
fixed end. The system reached equilibrium at approximately (cid:1872) ≅ 10(cid:2872)(cid:1872)(cid:3046) (see Fig. 6.5). The
measured bending rigidity is close to the experimentally measured value (cid:2018) = 5.2 × 10(cid:2879)(cid:2870)(cid:2873)Nm(cid:2870)
and it corresponds to a persistence length of approximately (cid:1864)(cid:3043) ≅ 121 μm (292). Because of the
large persistence length, the numerical results show that a single HbS fiber subjected to thermal
forces behaves similarly to a stiff rod as it fluctuates about its central axis (see Fig. 6.4).
Figure 6.4. Characteristic configuration of the HbS fiber model.
137
Figure 6.5. Variation with time of the bending rigidity of the HbS fiber model.
The probability density of the deviations at the middle point of the HbS fiber can be
approximated by a normalized Gaussian distribution (solid curve in Fig. 6.6) given by
(cid:1842)((cid:2012)(cid:1873)) =
(cid:2869)
(cid:3493)(cid:2870)(cid:3095)〈(cid:3083)(cid:3048)(cid:3118) 〉
(cid:1857)((cid:2879) (cid:3331)(cid:3296)(cid:3118)
(cid:3118)〈(cid:3331)(cid:3296)(cid:3118) 〉
) ,
(6.12)
where the (cid:2012)(cid:1873) is the HbS fiber midpoint displacement and 〈(cid:2012)(cid:1873)(cid:2870)〉(cid:2869) (cid:2870)⁄ ≅ 1.4 (cid:2026). The result is in
agreement with the theoretical prediction that the probability density of the deviations of a semi-
flexible rod follow the Gaussian distribution in the case of thermal fluctuations (310).
138
Figure 6.6. The distribution of HbS fiber midpoint displacements and the associated normalized
Gaussian probability distribution.
6. 3.1.2 Torsional rigidity
As for the torsional rigidity, it has been shown (290) that
〈∆(cid:2019)(cid:2870)〉 = (cid:3038)(cid:3251)(cid:3021)(cid:3090)(cid:3119)
(cid:3030)(cid:3095)(cid:3118) [1 − (cid:3013)
(cid:3090)
(1 − (cid:1857)(cid:2879)(cid:3338)
(cid:3261))] ,
(6.13)
where ∆(cid:2019) = (cid:2019)∆(cid:2016)/(cid:2024), and (cid:2019) = 135 nm is half of the average pitch length for the HbS fiber, (cid:1986)(cid:2016) is
the twisted angle in half pitch length, and (cid:1838) is the length of the HbS fiber. By substituting (cid:1986)(cid:2019) into
Eq. (6.13), we obtain (cid:1855) as
(cid:1855) = (cid:3038)(cid:3251)(cid:3021)(cid:3090)
〈∆(cid:3087)(cid:3118)〉 [1 − (cid:3013)
(cid:3090)
(1 − (cid:1857)(cid:2879)(cid:3338)
(cid:3261))] .
(6.14)
139
Fig. 6.7 shows that the selected value (cid:1837)(cid:3047) = 500 (cid:2013) results in a torsional rigidity of the hemoglobin
fiber model (cid:1855) ≅ 6.5 × 10(cid:2879)(cid:2870)(cid:2875)J m, which is very close to the experimentally measured value of
(cid:1855) = 6 × 10(cid:2879)(cid:2870)(cid:2875)J m (292).
Figure 6.7. Variation with time of the torsional rigidity of the HbS fiber model.
6. 3.1.3 Effect of the bending and torsional potentials on the HbS fiber model
Finally, we ensure that the contribution to bending and torsional rigidity results mainly from the
special potentials (cid:1847)(cid:3029) and (cid:1847)(cid:3047) and not from the FENE spring potential between the particles. When
only the spring and the truncated L-J potentials are implemented, the shape of the HbS fiber is
sinusoidal and strongly twisted (see Fig. 6.8). Also, it is found that the amplitude of the HbS
fiber fluctuations is small and that the fluctuations are limited only about the central axis
(marked with dashed line in the Fig. 6.8) of the fiber and no bulk motion is observed.
140
Figure 6.8. Example of the behavior of the HbS fiber model without applying the bending and
the torsional potentials.
According to (290), Eq. (6.10) is derived for semi-flexible fibers which are relatively stiff. Since
the fiber model without the special bending and torsional potentials is highly flexible, the Eq.
(6.10) cannot be employed in the calculation of the bending rigidity of this fiber. Instead, we use
the equation
〈(cid:2200)(cid:3032)(cid:3032)
(cid:2779) 〉 = 2(cid:1864)(cid:3043)(cid:1838) − 2(cid:1864)(cid:3043)
(cid:2870)[1 − (cid:1857)(cid:1876)(cid:1868) (cid:3436)− (cid:3013)
(cid:3039)(cid:3291)
(cid:3440)],
(6.15)
to obtain the persistence length (cid:1864)(cid:3043) and then to calculate the bending rigidity from the expression
(11) (126). (cid:2200)(cid:2187)(cid:2187) is the end-to-end displacement vector, and (cid:1838) is the length of the HbS fiber. Based
on the Eqs. (6.14), and (6.15), the values of bending rigidity and torsional rigidity of HbS fiber
model without the bending (cid:1847)(cid:3029) and torsional (cid:1847)(cid:3047) potentials are shown in Fig. 6.9A and Fig. 6.9B
respectively.
141
Figure 6.9. Variation with time of (A) the bending rigidity and (B) the torsional rigidity of the
HbS fiber model without applying the bending and the torsional potentials.
The measured bending rigidity is approximately 0.21 × 10(cid:2879)(cid:2870)(cid:2875) N m(cid:2870), which is about 1/2000 of
the actual value of the HbS fiber ( (cid:2018) = 5.2 × 10(cid:2879)(cid:2870)(cid:2873)N m(cid:2870) (292)). The torsional rigidity is
approximately 6 × 10(cid:2879)(cid:2870)(cid:2877)J m , which
is 1/100 of
the experimental value ( c = 6 ×
10(cid:2879)(cid:2870)(cid:2875)J m (292)). The results above indicate that the spring potential has little effect on the
model's bending and torsion rigidity. It is noted that the proposed HbS fiber model is strongly
anisotropic since the experimentally measured torsional rigidity is significantly less than the
142
bending rigidity, while for isotropic materials these properties are on the same order of
magnitude.
6.3.2 Modeling the zippering of two HbS fibers
Individual HbS fibers interact with each other to form various X-shaped junctions, Y-shaped
branches, and side-to-side coalescence ("zippering") cross-links (294, 311). We apply the
developed model to study the formation of bundles by zippering fibers. Two cases of HbS fibers
zippering are simulated. In the first case, two fibers are initially parallel to each other and then
come into contact as shown in Fig. 6.10. In the second case, two fibers in a Y-shaped cross-link
zippered from their contacting tips (see Fig. 6.13). The interfiber force is represented by a L-J
potential applied between particles belonging to different fibers.
Figure 6.10. (A) Initial states of two parallel HbS fibers before zippering. (B) Intermediate state
of two initially parallel HbS fibers zippering. (C) Equilibrium configuration of two initially
parallel HbS fibers. The value of the L-J parameter k is 200.
The minimum value of the parameter k, which adjusts the interaction between the fibers, is
determined by assuming that the interfiber forces have the same origin as the forces between the
particles comprised by the fiber. In this case, the FENE potential between the particles that form
the fiber and the L-J potential must have the same curvature at the equilibrium, meaning that
143
k(cid:2013) = (cid:1837)(cid:3046)(cid:2026) (36 2(cid:2870)/(cid:2871))
⁄
and since (cid:1837)(cid:3046) = 3600 (cid:2013) (cid:1870)(cid:2868)⁄ we obtain that k ≅ 60. According to (44), there
are two main contributions to the attraction energy between zippering fibers. One is due to Van
der Waals forces and the second is due to depletion forces. The depletion energy was computed
to be approximately two times the Van der Waals energy. Based on the previous estimations, we
explored the interaction between two fibers for the L-J parameter k varying from a very small
value of k = 5 to a very large value of k = 500.
144
Figure 6.11. (A) Position (cid:1876) and (B) speed of the fiber junction (cid:1874) during the zippering process of
two parallel HbS fibers.
In the first case, two fibers AB and CD are initially placed in parallel arrangement as seen in Fig.
6.10A with an interaction parameter k = 200, which ensures zippering. At the beginning of the
simulation, they fluctuate under brownian forces. Fig. 6.10B shows that the tip A approaches and
subsequently touches the body of the fiber CD. Then, the two fibers begin to progressively
merge from the contact point to form a thicker fiber in Fig. 6.10C. The non-symmetrical
behavior of the two ends A and D is due to random fluctuations. One of the two ends, in this case
point A, approaches the neighboring fiber closer than (cid:1870)(cid:3030)(cid:3048)(cid:3047) resulting to a broken symmetry. As for
the position and speed of the fiber junction during the zippering process, Fig. 6.11A and B show
that the speed of the zippering is relatively fast at the beginning of the zippering process and at a
critical distance gradually decreases to zero when the two fibers reach the equilibrium state.
Figure 6.12. Equilibrium states of two parallel HbS fibers zippered under different interfiber
forces. The L-J parameter k was selected to be (A) k = 100, (B) k = 50, (C) k = 10, (D) k = 5.
The behavior of the fiber pair for different values of parameter k is explored next. Knowing that
in the case of k = 200 the fibers zipper tightly, we decrease the value of k to k = 100. Again the
two fibers finally move like a single fiber (see Fig. 6.12A). When k = 50 , the fibers still "zipper"
145
together but the attraction is not large enough to prevent relative rotation (see Fig. 6.12B). For
k = 10 (see Fig. 6.12C), the two fibers still stay together but relative rotation and instantaneous
dissociation are observed between the two fibers which no longer behave as a single fiber. When
k is finally reduced to k = 5 (see Fig. 6.12D), no stable zippering is achieved and two fibers
separate easily under thermal fluctuations. The results show that because zippered fibers are
stable for k ≥ 100, the stability of a polymerized bundle of fibers is due to both Van der Waals
and depletion forces.
Figure 6.13. (A) Initial state of two HbS fibers before zippering. (B) Intermediate state of the
two HbS fibers during the "Y" shape zippering. (C) Equilibrium configuration of the two HbS
fibers
Next, we study how the strength of interactions drives zippering of HbS fibers in frustrated
structures formed by three fibers. In the first simple case shown in Fig. 6.13, two fibers AB and
CD are arranged to form a "V" shape junction. Points A and C represent the attachments of the
two fibers along a third fiber and they are thus fixed, while their two ends B and D are free to
move. The value of the parameter k, in the expression (8) of the L-J potential between different
fibers is 500. The initial temperature increases and it reaches the final value of (cid:1846) = 1 at the
2 × 10(cid:2872) timestep. The attractive force between the two fibers is turned on after the 4 × 10(cid:2872)
146
timestep to allow the system to reach the equilibrium. Then, the two fibers start zippering from
points B and D. At the initial steps, the junction point quickly moves toward points A and C.
However, as the zippering progresses (see Fig. 6.13B), the speed of the zippering reduces
significantly and finally the system reaches the equilibrium state, as illustrated in Fig. 6.13C.
After zippering, the two fibers move together as a single thicker fiber. The interfiber attractive
forces are balanced by the bending energy stored in the two fibers (44, 294). The position (cid:1876) of
the junction point during zippering is shown in Fig. 6.14A as a function of time and the variation
of the tip-velocity (cid:1874) with time is shown in Fig. 6.14B. Initially, the two end points snap together
and the tip advances very fast but gradually the velocity decreases and zippering ceases at
approximately 55 (cid:2026). Next, the effect of the interfiber forces during the fibers zippering process is
examined. The parameter k of the L-J potential is changed to 100, 200, 300 and 400 and the final
configurations of the zippered fibers are shown in Figs. 6.15A, B, C and D respectively. Fig.
6.15E clearly demonstrates that when fibers are constrained in their relative motion then the
interaction energy has to increase substantially to cause partial zippering.
147
Figure 6.14. (A) Position (cid:1876) and (B) speed of the fiber junction (cid:1874) during the "Y" shape zippering
process.
In a variation of the above numerical experiment, the two fibers are cross-linked to a third fiber
while the attractive interfiber forces between the two inclined fibers and between an inclined
fiber and the third fiber are governed by the same L-J potential (see Fig. 6.16). The values of the
parameter k of the L-J potential are 200, 300, 400 and 500 and the final configurations of the
zippered fibers are shown in Figs. 6.16A, B, C and D respectively. The behavior is similar as in
148
the previous case where point A and C were completely immobilized. It is noted that zippering
does not induce relative sliding of the attaching points between EF and AB, CD fibers. By
comparing Figs. 6.15E and 16E, it is obvious that a higher attractive force is required to generate
zippering in the three-fiber configuration. The reason is that the constrained points are not at the
ends of the fibers resulting to a shorter effective fiber length. The cases shown in Figs. 6.15 and
6.16 demonstrate that frustration in HbS fibers can cause partial zippering and consequently the
polymerization of HbS fibers can deviate from rod-like shapes and branch out to more irregular
configurations resulting to the variety of shapes of deoxygenated sickle cells.
Figure 6.15. Equilibrium states of two HbS fibers zippered as "Y" shapes under different
interfiber forces. The L-J parameter k is selected to be (A) k = 100, (B) k = 200, (C) k = 300, (D)
149
k = 400. (E) The final position of the fiber junction (cid:1876) measured from point D for different
interfiber forces.
Figure 6.16. Equilibrium states of two HbS fibers zippered in a "Y" shape as a result of a third
fiber crossing the other two fibers near to their ends for different interfiber forces. The L-J
parameter k is selected to be (A) k = 200, (B) k = 300, (C) k = 400, (D) k = 500. (E) The final
position of the fiber junction (cid:1876) measured from point D for different interfiber forces.
150
6.3.3 Bending of two zippered fibers
In this part, we show that bending of zippered HbS fibers can cause partial separation.
This is another mechanism that could generate irregular shapes of deoxygenated sickle cells via
polymerization of the partially separated fibers. The two zippered fibers are initially maintained
at equilibrium state and then the two ends of one fiber (C and D) are compressed until their
initial distance is reduced by 10% (see Fig. 6.17). Point B of the fiber AB is forced to move
together with point D while point A is free. At the beginning of the compression, the two fibers
are bent together because the bending force is balanced by the attractive force. When the
maximum attractive force, which is regulated by the parameter k, becomes equal to the bending
force required for fiber AB, the free end A separates from the fiber CD, and the two fibers begin
to unzipper. Figs. 6.17A, B, C and D correspond to k equals 400, 300, 200 and 100 respectively.
Fig.6.17E illustrates the dependence of the degree of separation of the two fibers on the
magnitude of the attractive force between them. In another approach shown in Fig. 6.18, the
parameter k is kept constant (k = 200) while the end points of the fiber CD are gradually
compressed causing buckling. As the compression increases, the separation between the two
fibers increases non-linearly showing that the expected interaction energy cannot restrain bent
fibers from separation. At 20% decrease of the initial distance between the end points C and D,
the two fibers are almost completely separated. As a result, polymerization of initially free fibers
can cause bending and consequently partial separation and generation of irregular fiber networks
and sickle cell shapes.
151
Figure 6.17. Equilibrium states of two zippered HbS fibers under compression for the same final
deformation ratio e = 0.2 and at different interfiber forces. The L-J parameter k is selected to be
(A) k = 400, (B) k = 300, (C) k = 200, (D) k = 100. (E) Separations of the two fibers measured
from point C, at different interfiber forces.
152
Figure 6.18. Equilibrium states of two zippered HbS fibers under compression at different final
deformation ratios and for the same interfiber force (k = 200). The deformation ratio e is
selected to be (A) e = 0.02, (B) e = 0.05, (C) e = 0.15, (D) e = 0.2. (E) Separations of the two
fibers measured from point C at different deformation ratios.
6. 4. Summary
We model single HbS fibers using a quadruple chain of particles and the hexagonally shaped
cross-section of the HbS fiber is coarse grained into four particles. The motions of all the
particles are governed by the Langevin equation. In order to simulate the thermal behaviors of
HbS fibers, five different interaction potentials are applied between the particles, namely a FENE
potential, a truncated L-J potential, a bending potential, a torsional potential and a L-J potential.
153
By employing these five potentials, the proposed model is able to derive the experimentally
measured bending rigidity, and the torsional rigidity of a single HbS fiber. Then, the model is
used in the study of the zippering process of initially parallel free fibers and of frustrated fibers.
Finally, the behavior of zippered fibers under compression is explored. The results show that
while low interaction energy between free fibers is enough to cause zippering, frustrated fibers or
fibers under compression require much higher interaction energy to remain zippered. This means
that fiber frustration and compression can result to partial unzippering of bundles of polymerized
HbS fibers. Continuous polymerization of the unzippered fibers via heterogeneous nucleation
and additional unzippering under compression can explain the formation of HbS fiber networks
and consequently the wide variety of shapes of deoxygenated sickle cells.
Appendix 6.A
The single hemoglobin fiber is modeled as four tightly bonded chains, each of which is
composed of 100 soft particles (Fig. 6.1D). Adjacent particles that belong to the same chain (e.g.,
D and E in Fig. 6.1D) and particles that belong to different chains that are initially in the same
cross-sectional plane (A-B, A-C, A-D, B-C, B-D and C-D in Fig. 6.1C) interact via the FENE
potential
(cid:1847)(cid:3046) = − (cid:2869)
(cid:2870)
(cid:1837)(cid:3046) (cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) ln [1 − ((cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)
∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
)(cid:2870)],
which generates a force
(cid:1832) = −(cid:1837)(cid:3046)
⁄
(cid:3435)(cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)(cid:3439) ∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
(cid:2869)(cid:2879)(cid:3427)(cid:3435)(cid:3045)(cid:3284)(cid:3285)(cid:2879)(cid:3045)(cid:3116)(cid:3439) ∆(cid:3045)(cid:3288)(cid:3276)(cid:3299)
⁄
(cid:3118).
(cid:3431)
154
(6.A.1)
(6.A.2)
In the simulation, the deformations of the HbS fibers are small. In this case the FENE potential
can be approximated by the expression
(cid:1847)(cid:3046) = (cid:2869)
(cid:2870)
(cid:1837)(cid:3046)/(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) (cid:3435)(cid:1870)(cid:3036)(cid:3037) − (cid:1870)(cid:2868)(cid:3439)(cid:2870)
,
Which corresponds to a harmonic potential with a spring constant of
(cid:1837)(cid:3046)(cid:3043) = (cid:1837)(cid:3046)/(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051).
(6.A.3)
(6.Α.4)
For small deformations, it can be assumed that the relation between the applied stress (cid:2028) and the
applied strain (cid:2013) is linear
(cid:2028) = (cid:1831)(cid:2013) ,
(6.A.5)
where E is the Young's modulus. The Eq. (6.A.5) can be rewritten as
(cid:1832)/(cid:1827) = (cid:1831)(cid:1986)(cid:1838)/(cid:1838) ,
(6.A.6)
where (cid:1832) is the applied force along the fiber, (cid:1827) = (cid:2024)(cid:1870)(cid:2868)
(cid:2870) is the cross-sectional area, (cid:1986)(cid:1838) is the change
of the spring's length, and (cid:1838) = (cid:1840)(cid:1870)(cid:2868) is the total length of the fiber, where (cid:1840) is the number of
particles along one of the four chains. The Young's modulus is expressed as
155
(cid:1831) = (cid:1832)(cid:1838)/(cid:1827)(cid:1986)(cid:1838) .
(6.A.7)
Approximating the FENE potential with the harmonic potential for small deformations, the force
can be calculated by
(cid:1832) = (cid:1837)(cid:3046)(cid:3043)
(cid:3047) ∗ (cid:1986)(cid:1838) ,
(6.A.8)
where (cid:1837)(cid:3046)(cid:3043)
(cid:3047) is the total spring constant of the whole fiber. Since the fiber model comprises 4
groups of springs in parallel and each group has (cid:1840) springs in series. The total spring constant of
the whole fiber is approximately
(cid:1837)(cid:3046)
(cid:3047) = 4(cid:1837)(cid:3046)(cid:3043)/(cid:1840) .
Substituting the Eqs. (6.A.9) and (6.A.8) into the Eq. (6.A.7), we obtain
(cid:1837)(cid:3046)(cid:3043) = (cid:1831)(cid:2024)(cid:1870)(cid:2868)/4.
(6.A.9)
(6.A.10)
If the fiber is approximated by a cylindrical rod, then the bending rigidity is related to the
Young's modulus by
(cid:1831) = (cid:2018) (cid:1835)⁄ = 4(cid:2018)/(cid:2024)(cid:1870)(cid:2868)
(cid:2872) ,
(6.A.11)
156
where (cid:1835) = (cid:2024)(cid:1870)(cid:2868)
(cid:2872) 4⁄ is the cross-sectional moment of inertia, and (cid:2018) is the bending rigidity.
Substituting Eq. (6.A.11) into Eq. (6.A.10), we obtain the relation between the spring constant
and (cid:2018)
(cid:1837)(cid:3046)(cid:3043) = (cid:2018)/(cid:1870)(cid:2868)
(cid:2871) .
(6.A.12)
The expected value of the bending rigidity is approximately (cid:2018) = 5 × 10(cid:2879)(cid:2870)(cid:2873)Nm(cid:2870) , which
corresponds to
(cid:2018) ≅ 12000 (cid:2013)(cid:1870)(cid:2868),
where (cid:2013) = (cid:1863)(cid:3029)(cid:1846) is the energy unit. Therefore the spring constant is
(cid:1837)(cid:3046)(cid:3043) = 12000 (cid:2013) (cid:1870)(cid:2868)
(cid:2870)⁄ .
(6.A.13)
(6.A.14)
Βy substituting the Eq. (6.A.14) into the Eq. (6.A.4) and by using that (cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) = 0.3(cid:1870)(cid:2868), we
obtain that
(cid:1837)(cid:3046) = (cid:1837)(cid:3046)(cid:3043)(cid:1986)(cid:1870)(cid:3040)(cid:3028)(cid:3051) = 3600 (cid:2013) (cid:1870)(cid:2868)⁄ .
(6.Α.15)
157
Chapter 7.
Modeling Sickle Hemoglobin Fibers as One Chain of Coarse-
Grained Particles
ABSTRACT
Sickle cell disease (SCD) is caused by a single point mutation in the beta-chain hemoglobin gene,
resulting in the presence of abnormal hemoglobin S (HbS) in the patients' red blood cells
(RBCs). In the deoxygenated state, the defective hemoglobin tetramers polymerize forming stiff
fibers which distort the cell and contribute to changes in its biomechanical properties. Because
the HbS fibers are essential in the formation of the sickle RBC, their material properties draw
significant research interests. Here, a solvent-free coarse-grain molecular dynamics (CGMD)
model is introduced to simulate single HbS fibers as a chain of particles. First, we show that the
proposed model is able to efficiently simulate the mechanical behavior of single HbS fibers.
Then, the zippering process between two HbS fibers is studied and the effect of depletion forces
is investigated. Simulation results illustrate that depletion forces play a role comparable to direct
fiber-fiber interaction via Van der Waals forces. This proposed model can greatly facilitate
studies on HbS polymerization, fiber bundle and gel formation as well as interaction between
HbS fiber bundles and the RBC membrane.
158
7.1. Introduction
The erythrocytes of the SCD patients contain defective HbS, in which a charged surface group
glu at β6 is replaced by a hydrophobic group val, inducing polymerization of deoxygenated HbS
at high enough concentrations (27, 283). The stiff polymerized HbS fibers (see Fig. 7.1A) can
create protrusions and cause vesiculation of the RBC membrane (35). Electron microscopy and
X-ray crystallography studies revealed that a single HbS fiber consists of 14 filaments arranged
in 7 double helical strands twisted about a common axis in a rope-like fashion (see Fig. 7.1B)
(312). The average radius of a single HbS fiber is approximately 11 nm. The mean helical path
along the fiber, defined as the length over which the fiber twists through 180º, is about 270 nm.
Measurements of the bending rigidities and persistence lengths of HbS fibers were conducted via
the differential interference contrast (DIC) microscopy (290) and cryo-electron microscopy (291,
292), which showed that the bending rigidity of a single HbS fiber was approximately 5.2×10-25
N m2 and the torsional rigidity was approximately 6×10-27 J m (292).
159
Figure 7.1. (A) Schematic of HbS fibers in a sickle cell. (B) A single HbS fiber comprises 7
pairs of double strands of hemoglobin tetramers. The pairs are shown as same-color agents. It is
drawn after (27). (C) Each bead represents the entire cross-section of a single HbS fiber. (D) Side
view of the HbS fiber model.
The formation of X-shaped and Y-shaped cross-links, along with zippering of HbS fibers affects
the rheology and stiffness of deoxygenated sickle RBCs and consequently the pathophysiology
of SCD (313). Various analytical schemes based on Van der Waals and depletion forces have
been proposed to estimate the lateral attraction between hemoglobin fibers and the results show
that the depletion forces have an equivalent contribution with the Van der Waals forces.
However, the experimentally obtained binding energy is lower than the analytically computed
value (44, 314). This discrepancy was attributed to several possible reasons such as existence of
fiber tensile stress, presence of non-steric repulsive forces between the fibers, aggregation of
HbS monomers which reduces the solution osmotic pressure, steric frustration due to the helical
nature of the fibers, and non-zero surface separation because of thermal fluctuations (44).
Research on the mechanical behavior of HbS fibers so far is mainly based on analytical methods
and experimental observations. Molecular dynamics simulations have only been employed in
investigations related to the contact points of the double strands in HbS fiber formation and the
crystal structure of HbS (295, 296). Here, a CGMD model of single HbS fibers is introduced.
The parameters used in the CGMD potentials are validated by the experimentally measured
bending and torsional rigidities of HbS fibers (290, 292). The proposed model follows a top-
down approach. It is much simpler than multiscale coarse-graining strategies which usually
160
consider atomistic details while the CG force field is parameterized to match the atomistic forces
and predict the thermodynamic behavior of the system (299-303). Similar approaches have been
successfully used in the representation of the spectrin cytoskeleton of erythrocyte membrane
(109) and in the study of cross-linked actin-like networks (297, 298). A four chain model for the
hemoglobin filaments has been proposed earlier by the authors (315). While the overall strategy
is similar with the model introduced here, there are significant differences that considerably
improve the present model. The cross-section of the HbS fiber in the present model is
represented by one CG particle instead of four particles. Also, the introduction of torsional
rigidity is more straightforward while in the previous model the torsional rigidity originated from
the interactions between the four chains. The present model is applied to investigate the
interaction between two HbS fibers and the fiber zippering process. The effects of the Van der
Waals force and the depletion force on the HbS fiber zippering are explored.
7.2. Model and method
A 2 μm long HbS fiber with diameter of 20 nm is modeled as one chain of 100 soft particles (see
Fig. 7.1C and Fig. 7.1D). The diameter of each particle d0 is 20 nm. The translational motion of
the particles is governed by the Langevin equation
m
i
r
i
2
2
d
dt
F
i
f
d
r
i
dt
B
F ,
i
(7.1)
where mi represents the mass of the ith particle, f is the friction coefficient and it is identified to
be 100 mi/ts, where ts is the time scale. ri is the position vector of the ith particle, and t is time
(304). Fi = -∂U/∂ri is the deterministic force produced by the implemented total potential U,
which is introduced below. Fi
B is related to the environmental Gaussian white noise and it obeys
the fluctuation-dissipation theorem
161
B
F
i
0
,
B
F F
i
B
j
ij
2 B
k Tf
t
,
(7.2)
(7.3)
where kB is the Boltzmann's constant, T = 300 K is the absolute environmental temperature, δij is
the Kronecker delta, and Δt is the time-step (98, 305). The energy unit is kBT. The time step for
the numerical solution of the Langevin equation is chosen to be Δt = 0.001ts. Deviations from
that temperature are corrected with a decay time td = mi/f. By choosing f = 100 mi/ts, the decay
time becomes td = ts/100 =10Δt, meaning that the system corrects the temperature variations in
about 10 time steps.
Figure 7.2 (A) Bending angle θ and (B) twist angle ΔΦ implemented in the HbS fiber model.
The rotational motion of the particles is simplified. Each particle only rotates in the plane
perpendicular to the vector defined by the centers of this particle and the next one towards the
end of the fiber (see Fig. 7.2A). The rotational angle Φ (see Fig. 7.2B) is controlled by the
Langevin equation for the rotational Brownian motion
( )
dM t
i
dt
T
i
i
( )
t
i
( )
t
,
162
(7.4)
where Mi = I ωi is the angular momentum of the ith particles (316), I is the moment of inertia and
ω = dΦ(t)/dt is the angular velocity. Ti is the torque generated by the torsional potential, which is
introduced below. ζωi(t) and λ(t) are the frictional and white noise torque, respectively. ζ is the
friction coefficient and it is chosen based on the assumption that the Langevin equation for the
translational and rotational motion should have the same decay time, td = mi/f = Ii / ζ. Then, the
friction coefficient is computed to be ζ = 10miσ2/ts, where the length unit is σ = 2-1/6d0. The white
noise torque follows the properties below,
i t
0
( )
i
( )
t
( )
t
j
2
k T
B
ij
t
.
(7.5)
(7.6)
A FENE potential (see Fig. 7.3) is introduced between adjacent particles that belong to the same
chain (e.g., A and B in Fig. 7.1D). The FENE potential is expressed as
U
s
1
2
K d
s
max
ln 1
d
0
d
ij
d
max
2
,
(7.7)
where, Ks, is related to the stiffness Ksp = Ks/Δdmax of the harmonic potential which has the same
stiffness with the FENE potential at equilibrium. The value of Ks that generates the Young's
modulus of the HbS fiber is approximately Ks = Ksp Δdmax = 28800kBT/d0. dij and d0 are the
distance and the equilibrium distance between particles i and j, respectively. In the simulations,
the maximum extension allowed between two particles is set to be Δdmax= 0.3d0.
U
b
1
2
K
b
max
ln 1
0
max
2
,
(7.8)
where Kb is the parameter that directly regulates the bending stiffness of the HbS fiber, and (cid:2016) is
the angle formed by three consecutive particles in the same chain, as illustrated in Fig. 7.2A. (cid:2016)0
163
is the equilibrium angle and it is chosen to be π, meaning that the three consecutive particles are
initially located in-line. Δ(cid:2016)max is the maximum allowed bending angle between two particles and
is set to 0.3(cid:2016)0. Kb is selected via a trial and error process in order to produce a bending rigidity
identical to the experimental value and it is determined to be Kb = 8×103 kBT.
Figure 7.3 The finitely extendable nonlinear elastic (FENE) potential applied between
consecutive fiber particles, the corresponding harmonic potential, and the Lennard-Jones
potential applied between particles belonging to different HbS fibers. The additional bending and
torsional FENE potentials employed in the simulations have similar behavior with the FENE
potential shown here.
The torsional rigidity of the proposed model is introduced via a torsional FENE potential
between neighboring particles in the chain. The potential is expressed as
U
t
1
2
K
t
max
ln 1
max
2
,
(7.9)
where Kt adjusts the torsional stiffness of the HbS fiber, and ∆Φ is the angle between two
directional vectors defined in two consecutive particles, as illustrated in Fig. 7.2B. ΔΦmax is the
164
maximum allowed twist angle between two consecutive particles and is set to be 0.3. Kt is tuned
to match the model against experimentally obtained values of the torsional rigidity of single HbS
fibers (44) and it is identified to be Kt = 8.1×103 kBT.
The interactions between neighboring HbS fibers are represented by a L-J potential between
particles from different fibers (see Fig. 7.3). The expression of L-J potential is given by
U r
(
ij
LJ
)
4
k
r
ij
12
6
r
ij
0
k
r
ij
r
cut
r
ij
r
cut
,
(7.10)
where rij is the distance between particles i and j. The parameter k adjusts the force between two
HbS fibers. The value k = 70 is used for the plot in Fig. 7.3. The distance between two particles,
beyond which the attractive force is neglected, is rcut = 2.5σ.
A coupling torque Tc is introduced between two particles from different fibers to represent the
resistance to the rotational motion when the two fibers are zippered. The torque is assumed to be
cT
T e
0
,
1
(7.11)
where T0 is chosen to be 10 kBT, which is on the same scale of the torque generated by the FENE
torsional potential. α is the relative rotational angle between two neighboring bonded particles,
measured after the instance of the two particles zippering.
The molecular weight of each hemoglobin tetramer is approximately 64.5 KDa. Thus, each
coarse-grained particle has a mass of mi ≈ 6×10-21kg representing a cluster of 56 hemoglobin
tetramers. In the Langevin equation, the friction coefficient is chosen to be f = 100 mi/ts while f
165
also can be expressed as a function of the particle diameter d0 and the medium viscosity η ( f =
3πηd0). The viscosity of Hb solution with concentration of c = 24.4g/dl used by (44) is found to
be η ≈ 3×10-3 kg/m s (317). By solving the equation 100 mi/ts = 3πηd0, the timescale of the
simulation is found to be ts ≈ 1×10-9 s. The spring constant for the harmonic potential Ksp can be
expressed in terms of the bending rigidity κ, as
K
sp
d
16 /
3
0
. A relevant time-scale for the
t
vibration of the HbS fibers can be estimated as c
m K
2
/
556.4ps
sp
. The time-scale for the
fluctuation of the rotational motion of the HbS fibers can be similarly obtained by the expression
t
r
I K
2
/
max
t
560.6ps
, which is close to tc.
The numerical method employed for the integration of Eq. (7.1) is a modified version of the
leapfrog algorithm and it is given by
v
i
(
t
t
/ 2)
( )
t
t
v
i
/ 2 ( )
t
a
i
)
t
( )
t
t
r
i
v
i
(
t
t
/ 2)
r
i
(
t
v
i
(
t
a
i
(
t
)
t
v
i
(
t
t
/ 2)
t
/ 2 ( )
t
a
i
7.12(c)
7.12(a)
7.12(b)
7.12(d)
7.12(e)
)
t
( (
t
a r
i
i
t
),
v
i
(
t
t
))
v
i
(
t
)
t
v
i
(
t
t
/ 2)
t
/ 2 (
t
a
i
t
)
.
The total force applied on a particle depends on the velocity of the particle. Thus, a prediction is
made for the new velocity, which is denoted as , and it is corrected in the last step (106, 315).
The same strategy was followed for the integration of the rotational Langevin equation.
7.3. Results and discussion
7.3.1 Measurements of material properties of HbS fiber
166
7.3.1.1 Bending rigidity
The bending modulus of semi-flexible fibers can be derived from the thermally driven
fluctuations (126). One method is to measure the mean-squared amplitude of each dynamical
mode of vibrations and by using the principle of equipartition of energy to extract the bending
rigidity for each mode independently (318). However, since HbS fibers are stiff with a large
persistence length of approximately 120 -240 μm while the simulated fibers have a total length of
2 μm, only small bending is observed in the simulations. This leads to a large uncertainty when
the motion is decomposed into Fourier modes. In order to measure the bending rigidity of HbS
fibers under thermal fluctuation, a method that relates the fiber bending rigidity with the normal
deviation of the fiber end point through the expression
3
Bk TL
( )
u L
2
3
,
(7.13)
is followed (44, 314). L is the total length of the HbS fiber and
u L
(
)
u L
(
)
u L
(
)
is the
normal deviation of the fiber end measured from its average position during the simulation to
eliminate the rigid body motion effects and to only account for thermal fluctuations. Once the
bending rigidity is obtained, the persistence length can be easily calculated by the expression
p
l
k T
B
.
(7.14)
167
Figure. 7.4. The distribution of HbS fiber end displacements and the associated normalized
Gaussian probability distribution.
The probability density of the deviations from the equilibrium at the end of the HbS fiber can be
approximated by a normalized Gaussian distribution (solid curve in Fig. 7.4) given by
(
P u
)
1
2
u
2
exp
2
u
u
2
2
,
(7.15)
where δu is the HbS fiber end displacement, and
1/ 2
2
u
8
. The result is consistent with the
theoretical prediction that the probability density of the deviations of a semi-flexible rod follows
the Gaussian distribution in the case of thermal fluctuations (310) and it indicates that the HbS
fiber has reached thermal equilibrium. The value of the bending rigidity calculated via the
equation (16) is κ = 5.3×10-25 N m2, which is close to the experimentally measured value of κ =
5.2×10-25 N m2 (292). The obtained bending rigidity corresponds to a persistence length of
approximately
pl
121
μm. The numerical results show that, because of the large persistence
length, a single HbS fiber subjected to thermal forces behaves similarly to a stiff rod as it
fluctuates about its central axis (see Fig. 7.5). It is noted that when a longer fiber of 5 μm length
168
is tested, the computed bending rigidity has the same value with the bending rigidity of the 2 μm
long fiber, meaning that the result is size independent at least up to 5 μm.
Figure 7.5. Characteristic configuration of the HbS fiber model. The colors of the particles
represent the value of the rotation angle Φ, the blue color corresponds to Φ = 0 and the red color
corresponds to Φ = 0.72. The color bar is shown at the top of the figure.
7.3.1.2 Torsional rigidity
The torsional rigidity c of the HbS fiber can be obtained via the expression
2
3
Bk T
2
c
1
L
1
L
e
,
(7.16)
where Δλ = λΨ/π, λ = 135 nm is half of the average pitch length for the HbS fiber, Ψ is the
accumulated relative rotation between the first fixed particle and the 7th particle located at
approximately half of the pitch length, and L is the length of the HbS fiber (290). By substituting
Δλ into equation (16), the torsional rigidity c is found to be
c
Bk T
1
2
Ψ
L
1
e
L
.
(7.17)
169
Figure. 7.6. The distribution of the accumulation of the twist angle at half of the pitch length (7th
chain particle) Ψ and the associated normalized Gaussian probability distribution
Fig. 7.6 shows that the twist angle through half of the pitch length Ψ also follows the Gaussian
distribution with a mean twist angle of 0.293, resulting in a torsional rigidity of the HbS fiber
model
c
6.5 10
27
J m, which is very close to the experimentally measured value of
c
6.0 10
27
J m (292). As in the case of bending rigidity, it is found that a longer fiber of 5 μm
exhibits the same torsional rigidity with the shorter one when the same parameters are used. It is
also worth mentioning that the proposed HbS fiber model is strongly anisotropic since the
torsional rigidity is significantly lower than the bending rigidity.
7.3.2 Modeling the zippering of two HbS fibers
The developed CGMD model is employed to study the effect of the Van der Waals and the
depletion forces on the zippering of two fibers in a Y-shaped cross-link (see Fig. 7.7). The Van
der Waals interaction is represented by a L-J potential applied between particles belonging to
different fibers. The effect of the depletion force is measured by the contribution of HbS
monomers on the fiber zippered length. The concentration of the hemoglobin solution is chosen
170
to be c = 24.4 g/dl, which is the value used in the previous experimental work (44). The HbS
monomer particles are assumed to have the same size (d0 = 20 nm) and same mass (6 ×10-21kg)
of the fiber particles. Therefore, the concentration of the solution c = 24.4 g/dl corresponds to
HbS monomers density of ρ = 0.23 particles/σ3 in the simulation space (a box with the size of
140σ × 50σ × 30σ). The height of the simulation space is chosen to be 30σ (about 540 nm),
which is more than two times thicker than the fiber interaction space observed in the experiments
(44, 313). In this section, we applied the Berendsen thermostat to control the temperature of the
system. In addition, the effect of monomer aggregation on the fiber zippering is studied by
reducing the density of the free monomer particles to ρ = 0.115particles/σ3.
Figure 7.7. (A) Initial 3D configuration of two HbS fibers (red particles) arranged in "Y" shape.
(B) Equilibrium 3D configuration of the two HbS fibers after zippering. The value of the L-J
parameter is k = 30. (C) Initial 3D configuration of two HbS fibers arranged in "Y" shape
surrounded by HbS monomer particles (white particles). (D) Equilibrium 3D configuration of the
two HbS fibers surrounded by HbS monomer particles after zippering. The value of the L-J
parameter is k = 30 and the density of the HbS monomers is ρ = 0.23 particles/σ3. It is observed
that the zippered length of two fibers is increased by 7σ compared to the (B) due to the presence
of the HbS monomer particles.
171
The case of two fibers arranged initially in a "V" shape junction without any HbS monomers is
first studied (see Fig. 7.7A). Points A and C represent the points where the measured fibers are
bonded stably to a third fiber and they are thus fixed. The other two ends B and D are free to
move. The HbS fibers interact through physical and chemical forces including direct electrostatic
effects, Van der Waals forces and the entropic depletion forces between the free monomers and
the fibers. In addition, there are small hydrophobic forces applied on patches on the surfaces of
fiber aggregates (44). The attractive energy between two fibers is represented by a L-J potential
(Eq. (7.10)) and it is adjusted by varying the value of the parameter k in the expression (7.10).
After zippering (see Fig. 7.7B), the two fibers move together as a Y-shaped branch. In the
equilibrium state, the interfiber attractive forces are balanced by the bending energy stored in the
two fibers (44, 313). The zippered lengths for various values of k are illustrated by the black line
in Fig. 7.8. It is observed that for low k values, the zippering is not significant. As k increases,
the slope of the zippered length with respect to energy decreases since the effect of the
constraints on the fiber ends A and C becomes more pronounced. Secondly, the entropic effect of
the free HbS monomers on the zippering of the two HbS fibers is investigated (Figs. 7C and 7D).
It is assumed that there are only repulsive forces between HbS monomers and between HbS
monomers and fiber particles. The repulsive forces result from the repulsive part of the L-J
potential (Eq. (7.10)) and it creates an excluded volume for HbS monomers and fibers. The effect
of the depletion force on the zippered lengths of two fibers for two different densities of the free
HbS monomers is shown in Fig. 7.8. The depletion force has a dramatic effect when the
attractive force is low. For example, when ρ = 0.23particles/σ3 (red line), contributions of the
depletion force to the fiber zippering is more than the Van der Waals force when k <= 15. As the
attractive energy increases, the effect of the depletion force attenuates.
172
Figure 7.8. Zippered length ratios of two fibers (Lzip/L) for different attractive forces between the
HbS fibers. The solid lines describe the ratios of the zippered length to the fiber length for
different values of the parameter k. The solid lines in different colors represent different densities
of HbS monomers.
A main question raised from experimental studies is how much the aggregation of the HbS
monomers reduces the depletion force (44, 314). To investigate it, we assume that 50% of the
monomers aggregate resulting to a reduced HbS monomer density ρ = 0.115particles/σ3. The
corresponding zippering lengths of two fibers for various values of k are illustrated by the green
lines in Fig. 7.8. For k < 30, the reduction in the monomer density results in a decrease in the
additional zippering caused by the depletion effect. As the k increases, however, the drop of the
monomer density has smaller effect on the reduction of the fiber zippering because the attractive
forces between the two fibers become dominant. This result indicates that the effect of monomer
aggregation on the zippered length is larger at lower values of the interaction force between the
fibers.
173
In summary, a CGMD model with high computational efficiency is proposed to simulate single
HbS fibers as one chain of particles. The model is able to derive the experimentally measured
bending and torsional rigidities of single HbS fibers. In addition, it is employed to investigate the
effect of the Van der Waals force and the depletion force on the zippering of two HbS fibers to
improve our understanding on the HbS fiber aggregation in SCD. The present model has the
potential to significantly facilitate the studies of HbS polymerization, fiber aggregation and gel
formation. The present model could be combined with a hybrid method of dynamic coarse-
graining, to simulate HbS polymerization. Since the HbS polymerization process is governed by
monomer addition as opposed to oligomer addition (319), the polymerization has to be studied at
the atomistic level. Therefore, the tip of a HbS fiber, where polymerization occurs, could be
simulated at the atomistic level while the already polymerized HbS fiber could be simulated by
the proposed coarse-grained model to reduce the overall computational cost. A similar technique
has been introduced in phospholipid bilayer models (320, 321).
174
Chapter 8.
Epilogue
We have developed a two-component CGMD model for RBC membranes. In this model, the
lipid bilayer and the cytoskeleton are considered as an ensemble of discrete particles that interact
through a direction-dependent pair potential. Three types of coarse-grained particles are
introduced, representing a cluster of lipid molecules, actin junctions and band-3 complexes. The
large grain size extends the accessible length and time scales of the simulations to ~ µm and ~
ms. By tailoring only a few parameters of the inter-grain interaction potential, the model can be
stabilized into a fluid phase manifested by the free diffusion of the particles and the reproduction
of the essential thermodynamic properties of the RBC membrane. An important feature of the
proposed model is the combination of the spectrin network with the lipid bilayer, which
represents a more compete representation of the RBC membrane compared to one-component
CGMD model and to most of the continuum models. This model allows us to study the behavior
of the membrane under shearing. The behavior of the model under shearing at different strain
rates illustrates that at low strain rates up to 0.001d/ts, the developed shear stress is mainly due to
the spectrin network and it shows the characteristic non-linear behavior of entropic networks,
while the viscosity of the fluid-like lipid bilayer contributes to the resulted shear stress at higher
strain rates. Decrease of the spectrin network connectivity results to significant decrease of the
shear modulus of the membrane, which demonstrates that the cytoskeleton carries most of the
load applied on the cell while the lipid bilayer only functions as 2D fluid. The values of the shear
moduli measured from the membrane with reduced spectrin network connectivity are in a good
agreement with previous experimental, theoretical, and numerical results.
175
We introduce a particle-based model for the erythrocyte membrane that accounts for the most
important structural components of the membrane, including the lipid bilayer, the spectrin
network, and the proteins that play an important role in the anchoring of the spectrin cortex to the
lipid bilayer, as well as the band-3 proteins. In particular, five types of CG particles are used to
represent actin junctional complexes, spectrin, glycophorin, immobile band-3 protein, mobile
band-3 protein and an aggregation of lipids. We first demonstrate that the model captures the
fluidic behavior of the lipid bilayer and then that it reproduces the expected mechanical material
properties of bending rigidity and shear modulus of the RBC membrane. The timescale of our
simulations, which is found to be ts ~ 3×10-6s, is inferred by comparing the viscosity of the
membrane model to experimentally measured values. Then, the self-consistency of the model
with respect to the timescale is tested by comparing the computed vibration frequency of the
spectrin filaments and lipid membrane to analytically obtained values. We confirm that vibration
frequency of the spectrin filaments and lipid membrane measured from the proposed membrane
model, are also in agreement with experimental values. At last, we study the interactions
between the cytoskeleton and the lipid bilayer, and measure the pressure applied on the
membrane by the spectrin filaments. We also investigate how disruption of the connection
between the spectrin network and the lipid bilayers, which simulates defects in HS, and rupture
of the dimer-dimer association, which simulates defects in HE, affect the pressure exerted on the
lipid bilayer. We show that overall the introduction of defects in the spectrin network or in the
vertical connection between the lipid bilayer and the cytoskeleton results in lower pressure,
which is consistent with prediction in (155). In addition, we find that the defects related to HE
have a stronger effect than the defects related to HS. This result implies that diffusion of band-3
proteins in RBCs from patients with HS and HE is enhanced compared to the normal RBCs.
176
Moreover, elliptocytes exhibit more prominent diffusion of band-3 proteins than spherocytes.
Both conclusions are supported by experimental results. The level of attraction forces between
the lipid bilayer and the membrane cytoskeleton is another important parameter that regulates the
pressure applied by a spectrin filament to the lipid bilayer. We show that as the attractive force
increases, it causes an overall increase in the pressure and it diminishes the differences in the
pressure generated by membrane protein defects and, consequently, the differences in the
diffusion of band-3 proteins in normal and defective erythrocytes. Since this finding is not
supported by experimental results, we conjecture that the attractive force between the lipid
bilayer and the spectrin filaments should be low, resulting in a membrane model where the
filaments are not completely attached to the lipid bilayer. A detailed study of the band-3
diffusion in this model and direct comparison with experimental results is necessary in order to
form a more accurate picture of how the model regulates diffusion. Because of the explicit
representation of the lipid bilayer and the cytoskeleton, the proposed model can be potentially
used in the investigation of a variety of membrane related problems in RBCs in addition to
diffusion. For example, membrane loss through vesiculation and membrane fragility in
spherocytosis and elliptocytosis, interaction between hemoglobin fibers and RBC membrane in
sickle cell disease, and RBC adhesion are problems where the applications of the proposed
membrane model could be beneficial.
We apply a two-component RBC membrane model to study band-3 diffusion in the normal RBC
membrane and in the membranes with defective vertical and horizontal interactions. We measure
the band-3 diffusion coefficients and quantify the relation between the band-3 diffusion
coefficients and percentage of protein defects. Our measurements show that the band-3 diffusion
177
coefficients are increased by 8 times when connectivity between the spectrin filaments and
immobile band-3 particles Cvertical is decreased from 100% to 0%, whereas the band-3 diffusion
coefficients can be boosted by 20-40 times with significantly reduced cytoskeleton connectivity
Chorizontal. These results are in agreement with the experimental measurements of the band-3
diffusion in the HS and HE RBCs (112, 115, 147, 197). By comparing the measured band-3
diffusions coefficients, we demonstrate that cytoskeleton connectivity is the major determinant
of the lateral diffusivity of band-3. In addition, we quantify the relations between the anomalous
diffusion exponent and the percentage of protein defects in the horizontal interactions. We find
that the cytoskeleton has a small effect on the anomalous diffusion exponent, comparing to the
obstructions due to the lipid rafts or protein domains. At last, we introduce attractive forces
between the spectrin filaments and the lipid bilayer, and study the effects of the attractive forces
on the band-3 diffusion. We find that application of the attractive forces slows down the band-3
diffusion in the normal and defective RBC membrane. Especially, large attractive forces can
prevent the band-3 hop diffusion in the normal RBC membrane and in the membrane with
protein defects in the vertical interactions. But, in the cases of the membrane with protein defects
in the horizontal interactions, the band-3 diffusion coefficients become independent of the
attractive forces when attractive forces are large. This is because the band-3 particles travel to
different cytoskeleton compartments through the broken spectrin filaments. Moreover, we show
that the band-3 diffusion coefficients measured at the large attractive forces generally agree with
the percolation analysis in (116). By comparing the band-3 diffusion coefficients from our
simulation with the experimental measured band-3 diffusion coefficients in HS and HE (112, 115,
147, 197), we estimated the scale of the effective attractive force between the spectrin filaments
and lipid bilayer is at least 20 times smaller than the binding forces at the two major binding sites
178
at the immobile band-3 proteins and the glycophorin C. The approach used here for the
simulation of band-3 diffusion in the defective RBC membrane provides a basis for the future
study of the mechanism of the membrane vesiculation in HS and HE.
We model single HbS fibers using a quadruple chain of particles and the hexagonally shaped
cross-section of the HbS fiber is coarse grained into four particles. The motions of all the
particles are governed by the Langevin equation. In order to simulate the thermal behaviors of
HbS fibers, five different interaction potentials are applied between the particles, namely a FENE
potential, a truncated L-J potential, a bending potential, a torsional potential and a L-J potential.
By employing these five potentials, the proposed model is able to derive the experimentally
measured bending rigidity, and the torsional rigidity of a single HbS fiber. Then, the model is
used in the study of the zippering process of initially parallel free fibers and of frustrated fibers.
Finally, the behavior of zippered fibers under compression is explored. The results show that
while low interaction energy between free fibers is enough to cause zippering, frustrated fibers or
fibers under compression require much higher interaction energy to remain zippered. This means
that fiber frustration and compression can result to partial unzippering of bundles of polymerized
HbS fibers. Continuous polymerization of the unzippered fibers via heterogeneous nucleation
and additional unzippering under compression can explain the formation of HbS fiber networks
and consequently the wide variety of shapes of deoxygenated sickle cells.
We apply a CGMD model with high computational efficiency to simulate single HbS fibers as
one chain of particles. The model is able to derive the experimentally measured bending and
torsional rigidities of single HbS fibers. In addition, it is employed to investigate the effect of the
179
Van der Waals force and the depletion force on the zippering of two HbS fibers to improve our
understanding on the HbS fiber aggregation in SCD. The present model has the potential to
significantly facilitate the studies of HbS polymerization, fiber aggregation and gel formation.
The present model could be combined with a hybrid method of dynamic coarse-graining, to
simulate HbS polymerization. Since the HbS polymerization process is governed by monomer
addition as opposed to oligomer addition (319), the polymerization has to be studied at the
atomistic level. Therefore, the tip of a HbS fiber, where polymerization occurs, could be
simulated at the atomistic level while the already polymerized HbS fiber could be simulated by
the proposed coarse-grained model to reduce the overall computational cost. A similar technique
has been introduced in phospholipid bilayer models (320, 321).
180
Chapter 9.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Peppiatt CM, Howarth C, Mobbs P, & Attwell D (2006) Bidirectional control of CNS
capillary diameter by pericytes. Nature 443(7112):700-704.
Mebius RE & Kraal G (2005) Structure and function of the spleen. Nat Rev Immunol
5(8):606-616.
Tse WT & Lux SE (1999) Red blood cell membrane disorders. British Journal of
Haematology 104(1):2-13.
Mohandas N & Gallagher PG (2008) Red cell membrane: past, present, and future. Blood
112(10):3939-3948.
Delaunay J (1995) Genetic disorders of the red cell membrane. Critical Reviews in
Oncology/Hematology 19(2):79-110.
Alberts B, Johnson, A., Lewis, J., Ralf, M., Roberts, K., Walter, P. (2002) Molecular
Biology of the Cell (Garland, New York).
Bennett V (1985) The Membrane Skeleton of Human-Erythrocytes and Its Implications
for More Complex Cells. Annu. Rev. Biochem. 54:273-304.
Bennett V & Baines AJ (2001) Spectrin and ankyrin-based pathways: Metazoan
inventions for integrating cells into tissues. Physiol. Rev. 81(3):1353-1392.
Helfrich W (1973) Elastic properties of lipid bilayers - theory and possible experiments.
Zeitschrift Fur Naturforschung C-a Journal of Biosciences C28(11-1):693-703.
Frenkel D & Smit B (2002) Understanding Molecular Simulation: from Algorithms to
Applications (Academic Press, San Diego) 2nd Ed.
Seifert U, Berndl K, & Lipowsky R (1991) Shape transformations of vesicles: Phase
diagram for spontaneous- curvature and bilayer-coupling models. Physical Review A
44(2):1182-1202.
12. Miao L, Seifert U, Wortis M, & HG. D (1994) Budding transitions of fluid-bilayer
vesicles-the effect of area-difference elasticity. physical Review E 49:5389-5407.
Lim HWG, Wortis M, & Mukhopadhyay R (2002) Stomatocyte-discocyte-echinocyte
sequence of the human red blood cell: Evidence for the bilayer-couple hypothesis from
membrane mechanics. Proceedings of the National Academy of Sciences of the United
States of America 99(26):16766-16769.
S. Tuvia, S. Levin, A. Bitler, & Korenstein. R (1998) Mechanical Fluctuations of the
Membrane–Skeleton Are Dependent on F-Actin ATPase in Human Erythrocytes. J. Cell
Biol. 141:1551.
Sens P & Gov N (2007) Force Balance and Membrane Shedding at the Red-Blood-Cell
Surface. Physical Review Letters 98(1):018102.
Eber S & Lux SE (2004) Hereditary spherocytosis - Defects in proteins that connect the
membrane skeleton to the lipid bilayer. Seminars in Hematology 41(2):118-141.
Gallagher PG (2004) Hereditary elliptocytosis: Spectrin and protein 4.1R. Semin.
Hematol. 41(2):142-164.
SoM
University
ed.virginia.edu/courses/path/innes/rcd/hemo.cfm.
http://www.med-
(2010)in
of
Virginia
181
13.
14.
15.
16.
17.
18.
19. Marchesi SL, Knowles WJ, Morrow JS, Bologna M, & Marchesi VT (1986) Abnormal
20.
Spectrin in Hereditary Elliptocytosis. Blood 67(1):141-151.
Tse WT, et al. (1990) Point Mutation in the Beta-Spectrin Gene Associated with Alpha-
I/74 Hereditary Elliptocytosis - Implications for the Mechanism of Spectrin Dimer Self-
Association. Journal of Clinical Investigation 86(3):909-916.
21. McMullin MF (1999) The molecular basis of disorders of the red cell membrane. J Clin
22.
23.
24.
25.
Pathol 52(4):245-248.
Anong WA, Weis TL, & Low PS (2006) Rate of rupture and reattachment of the band 3-
ankyrin bridge on the human erythrocyte membrane. Journal of Biological Chemistry
281(31):22360-22366.
Hoffbrand V, Moss P, & Pettit J (2006) Essential Haematology (Wiley-Blackwell) 5 Ed p
392.
Salomao M, et al. (2010) Hereditary spherocytosis and hereditary elliptocytosis: aberrant
protein sorting during erythroblast enucleation. Blood 116(2):267-269.
Eber S & Lux S, E. (2004) Hereditary spherocytosis-defects in proteins that connect the
membrane skeleton to the lipid bilayer. Seminars in Hematology 41(2):118-141.
29.
30.
32.
28.
31.
27.
26. Walensky LD, Mohandas N, & Lux SE (2003) Disorders of the Red Blood Cell
Membrane, Herediatery Spherocytosis. Blood - Principles and Practice in Hematology -
Herediatery Spherocytosis, eds Handin RI, Lux SE, & Stossel TP (Lippincott Williams &
Wilkins), Second Ed.
Ferrone FA (2004) Polymerization and sickle cell disease: A molecular view.
Microcirculation 11(2):115-128.
Dou Q & Ferrone FA (1993) Simulated formation of polymer domains in sickle
hemoglobin. Biophysical Journal 65(5):2068-2077.
Fixler J & Styles L (2002) Sickle cell disease. Pediatric Clinics of North America
49(6):1193-1210.
Edwards C, et al. (2005) A brief review of the pathophysiology, associated pain, and
psychosocial issues in sickle cell disease. International Journal of Behavioral Medicine
12(3):171-179.
Buchanan GR, DeBaun MR, Quinn CT, & Steinberg MH (2004) Sickle cell disease.
Hematology Am Soc Hematol Educ Program:35-47.
Steinberg MH & Brugnara C (2003) Pathophysiological-Based Approaches to Treatment
of Sickle Cell Disease. Annual Review of Medicine 54(1):89.
Steinberg MH & Rodgers GP (2001) Pathophysiology of sickle cell disease: Role of
cellular and genetic modifiers. Semin. Hematol. 38(4):299-306.
Rosse WF, Narla M, Petz LD, & Steinberg MH (2000) New Views of Sickle Cell Disease
Pathophysiology and Treatment. Hematology 2000(1):2-17.
Statius van Eps LW (1999) Sickle cell disease. Atlas of diseases of the kidney, ed Klahr S
(Wiley-Blackwell, Philadelphia, Pennsylvania), Vol 4, p 22.
Embury SH (1986) The Clinical Pathophysiology of Sickle Cell Disease. Annual Review
of Medicine 37(1):361-376.
Strouse JJ, et al. (2008) Hydroxyurea for Sickle Cell Disease: A Systematic Review for
Efficacy and Toxicity in Children. Pediatrics 122(6):1332-1342.
Noguchi CT & Schechter AN (1985) Sickle Hemoglobin Polymerization in Solution and
in Cells. Annual Review of Biophysics and Biophysical Chemistry 14(1):239-263.
33.
34.
35.
36.
37.
38.
182
39.
40.
41.
42.
43.
44.
45.
46.
47.
Turner MS, et al. (2002) Fluctuations in self-assembled sickle hemoglobin fibers.
Langmuir 18(19):7182-7187.
Li H, Ha V, & Lykotrafitis G (2012) Modeling sickle hemoglobin fibers as one chain of
coarse-grained particles. Journal of biomechanics 45(11):1947-1951.
Bluemke DA, Carragher B, Potel MJ, & Josephs R (1988) Structural analysis of polymers
of sickle cell hemoglobin: II. Sickle hemoglobin macrofibers. Journal of Molecular
Biology 199(2):333-348.
Dykes G, Crepeau RH, & Edelstein SJ (1978) 3-Dimensional reconstruction of fibers of
sickle-cell hemoglobin. Nature 272(5653):506-510.
Dykes GW, Crepeau RH, & Edelstein SJ (1979) Three-dimensional reconstruction of the
14-filament fibers of hemoglobin S. Journal of Molecular Biology 130(4):451-472.
Jones CW, Wang JC, Ferrone FA, Briehl RW, & Turner MS (2003) Interactions between
sickle hemoglobin fibers. Faraday Discussions 123:221-236.
Vekilov PG (2007) Sickle-cell haemoglobin polymerization: is it the primary pathogenic
event of sickle-cell anaemia? British Journal of Haematology 139(2):173-184.
Ferrone FA, Hofrichter J, & Eaton WA (1985) Kinetics of sickle hemoglobin
polymerization : II. A double nucleation mechanism. Journal of Molecular Biology
183(4):611-631.
Ferrone FA, Hofrichter J, Sunshine HR, & Eaton WA (1980) Kinetic studies on
photolysis-induced gelation of sickle cell hemoglobin suggest a new mechanism.
Biophysical Journal 32(1):361-380.
48. Mickols W, Maestre MF, Tinoco I, & Embury SH (1985) Visualization of oriented
hemoglobin S in individual erythrocytes by differential extinction of polarized light.
Proceedings of the National Academy of Sciences of the United States of America
82(19):6527-6531.
50.
51.
52.
49. Mickols WE, et al. (1988) The effect of speed deoxygenation on the percentage of
aligned hemoglobin in sickle cells – Application of differential polarization microscopy.
Journal of Biological Chemistry 263(9):4338-4346.
Aprelev A, et al. (2005) The Effects of Erythrocyte Membranes on the Nucleation of
Sickle Hemoglobin. Biophysical Journal 88(4):2815-2822.
Rottera MA, H.Chub, P. S. Lowb, & Ferronea FA (2010) Band 3 Catalyzes Sickle
Hemoglobin Polymerization. Biophysical Chemistry 146:55-59.
Kaul. D. K., M. E. Fabry, P. Windisch, S. Baez a, & Nagel RL (1983) Erythrocytes in
sickle-cell-anemia are heterogeneous
rheological and hemodynamic
characteristics. The Journal of Clinical Investigation 72(1):22-31.
Galkin O, et al. (2007) Two-Step Mechanism of Homogeneous Nucleation of Sickle Cell
Hemoglobin Polymers. Biophysical Journal 93(3):902-913.
Kaul DK, Fabry ME, & Nagel RL (1996) The pathophysiology of vascular obstruction in
the sickle syndromes. Blood Rev. 10(1):29-44.
Brandao MM, et al. (2003) Optical tweezers for measuring red blood cell elasticity:
application to the study of drug response in sickle cell disease. European Journal of
Haematology 70(4):207-211.
Chien S (1987) Red cell deformability and its relevance to blood flow. Annual Review of
Physiology 49:177-192.
Ballas SK & Mohandas N (1996) PATHOPHYSIOLOGY OF VASO-OCCLUSION.
Hematology/Oncology Clinics of North America 10(6):1221-1239.
their
56.
57.
in
53.
54.
55.
183
Lande WM, et al. (1988) The incidence of painful crisis in homozygous sickle cell
disease: correlation with red cell deformability. Blood 72(6):2056-2059.
Schaeffer JJW, et al. (1999) Depression, disease severity, and sickle cell disease. Journal
of Behavioral Medicine 22(2):115-126.
Abrams DJR, et al. (2005) Prolapse of the antero-superior leaflet of the tricuspid valve
secondary to congenital anomalies of the valvar and sub-valvar apparatus: A rare cause of
severe tricuspid regurgitation. Cardiology in the Young 15(4):417-421.
Adams RJ, Ohene-Frempong K, & Wang W (2001) Sickle Cell and the Brain.
Hematology 2001(1):31-46.
Hillery CA & Panepinto JA (2004) Pathophysiology of stroke in sickle cell disease.
Microcirculation 11(2):195-208.
Adams RJ (2007) Big strokes in small persons. Archives of Neurology 64:1567-1574.
Dichgans M (2007) Genetics of ischaemic stroke. Lancet Neurol. 6(2):149-161.
Kral MC, Brown RT, & Hynd GW (2001) Neuropsychological aspects of pediatric sickle
cell disease. Neuropsychology Review 11(4):179-196.
Prengler M, Pavlakis SG, Prohovnik I, & Adams RJ (2002) Sickle cell disease: The
neurological complications. Ann. Neurol. 51(5):543-552.
Schatz J, McClellan CB, Puffer ES, Johnson K, & Roberts CW
(2008)
Neurodevelopmental Screening in Toddlers and Early Preschoolers With Sickle Cell
Disease. Journal of Child Neurology 23(1):44-50.
Kinney TR, et al. (1999) Silent cerebral infarcts in sickle cell anemia: A risk factor
analysis. Pediatrics 103(3):640-645.
Canham PB (1970) The minimum energy of bending as a possible explanation of the
biconcave shape of the human red blood cell. J. Theoret. Biol 26:61 - 81.
Evans E (1974) Bending resistance and chemically induced moments in membrane
bilayers. Biophysical Journal 14:923-931.
Feng F & Klug WS (2006) Finite element modeling of lipid bilayer membranes. Journal
of Computational Physics 220(1):394-408.
Powers TR, Huber G, & Goldstein RE (2002) Fluid-membrane tethers: Minimal surfaces
and elastic boundary layers. Physical Review E 65(4):041901.
Lipowsky R (1998) Vesicles and biomembranes. Encyclopedia of Applied Physics,
(VCH Publishers, Weinheim and New York ), Vol 23, pp 199-222.
Seifert U (1997) Configurations of fluid membranes and vesicles. Advances in Physics
46(1):13-137.
Smondyrev AM & Berkowitz ML (1999) Molecular dynamics simulation of fluorination
effects on a phospholipid bilayer. Journal of Chemical Physics 111(21):9864-9870.
Feller SE (2000) Molecular dynamics simulations of lipid bilayers. Current Opinion in
Colloid & Interface Science 5(3-4):217-223.
Saiz L, Bandyopadhyay S, & Klein ML (2002) Towards an understanding of complex
biological membranes from atomistic molecular dynamics simulations. Bioscience
Reports 22(2):151-173.
Tieleman DP, Marrink SJ, & Berendsen HJC (1997) A computer perspective of
membranes: molecular dynamics studies of lipid bilayer systems. Biochimica Et
Biophysica Acta-Reviews on Biomembranes 1331(3):235-270.
184
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
Tu KC, Klein ML, & Tobias DJ (1998) Constant-pressure molecular dynamics
investigation of cholesterol effects
in a dipalmitoylphosphatidylcholine bilayer.
Biophysical Journal 75(5):2147-2156.
Noguchi H & Gompper G (2005) Shape Transitions of Fluid Vesicles and Red Blood
Cells in Capillary Flows. Proceedings of the National Academy of Sciences of the United
States of America 102:14159.
Boey SK, Boal DH, & Discher DE (1998) Simulations of the Erythrocyte Cytoskeleton at
Large Deformation. I. Microscopic Models. Biophysical Journal 75(3):1573-1583.
Venturoli M, Sperotto MM, Kranenburg M, & Smit B (2006) Mesoscopic models of
biological membranes. Physics Reports-Review Section of Physics Letters 437(1-2):1-54.
Drouffe JM, Maggs AC, & Leibler S (1991) Computer-Simulations of Self-Assembled
Membranes. Science 254(5036):1353-1356.
Goetz R, Gompper G, & Lipowsky R (1999) Mobility and elasticity of self-assembled
membranes. Physical Review Letters 82(1):221-224.
Kumar PBS, Gompper G, & Lipowsky R (2001) Budding dynamics of multicomponent
membranes. Physical Review Letters 86(17):3911-3914.
Noguchi H & Takasu M (2002) Adhesion of nanoparticles to vesicles: A Brownian
dynamics simulation. Biophysical Journal 83(1):299-308.
Yamamoto S, Maruyama Y, & Hyodo S (2002) Dissipative particle dynamics study of
spontaneous vesicle formation of amphiphilic molecules. Journal of Chemical Physics
116(13):5842-5849.
Farago O (2003) "Water-free" computer model for fluid bilayer membranes. Journal of
Chemical Physics 119(1):596-605.
Hofsass C, Lindahl E, & Edholm O (2003) Molecular dynamics simulations of
phospholipid bilayers with cholesterol. Biophysical Journal 84(4):2192-2206.
90. Marrink SJ & Mark AE (2003) Molecular dynamics simulation of the formation,
structure, and dynamics of small phospholipid vesicles. Journal of the American
Chemical Society 125(49):15233-15242.
Brannigan G, Philips PF, & Brown FLH (2005) Flexible lipid bilayers in implicit solvent.
Physical Review E 72(1):011915.
Cooke IR, Kremer K, & Deserno M (2005) Tunable generic model for fluid bilayer
membranes. Physical Review E 72(1):011506.
Laradji M & Kumar PBS (2005) Domain growth, budding, and fission in phase-
separating self-assembled fluid bilayers. Journal of Chemical Physics 123(22):224902.
92.
91.
93.
94. Markvoort AJ, Pieterse K, Steijaert MN, Spijker P, & Hilbers PAJ (2005) The bilayer-
vesicle transition is entropy driven. Journal of Physical Chemistry B 109(47):22649-
22654.
95. Wang ZJ & Frenkel D (2005) Modeling flexible amphiphilic bilayers: A solvent-free off-
96.
lattice Monte Carlo study. Journal of Chemical Physics 122(23):234711.
Brannigan G, Lin LCL, & Brown FLH (2006) Implicit solvent simulation models for
biomembranes. European Biophysics Journal with Biophysics Letters 35(2):104-124.
97. Markvoort AJ, van Santen RA, & Hilbers PAJ (2006) Vesicle shapes from molecular
98.
dynamics simulations. Journal of Physical Chemistry B 110(45):22780-22785.
Noguchi H & Gompper G (2006) Meshless membrane model based on the moving least-
squares method. Physical Review E 73(2):021903.
185
99. Marrink SJ, Risselada HJ, Yefimov S, Tieleman DP, & de Vries AH (2007) The
MARTINI force field: Coarse grained model for biomolecular simulations. Journal of
Physical Chemistry B 111(27):7812-7824.
100. Muller M, Katsov K, & Schick M (2006) Biological and synthetic membranes: What can
be learned from a coarse-grained description? Physics Reports 434(5-6):113-176.
101. Kohyama T (2009) Simulations of flexible membranes using a coarse-grained particle-
based model with spontaneous curvature variables. Physica a-Statistical Mechanics and
Its Applications 388(17):3334-3344.
102. Espanol P & Warren P (1995) Statistical-Mechanics of Dissipative Particle Dynamics.
Europhys. Lett. 30(4):191-196.
103. Hoogerbrugge PJ & Koelman J (1992) Simulating Microscopic Hydrodynamic
Phenomena with Dissipative Particle Dynamics. Europhys. Lett. 19(3):155-160.
104. Grafmuller A, Shillcock J, & Lipowsky R (2009) Dissipative particle dynamics of
tension-induced membrane fusion. Molecular Simulation 35(7):554-560.
105. Shillcock JC & Lipowsky R (2002) Equilibrium structure and lateral stress distribution of
amphiphilic bilayers from dissipative particle dynamics simulations. Journal of Chemical
Physics 117:5048-5061.
106. Groot RD & Warren PB (1997) Dissipative particle dynamics: Bridging the gap between
atomistic and mesoscopic simulation. The Journal of Chemical Physics 107(11):4423-
4435.
107. Brannigan G & Brown FLH (2004) Solvent-free simulations of fluid membrane bilayers.
The Journal of Chemical Physics 120(2):1059-1071.
108. Yuan H, Huang C, Li J, Lykotrafitis G, & Zhang S (2010) One-particle-thick, solvent-
free, coarse-grained model for biological and biomimetic fluid membranes. Physical
Review E 82(1):011905.
109. Li J, Lykotrafitis G, Dao M, & Suresh S (2007) Cytoskeletal dynamics of human
erythrocyte. Proceedings of the National Academy of Sciences of the United States of
America 104(12):4937-4942.
110. Pivkin IV & Karniadakis GE (2008) Accurate Coarse-Grained Modeling of Red Blood
Cells. Physical Review Letters 101(11):118105.
111. Li H & Lykotrafitis G (2012) Two-Component Coarse-Grained Molecular-Dynamics
Model for the Human Erythrocyte Membrane. Biophysical Journal 102(1):75-84.
112. Kodippili GC, et al. (2009) Imaging of the diffusion of single band 3 molecules on
normal and mutant erythrocytes. Blood 113:6237-6245.
113. N. Gov, J. Cluitmans, P. Sens, & Bosman GJCGM (2009) Cytoskeletal Control of Red
Blood Cell Shape: Theory and Practice of Vesicle Formation. Advances in Planar Lipid
Bilayers and Liposomes, (Elsevier Inc, Burlington), Vol 10, pp 95-119.
114. Schindler M, Koppel DE, & Sheetz. MP (1980) Modulation of membrane protein lateral
mobility by polyphosphates and polyamines. Proceedings of the National Academy of
Sciences of the United States of America 77:1457-1461.
115. Corbett JD, Agre P, Palek J, & Golan DE (1994) Differential control of band 3 lateral and
rotational mobility in intact red cells. The journal of Clinical Investigation 94(2):683-688.
116. Saxton MJ (1989) The spectrin network as a barrier to lateral diffusion in erythrocytes. A
percolation analysis. Biophysical Journal 55(1):21-28.
117. Boal DH (1994) Computer simulation of a model network for the erythrocyte
cytoskeleton. Biophys. J. 67:521–529. Biophysical Journal 67:521-529.
186
118. Auth T & Gov NS (2009) Diffusion in a Fluid Membrane with a Flexible Cortical
Cytoskeleton. Biophysical Journal 96(3):818-830.
119. McIntosh TJ & Simon SA (2006) Roles of Bilayer Material Properties in Function and
Distribution of Membrane Proteins. Annual Review of Biophysics and Biomolecular
Structure 35:177-198.
120. Discher DE, Boal DH, & Boey SK (1998) Simulations of the Erythrocyte Cytoskeleton at
Large Deformation. II. Micropipette Aspiration. Biophysical Journal 75(3):1584-1597.
121. Marrink SJ & Mark AE (2003) The mechanism of vesicle fusion as revealed by
the American Chemical Society
molecular dynamics simulations. Journal of
125(37):11144-11145.
122. Byers TJ & Branton D (1985) Visualization ofthe protein associations in the erythrocyte
membrane skeleton. Proceedings of the National Academy of Sciences, pp 6153-6157.
123. Shen BW, Josephs R, & Steck T, L. (1986) Steck TL. Ultrastructure of the intact skeleton
of the human erythrocyte membrane. Journal of Cell Biology 102:997-1006.
124. Bustamante C, Bryant Z, & Smith SB (2003) Ten years of tension: single-molecule DNA
mechanics. Nature 421(6921):423-427.
125. Marko JF, Siggia, E. D. (1995) Stretching DNA. Micromolecules 28:8759-8770.
126. Boal D (2002) Mechanics of the cell (Cambridge University Press, Cambridge, United
Kingdom).
127. Strey H, Peterson M, & Sackmann E (1995) Measurement of erythrocyte membrane
elasticity by flicker eigenmode decomposition. Biophysical Journal 69(2):478-488.
128. Gov N, Zilman AG, & Safran S (2003) Cytoskeleton Confinement and Tension of Red
Blood Cell Membranes. Physical Review Letters 90(22):228101.
129. Seifert U & Lipowsky R (1995) Morphology of Vesicles. Structure and dynamics of
membranes: From cells to vesicles, ed Hoff AJ (Elsevier, Amsterdam), Vol 1.
130. Allen M & Tildesley D (1987) Computer Simulations of Liquids (Clarendon Press, New
131.
York).
Imparato A, Shillcock JC, & Lipowsky R (2005) Shape fluctuations and elastic properties
of two-component bilayer membranes. Europhysics Letters 69:650-656.
132. Goetz R & Lipowsky R (1998) Computer simulations of bilayer membranes: Self-
assembly and interfacial tension. Journal of Chemical Physics 108:7397-7410.
133. Fahey PF & Webb WW (1978) Lateral diffusion in phospholipid bilayer membranes and
multilamellar liquid crystals. Biochemistry 17(15):3046-3053.
134. Nelson D, Piran T, & Weinberg S (2004) Statistical Mechanics of Membranes and
Surfaces (World Scientific, New Jersey) 2nd Ed.
135. Cooke IR & Deserno M (2005) Solvent-free model for self-assembling fluid bilayer
membranes: Stabilization of the fluid phase based on broad attractive tail potentials.
Journal of Chemical Physics 123:224710.
136. Mills JP, Qie L, Dao M, Lim CT, & Suresh S (2004) Nonlinear elastic and viscoelastic
deformation of the human red blood cell with optical tweezers. Mechanics and Chemistry
of Biosystems 1:169-180.
137. Liu SC, L. H. Derick, & Palek J (1987) Visualization of the hexagonal lattice in the
erythrocyte-membrane skeleton. The Journal of Cell Biology 104:527-536.
138. Liu SC, L. H. Derick, P. Agre, & Palek. J (1990) Alteration of the erythrocyte-membrane
in hereditary spherocytosis,hereditary elliptocytosis, and
skeletal ultrastructure
yropoikilocytosis. Blood 76:198-205.
187
139. Liu FJ, Burgess H, Mizukami, & Ostafin A (2003) Sample preparation and imaging of
erythrocyte cytoskeleton with the atomic force microscopy. Cell Biochemistry and
Biophysics 38:251-270.
140. Park YK, et al. (2010) Metabolic remodeling of the human red blood cell membrane.
Proceedings of the National Academy of Sciences pp 1289-1294.
141. Dao M, Li J, & Suresh S (2006) Molecularly based analysis of deformation of spectrin
network and human erythrocyte. Materials Science and Engineering: C 26(8):1232-1244.
142. Li J, Dao M, Lim CT, & Suresh S (2005) Spectrin-level modeling of the cytoskeleton and
optical tweezers stretching of the erythrocyte. Biophysical Journal 88(5):3707-3719.
143. Swihart AH, Mikrut JM, Ketterson JB, & MacDonald RC (2001) Atomic force
microscopy of the erythrocyte membrane skeleton. Journal of Microscopy 204:212-225.
144. Takeuchi M, Miyamoto H, Sako Y, & Kusumi A (1998) Structure of the erythrocyte
membrane skeleton as observed by atomic force microscopy. Biophysical Journal
74:2171-2183.
145. An X & Mohandas N (2008) Disorders of red cell membrane. British Journal of
Haematology 141(3):367-375.
146. Delaunay J (2007) The molecular basis of hereditary red cell membrane disorders. Blood
Reviews 21(1):1-20.
147. Sheetz MP, Schindler M, & Koppel DE (1980) Lateral mobility of integral membrane
proteins is increased in spherocytic erythrocytes. Nature 285(5765):510-512.
148. Tsuji A & Ohnishi S (1986) Restriction of the lateral motion of band 3 in the erythrocyte
membrane by the cytoskeletal network: dependence on spectrin association state.
Biochemistry 25(20):6133-6139.
149. Tsuji A, Kawasaki K, Ohnishi S, Merkle H, & Kusumi A (1988) Regulation of band 3
mobilities in erythrocyte ghost membranes by protein association and cytoskeletal
meshwork. Biochemistry 27(19):7447-7452.
150. Kusumi A & Sako Y (1996) Cell surface organization by the membrane skeleton.
Current Opinion in Cell Biology 8(4):566-574.
151. Gov NS & Safran SA (2005) Red blood cell membrane fluctuations and shape controlled
by ATP-induced cytoskeletal defects. Biophysical Journal 88(3):1859-1874.
152. Helfrich W (1975) Out-of-plane fluctuations of
lipid bilayers. Zeitschrift Fur
Naturforschung C-a Journal of Biosciences 30(6):841-842.
153. Bennett WFD & Tieleman DP (2013) Computer simulations of lipid membrane domains.
Biochimica et Biophysica Acta (BBA) - Biomembranes 1828(8):1765-1776.
154. Peng Z, et al. (2013) Lipid bilayer and cytoskeletal interactions in a red blood cell.
Proceedings of the National Academy of Sciences.
155. Auth T, Safran SA, & Gov NS (2007) Filament Networks Attached to Membranes:
Cytoskeletal Pressure and Local Bilayer Deformation. New Journal of Physics 9:430.
156. Zeman K, Engelhard H, & Sackmann E (1990) Bending undulations and elasticity of the
erythrocyte membrane: effects of cell shape and membrane organization. European
Biophysics Journal 18(4):203-219.
157. Boal D (2012) Mechanics of the cell (Cambridge University Press, Cambridge).
158. Mombers C, De Gier J, Demel RA, & van Deenen LLM (1980) Spectrin-phospholipid
interaction: A monolayer study. Biochimica et Biophysica Acta (BBA) - Biomembranes
603(1):52-62.
188
159. Bonnet D & Begard E (1984) Interaction of anilinonaphtyl labeled spectrin with fatty
acids and phospholipids: A fluorescence study. Biochemical and Biophysical Research
Communications 120(2):344-350.
160. Takeshita K, Macdonald RI, & Macdonald RC (1993) Band 4.1 Enhances Spectrin
to Phosphatidylserine Vesicles. Biochemical and Biophysical Research
Binding
Communications 191(1):165-171.
161. MacDonald RI (1993) Temperature and ionic effects on the interaction of erythroid
spectrin with phosphatidylserine membranes. Biochemistry 32(27):6957-6964.
162. Bitbol M, Dempsey C, Watts A, & Devaux PF (1989) Weak interaction of spectrin with
phosphatidylcholine-phosphatidylserine multilayers: A 2H and 31P NMR study. FEBS
Letters 244(1):217-222.
163. Bialkowska K, Zembroń A, & Sikorski AF (1994) Ankyrin inhibits binding of
erythrocyte spectrin to phospholipid vesicles. Biochimica et Biophysica Acta (BBA) -
Biomembranes 1191(1):21-26.
164. Ohanian V, et al. (1984) Analysis of the ternary interaction of the red cell membrane
skeletal proteins, spectrin, actin, and 4.1. Biochemistry 23(19):4416-4420.
165. Nakao M (1990) Erythroid Cells (Blood Cell Biochemistry) (Springer) 1 Ed.
166. Wang DN, V E Sarabia, Reithmeier RA, & Kühlbrandt W (1994) Three-dimensional map
of the dimeric membrane domain of the human erythrocyte anion exchanger, Band 3. The
EMBO Journal 13(14):3230-3235.
167. Zhang D, A. Kiyatkin, J. T. Bolin, & Low PS (2000) Crystallographic structure and
functional interpretation of the cytoplasmic domain of erythrocyte membrane band 3.
Blood 96:2925-2933.
168. Anong W, A.,, Weis T, L., , & Low PS (2006) Rate of Rupture and Reattachment of the
Band 3-Ankyrin Bridge on the Human Erythrocyte Membrane. The Journal of Biological
Chemistry 281:22360-22366.
169. Betz T, Lenz M, Joanny J-F, & Sykes C (2009) ATP-dependent mechanics of red blood
cells. Proceedings of the National Academy of Sciences.
170. Gov NS (2007) Active elastic network: Cytoskeleton of the red blood cell. Physical
Review E 75(1):011921.
171. Hochmuth RM & Waugh RE (1987) Erythrocyte membrane elasticity and viscosity.
Annu Rev Physiol 49:209-219.
172. Cheung KS & Yip S (1991) Atomic-level stress in an inhomogeneous system. Journal of
Applied Physics 70(10):5688-5690.
173. Fujiwara T, Ritchie K, Murakoshi H, Jacobson K, & Kusumi A (2002) Phospholipids
undergo hop diffusion in compartmentalized cell membrane The Journal of Cell Biology
157:1071-1082.
174. Saxton MJ (1990) The membrane skeleton of erythrocytes. A percolation model.
Biophysical Journal 57(6):1167-1177.
175. Saxton MJ (1990) The membrane skeleton of erythrocytes: models of its effect on lateral
diffusion. International Journal of Biochemistry 22(8):801-809.
176. Saxton MJ (1995) Single-particle tracking: effects of corrals. Biophysical Journal
69(2):389-398.
177. Boal DH & Boey SK (1995) Barrier-free paths of directed protein motion in the
erythrocyte plasma membrane. Biophysical Journal 69(2):372-379.
189
178. Tomishige M, Sako Y, & Kusumi A (1998) Regulation Mechanism of the Lateral
Diffusion of Band 3 in Erythrocyte Membranes by the Membrane Skeleton. The Journal
of Cell Biology 142(4):989-1000.
179. Lamb TD (1996) Gain and kinetics of activation in the G-protein cascade of
phototransduction. Proc Natl Acad Sci U S A 93:566-570.
180. Chan. PY, et al. (1991) Influence of receptor lateral mobility on adhesion strengthening
between membranes containing LFA-3 and CD2. J Cell Biol 115:245-255.
181. Dustin ML, Ferguson LM, Chan PY, Springer TA, & Golan DE (1996) Visualization of
CD2 interaction with LFA-3 and determination of the two-dimensional dissociation
constant for adhesion receptors in a contact area. J Cell Biol 132:465-474.
182. Saxton MJ (1999) Lateral diffusion of lipids and proteins. Curr. Top. Membr. 48:229-282.
183. Singer SJ & Nicolson GL (1972) The fluid mosaic model of the structure of cell
184.
membranes. Science 175(4023):720-731.
Jovin TM & Vaz WLC (1989) [27] Rotational and translational diffusion in membranes
measured by fluorescence and phosphorescence methods. Methods in Enzymology, ed
Sidney Fleischer BF (Academic Press), Vol Volume 172, pp 471-513.
185. Feder TJ, Brust-Mascher I, Slattery JP, Baird B, & Webb WW (1996) Constrained
diffusion or immobile fraction on cell surfaces: a new interpretation. Biophysical Journal
70(6):2767-2773.
186. Kenworthy AK, et al. (2004) Dynamics of putative raft-associated proteins at the cell
surface. J.Cell Biol 165:735-746.
187. Murase K, et al. (2004) Ultrafine Membrane Compartments for Molecular Diffusion as
Revealed by Single Molecule Techniques. Biophysical Journal 86(6):4075-4093.
188. Kusumi A, Ike H, Nakada C, Murase K, & Fujiwara T (2005) Single-molecule tracking
of membrane molecules: plasma membrane compartmentalization and dynamic assembly
of raft-philic signaling molecules. Seminars in Immunology 17(1):3-21.
189. Douglass AD & Vale RD (2005) Single-Molecule Microscopy Reveals Plasma
Membrane Microdomains Created by Protein-Protein Networks that Exclude or Trap
Signaling Molecules in T Cells. Cell 121(6):937-950.
190. Saxton MJ & Jacobson K (1997) SINGLE-PARTICLE TRACKING: Applications to
Membrane Dynamics. Annual Review of Biophysics & Biomolecular Structure 26(1):373.
191. Saxton MJ (1990) Lateral diffusion in a mixture of mobile and immobile particles. A
Monte Carlo study. Biophysical Journal 58(5):1303-1306.
192. Saxton MJ (1994) Anomalous diffusion due to obstacles: a Monte Carlo study.
Biophysical Journal 66(2):394-401.
193. Saxton MJ (1996) Anomalous diffusion due to binding: a Monte Carlo study. Biophysical
Journal 70(3):1250-1262.
194. Bates IR, et al. (2006) Membrane Lateral Diffusion and Capture of CFTR within
Transient Confinement Zones. Biophysical Journal 91(3):1046-1058.
195. Nicolau Jr DV, Hancock JF, & Burrage K (2007) Sources of Anomalous Diffusion on
Cell Membranes: A Monte Carlo Study. Biophysical Journal 92(6):1975-1987.
196. Kusumi A, Sako Y, & Yamamoto M (1993) Confined lateral diffusion of membrane
receptors as studied by single particle tracking (nanovid microscopy). Effects of calcium-
induced differentiation in cultured epithelial cells. Biophysical Journal 65(5):2021-2040.
190
197. Cho MR, Eber SW, Liu S-C, Lux SE, & Golan DE (1998) Regulation of Band 3
Rotational Mobility by Ankyrin in Intact Human Red Cells†. Biochemistry 37(51):17828-
17835.
198. Brown FLH, Leitner DM, McCammon JA, & Wilson KR (2000) Lateral Diffusion of
Membrane Proteins in the Presence of Static and Dynamic Corrals: Suggestions for
Appropriate Observables. Biophysical Journal 78(5):2257-2269.
199. Brown FLH (2003) Regulation of Protein Mobility via Thermal Membrane Undulations.
Biophysical Journal 84(2):842-853.
200. Kenkre VM, Giuggioli L, & Kalay Z (2008) Molecular motion in cell membranes:
Analytic study of fence-hindered random walks. Physical Review E 77(5):051907.
201. Brown FLH (2008) Elastic Modeling of Biomembranes and Lipid Bilayers. Annual
Review of Physical Chemistry 59:685-712.
202. Cohen AM, Liu SC, Derick LH, & Palek J (1986) Ultrastructural studies of the
interaction of spectrin with phosphatidylserine liposomes. Blood 68:920-926.
203. Maksymiw R, Sui SF, Gaub H, & Sackmann E (1987) Electrostatic coupling of spectrin
dimers to phosphatidylserine containing lipid lamellae. Biochemistry 26(11):2983-2990.
204. An X, et al. (2003) Phosphatidylserine Binding Sites in Erythroid Spectrin: Location and
Implications for Membrane Stability†. Biochemistry 43(2):310-315.
205. Grzybek M, et al. (2006) Spectrin–phospholipid interactions: Existence of multiple kinds
of binding sites? Chemistry and Physics of Lipids 141(1–2):133-141.
206. Lux SE (1999) Review: Red Blood Cell Membrane Disorders. British Journal of
Haematology, 104:2-13.
207. Lux SE, et al. (1990) Hereditary spherocytosis associated with deletion of human
erythrocyte ankyrin gene on chromosome 8. Nature 345(6277):736-739.
208. Kodippili GC, et al. (2012) Analysis of the mobilities of band 3 populations associated
with ankyrin protein and junctional complexes in intact murine erythrocytes. J Biol Chem.
287:4129-4138.
209. Sheetz MP & Casaly J (1980) 2,3-Diphosphoglycerate and ATP dissociate erythrocyte-
membrane skeletons. Journal of Biological Chemistry 255:9955-9960.
210. Li H, Zhang Y, Ha V, & Lykotrafitis G (2016) Modeling of band-3 protein diffusion in
the normal and defective red blood cell membrane. Soft matter 12(15):3643-3653.
211. Li H & Lykotrafitis G (2014) Erythrocyte membrane model with explicit description of
the lipid bilayer and the spectrin network. Biophysical Journal 107:1-12.
212. McQuarrie DA (1976) Statistical Mechanics (Harper & Row, New York).
213. Suresh S, et al. (2005) Connections between single-cell biomechanics and human disease
states: gastrointestinal cancer and malaria. Acta Biomaterialia 1(1):15-30.
214. Golan DE & Veatch. W (1980) Lateral mobility of band 3 in the human erythrocyte
membrane studied by fluorescence photobleaching recovery: evidence for control by
cytoskeletal interactions. Proceedings of the National Academy of Sciences of the United
States of America 77:2537-2541.
215. Ritchie K, Iino R, Fujiwara T, Murase K, & A. K (2003) The fence and picket structure
of the plasma membrane of live cells as revealed by single molecule techniques
Molecular Membrane Biology 20(1):13-18.
216. Saxton MJ (1987) Lateral diffusion in an archipelago. The effect of mobile obstacles.
Biophysical Journal 52(6):989-997.
191
217. Saxton MJ (1993) Lateral diffusion in an archipelago. Single-particle diffusion.
218.
Biophysical Journal 64(6):1766-1780.
Javanainen M, et al. (2013) Anomalous and normal diffusion of proteins and lipids in
crowded lipid membranes. Faraday Discuss. 161:397-417.
219. Saxton MJ (2007) A Biological Interpretation of Transient Anomalous Subdiffusion. I.
Qualitative Model. Biophysical Journal 92(4):1178-1191.
220. Forstner MB, Martin DS, Rückerl F, Käs JA, & Selle C (2008) Attractive membrane
domains control lateral diffusion. Physical Review E 77(5):051906.
221. Lillemeier BF, J. R. Pfeiffer, Z. Surviladze, B. S. Wilson, & M.Davis. M (2006) Plasma
membrane-associated proteins are clustered into islands attached to the cytoskeleton.
proceedings of the National Academy of Sciences 103:18992-18997.
222. Essam JW (1972) Percolation and cluster size. In Phase Transitions and Critical
Phenomena, (Academic Press, London and New York).
223. Willekens FLA, et al. (2003) Hemoglobin loss from erythrocytes in vivo results from
spleen-facilitated vesiculation pp 747-751.
224. Willekens FLA, et al. (2008) Erythrocyte vesiculation: a self-protective mechanism?
British Journal of Haematology 141(4):549-556.
225. Xiong Z, Oriss TB, Cavaretta JP, Rosengart MR, & Lee JS (2012) Red cell microparticle
enumeration: validation of a flow cytometric approach. Vox Sanguinis 103(1):42-48.
226. Barteneva NS, et al. (2013) Circulating microparticles: square the circle. BMC Cell Biol.
14:23.
227. Roux A, et al. (2005) Role of curvature and phase transition in lipid sorting and fission of
membrane tubules. EMBO J. 24:1537.
228. Tsai F-C & Chen H-Y (2008) Adsorption-induced vesicle fission. Physical Review E
78(5):051906.
229. Gov N, J. Cluitmans, P. Sens, & Bosman GJCGM (2009) Cytoskeletal Control of Red
Blood Cell Shape: Theory and Practice of Vesicle Formation. Advances in Planar Lipid
Bilayers and Liposomes, (Elsevier Inc, Burlington), Vol 10, pp 95-119.
230. Liu SC, L. H D, M. A D, & J P (1989) Separation of the lipid bilayer from the membrane
skeleton during discocyte-echinocyte transformation of human erythrocyte ghosts. Eur J
Cell Biol 49:358-365.
231. Fricke K & Sackmann E (1984) Variation of frequency spectrum of the erythrocyte
flickering caused by aging, osmolarity, temperature and pathological changes. Biochim
Biophys Acta 803(3):145-152.
232. Allan D, Billah MM, Finean JB, & Michell RH (1976) Release of diacylglycerol-
enriched vesicles from erythrocytes with increased intracellular [Ca2+]. Nature
261(5555):58-60.
233. Elgsaeter A, Shotton DM, & Branton D (1976) Intramembrane particle aggregation in
erythrocyte ghosts. II. The influence of spectrin aggregation. Biochimica et Biophysica
Acta (BBA) - Biomembranes 426(1):101-122.
234. Lutz HU, Liu SC, & Palek J (1977) Release of spectrin-free vesicles from human
erythrocytes during ATP depletion. I. Characterization of spectrin-free vesicles. J. Cell
Biol. 73:548-560.
235. Hägerstrand H & Isomaa B (1994) Lipid and protein composition of exovesicles released
from human erythrocytes following treatment with amphiphiles. Biochim Biophys Acta.
1190(2):409-415.
192
236. Alaarg A, Schiffelers R, van Solinge WW, & Van Wijk R (2013) Red blood cell
vesiculation in hereditary hemolytic anemia. Frontiers in Physiology 4.
237. Gallagher PG (2004) Hereditary elliptocytosis: spectrin and protein 4.1R. Seminars in
Hematology 41:142-164.
238. Goodman SR, Shiffer KA, Casoria LA, & Eyster ME (1982) Identification of the
molecular defect in the erythrocyte membrane skeleton of some kindreds with hereditary
spherocytosis. Blood 60(3):772-784.
239. Liu S-C, Palek J, & Prchal TJ (1982) Defective spectrin dimer-dimer association with
hereditary elliptocytosis. Proceedings of the National Academy of Sciences of the United
States of America 79(6):2072-2076.
240. Palek J (1985) Hereditary elliptocytosis and related disorders. Clinics in haematology
241.
1:45-87.
Iglic A, S. S, & B. Z (1995) Depletion of Membrane Skeleton in Red Blood Cell Vesicles.
Biophysical Journal 69:274-279.
242. Reliene R, et al. (2002) Splenectomy prolongs in vivo survival of erythrocytes differently
in spectrin/ankyrin- and band 3–deficient hereditary spherocytosis pp 2208-2215.
243. Bevers EM, et al. (1992) Defective Ca(2+)-induced microvesiculation and deficient
expression of proco-agulant activity in erythrocytes from a patient with a bleeding
disorder: a study of the red blood cells of scott syndrome. Blood 79:380-388.
244. Greenwalt TJ (2006) The how and why of exocytic vesicles. Transfusion 46(1):143-152.
245. Salzer U, et al. (2008) Vesicles generated during storage of red cells are rich in the lipid
246.
raft marker stomatin. Transfusion 48(3):451-462.
de Vooght KM, et al. (2013) Extracellular vesicles in the circulation: are erythrocyte
microvesicles a confounder in the plasma haemoglobin assay? Biochem Soc Trans.
41(1):288-292.
247. Tinevez J-Y, et al. (2009) Role of cortical tension in bleb growth. Proceedings of the
National Academy of Sciences 106(44):18581-18586.
248. Young J & Mitran S (2010) A numerical model of cellular blebbing: A volume-
conserving, fluid–structure interaction model of the entire cell. Journal of Biomechanics
43(2):210-220.
249. Sheetz MP & Singer SJ (1974) Biological Membranes as Bilayer Couples. A Molecular
Mechanism of Drug-Erythrocyte Interactions. Proceedings of the National Academy of
Sciences 71(11):4457-4461.
250. Sheet MP & Singer SJ (1974) Biological membranes as bilayer couples. A molecular
mechanism of drug-erythrocyte interactions. Proc Natl Acad Sci U S A 71:4457-4461.
251. Miao L, Seifert U, Wortis M, & Döbereiner HG (1994) Budding transitions of fluid-
bilayer vesicles-the effect of area-difference elasticity. Physical Review E 49:5389-5407.
252. Waugh RE (1996) Elastic energy of curvature-driven bump formation on red blood cell
253.
membrane. Biophysical Journal 70(2):1027-1035.
Iglič A (1997) A possible mechanism determining the stability of spiculated red blood
cells. Journal of Biomechanics 30(1):35-40.
254. Mukhopadhyay R, Gerald Lim HW, & Wortis M (2002) Echinocyte Shapes: Bending,
Stretching, and Shear Determine Spicule Shape and Spacing. Biophysical Journal
82(4):1756-1772.
255. Reynwar BJ, et al. (2007) Aggregation and vesiculation of membrane proteins by
curvature-mediated interactions. Nature 447(7143):461-464.
193
256. Blood PD & Voth GA (2006) Direct observation of Bin/amphiphysin/Rvs (BAR)
domain-induced membrane curvature by means of molecular dynamics simulations.
Proceedings of the National Academy of Sciences 103(41):15068-15072.
257. Zimmerberg J & Kozlov MM (2006) How proteins produce cellular membrane curvature.
Nat Rev Mol Cell Biol 7(1):9-19.
258. Spangler EJ, Harvey CW, Revalee JD, Kumar PBS, & Laradji M (2011) Computer
simulation of cytoskeleton-induced blebbing in lipid membranes. Physical Review E
84(5):051906.
259. Edidin M (2003) The state of lipid rafts: From model membranes to cells. Annual Review
of Biophysics and Biomolecular Structure 32(32):257-283.
260. Simons K & W.L.C. Vaz (2004) Model systems, lipid rafts, and cell membranes. Annual
261.
Review of Biophysics and Biomolecular Structure 33:269-295.
Jacobson K, Mouritsen OG, & Anderson RGW (2007) Lipid rafts: at a crossroad between
cell biology and physics. Nat Cell Biol 9(1):7-14.
262. Kamata K, Manno S, Ozaki M, & Takakuwa Y (2008) Functional evidence for presence
of lipid rafts in erythrocyte membranes: Gsα in rafts is essential for signal transduction.
American Journal of Hematology 83(5):371-375.
263. Ursell TS, Klug WS, & Phillips R (2009) Morphology and interaction between lipid
domains. Proceedings of the National Academy of Sciences 106(32):13301-13306.
264. Yuan H, Huang C, & Zhang S (2011) Membrane-Mediated Inter-Domain Interactions.
BioNanoSci. 1(3):97-102.
265. Simons K & Sampaio JL (2011) Membrane organization and lipid rafts. Cold Spring
Harb Perspect Biol 3(10):a004697.
266. Li H & Lykotrafitis G (2015) Vesiculation of healthy and defective red blood cells.
Physical Review E 92(1):012715.
267. Sunil Kumar PB, Gompper G, & Lipowsky R (2001) Budding Dynamics of
Multicomponent Membranes. Physical Review Letters 86(17):3911-3914.
268. Laradji M & Sunil Kumar PB (2004) Dynamics of Domain Growth in Self-Assembled
Fluid Vesicles. Physical Review Letters 93(19):198105.
269. Laradji M & Sunil Kumar PB (2005) Domain growth, budding, and fission in phase-
separating self-assembled fluid bilayers. The Journal of Chemical Physics 123(22):-.
270. Laradji M & Kumar PB (2006) Anomalously slow domain growth in fluid membranes
with asymmetric transbilayer lipid distribution. Physical Review E 73:040901.
271. Yanagisawa M, Imai M, Masui T, Komura S, & Ohta T (2007) Growth Dynamics of
Domains in Ternary Fluid Vesicles. Biophysical Journal 92(1):115-125.
272. Ramachandran S, M. Laradji, & Kumar PBS (2009) Lateral Organization of Lipids in
Multi-component Liposomes. Journal of the Physical Society of Japan 78(4):041006.
273. Tuvia S, S. Levin, A. Bitler, & Korenstein R (1998) Mechanical Fluctuations of the
Membrane–Skeleton Are Dependent on F-Actin ATPase in Human Erythrocytes. J. Cell
Biol. 141:1551.
274. Backman L, Jonasson JB, & Hörsted P (1998) Phosphoinositide metabolism and shape
control in sheep red blood cells. Molecular Membrane Biology 15(1):27-32.
275. Sheetz MP, Sable JE, & Döbereiner H-G (2006) CONTINUOUS MEMBRANE-
CYTOSKELETON ADHESION REQUIRES CONTINUOUS ACCOMMODATION
TO LIPID AND CYTOSKELETON DYNAMICS. Annual Review of Biophysics and
Biomolecular Structure 35(1):417-434.
194
276. Evans EA & Skalak R (1980) Mechanics and thermodynamics of biomembranes (CRC
Press, 1980) p 254.
277. Westerman M, et al. (2008) Microvesicles in haemoglobinopathies offer insights into
mechanisms of hypercoagulability, haemolysis and the effects of therapy. British Journal
of Haematology 142(1):126-135.
278. Knowles DW, Tilley L, Mohandas N, & Chasis JA (1997) Erythrocyte membrane
vesiculation: Model for the molecular mechanism of protein sorting. Proceedings of the
National Academy of Sciences 94(24):12969-12974.
279. Embury SH (2004) The Not-So-Simple Process of Sickle Cell Vasoocclusion.
Microcirculation 11(2):101 - 113.
280. Routhieaux J, Sarcone S, & Stegenga K (2005) Neurocognitive Sequelae of Sickle Cell
Disease: Current Issues and Future Directions. Journal of Pediatric Oncology Nursing
22(3):160-167.
281. Zennadi R, Chien A, Xu K, Batchvarova M, & Telen MJ (2008) Sickle red cells induce
adhesion of lymphocytes and monocytes to endothelium. Blood:blood-2008-2001-134346.
282. Zermann DH, Lindner H, Huschke T, & Schubert J (1997) Diagnostic value of natural fill
cystometry in neurogenic bladder in children. Eur Urol 32(2):223-228.
283. Christoph GW, Hofrichter J, & Eaton WA (2005) Understanding the shape of sickled red
cells. Biophys J 88(2):1371-1376.
284. Carragher B, Bluemke DA, Gabriel B, Potel MJ, & Josephs R (1988) Structural analysis
of polymers of sickle cell hemoglobin. I. Sickle hemoglobin fibers. J Mol Biol
199(2):315-331.
285. Carragher B, et al. (1988) Structural analysis of polymers of sickle cell hemoglobin. III.
Fibers within fascicles. J Mol Biol 199(2):383-388.
286. Bluemke DA, Carragher B, Potel MJ, & Josephs R (1988) Structural analysis of polymers
of sickle cell hemoglobin. II. Sickle hemoglobin macrofibers. J Mol Biol 199(2):333-348.
287. Turner MS, Briehl RW, Ferrone FA, & Josephs R (2003) Twisted Protein Aggregates and
Disease: The Stability of Sickle Hemoglobin Fibers. Physical Review Letters
90(12):128103.
288. Ferrone FA, Hofrichter J, Sunshine HR, & Eaton WA (1980) Kinetic-studies on
photolysis-induced gelation of sickle-cell hemoglobin suggest a new mechanism.
Biophysical Journal 32(1):361-380.
289. Mickols W, Maestre MF, Tinoco I, & Embury SH (1985) Visualization of oriented
hemoglobin-S in individual erythrocytes by differential extinction of polarized-light.
Proceedings of the National Academy of Sciences of the United States of America
82(19):6527-6531.
290. Wang JC, et al. (2002) Micromechanics of isolated sickle cell hemoglobin fibers:
bending moduli and persistence lengths. Journal of Molecular Biology 315(4):601-612.
291. Lewis MR, Gross LJ, & Josephs R (1994) Cryoelectron Microscopy of Deoxy-Sickle
Hemoglobin Fibers. Microsc Res Techniq 27(5):459-467.
292. Turner MS, Briehl RW, Wang JC, Ferrone FA, & Josephs R (2006) Anisotropy in Sickle
Hemoglobin Fibers from Variations in Bending and Twist. Journal of Molecular Biology
357(5):1422-1427.
293. Knowles TP, et al. (2007) Role of intermolecular forces in defining material properties of
protein nanofibrils. Science 318(5858):1900-1903.
195
294. Briehl RW & Guzman AE (1994) Fragility and structure of hemoglobin S fibers and gels
and their consequences for gelation kinetics and rheology. in Blood, pp 573-579.
295. Prabhakaran M & Michael EJ (1993) Molecular dynamics of sickle and normal
hemoglobins. Biopolymers 33(5):735-742.
296. Roufberg A & Ferrone FA (2000) A model for the sickle hemoglobin fiber using both
mutation sites. Protein Science 9(5):1031-1034.
297. Kim T, Hwang W, & Kamm RD (2009) Computational Analysis of a Cross-linked Actin-
like Network. Experimental Mechanics 49(1):91-104.
298. Kim T, Hwang W, Lee H, & Kamm RD (2009) Computational Analysis of Viscoelastic
Properties of Crosslinked Actin Networks. Plos Comput Biol 5(7):1-12.
299. Berendsen HJC (2010) Concluding remarks. Faraday Discussions 144:467-481.
300. Ayton GS, Lyman E, & Voth GA (2010) Hierarchical coarse-graining strategy for
protein-membrane systems to access mesoscopic scales. Faraday Discussions 144:347-
357.
301. Ayton GS, Noid WG, & Voth GA (2007) Systematic coarse graining of biomolecular and
soft-matter systems. Mrs Bulletin 32(11):929-934.
302. Buehler MJ & Yung YC (2009) Deformation and failure of protein materials in
physiologically extreme conditions and disease. Nature Materials 8(3):175-188.
303. Das A & Andersen HC (2010) The multiscale coarse-graining method. V. Isothermal-
isobaric ensemble. Journal of Chemical Physics 132(16):164106.
304. Reif F ed (1965) Fundamentals of Statistical and Thermal Physics (McGraw-Hill, New
York, NY.).
305. Underhill PT & Doyle PS (2004) On the coarse-graining of polymers into bead-spring
chains. Journal of Non-Newtonian Fluid Mechanics 122(1-3):3-31.
306. Berendsen HJC, Postma JPM, Van Gunsteren WF, DiNola A, & Haak JR (1984)
Molecular dynamics with coupling to an external bath. Journal of Chemical Physics
81:3684 - 3690.
307. Kasas S, et al. (2004) Mechanical properties of microtubules explored using the finite
elements method. Chemphyschem 5(2):252-257.
308. Groot RD & Warren PB (1997) Dissipative particle dynamics: Bridging the gap between
atomistic and mesoscopic simulation. Journal of Chemical Physics 107(11):4423-4435.
309. Gittes F, Mickey B, Nettleton J, & Howard J (1993) Flexural rigidity of microtubules and
actin-filaments measured from thermal fluctuations in shape. Journal of Cell Biology
120(4):923-934.
310. Rubinstein M & Colby RH (2003) Polymer physics (Oxford University Press, Oxford ;
New York) pp xi, 440 p.
311. Samuel RE, Salmon ED, & Briehl RW (1990) Nucleation and growth of fibers and gel
formation in sickle-cell hemoglobin. Nature 345(6278):833-835.
312. Magdoff-Fairchild B & Chiu CC (1980) Double filaments in fibers and crystals of
deoxygenated hemoglobin S. Biophysical Journal 32(1):436-438.
313. Briehl RW & Guzman AE (1994) Fragility and structure of hemoglobin S fibers and gels
314.
and their consequences for gelation kinetics and rheology. Blood 83:573-579.
Jones CW, Wang JC, Briehl RW, & Turner MS (2005) Measuring Forces between
Protein Fibers by Microscopy. Biophysical Journal 88(4):2433-2441.
196
315. Li H & Lykotrafitis G (2011) A coarse-grain molecular dynamics model for sickle
hemoglobin fibers. Journal of the mechanical behavior of biomedical materials 4(2):162-
173.
316. Coffey WT, Kalmykov YP, & Titov SV (2002) Langevin equation method for the
rotational Brownian motion and orientational relaxation in liquids Journal of Physics A:
Mathematical and General 35:6789-6803.
317. Ross P, D. & Minton A, P. (1977) Hard Quasispherical Model for The Viscosity of
Hemoglobin Solution. BioChemical and Biophysical Research Communications
76(4):971-976.
318. Gittes F, Mickey B, Nettleton J, & Howard J (1993) Flexural rigidity of microtubules and
actin-filaments measured from thermal fluctuation in shape. Journal of Cell Biology
120(4):923-934.
319. Aprelev A, Liu Z, & Ferrone Frank A (2011) The Growth of Sickle Hemoglobin
320.
Polymers. Biophysical Journal 101(4):885-891.
Izvekov S & Voth GA (2005) A Multiscale Coarse-Graining Method for Biomolecular
Systems. Journal of Physical Chemistry B 109(7):2469-2473.
321. Shi Q, iZvekov S, & Voth GA (2006) Mixed Atomistic and Coarse-Grained Molecular
Dynamics: Simulation of a Membrane-Bound Ion Channel. Journal of Physical
Chemistry B 110(31):15045–15048.
197
|
1902.02303 | 1 | 1902 | 2019-02-06T17:56:51 | Aggregation dynamics of active cells on non-adhesive substrate | [
"physics.bio-ph",
"cond-mat.soft",
"nlin.CG",
"q-bio.CB"
] | Cellular self-assembly and organization are fundamental steps for the development of biological tissues. In this paper, within the framework of a cellular automata model, we address how an ordered tissue pattern spontaneously emerges from a randomly migrating single cell population without the influence of any external cues. This model is based on the active motility of cells and their ability to reorganize due to cell-cell cohesivity as observed in experiments. Our model successfully emulates the formation of nascent clusters and also predicts the temporal evolution of aggregates that leads to the compact tissue structures. Moreover, the simulations also capture several dynamical properties of growing aggregates, such as, the rate of cell aggregation and non-monotonic growth of the aggregate area which show a good agreement with the existing experimental observations. We further investigate the time evolution of the cohesive strength, and the compactness of aggregates, and also study the ruggedness of the growing structures by evaluating the fractal dimension to get insights into the complexity of tumorous tissue growth which were hitherto unexplored. | physics.bio-ph | physics |
Aggregation dynamics of active cells on
non-adhesive substrate
Debangana Mukhopadhyay1, Rumi De1*,
Department of Physical Sciences, Indian Institute of Science Education and Research
Kolkata, Mohanpur 741246, India.
E-mail: [email protected]
December 2018
Abstract. Cellular self-assembly and organization are fundamental steps for the
development of biological tissues. In this paper, within the framework of a cellular
automata model, we address how an ordered tissue pattern spontaneously emerges
from a randomly migrating single cell population without the influence of any external
cues. This model is based on the active motility of cells and their ability to reorganize
due to cell-cell cohesivity as observed in experiments. Our model successfully emulates
the formation of nascent clusters and also predicts the temporal evolution of aggregates
that leads to the compact tissue structures. Moreover, the simulations also capture
several dynamical properties of growing aggregates, such as, the rate of cell aggregation
and non-monotonic growth of the aggregate area which show a good agreement with
the existing experimental observations. We further investigate the time evolution of the
cohesive strength, and the compactness of aggregates, and also study the ruggedness
of the growing structures by evaluating the fractal dimension to get insights into the
complexity of tumorous tissue growth which were hitherto unexplored.
1. Introduction
Formation and development of tissues through self-assembly and organization of living
cells are intriguing and complex phenomena [1, 2].
It is still not well understood
how ordered tissue structures spontaneously develop from orchestrated response of
interacting multi-cellular components. Understanding the process of tissue organization,
thus, would immensely benefit diverse areas of developmental biology, wound healing,
cancer therapy, tissue engineering and even organ printing to name a few [1 -- 9].
In nature, numerous examples of self-assembled aggregation processes can be found
both in living as well as non-living systems [10], such as, formation of snowflakes [11],
cloud formation [12], coagulation of colloids [13], aggregation of proteins [14], swarming
of bacteria [15 -- 17], flocking of birds [18] etc. Several experimental, theoretical, and
computer simulation studies have been carried out which reveal a great deal about
the structural and dynamical aspects of self-assembly and aggregation processes in
passive systems [19, 20]. For an example, studies on diffusion-limited aggregations
2
provide a deep understanding into the process of snowflakes like branched dendritic
patterns formation [21 -- 25]. Moreover, it has been found that variety of inter-particle
interactions, reaction mechanisms, coalescence rates, and other factors play a crucial
role in determining the dynamics of aggregates [19, 20].
There have also been many efforts to understand the assembly and organization
processes of living tissues, such as, how tissues spread [26 -- 28], how sorting takes
place in a multicellular system [29 -- 32], or how tumors develop and grow [4, 33]. An
insight into the complex tissue organizations could be obtained based on differential
adhesion hypothesis (DAH) proposed by Steinberg et. al. [34 -- 36]. According to DAH,
motile and cohesive cells spontaneously tend to reorganize to maximize cell-cell cohesive
binding strength and to minimize the interfacial energy of aggregates. DAH, thus,
provides an important route to the formation of ordered tissue patterns from collective
interactions of multi-cellular components. There are also quite a few theoretical and
computational studies to understand the complex behaviours of tissues, for examples,
Graner and Glazier have developed theoretical model to investigate the phenomenon
of cell sorting [31]; Sun and Wang's group have simulated fusion of multicellular
aggregates [37]; Flenner et. al have modelled temporal shape evolution of multicellular
systems [30]. Moreover, reaction-diffusion mechanisms have also been shown to influence
the formation of many tissue patterns, such as rapidly growing embryo or stem cell
aggregate [1], aggregation of amoebae motion [38].
Interestingly, in recent experiments, performed by Douezan and Brochard-Wyart,
it has been observed that randomly migrating cells deposited on non-adhesive surface
spontaneously form closely packed compact aggregates [7]. There are other experiments
which also exhibit the tendency of cellular aggregates to spontaneously form compact
tissue structures acquiring minimum area [28,39 -- 42]. Examples include, rounding up of
Hydra aggregates into circular shapes in 2D [43] or chick embryonic cellular aggregates
evolving into spherical shapes in 3D [44]; in all these cases, cell-cell adhesion bonds
have been found to play a significant role in shaping up the structure. A key step
to get insights into these processes is to understand what drives cell-cell attraction to
form compact multi-cellular aggregates. Several hypotheses have been proposed, in
particular, whether cell-cell attraction is based on chemical signalling [45] or whether
cellular communications are mediated by extracellular matrix - where deformations
created in the substrate by one cell can be detected by an adjacent cell [46], or the
attraction is mediated due to haptotaxis, i.e., directional cell motility up a gradient in
the substrate [47]. Motivated by the experimental findings, in this paper, we address how
an ordered tissue pattern spontaneously emerges from a seemingly disordered single cell
population without the influence of any external physical or chemical forces. We show
that a suitably constructed cellular automata model based on the active motility and
local reorganization of cells can successfully capture the formation of nascent clusters and
predict the temporal evolution of aggregates that leads to the compact tissue structures
as observed in experiments. The crux of our model is that the sole consideration of
the cellular tendency of forming bonds with neighbouring cells to maximize cohesive
3
strength, is sufficient for the emergence of compact tissue pattern. Our study thus
shows that the presence of an external cue or a mediator or a gradient is not critical
to these types of aggregations which was hitherto unexplored. Our theory also reveals
the existence of two distinct time scales in such cellular aggregation processes - one
is fast time scale associated with the diffusion of cells and another much slower time
scale associated with the tissue compaction process that involves breaking of cell-cell
cohesive bonds and making of new bonds. Interestingly, we find that the difference in
the tendency of cell types to self-organize to increase the binding strength plays a crucial
role in determining the time scale and the structure of the tissue pattern.
Apart from emulating the structural evolution, our model also captures several
dynamical properties of cellular aggregates [4, 7, 28], such as, the rate of aggregation,
i.e., how the number of cell clusters evolve with time as the aggregation takes place.
Moreover, as observed in experiments, our model also predicts the non-monotonic growth
of the surface area of the growing aggregates on non-adhesive substrates. We further
investigate, how the cellular cohesive binding strength and the overall compactness of the
growing aggregate evolve as time progresses which remains so far unexplored. Besides,
we also study how the ruggedness of the growing aggregate structure changes towards
a smooth compact structure by evaluating the fractal dimension of the aggregate which
can be useful in comparing normal versus deceased tissue growth as revealed by the
recent experiments [48, 49].
2. Cellular automata modelling
Figure 1. (a) Illustration of randomly deposited cells on a lattice surface. Shaded
areas represent lattice sites occupied by cells and each dot represents the center of
mass of the cell.(b) Schematic of eight neighbouring sites of a cell.
In this section, we now discuss the formulation of our cellular automata model in
details. First, we start with N0 cells randomly deposited on a (L × L) square lattice as
illustrated in Fig. 1(a). Each lattice site is, thus, either occupied by a cell or remains
empty. Each cell is considered to have eight nearest neighbouring sites as depicted in
Fig. 1(b) to form a close packed structure in a square lattice. In our model, cells can
diffuse to any of its empty neighbouring sites. On the way, if it finds another immediate
neighbouring cell then sticks to it due to cell-cell binding affinity and starts moving
(a)(b) 4
together as a whole cluster. Here, the sticking probability is considered to be 1. Thus,
at any instant, there are N (t) number of cell clusters. It is to be noted that, in our
simulation, we refer a single cell also as a cluster. Thus, a cluster can consist of a single
cell or multiple cells. In our model, at each simulation step, N (t) clusters are picked up
at random one by one and then the cluster is given the opportunity to (i) either diffuse
with a motility probability, Pm(n), (ii) or locally reorganize with a rolling probability,
Pr(n), within its own cluster so that cell-cell cohesive binding strength increases, (iii)
otherwise, just stay put.
Motility criteria: The motility probability of a cluster is considered to be
dependent on its size as Pm(n) = µ
n; where n is the number of cells in the cluster and µ
is the proportionality constant. In our simulations, µ is taken as 1. As time progresses,
diffusing clusters collide and merge to form a bigger cluster. Thus, the motility of the
cluster decreases with accumulation of more number of cells. The dependence of the
probability, Pm(n) on 1/n could simply be attributed to the decreasing tendency of
diffusion of the cluster as it gathers more mass, since the diffusion coefficient, D ∝ 1
m;
where m is the mass to the cluster and hence, proportional to the number of cells, n, of
the cluster [50]. Moreover, as found in experiments, the cluster motility gradually slows
down with increase in size and eventually stops moving after it reaches a critical size
(say, nc number of cells).
Local reorganization: Now, we discuss in detail the local reorganization of cells
within its own cluster via rolling process. As cells in a cluster are bound together by
strong cell-cell adhesion, local reorganization process requires breaking of existing cell-
cell cohesive bonds and again making of new bonds. Thus, this process is observed to
be much slower process compared to diffusion. Therefore, in our model, we consider
an additional probability factor, Pr(n) = βn to incorporate the slower compactification
process, where the rolling parameter β denotes the tendency of the cell type to locally
reorganize due to the binding affinity. The cell rearranges its position relative to the
neighbouring cells so that it is surrounded by the maximal possible neighbours and
hence, increases the cell-cell binding strength. As the cluster size gets bigger, probability
of rolling increases with increase in number of cells, n, in the cluster.
In our model, the cell-cell binding interaction energy is described as,
E = −1n1 − 2n2 − 3n3;
(1)
where n1, n2, and n3 are the number of first, second and third nearest neighbouring
cells and 1, 2, and 3 are the weight factors given to its neighbouring binding sites
respectively. Here, we assume that the interaction strength between a cell and its first
nearest neighbours is maximum as they are adjacent to each other and the interaction
strength gradually decreases from second to third neighbours.
It is also observed in
experiments that cell-cell interaction can extend beyond first nearest neighbours, since
apart from the physical contact, cell-cell attraction could be driven by chemical signalling
i.e. cells secrete chemoattractants into the medium.
Now, at each simulation step, for a randomly chosen cluster, if the motility criteria,
5
Pm(n), are not satisfied, then we check for the rolling probability criterion, Pr(n). Once
it is satisfied then we check whether the local reorganization can take place. We, first,
calculate the binding interaction energy, E0, of the cell in its current location following
(1). Next, we choose randomly an empty 1st nearest neighbouring site for its
Eq.
possible relocation and calculate its binding energy, En, at the new position.
If the
corresponding binding energy, En, associated with the new location is lowered compare
to its current configuration, E0, so that it leads to an over all increase of cellular binding
strength due to more cell-cell interactions, then the move is readily accepted. However,
if the energy, En, becomes higher, then a random number, r, is generated from the
uniform distribution over the interval [0, 1], and the new configuration is accepted with
a probability such that r < exp ( − ∆E), where ∆E = En − Eo, is the corresponding
change in cellular interaction energy. In simulations, we have considered, 1 = 3, 2 = 2,
and 3 = 1.
2.1. Numerical algorithm
Details of numerical steps of our cellular automata model are summarized as follows:
(i) N0 cells are, initially, deposited at random on a L×L two dimensional square lattice
as shown in Fig. 1(a). (As single cell is also referred as a single cell cluster; thus,
initially, the number of clusters, N (t) = N0.)
(ii) At each simulation step, N (t) clusters are randomly picked up one by one (once a
cell of a cluster is randomly chosen, no other cells of that particular cluster will be
chosen at that step). Then, we record the number of cells, n, of the chosen cluster.
Next, we check whether the motility or the local reorganization or the stay put
criterion of the cluster is satisfied.
(iii) Motility criteria: a random number R1 is generated from the uniform distribution
over the interval [0, 1]. If R1 ≤ Pm(n) = 1
n and the size of the cluster is less than a
critical number nc, then the whole cluster moves in any of the empty first nearest
neighbouring sites.
(iv) Criteria of local rearrangement :
if the motility criteria are not satisfied then
we check for the rolling criteria. We call another random number R2 and if
R2 ≤ Pr(n) = βn, then the chosen cell of that cluster given an opportunity to locally
rearrange within its own cluster to be surrounded by more number of neighbours.
(v) Now,
for reorganisation process to occur, an empty site in its first nearest
neighbouring region is randomly chosen. Then, we calculate the binding energy,
En, in its possible new location and the energy, E0, in its current location.
If
En ≤ Eo, the new location is readily accepted. Otherwise, if En > Eo, then we
call a random number (r) from a uniform distribution [0, 1], if r < exp(−∆E)
where ∆E = En − Eo, then the move is accepted except if the move disintegrates
the cluster then that move is not allowed (as illustrated in Figure ??(a) & (b) in
supplementary information).
6
(a) T=0
(b) T=100
(c) T=5000
Figure 2. Time evolution of diffusion limited aggregation of cells. (a) Presents
a snapshot at simulation time T = 0, here 2000 cells are deposited randomly on
(200 × 200) lattice surface. (b) While diffusing, cells collide and stick to each other to
form aggregates, one such snapshot at T = 100. (c) Shows irregular branched cellular
aggregates at T = 5000. The inset focuses one such irregular shaped cell aggregate.
(vi) If both the motility and the rolling criteria are not satisfied, then the cluster just
stays put.
3. Results and discussions
We now investigate the dynamics of initiation, growth, and evolution of cellular
aggregates on non interacting surface following the cellular automata model described
in the previous section.
In our simulations, we start from a population of randomly
distributed single cells spread over a 2D surface. While diffusing on the surface, cells
collide, stick to each other, and thus form small nascent clusters of two or three cells
clusters. These clusters then grow with accumulation of more colliding cells or clusters
and give rise to large aggregates. Our simulations capture many dynamical properties
of growing aggregates, those we discuss in detail in the following subsections. We have
performed simulations for a wide range of parameter values and results presented here
after averaging over many such simulations.
3.1. Aggregation of cells due to diffusion
We, first, study the effect of simply diffusion into the aggregation of cells. The time
evolution of cell aggregation due to diffusion is shown in Figs. 2(a)-(c). Figure 2(a) shows
the initial deposition of 2000 cells on a 200 × 200 lattice. These cells diffuse to any of
its empty first nearest neighbouring sites with a probability Pm(n) = µ/n. On the way,
when they come across with other cells or clusters, they stick irreversibly and form bigger
clusters. Figure 2(b) shows such an intermediate time step of clustering of cells. As
time progresses, these clusters grow with gathering of more diffusing clusters. However,
the cluster motility gradually slow down with increase in cluster size and eventually
stop moving after reaching a critical size, nc, as observed in experiment [7]. Figure
2(c) presents a snapshot of cellular aggregates acquiring irregular branched structures
as predicted by diffusion limited aggregation studies [21 -- 23]. The irregular shapes arise,
7
since the diffusing cells/clusters only access the protruding exterior regions as they
irreversibly stick to the immediate neighbours, thus, do not roll down to the interior
of the cluster. Simulation results are presented here for nc = 10 as the critical size of
the cluster is found of the order of ten in experiment [4, 7]. We have also carried out
simulations with nc = 20 and 30; the size of aggregates becomes bigger and develops
more branched structure; however, the qualitative nature of aggregates remains the
same.
3.2. Aggregation: effect of local reorganisation of cells
Next we investigate, in addition with diffusion, the influence of local reorganization of
cells due to strong cell-cell binding affinity on the dynamics of aggregate formation.
As observed in experiments, aggregates are grown generally compact in nature, where,
cell-cell interaction to increase the cohesive binding strength plays a crucial role in
smoothening out the irregular structures. We start with N0 = 2000 cells placed
randomly on 2D square lattice of size L = 200.
In this case, cells or clusters may
diffuse with a probability, Pm(n) = µ/n, otherwise may opt to locally reorganize with
a rolling probability, Pr(n) = βn or just stay put as described in the previous section.
Figures 3(a)-(c) present one such simulation results of temporal evolution of aggregates
formation (keeping µ = 1 and β = 0.05). From these figures, it is clearly seen that as
time progresses, aggregates grow in size and also gradually become compact due to local
rearrangements of clustering cells. After a sufficiently long time, aggregates become fully
compact and one such compact structure is shown in the inset of Fig. 3(c).
Moreover, we have carried out simulations for different cell density, ρ = N0/(L×L),
(keeping N0 = 500, 1000, 2000, 10000, and L = 200) and studied the effect of varying
density in the formation of cellular aggregates. Under low densities, cells/clusters diffuse
for longer time to find another cell or cluster to bind together and hence, the clustering
(a) T=0
(b) T=100
(c) T=4000
Figure 3. Time evolution of aggregates formation due to diffusion and local
reorganizations of cells. (a) Presents a snapshot of random deposition of 2000 cells
on a (200×200) lattice at simulation time T = 0. (b) A snapshot of growing aggregates
at T = 100. Clusters initially grow as irregular shaped structure as shown in the inset.
(c) Shows compact cellular aggregates after a sufficiently long time at T = 4000.
The compact shape develops due to rearrangement of clustering cells to maximize the
binding strength. The inset shows such a compact cell aggregate. (Results presented
here for nc = 10 and β = 0.05.)
8
process also takes longer time. However, for higher densities, diffusing cells find other
cells in its immediate vicinity. So, the growth process becomes faster and the cluster
size also gets bigger due to availability of more number of cells. On the other hand,
having a large cluster size, local cell-cell rearrangements take longer time and thus, the
overall compactification process becomes slower. Moreover, the compactification time
strongly depends on the nature of cell types, cellular tendency of making and breaking
adhesion bonds etc., in our simulations, which is represented by the variation in rolling
coefficient, β. We discuss these properties of the cellular aggregation in more details in
the following subsections.
3.3. Rate of cell aggregation
In this section, we investigate the rate of aggregation, i.e., number of clusters formation
as a function of time. Clusters mainly form due to the random collisions of diffusing
clusters. Thus, at any instant, if there are N (t) number of clusters of single cell or
multiple cells, then the occurrence of number of collisions for one cluster is (N − 1) as
there are other (N−1) surrounding clusters. Thus, at any time instant t, since all clusters
randomly move and collide with other; so the total number of collisions considering all
clusters movement is proportional to N (N − 1). Thus, the time evolution of the number
of clusters can be described by the equation,
dN (t)
dt
= −KN (t)(N (t) − 1) ≈ −KN (t)2,
(2)
where K is the rate constant which depends on the size and motility of the cell type. [7]
(a)
(b)
Figure 4. Time evolution of number of clusters.(a) Comparison of simulation
result of rate of aggregation (( ), keeping ρ = 0.0125, β = 0.05) with theoretical model
prediction, solution of Eq. 2, shown by the dashed curve. (b) Formation of number
of aggregates as a function of time for different initial cluster density, ρ = 0.0125( ),
0.025( ), and 0.05( ).
From our simulation, we evaluate the rate of cell aggregation, i.e., the rate of change
of number of clusters during the aggregation process. We find as the clusters while
diffusing collide with each other and merge to form bigger aggregates and hence, the
00.20.40.60.81050010001500N/ N0TSimulationTheory00.20.40.60.81050010001500N/ N0Tρ=0.05ρ=0.025ρ=0.0125 9
number of clusters decreases with time. Besides, since the aggregate stops moving after
reaching a critical size (as its mass gets heavier), number of clusters eventually reaches
to a steady value. Figure 4(a) shows our simulation results of time evolution of number
of clusters which is in good agreement with recent experimental observations [7] (as it
is also seen from Figure ?? provided in the supporting information). We also compare
our simulation results with the theoretical prediction of Eq. (2). As seen from Fig.
4(a), the theoretical prediction agrees quite well with our simulations as well as with
the experimental observations by tuning only one fitting parameter, the rate constant
K.
Moreover, we also study the effect of cell density variation on the rate of cell
aggregation. Figure 4(b) shows the simulation results for three different densities
ρ = 0.0125, 0.025, and 0.05 (with initial cell number, N0 = 500, 1000, 2000 and L = 200).
As shown in Fig. 4(b), for higher density, the decrease in the number of clusters is faster;
since chances of finding neighbouring cells/clusters are higher, the rate of merging of
clusters is also high. In other words, the growth of aggregates happens faster for higher
cell density.
3.4. Evolution of surface area of an aggregate
Resent experimental studies show that the surface area of the growing aggregates on non-
adhesive substrate exhibits a non-monotonic evolution unlike the monotonic increase in
the area as observed in case of spreading of cell aggregates [27].
In order to get an
insight into the dynamics, we also investigate the area evolution of the aggregates. In
our simulations, we consider two methods to estimate the surface area of an aggregate.
In one method, we calculate the perimeter of the cluster, which in turn provides an
estimation of the projected area and in another, we calculate the radius of gyration of
the cluster to evaluate the surface area (details have been given as supporting materils).
We have studied the time evolution of the surface area averaged over many
aggregates for different cell density, ρ = 0.0125, 0.025, 0.05 and rolling tendency of
varied cell type given by β = 0.0005, 0.005, 0.05 as shown in Figure 5 (a) and 5(b)
respectively. We find, as observed in experiment, [7] at the early stages of the growth
process, the surface area of the cluster increases rapidly due to collisions and merging of
clusters. However, once the aggregate reaches a critical size, it stops moving and other
than occasional joining of randomly wandering of small clusters, the cells reorganize
among themselves to be surrounded by the maximum possible neighbours, and hence,
the surface area start decreasing due to the compactification process and finally reaches
to a steady state value giving rise to the most compact structure.
Insets of Fig. 5
(a) show an irregular shaped structure that arises due to joining of diffusing cells at
early times and an compact structure due to self-organization of the clustering cells at
a longer period of time. Importantly, our model could capture the non-linear nature of
the cluster area evolution as found in experiments which could also be seen from Figure
?? in the supplementary information.
10
(a)
(b)
Figure 5.
Time evolution of the surface area of cellular aggregate (a)
Simulation results for different initial cluster density, ρ = 0.0125( ), 0.025( ), 0.05( ).
Left panel inset shows the development of irregular shaped structure at some initial
time and right panel inset shows the compactified structure at a longer period of time.
(b) Shows the effect of rolling coefficient on the surface area evolution for β = 0.0005( ),
0.005( ), and 0.05( ). Theoretical prediction, solution of Eq.
(3), is shown by the
dashed curve (τc has been varied for different β values and while keeping K constant).
The time evolution of the cluster area, A(t), during the aggregation process can
also be theoretically modeled following earlier studies [39, 40] as,
dA(t)
dt
= KN (t)A(t) − 1
τc
[A(t) − Af inal];
(3)
where N (t) is the number of clusters at any instant t. Here, the first term represents
the area increase due to diffusive collision and merging of clusters and the second term
represents the decrease in area with a characteristic relaxation time, τc, which is related
to the tendency of reorganization of cells to reach to the most compact structure of an
area Af inal.
We now compare simulation results of our cellular automata model with the
theoretical prediction. We numerically solve Eq. (3) substituting N (t) from Eq. (2)
keeping K constant and varying values of τc and then compare it with different values of
β (= 0.05, 0.005, and 0.0005). As seen from the Fig. 5 (b), in our simulation, the rolling
coefficient, β, plays the role as of 1/τc in Eq. (3). For low values of β, since the chances of
cellular rolling/reorganization decreases, sparse branching structure prevails for longer
period of time, thus, the coalescence time, τc becomes higher. On the other hand,
increasing β value increases the rate of coalescence and thus, the cluster compactifies
faster.
3.5. Compactness of cellular aggregate
We further investigate how the cellular cohesive strength evolves as the cluster grows
in time. In our simulations, it is measured by estimating the adhesion junctions formed
between cells, i.e., the total number of bonds formed among the constitutive cells within
02040608001000200030004000ATρ=0.05ρ=0.025ρ=0.0125 20 40 60 80 100 0 1000 2000 3000ATβ=0.05β=0.005β=0.0005 11
the aggregate. As time progresses, the binding strength increases due to more number
of cells/clusters joining the aggregate. On the other hand, addition of new cells in the
cluster due to diffusion makes the aggregate to spread out. However, aggregates slowly
become compact due to local reorganization of cells within the cluster to minimize the
binding free energy by forming maximum possible bonds with the neighbouring cells.
In our model, we define the compactness of an aggregate, Cr, as total number of bonds
in a cluster. It has been, further, normalized to differentiate the increase in number
of bonds due to addition of diffusing cells or due to local rearrangement of the relative
position of the cells to maximize the binding strength as defined by Cnr,
Cnr =
T otal no. of bonds in a cluster of n cells
Bonds f or making 1D chain of n cells
(4)
Thus, Cr has information about the total number of bonds in the aggregate, on the
other hand, Cnr represents that given a number of cells in an aggregate, how closely
cells are packed together.
(a)
(b)
Figure 6. Time evolution of the compactness of cellular aggregate. (a)
Shows the effect of different initial cluster density, ρ = 0.0125( ), 0.025( ), and
0.05( ). Corresponding Cnr, for different density is shown in inset.(b) Plot shows
the dependence of the aggregate compactness on the variation of rolling coefficient for
β = 0.0005( ), 0.005( ), and 0.05( ).
We have simulated the time evolution of the compactness, Cr, of an aggregate for
different cell density, ρ and varying rolling coefficient, β averaged over many aggregates
shown in Figs. 6(a) and (b) respectively. As seen from the plot, the compactness, i.e.,
the binding strength of an aggregate increases as the cluster grows with time. The
initial rapid increase is due to joining of diffusing clusters then the increase happens
mainly due to the local reorganization of cells that involves breaking and making of
new bonds; thus, it occurs at a much slower rate compare to diffusion. Moreover, as
cell density increases, the number of cells in an aggregate also increases and hence, the
total number of bonds, i.e., compactness also becomes higher. Further, as seen from
the inset of Fig. 6(a), since Cnr is normalized by the cluster size, all curves for different
cell density collapse into a single curve.
05010015020025001000200030004000CrTρ=0.05ρ=0.025ρ=0.012512301000200030004000CnrT5010015020001000200030004000CrTβ=0.0005β=0.005β=0.05 12
3.6. Fractal geometry of the aggregate
Recent studies have shown that the fractal geometry can be useful to understand the
underlying mechanisms of tissue growth as living tissues are spatially heterogeneous and
thus, exhibit fractal pattens [48]. Moreover, it has been observed that fractal analysis
may provide an efficient way to distinguish normal versus cancerous tissue growth. We
have, therefore, characterized the fractal dimension of growing aggregates to get insights
into the complexity of the structure. In our simulations, the fractal dimension, D, has
been estimated from the radius of gyration, Rg, of the evolving cluster using the following
relationship [21, 25],
Rg ∼ n1/D;
(5)
where n is the number of cells in the cluster. As discussed in the previous section,
at early times, the cluster grows due to numerous collisions between the diffusing
cells/clusters; thus initially, it develops a branched sparse structure and gradually self-
organization of clustering cells gives rise to the compact structure. In our simulations,
we start calculating the fractal dimension after the aggregate has grown to a branched
structure and then study as time progresses, how the dimension changes due to the
compactification process. The time evolution of the fractal dimension, D, averaged
over many such aggregates is presented in Fig. 7. As seen from the figure, at the
initial stage, the fractal dimension turns out to be similar to the diffusion limited
aggregates [21, 22, 51], however, it slowly increases as the clustering cells relocate to the
energetically favourable binding sites, and as a result, also compactifies the aggregate.
Eventually, the fractal dimension approaches to two for the most compact aggregated
structure, as it is expected since the simulation is done on a 2D surface. Moreover,
as seen from Fig. 7, with increase in rolling coefficient, β, cluster cells to have higher
chances to find their favourable binding sites; thus the compaction process becomes
Figure 7. Time variation of fractal dimension of a growing cellular aggregate for
different rolling coefficient, β = 0.0005( ), 0.005( ), and 0.05( ). Here, the initial
cluster density is kept constant at ρ = 0.05.
1 1.5 2 1000 2000 3000 4000DTβ=0.05β=0.005β=0.0005 13
quicker and hence, the fractal dimension of the aggregate also increases faster. On the
other hand, it has been it has been observed that the fractal dimension of cancerous
tissue increases with increase with progress in cancer stages as the heterogeneity changes
due to accumulation of more masses. Interestingly, our study reveals that the difference
in the tendency of cell types to increase binding strength also plays a crucial role in
determining the structure of the tissue pattern which can be tested further by suitable
experiments.
4. Conclusion
We have developed a cellular automata model to study the aggregation dynamics of a
seemingly disordered tissue cell population in the absence of any external mediator. This
model based on the active motility and local reorganization of cells could successfully
capture the structural and temporal evolution of aggregates that leads to the compact
tissue structures as observed in experiments.
Importantly, our study shows that the
sole consideration of the cellular tendency of forming bonds with the neighbouring
cells to maximize cohesive strength is sufficient for the spontaneous emergence of
compact tissues. Moreover, it provides several insights into the dynamics of the cell
aggregation process.
It reveals the existence of two distinct time scales - one fast
time scale associated with the diffusion of cells and another much slower time scale
associated with the tissue compaction process that involves breaking of cell-cell cohesive
bonds and making of new bonds leading to local reorganization of cells. Besides, as
found in experiments, our simulation results also successfully predicts many dynamical
properties of the growing aggregates, such as, the rate of cell aggregation, the non-linear
evolution of the surface area, the binding strength and the compactness of the growing
aggregate [4, 7, 28]. Moreover, we have demonstrated that the variation in tendency of
rolling and reorganization of cells have a profound effect on the formation of tissue shapes
and structures and it could be further tested by carrying out suitable experiments. Our
theoretical model in essence is of a generic nature and hence can be extended to other
systems with suitable modifications. Moreover, since tissue development is quite a
complex process and current tissue engineering procedures are still very experimental
and also expensive; thus, simple theoretical and computational model studies are
envisaged to facilitate the understanding of how individual cells organize into tissues
much like as it has been done in passive growth processes which once seemed to be a
difficult prospect.
5. Acknowledgments
The authors acknowledge the financial support from Science and Engineering Research
Board (SERB), Grant No.
SR/FTP/PS-105/2013, Department of Science and
Technology (DST), India.
14
6. References
[1] Sasai Y. Next-Generation Regenerative Medicine: Organogenesis from Stem Cells in 3D Culture.
2013 Cell Stem Cell 12 520 - 530
[2] Whitesides GM, Grzybowski B. Self-Assembly at All Scales. 2002 Science 295 2418 -2421
[3] Jakab K, Norotte C, Marga F, Murphy K, Vunjak-Novakovic G, Forgacs G. Tissue engineering by
self-assembly and bio-printing of living cells. 2010 Biofabrication 2 022001
[4] Gonzalez-Rodriguez D, Guevorkian K, Douezan S, Brochard-Wyart F. Soft Matter Models of
Developing Tissues and Tumors. 2012 Science 338 910 -917
[5] Mironov V, Visconti RP, Kasyanov V, Forgacs G, Drake CJ, Markwald RR. Organ printing: Tissue
spheroids as building blocks. 2009 Biomaterials 30 2164 - 2174
[6] Neagu A, Jakab K, Jamison R, Forgacs G. Role of physical mechanisms in biological self-
organization. 2005 Physical review letters 95 178104
[7] Douezan S, Brochard-Wyart F. Active diffusion-limited aggregation of cells. 2012 Soft Matter 8
784-788
[8] De R, Zemel A, Safran SA. 2010 Theoretical Concepts and Models of Cellular Mechanosensing.
Methods Cell Biol. 98 143-175
[9] De R. 2018 A general model of focal adhesion orientation dynamics in response to static and cyclic
stretch, Communications Biology 1 81
[10] Ben-Jacob E, Garik P. The formation of patterns in non-equilibrium growth. 1990 Nature 343 523
[11] Langer JS. Instabilities and pattern formation in crystal growth. 1980 Rev Mod Phys. 52 1-28
[12] Blando J, Turpin B. Secondary organic aerosol formation in cloud and fog droplets: A literature
evaluation of plausibility. 2000 Atmospheric Environment 34(10) 1623-1632
[13] Weber AP, Baltensperger U, Gggeler HW, Tobler L, Keil R, Schmidt-Ott A. Simultaneous in-
situ measurements of mass, surface andmobilitydiameter of silver agglomerates. 1991 Journal of
Aerosol Science.22 S257 - S260
[14] Bence NF, Sampat RM, Kopito RR. Impairment of the Ubiquitin-Proteasome System by Protein
Aggregation. 2001 Science 292(5521) 1552 -- 1555
[15] Ben-Jacob E, Schochet O, Tenenbaum A, Cohen I, Czirk A, Vicsek T. Generic modelling of
cooperative growth patterns in bacterial colonies. 1994 Nature 36846.
[16] Matsushita M, Wakita J, Itoh H, R`afols I, Matsuyama T, Sakaguchi H, et al. Interface growth
and pattern formation in bacterial colonies. 1998 Physica A: Statistical and Theoretical Physics
249(1-4) 517-524
[17] Tsimring L, Levine H, Aranson I, Ben-Jacob E, Cohen I, Shochet O, et al. Aggregation Patterns
in Stressed Bacteria. 1995 Phys Rev Lett. 75 1859-1862
[18] Vicsek T, Zafeiris A. Collective motion. 2012 Physics Reports. 517(3) 71 - 140
[19] Meakin P. 1998 Fractals, scaling, and growth far from equilibrium ( Cambridge nonlinear science
series 5 Cambridge, U.K. ; New York : Cambridge University Press)
[20] Barab´asi AL, Stanley HE. 1995 Fractal concepts in surface growth (Cambridge University Press)
[21] Meakin P. Formation of Fractal Clusters and Networks by Irreversible Diffusion-Limited
Aggregation. 1983 Phys Rev Lett. 51 1119-1122
[22] Witten TA, Sander LM. Diffusion-limited aggregation. 1983 Phys Rev B. 27 5686-5697
[23] Meakin P. Computer simulation of cluster-cluster aggregation using linear trajectories: Results
from three-dimensional simulations and a comparison with aggregates formed using brownian
trajectories. 1984 Journal of Colloid and Interface Science. 102(2) 505 - 512
[24] Vicsek T. Pattern Formation in Diffusion-Limited Aggregation. 1984 Phys Rev Lett. 53 2281-2284
[25] Kolb M, Botet R, Jullien R. Scaling of Kinetically Growing Clusters. 1983 Phys Rev Lett. 51
1123-1126
[26] Ryan PL, Foty RA, Kohn J, Steinberg MS. Tissue spreading on implantable substrates is a
competitive outcome of cell -- cell vs. cell -- substratum adhesivity. 2001 Proceedings of the National
Academy of Sciences 98(8) 4323-4327
15
[27] Douezan S, Guevorkian K, Naouar R, Dufour S, Cuvelier D, Brochard-Wyart F. Spreading
dynamics and wetting transition of cellular aggregates. 2011 Proceedings of the National Academy
of Sciences of the United States of America 108 18 7315-20
[28] Guo Wh, Frey MT, Burnham NA, Wang Yl. Substrate Rigidity Regulates the Formation and
Maintenance of Tissues. 2006 Biophysical Journals. 90(6) 2213-2220
[29] Jakab K, Neagu A, Mironov V, Markwald RR, Forgacs G. Engineering biological structures of
prescribed shape using self-assembling multicellular systems. 2004 Proceedings of the National
Academy of Sciences 101(9) 2864-2869
[30] Flenner E, Janosi L, Barz B, Neagu A, Forgacs G, Kosztin I. Kinetic Monte Carlo and cellular
particle dynamics simulations of multicellular systems. 2012 Phys Rev E. 85 031907
[31] Graner F, Glazier JA. Simulation of biological cell sorting using a two-dimensional extended Potts
model. 1992 Phys Rev Lett. 69 2013-2016
[32] Beatrici CP and Brunnet LG. Cell sorting based on motility differences. 2011 Phys Rev E. 84
031927
[33] Tracqui P. Biophysical models of tumour growth. 2009 Reports on Progress in Physics 72(5)
056701
[34] Steinberg MS. Reconstruction of Tissues by Dissociated Cells. 1963 Science 141(3579) 401-408
[35] Steinberg MS. Differential adhesion in morphogenesis : a modern view. 2007 Current Opinion in
Genetics & Development 17(4) 281-286
[36] Foty RA, Steinberg MS. The differential adhesion hypothesis:
a direct evaluation. 2005
Developmental Biology 278(1) 255 - 263
[37] Sun Y, Wang Q. Modeling and simulations of multicellular aggregate self-assembly in biofabrication
using kinetic Monte Carlo methods. 2013 Soft Matter 9 2172
[38] Vasiev BN, Hogeweg P and Panfilov AV. Simulation of Dictyostelium Discoideum Aggregation via
Reaction-Diffusion Model. 1994 Phys Rev Lett. 73 3173-3176
[39] Koch W, Friedlander SK. The effect of particle coalescence on the surface area of a coagulating
aerosol. 1990 Journal of Colloid and Interface Science 140(2) 419 - 427
[40] Hiram Y, Nir A. A simulation of surface tension driven coalescence. 1983 Journal of Colloid and
Interface Science 95(2) 462- 470
[41] Vicsek T. Formation of solidification patterns in aggregation models. 1985 Phys Rev A. 32 3084-
3089
[42] Huang SY, Zou XW, Shao ZG, Tan ZJ, Jin ZZ. Particle-cluster aggregation on a small-world
network. 2004 Phys Rev E. 69 067104
[43] Rieu JP and Sawada Y. Hydrodynamics and cell motion during the rounding of two dimensional
hydra cell aggregates. 2002 The European Physical Journal B - Condensed Matter and Complex
Systems 27(1) 167-172
[44] Mombach JCM, Robert D, Graner F, Gillet G, Thomas GL, Idiart M, et al. Rounding of aggregates
of biological cells: Experiments and simulations. 2005 Physica A: Statistical Mechanics and its
Applications 352(2) 525 - 534
[45] Weijer, Cornelis J. Collective cell migration in development. 2009 Journal of Cell Science 122(18)
3215-3223
[46] van Oers RFM, Rens EG, LaValley DJ, Reinhart-King CA, Merks RMH Mechanical Cell-Matrix
Feedback Explains Pairwise and Collective Endothelial Cell Behavior In Vitro. 2014 PLOS
Computational Biology 10(8) 1-14
[47] Weber, Michele and Hauschild, Robert and Schwarz, Jan and Moussion, Christine and de
Vries, Ingrid and Legler, Daniel F. and Luther, Sanjiv A. and Bollenbach, Tobias and Sixt,
Michael Interstitial Dendritic Cell Guidance by Haptotactic Chemokine Gradients. 2013 Science
339(6117) 328-332
[48] Baish, James W. and Jain, Rakesh K. Fractals and Cancer. 2006 Cancer Research 60(14) 3683-
3688
[49] C L Campbell, K Wood, C T A Brown and H Moseley Monte Carlo modelling of photodynamic
therapy treatments comparing clustered three dimensional tumour structures with homogeneous
tissue structures. 2016 Physics in Medicine & Biology 61 4840
[50] Rob Phillips JT Jane Kondev. 2013 Physical Biology of the Cell (Garland Science)
[51] Sander E, Sander LM and Ziff RM Fractals and Fractal Correlations. 1994 Comput Phys. 8(4)
420-425
16
|
1710.07413 | 1 | 1710 | 2017-10-20T04:38:38 | Using a hydrogen-bond index to predict the gene-silencing efficiency of siRNA based on the local structure of mRNA | [
"physics.bio-ph",
"q-bio.BM",
"q-bio.GN"
] | The gene silencing effect of short interfering RNA (siRNA) is known to vary strongly with the targeted position of the mRNA. A number of hypotheses have been suggested to explain this phenomenon. We would like to test if this positional effect is mainly due to the secondary structure of the mRNA at the target site. We proposed that this structural factor can be characterized by a single parameter called "the hydrogen bond (H-b) index", which represents the average number of hydrogen bonds formed between nucleotides in the target region and the rest of the mRNA. This index can be determined using a computational approach. We tested the correlation between the H-b index and the gene-silencing effects on three genes (Bcl-2, hTF and cyclin B1) using a variety of siRNAs. We found that the gene-silencing effect is inversely dependent on the H-b index, indicating that the local mRNA structure at the targeted site is the main cause of the positional effect. Based on this finding, we suggest that the H-b index can be a useful guideline for future siRNA design. | physics.bio-ph | physics | Published Jul 2004 in Biochem Biophys Res Comms, vol.320 (2) p. 622
doi.org/10.1016/j.bbrc.2004.04.027
Using a hydrogen-bond index to predict the gene-silencing efficiency of
siRNA based on the local structure of mRNA
Kathy Q. Luo1,2 and Donald C. Chang1
1 Department of Biology and 2 Department of Chemical Engineering, the Hong Kong University
of Science and Technology, Clear Water Bay, Hong Kong, China
Key words: RNA interference, siRNA, gene silencing, mRNA structure, H-bond.
The gene silencing effect of short interfering RNA (siRNA) is known to vary
strongly with the targeted position of the mRNA. A number of hypotheses have
been suggested to explain this phenomenon. We would like to test if this
positional effect is mainly due to the secondary structure of the mRNA at the
target site. We proposed that this structural factor can be characterized by a
single parameter called "the hydrogen bond (H-b) index", which represents the
average number of hydrogen bonds formed between nucleotides in the target
region and the rest of the mRNA. This index can be determined using a
computational approach. We tested the correlation between the H-b index and
the gene-silencing effects on three genes (Bcl-2, hTF and cyclin B1) using a
variety of siRNAs. We found that the gene-silencing effect is inversely
dependent on the H-b index, indicating that the local mRNA structure at the
targeted site is the main cause of the positional effect. Based on this finding, we
suggest that the H-b index can be a useful guideline for future siRNA design.
INTRODUCTION
RNA interference (RNAi) is an evolutionally conserved mechanism for repressing targeted
gene expression [1]. It exerts its silencing effect by mediating sequence-specific mRNA
degradation. In this process, double-stranded RNA (dsRNA) molecules are cleaved by a
ribonuclease (Dicer) into 21-23 bp fragments called "short interfering RNA" (siRNA) [2-4]. The
siRNA in turn binds to a protein complex called "RNA-induced silencing complex (RISC)", and
targets it to mRNAs that have complementary sequence with the siRNA [5-7]. Then, the
targeted mRNAs are destroyed through cleavage by RISC [8].
Application of siRNA has recently become an important tool for suppressing the expression
of specific genes. The efficiency of gene silencing, however, varied significantly between
siRNAs targeted to different positions of a gene [4,9-12]. At present, there is still a lack of clear
understanding on the mechanisms that determine the gene silencing efficiency of a given siRNA.
A number of hypotheses have been proposed in the literature, including: (a) Local protein
factor(s) on the mRNA may cause the positional effect [10]. (b) The local structure of the
targeted mRNA may affect the accessibility of the siRNA [7,11,12]. (c) Factors such as
sequence-dependent mRNA product
release or differential efficiency of 5' siRNA
phosphorylation may influence the efficacy of the siRNA [13]. Among these different proposals,
we think the structural factor may be the most important one. Two earlier comparative studies
between siRNA and antisense oligonucleotides (ASO) had suggested that the local structure of
the mRNA had a strong effect on the suppression of gene expression [7, 12]. One of these
studies utilized scanning oligonucleotide arrays to screen ASO that hybridized strongly to the
mRNA [12]. The local structure was not directly visualized in this case. In the other study,
*Corresponding author. Email: [email protected]
1
siRNAs targeted to two local structures (a stem and a loop) were analysed using a computational
approach [7]. Other more general structures have not yet been examined.
We have two major objectives in this study. First, we would like to find a quantitative
parameter that can characterize the accessibility of the siRNA to the targeted mRNA based on a
structural consideration. Second, we would like to test the structural hypothesis by examining
the correlation between the gene-silencing effect and the accessibility of the siRNA in a large
variety of local mRNA structures. The efficiency of mRNA degradation by the RISC should be
dependent on accessibility of the siRNA to the target region of the mRNA. Our hypothesis is
that, since nucleotides in the mRNA can often form hydrogen bonds (i.e., becoming double-
stranded) with other nucleotides in the same mRNA molecule, if the target region of the mRNA
has a more loosen structure (i.e. less hydrogen bonding), it will be easier for the siRNA to bind
with the targeted mRNA through base-pairing. This means that the mRNA accessibility can be
quantified using an H-b index (see below). Thus, if the positional effect of siRNA is determined
mainly by the local structure of mRNA at the target site, siRNAs with a lower H-b index should
have a higher efficiency in suppressing gene expression. Otherwise, alternative causes such as
local protein factors may be more important. Here, we reported results of several experimental
tests using three different genes, including Bcl-2, hTF and cyclin B.
MATERIALS AND METHODS
1. Preparation of siRNA duplexes
21-mer RNA oligonucleotides were synthesized by Dharmacon Inc. (Lafayette, CO). siRNA duplexes in the 2'-
deprotected and desalted form were dissolved in a 1X universal buffer (provided by the compony) in concentration
of 20 μM.
2. Determination of mRNA structure
The possible structures of a mRNA molecule was obtained using the "Mfold" web server, which provides
several closely related softwares for predicting the secondary structure of single stranded nucleic acids [14]. The
gateway for the Mfold web server is http://www.bioinfo.rpi.edu/ applications/mfold. The accession numbers of
genes used for Mfold analysis are the following: human Bcl-2 (AX057146), human Tissue Factor (hTF) (M16553)
and human cyclin B1 (NM_031966).
3. Electroporation and Western blotting
The siRNA was introduced into cells using an electroporation method based on our previously study [15]. HeLa
cells were grown in Minimum Essential Medium (MEM) supplemented with 10% fetal bovine serum (FBS). Cells
were then trypsinised and washed once with poration medium (PM), which contained 260 mM mannitol; 5 mM
sodium phosphate; 10 mM potassium phosphate; 1 mM MgCl2 and 10 mM HEPES (pH 7.3). A proper amount of
cells resuspended in PM was added with 2-4 μl of rhodamine-dextran (10 kDa) (1 mM), 8-10 μl of siRNA duplexes
(20 μM) and 3 μg of cyclin B-GFP plasmid DNA (only in the co-transfection experiments) to make up a total
volume of 100 μl. The cell mixture was transferred into a Gene Pluser® Cuvette (0.1 cm electrode) (Bio-Rad) and
electroporated using the GENE PULSER® II RF Module (Bio-Rad) under the conditions: 120 V and 10 RF pulses
with duration of 2.2 ms. Following electroporation, cells were transferred into a 100-mm culture plate containing
MEM plus 10% FBS and cultured at 37oC in a CO2 incubator. After 24 to 36 hrs, cells were collected and analyzed
using the standard Western blot technique. The antibodies used included: anti-Bcl2 (c-2) monoclonal antibody
(Santa Cruz); anti-ß tubulin (654162) monoclonal antibody (Calbiochem); anti-cyclin B1 (GNS-11) monoclonal
antibody (BD PharMingen); anti-Cdc2 monoclonal antibody (Santa Cruz) and anti-GFP polyclonal antibody
(Molecular Probes). Reactive bands were detected using the ECL™ system (Amersham) and their intensity was
measured using the MetaMorph software (Universal Imaging, West Chester, PA).
4. Imaging techniques
After cyclin B-GFP plasmid (with or without siRNA) was introduced into HeLa cells by electroporation, cells
were grown on an observation chamber that contained a glass coverslip at the bottom. After 24 or 36 hours, cells
were washed twice with culture medium. Their fluorescent images were then recorded using a fluorescence
microscope (Axiovert 35, Zeiss) equipped with a cooled CCD camera (MicroMax, by Princeton Instruments) and
processed digitally using the MetaMorph software.
RESULTS
2
1. The gene silencing efficiency of a siRNA is dependent on its targeting position in the gene
To demonstrate that the gene silencing effect of RNAi depends on the targeting region, we
synthesized two siRNA duplexes against different regions of Bcl-2 gene. The siRNA duplex was
introduced into HeLa cells using an electroporation method [15]. The effect of siRNA on Bcl-2
protein level was evaluated 48 hrs later by Western blot analysis using antibody against Bcl-2
(Fig. 1A). We found that the siRNA of Bcl-2-N (which targets to the N-terminus of Bcl-2
between amino acids 51-69) reduced the Bcl-2 level to 39% of the control, while siRNA of Bcl-
2-C (which targets to the C-terminus of Bcl-2 between amino acids 429-447) had little effect in
reducing the level of Bcl-2 protein (Fig. 1B). No reduction of β-tubulin protein level was
observed, suggesting that the silencing effect generated by the siRNA of Bcl-2-N was gene-
specific. This result showed that siRNAs against different regions of Bcl-2 gene can have
different silencing effects. We think that this result is due to the differences in the secondary
structures of Bcl-2 mRNA at the targeted sites. Indeed, using the Mfold program [14], we found
a loop structure in the mRNA at the region targeted by Bcl-2-N, while a hairpin structure was
seen in the mRNA complementing to Bcl-2-C (Fig. 1C). The difference in their local structures
may explain why the two siRNAs had different gene silencing efficiency.
Figure 1. Suppression of Bcl-2 gene expression. (A) Western blot analysis on HeLa
cells electroporated with or without siRNA duplexes Bcl-2-N or Bcl-2-C. (B) Based
on results of the Western blot, the expression levels of the Bcl-2 gene (normalised by
ß-tubulin protein level) were determined under conditions with or without the
application of the siRNAs. (C) The local secondary structures of Bcl-2 mRNA at the
regions targeted by the siRNAs Bcl-2-N and Bcl-2-C. The nucleotides targeted by
siRNA are colored in orange. The RNA structures were generated using the Mfold
software.
2. The secondary structure of the mRNA at the siRNA target region can be characterized by
an H-b index
The above explanation, however, is not totally satisfactory. Since as many as 21 possible
secondary structures can be generated for the Bcl-2 mRNA using the Mfold software, one
3
cannot rely on a single predicted structure of the mRNA to explain the effect of a given siRNA.
Instead, one needs to have a quantitative parameter that can characterize the over-all structural
effects. Thus, we propose to introduce a parameter called the "H-b index", which represents the
average number of hydrogen bonds formed in all possible secondary structures of Bcl-2 mRNA
at the region targeted by the siRNA (as determined using a computational approach). The H-b
index is defined by the following formula:
H-b index =
(PH-bond formation x NHB )i
(1)
Where, i is the index representing each mRNA nucleotide in the region targeted by the siRNA,
PH-bond formation is the probability that the ith nucleotide can form H-bonds with other nucleotides
within the same mRNA. PH-bond formation is calculated based on all possible structures of a mRNA
molecule predicted by the Mfold software, i.e.,
Number of structures that
PH-bond formation = 1
the ith nucleotide is in a single strand
Total number of possible structures
(2)
of the mRNA molecule
NHB is the number of hydrogen bonds that the ith nucleotide can form. NHB equals 3 for
nucleotides of G or C, and 2 for A or U. The possible structures of a mRNA molecule was
obtained using the "Mfold" web server, which provides several closely related softwares for
predicting the secondary structure of single stranded nucleic acids [14]. The gateway for the
Mfold web server is http://www.bioinfo.rpi.edu/ applications/mfold. Table 1 shows an example
of calculating the H-b index for the Bcl-2-N siRNA based on the Mfold analysis of the human
Bcl-2 mRNA.
Table 1. Calculation of the H-b index for the siRNA Bcl-2-N based on
21 secondary structures of Bcl-2 mRNA predicted by the Mfold program
Position and nucleotide
# of single strand
Probability of H-bond
of siRNA
structuresa
formationb
Average # of
H-bond formed
51 G
52 U
53 A
54 C
55 A
56 U
57 C
58 C
59 A
60 U
61 U
62 A
63 U
64 A
65 A
66 G
67 C
68 U
69 G
15
15
20
2
2
1
5
12
12
12
16
16
12
15
15
2
2
10
5
0.29
0.29
0.05
0.90
0.90
0.95
0.76
0.43
0.43
0.43
0.24
0.24
0.43
0.29
0.29
0.90
0.90
0.52
0.76
0.86
0.57
0.10
2.71
1.81
1.90
2.29
1.29
0.86
0.86
0.48
0.48
0.86
0.57
0.57
2.71
2.71
1.05
2.29
% of GC= 36.8%
H-b index = 24.95
Note: (a) Number of RNA structures having the nucleotide being single stranded. (b) The
definition of "probability of H-bond formation" is given in equation (2).
4
191i
Using the same method, we have also calculated the H-b index for the siRNA of Bcl-2-C. The
results, together with other relevant structural information, are summarized in Table 2. The H-b
index of Bcl-2-N (24.95) was found to be significantly lower than that of Bcl-2-C (32.38).
According to our hypothesis, a lower H-b index would indicate a better accessibility of the
siRNA to the targeted mRNA. Thus, the observation that Bcl-2-N was more effective in
silencing the Bcl-2 gene than Bcl-2-C is consistent with the prediction of our hypothesis.
Table 2. Properties of siRNAs targeted to the human Bcl-2 gene
siRNA name
siRNA sequence
Position RNA structure % of GC H-b index
Bcl-2-N
Bcl-2-C
GUACAUCCAUUAUAAGCUG
51-69
CUGGGGGAGGAUUGUGGCC
429-447
loop
hairpin
36.8
68.4
24.95
32.38
Note: The H-b index was determined using eq. (1) and based on the secondary structures of Bcl-2 mRNA predicted
by the Mfold program (14).
3. H-b index is highly correlated with the gene silencing efficiency of siRNA in human tissue
factor gene
In order to test the correlation between the H-b index and the siRNA efficiency more
extensively, we have analysed the positional effects of 14 different siRNAs in silencing the
human tissue factor (hTF) gene based on a recent study reported by Holen et al [10]. Figure 2
shows the local secondary structures of the hTF mRNA in regions targeted by various siRNAs.
(Here, siRNA-targeted nucleotides are highlighted in orange color). Based on structures
generated from the Mfold software, we calculated the H-b index for every siRNA designed for
the hTF gene (Table 3). Also, based on the experimental results reported by Holen et al [10], we
calculated the gene silencing efficiency of each siRNA on both the endogenous hTF gene and
the over-expressed hTF-luciferase (hTF-Luc) gene. These results are summarized in Table 3.
Table 3. Comparison of H-b index and efficiency of gene silencing
siRNA
name
77i
167i
256i
372i
459i
478i
562i
929i
158i
161i
164i
167i
170i
173i
176i
siRNA
sequence
uggagaccccugccuggcc
gcgcuucaggcacuacaaa
cccgucaaucaagucuaca
gaagcagacguacuuggca
cuccccagaguucacaccu
uaccuggagacaaaccucg
cggacuuuagucagaagga
gcuggaaggagaacucccc
agguggccggcgcuucagg
uggccggcgcuucaggcac
ccggcgcuucaggcacuac
gcgcuucaggcacuacaaa
cuucaggcacuacaaauac
caggcacuacaaauacugu
gcacuacaaauacuguggc
RNA
structure
stem
3 loops
stem
1 large loop
stem
hairpin
hairpin
hairpin
% of
GC
73.68
52.63
47.37
52.63
57.89
52.63
47.37
63.16
H-b
index
41.4
24.7
35.1
18.0
41.4
20.8
29.4
28.3
stem
stem
stem + loop
loops
loops
loops
1 large loop
73.68
73.68
68.42
52.63
42.11
42.11
47.37
36.5
35.1
29.0
24.7
20.2
22.4
28.1
% Reduction
in hTF-Luca
% Reduction
in hTFb
Average
reduction
13
80
71
84
22
16
50
19
27
26
58
79
59
68
44
5
85
43
76
2
13
13
5
0
26
71
87
34
68
40
9.1
82.3
57.0
80.1
12.0
14.5
31.3
11.9
13.5
25.8
64.3
83.2
46.6
67.9
41.8
Note:(a) Data were based on a previous study reported by (10). (b) Reduction of endogenous hTF mRNA
levels were measured based on northern blot analysis of hTF mRNA [Fig. 3A & 3B in (10)]. The mRNA levels
were normalised to a loading control (GAPDH) and standardized to mock-transfected cells.
5
The data in Table 3 were organized into two groups: The first group consists of siRNAs
targeted to different parts of the hTF gene, while the second group consists of overlapping
siRNAs targeted to neighboring regions of one specific site in the hTF gene (Figs. 2B & 2I).
From results shown in group #1, one can easily see that: (1) The gene silencing effects was not
correlated with the order of position of the siRNA-targeting region. (2) All siRNAs containing a
hairpin structure were generally not very effective in gene silencing. (3) For those siRNAs that
had a significant effect of gene silencing, their H-b index values were relatively low.
Figure 2. Local secondary structures of hTF (human tissue factor) mRNA. All local
structures were from the predicted structure #1 using the Mfold program. Panels A-H shows the
structures targeted by siRNA of 77i (A), 167i (B), 256i (C), 372i (D), 459i (E), 478i (F), 562i (G)
and 929i (H). Here, the siRNA targeting sequence was colored in orange. Panel I shows the
nucleotide sequence of hTF mRNA at the targeting site surrounding siRNA 158i-176i. Seven
siRNAs were generated by shifting the nucleotide sequence in increments of 3. Their structures
can be seen in Panel B.
Next, we conducted a more quantitative analysis by plotting the percentage of hTF gene
reduction versus the H-b index for all non-hairpin siRNAs in group #1 (Fig. 3A). It gave an
almost linear relationship (the coefficient of determination R2 = 0.879). (Note: A perfect linear
correlation would give R2 = 1.0). One may question the validity of our analysis by pointing out
6
that, since a siRNA having a higher % of GC content may in general tend to have a larger H-b
index, the results shown in Fig. 3A may simply indicate that the gene-silencing effect is related
to the GC content of the siRNA. Thus, for purpose of comparison, we have also plotted the
percentage of hTF gene reduction versus the % of GC content of the same siRNAs (Fig. 3B).
The correlation was far less clear (R2 = 0.407). These results suggest that the efficiency of gene
silencing for hTF had a higher correlation with the local mRNA structure than with the GC
content.
This conclusion is further supported by data from group #2 of Table 3, where seven
overlapping siRNAs were designed to target to one specific region of the hTF gene (Figs. 2B &
2I). We have examined the correlation between the percentage of reduction in the reporter gene
(hTF-Luc) and the H-b index (Fig. 3C) or the % of GC content (Fig. 3D). Again, a higher linear
regression (R2=0.706) was found in the H-b index, while a lesser correlation (R2=0.449) was
seen between the effect of RNA interference and the GC content.
Figure 3. Relationship between siRNA efficacy and the H-b index or GC content. Using
data summarized in Table 3, we analyzed the gene silencing effect vs the H-b index (A, C & E)
or the GC content (B, D). The gene silencing efficiency was based on the measured reduction
of expression either in the endogenous hTF gene (A, B) or in the hTF-Luc reporter gene (C, D
& E). Panels A and B were based on data of non-hairpin siRNAs listed in group #1 of Table 3.
Panels C and D were based on data of seven overlapping siRNA listed in group #2 of Table 3.
Panel E shows the combined result from 10 non-hairpin siRNAs listed in Table 3.
7
Finally, we combined both Group #1 and Group #2 data and plotted the efficiency of gene
silencing (based on % of suppression in expressing the hTF-Luc reporter gene) against the H-b
index for ten siRNA samples that did not contain a hairpin structure* (77i, 158i, 161i, 164i, 167i,
170i, 173i, 176i, 372i and 459i) (Fig. 3E). We found a highly significant correlation between the
gene silencing effect and the H-b index (R2 = 0.863), indicating that the gene-silencing
efficiency is largely dependent on the secondary structure of the mRNA at the target site. From
these results, we concluded that: (1) siRNA having a lower H-b index in general would have a
higher gene-silencing efficiency, and (2) siRNA forming a hairpin structure usually would be
less effective in gene-silencing. (*Note: Fig. 3E did not include the data of siRNA 256i, which
hybridized with both the 5' and 3' ends of the mRNA. Since regulatory proteins binding to these
terminal regions may interfere with the local secondary structure, they could affect the siRNA-
mRNA interaction at this target site).
4. Testing the prediction of our model using siRNAs against different regions of cyclin B1
gene
In order to further test the above observations, we have designed several siRNAs against the
human cyclin B1 gene at different regions. Their properties are summarised in Table 4. These
siRNAs were specially targeted to loop (L1, L2), stem (S1) or hairpin (H1) structures of the
human cyclin B1 mRNA. These siRNAs have similar GC content but different H-b index values
(L1<L2<S1). If our structural hypothesis is correct, L1 should have the highest efficiency in
silencing the cyclin B1 gene, L2 should have an intermediate effect, and S1 and H1 should be
least effective in gene silencing.
Table 4. Properties of siRNAs targeted to the human cyclin B1 gene
siRNA
name
siRNA sequence
Position
RNA structure % of GC H-b index
L1
L2
S1
CAGCUACUGGAAAAGUCAU 383-401
GAGCCAUCCUAAUUGACUG 782-800
CCUGAGCCUAUUUUGGUUG 526-544
loop
loop
stem
H1
AGCCCAAUGGAAACAUCUG 559-577
hairpin
42.1%
47.4%
47.4%
47.4%
17.06
25.06
35.66
25.38
In this study, siRNA of L1, L2, S, or H1 was introduced into HeLa cells by electroporation.
After 32 or 48 hrs, cell extracts were collected and analyzed by SDS-PAGE followed with
Western blot analysis using antibody against cyclin B1 (Fig. 4A). The amount of cyclin B1
protein was determined quantitatively by measuring the intensity of protein bands of cyclin B
which was normalized by the level of Cdc2 detected in the same protein blot. The normalised
results are shown in Figure 4B. At 32 hrs after electroporation, application of the L1 siRNA was
found to produce a very strong gene silencing effect; it reduced the expression level of the
endogenous cyclin B1 gene to 6% of the control. By comparison, application of the L2 siRNA
only reduced the cyclin B1 level to 25%. The siRNAs of S1 and H1, on the other hand, did not
show any significant gene silencing activity. Similar results were obtained from cells collected
at 48 hrs after applying the various siRNAs (data not shown). These results demonstrated that a
low H-b index value was indeed correlated with a high gene-silencing effect, and siRNA
containing a hairpin structure was generally ineffective.
8
Figure 4. Gene silencing effects of siRNAs on the endogenous and exogenous expression of
cyclin B1. (A) Western blot analysis of cyclin B1 protein in HeLa cells at 32 hrs after
electroporation with various siRNA duplexes. Control cells were electroporated with buffer only.
Protein levels of Cdc2 in the same samples were shown here as loading controls. (B) The relative
levels of the expressed cyclin B1 under various conditions were determined and normalised by
the levels of Cdc2. (C) The expressed level of cyclin B1-GFP (with or without L1 siRNA) at 24
and 36 hrs after gene transfection were analysed by Western blot analysis. Control sample had
neither siRNA nor cyclin B1-GFP. (D) Fluorescence images showing the expression of cyclin
B1-GFP with or without applying the L1 siRNA. The images for rhodamine indicated that all
cells under study were successfully electroporated. Scale bar: 10 μm.
To further verify our experimental results, we have examined the effects of different siRNAs
on repressing the expression of an exogenous cyclin B1-GFP fusion gene. Plasmid DNA
containing the cyclin B1-GFP gene was introduced into HeLa cells by electroporation with or
without the L1 siRNA. The efficiency of siRNA up-take was monitored by co-electroporation
with rhodamine-dextran (10 kDa). Results of gene expression assays based on Western blot
analysis are shown in Figure 4C. In the absence of L1 siRNA, cyclin B1–GFP was found to
express at a high level at 24 and 36 hrs after electroporation. In the presence of L1 siRNA, the
level of cyclin B1-GFP protein observed at 24 hrs reduced dramatically, and no cyclin B1-GFP
was detected at 36 hrs.
Similar findings were also obtained using an imaging technique. As shown in Figure 4D, in
the absence of L1 siRNA, a high level of cyclin B1-GFP fluorescent protein was observed in the
transfected HeLa cells at either 24 or 36 hrs after electroporation. However, in cells co-
electroporated with both cyclin B1-GFP and L1 siRNA, very little GFP fluorescence signal was
detected. These results again demonstrated that the expression of cyclin B1-GFP gene was
almost completely inhibited by the L1 siRNA (Fig. 4D).
DISCUSSION
RNA interference is a powerful approach for studying gene function in many organisms.
Thus, there is a strong interest to develop and improve this technique. At this time, a key
question is how to design siRNAs that can have high efficiency in gene silencing. We know that
the effectiveness of a siRNA is highly dependent on its target position [10,16]. The mechanism,
however, was not clear. Some hypothesized that local protein factors on different regions of
9
mRNA might cause the positional effect [10]. Others suggested that the activity of siRNA is
mainly affected by the secondary structure of mRNA at the target site [7,11,12]. Although these
two proposed mechanisms are not mutually exclusive, in this study, we found evidence
suggesting that the local structure of mRNA at the target site is a dominant factor. Furthermore,
we showed that one can use a single parameter, the H-b index, to characterize the overall
structural effects. This information, we believe, can be greatly helpful in optimizing the design
of effective siRNAs for silencing a specific gene.
The major advantage of using the H-b index is that one can bypass the cumbersome process
of guessing the correct secondary structure of a mRNA molecule at local regions. At present,
computer softwares can only predict a set of possible secondary structures for a given mRNA.
For example, using the Mfold program, we can obtain 32 predicted RNA structures for the
human cyclin B1 mRNA. It is difficult to know which one of the 32 possible structures
represents the real folding of the mRNA in a cell. In this work, we proposed to use a statistical
approach to solve this problem. By introducing the concept of the H-b index, we can use a single
parameter to reflect the overall probability for nucleotides within the siRNA targeting region to
form double stranded complex with other parts of the mRNA. A low value of the H-b index
would mean that most of the nucleotides within the target region are in single stranded structures
and thus are more likely to be accessible by the RISC/siRNA complex.
In addition to testing a variety of siRNAs targeting to different regions of the Bcl-2 and
cyclin B1 genes, we have also tested our hypothesis by re-analysing the data of an independent
study by Holen et al., who had examined the effects of a large number of siRNAs targeted to
different regions of the hTF gene [10]. We found a very significant correlation between the H-b
index and the siRNA efficacy (Fig. 3). These results strongly suggest that the H-b index can be a
useful indicator for predicting the gene silencing efficiency of siRNA.
Other than the H-b index, we also found that formation of a hairpin structure within the
siRNA target region can greatly reduce the efficiency of the siRNA. As shown in Table 3 and
Fig. 4A & B, low silencing efficiency was detected for most siRNAs targeting to sites with a
hairpin structure. We think this is because such siRNA may tend to form a hairpin structure by
itself and thus cannot be fully open. As a result, the RISC/siRNA complex will be less effective
in binding with the complementary mRNA.
In summary, based on results of this work, we proposed two guidelines for selecting siRNA
target sites for effective RNA interference: (1) It is preferable to choose a target region that has a
low H-b index (ideally less than 25). (2) One should avoid target regions where the mRNA can
form a hairpin structure. Besides these, one may also consider to stay away from target sites at
either the 5' or 3' ends of the mRNA. Since proteins involved in translational regulation or
mRNA processing may bind to these terminal regions, they could interfere with the siRNA-
mRNA interaction.
ACKNOWLEDGEMENTS
This work was supported by the Research Grants Council of Hong Kong (HKUST6109/01M,
HKUST6104/02M, and AoE) and HIA project of HKUST.
REFERENCES
[1] A. Fire, S. Xu, M.K. Montgomery, S.A. Kostas, S.E. Driver, C.C. Mello, Potent and specific genetic
interference by double-stranded RNA in Caenorhabditis elegans, Nature (1998) 806-811.
[2] E. Bernstein, A.A. Caudy, S.M. Hammond, G.J. Hannon, Role for a bidentate ribonuclease in the initiation step
of RNA interference, Nature (2001) 363-366.
[3] P.D. Zamore, T. Tuschl, P.A. Sharp, D.P. Bartel, RNAi: double-stranded RNA directs the ATP-dependent
cleavage of mRNA at 21 to 23 nucleotide intervals, Cell (2000) 25-33.
10
[4] S.M. Elbashir, J. Harborth, W. Lendeckel, A. Yalcin, K. Weber, T. Tuschl, Duplexes of 21-nucleotide RNAs
mediate RNA interference in cultured mammalian cells, Nature (2001) 494-498.
[5] S.M. Hammond, E. Bernstein, D. Beach, G.J. Hannon, An RNA-directed nuclease mediates post-transcriptional
gene silencing in Drosophila cells, Nature (2000) 293-296.
[6] S.M. Hammond, A.A. Caudy, G.J. Hannon, Post-transcriptional gene silencing by double-stranded RNA,
Nature Reviews Genetics (2001) 110-119.
[7] R. Kretschmer-Kazemi Far, G. Sczakiel, The activity of siRNA in mammalian cells is related to structural
target accessibility: a comparison with antisense oligonucleotides, Nucleic Acids Research (2003) 4417-4424.
[8] G.J. Hannon, RNA interference, Nature (2002) 244-251.
[9] J. Harborth, S.M. Elbashir, K. Bechert, T. Tuschl, K. Weber, Identification of essential genes in cultured
mammalian cells using small interfering RNAs, J Cell Sci (2001) 4557-4565.
[10] T. Holen, M. Amarzguioui, M.T. Wiiger, E. Babaie, H. Prydz, Positional effects of short interfering RNAs
targeting the human coagulation trigger Tissue Factor, Nucleic Acids Research (2002) 1757-1766.
[11] T.A. Vickers, S. Koo, C.F. Bennett, S.T. Crooke, N.M. Dean, B.F. Baker, Efficient reduction of target RNAs
by small interfering RNA and RNase H-dependent antisense agents. A comparative analysis, Journal of
Biological Chemistry (2003) 7108-7118.
[12] E.A. Bohula, A.J. Salisbury, M. Sohail, M.P. Playford, J. Riedemann, E.M. Southern, V.M. Macaulay, The
efficacy of small interfering RNAs targeted to the type 1 insulin-like growth factor receptor (IGF1R) is
influenced by secondary structure in the IGF1R transcript, Journal of Biological Chemistry (2003) 15991-
15997.
[13] A. Nykanen, B. Haley, P.D. Zamore, ATP requirements and small interfering RNA structure in the RNA
interference pathway, Cell (2001) 309-321.
[14] M. Zuker, Mfold web server for nucleic acid folding and hybridization prediction, Nucleic Acids Research
(2003) 3406-3415.
[15] D.C. Chang, Experimental strategies in efficient transfection of mammalian cells. Electroporation, Methods
Mol Biol (1997) 307-318.
[16] M. Miyagishi, M. Hayashi, K. Taira, Comparison of the suppressive effects of antisense oligonucleotides and
siRNAs directed against the same targets in mammalian cells, Antisense & Nucleic Acid Drug Development
(2003) 1-7.
11
|
1612.06901 | 1 | 1612 | 2016-12-20T22:13:02 | Emergent structures and dynamics of cell colonies by contact inhibition of locomotion | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.TO"
] | Cells in tissues can organize into a broad spectrum of structures according to their function. Drastic changes of organization, such as epithelial-mesenchymal transitions or the formation of spheroidal aggregates, are often associated either to tissue morphogenesis or to cancer progression. Here, we study the organization of cell colonies by means of simulations of self-propelled particles with generic cell-like interactions. The interplay between cell softness, cell-cell adhesion, and contact inhibition of locomotion (CIL) yields structures and collective dynamics observed in several existing tissue phenotypes. These include regular distributions of cells, dynamic cell clusters, gel-like networks, collectively migrating monolayers, and 3D aggregates. We give analytical predictions for transitions between noncohesive, cohesive, and 3D cell arrangements. We explicitly show how CIL yields an effective repulsion that promotes cell dispersal, thereby hindering the formation of cohesive tissues. Yet, in continuous monolayers, CIL leads to collective cell motion, ensures tensile intercellular stresses, and opposes cell extrusion. Thus, our work highlights the prominent role of CIL in determining the emergent structures and dynamics of cell colonies. | physics.bio-ph | physics | a
i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 1 - #1
Emergent structures and dynamics of cell colonies
by contact inhibition of locomotion
Bart Smeets ∗, Ricard Alert †, Jir´ı Pesek ∗ , Ignacio Pagonabarraga † , Herman Ramon ∗ and Romaric Vincent ‡
∗Division of Mechatronics, Biostatistics and Sensors (MeBioS) Kasteelpark Arenberg 30 - box 2456 3001 Leuven, Belgium,†Departament de F´ısica de la Mat`eria Condensada
& Universitat de Barcelona Institute of Complex Systems (UBICS), Facultat de F´ısica, Universitat de Barcelona, 08028 Barcelona, Spain, and ‡Universit´e Grenoble Alpes,
Commissariat `a l'´Energie Atomique (CEA) & Laboratoire d'´Electronique des Technologies de l'Information (LETI), Micro and Nanotechnology Innovation Centre (MINATEC),
F-38054 Grenoble, France
Submitted to Proceedings of the National Academy of Sciences of the United States of America
tion, which may ultimately have a deep impact on the overall
organization of the colony.
Such a repulsive interaction mediated by adhesion is indeed
present in many cell types upon cell-cell contact, and is known
as contact inhibition of locomotion (CIL) after Abercrombie
and Heaysman [19]. Upon a cell-cell collision, the cell front
adheres to the colliding cell, which hinders further cell protru-
sions. Subsequently, repolarization of the cell's cytoskeleton
creates a new front away from the adhesion zone, and the two
cells thus separate [20, 21]. This interaction has been shown to
be crucial in determining the collective behavior of cell groups
in several contexts [22]. For example, CIL guides the direc-
tional migration of neural crest cells [23], and also ensures the
correct dispersion of Cajal-Retzius cells in the cerebral cortex
[24] or of hemocytes in the embryo [25].
Here, we model cellular interactions by means of an attrac-
tion due to intercellular adhesion, and a soft repulsion associ-
ated to the reduction of cell-substrate adhesion area. In ad-
dition, CIL is modeled as an interaction orienting cell motil-
ity away from cell-cell contacts. We analytically show how
CIL acts as an effective repulsive force that hinders the for-
mation of cohesive cell monolayers or 3D tissues at increas-
ing cell-cell adhesion. We then explicitly predict the transi-
tions between non-cohesive, cohesive, and overlapped organi-
Significance
The regular distribution of mesenchymal cells, the formation of
epithelial monolayers or their collapse into spheroidal tumors
illustrate the broad range of possible organizations of cells in
tissues. Unveiling a physical picture of their emergence and dy-
namics is of critical importance to understand tissue morphogen-
esis or cancer progression. Although the role of cell-substrate
and cell-cell adhesion in the organization of cell colonies has
been widely studied, the impact of the cell-type-specific contact
inhibition of locomotion (CIL) remains unclear. Here, we in-
clude this interaction in simulations of active particles, and find
a number of structures and collective dynamics that recapitulate
existing tissue phenotypes. We give analytical predictions for the
epithelial-mesenchymal transition and the formation of 3D ag-
gregates as a function of cell-cell adhesion and CIL strengths.
Thus, our findings shed light on the physical mechanisms under-
lying multicellular organization.
Reserved for Publication Footnotes
Cells in tissues can organize into a broad spectrum of structures ac-
cording to their function. Drastic changes of organization, such as
epithelial-mesenchymal transitions or the formation of spheroidal ag-
gregates, are often associated either to tissue morphogenesis or to
cancer progression. Here, we study the organization of cell colonies
by means of simulations of self-propelled particles with generic cell-
like interactions. The interplay between cell softness, cell-cell adhe-
sion, and contact inhibition of locomotion (CIL) yields structures and
collective dynamics observed in several existing tissue phenotypes.
These include regular distributions of cells, dynamic cell clusters,
gel-like networks, collectively migrating monolayers, and 3D aggre-
gates. We give analytical predictions for transitions between non-
cohesive, cohesive, and 3D cell arrangements. We explicitly show
how CIL yields an effective repulsion that promotes cell dispersal,
thereby hindering the formation of cohesive tissues. Yet, in contin-
uous monolayers, CIL leads to collective cell motion, ensures tensile
intercellular stresses, and opposes cell extrusion. Thus, our work
highlights the prominent role of CIL in determining the emergent
structures and dynamics of cell colonies.
Self-propelled particles Cell-cell adhesion Contact Inhibition of Locomotion
Cell monolayers Collective motion
C ell colonies exhibit a broad range of phenotypes. In terms
of structure, collections of cells can arrange into distribu-
tions of single cells, assemble into continuous monolayers or
multi-layered tissues, or even form 3D agglomerates. In terms
of dynamics, cell motility may be simply absent, or produce
random, directed or collective migration of cells. Transitions
between these states of tissue organization are characteristic of
morphogenetic events and are also central to tumor formation
and dispersal [1, 2, 3, 4]. Therefore, a physical understand-
ing of the collective behavior of cell colonies will shed light
on the regulation of many multicellular processes involved in
development and disease.
However, a complete physical picture of multicellular or-
ganization is not yet available, partly due to the challenge of
modeling the complex interactions between cells. Here, we ad-
dress this problem by means of large-scale simulations of self-
propelled particles (SPP) endowed with interactions capturing
generic cellular behaviors. Models of SPP with aligning inter-
actions have been used to investigate collective cell motions in
tissue monolayers [5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18].
We extend this approach to unveil how the different structures
and collective dynamics of cell colonies emerge from cell-cell
interactions.
In addition to an excluded-volume repulsion, cells gener-
ally feature a short-range attraction as a consequence of their
active cortical contractility transmitted through cell-cell junc-
tions. With no additional interactions, this attraction would
typically lead to cohesive tissues. However, not all cell types
form cohesive tissues. Whereas epithelial cells tend to form
continuous monolayers, mesenchymal cells separate after divi-
sion despite the presence of cell-cell junctions. This observa-
tion calls for an extra effective repulsion to drive the separa-
www.pnas.org/cgi/doi/10.1073/pnas.0709640104
PNAS
December 20, 2016
vol. 113
no. 51
1–8
i
i
i
i
i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 2 - #2
i
i
zations of the colonies as a function of cell-cell adhesion and
CIL strength. In simulations, we identify states with different
structures and emergent dynamics, including ordered or dy-
namic arrangements of clusters, gel-like networks, active gas
and polar liquid states, and 3D aggregates. The results may
be interpreted in biological terms by associating each state
to common phenotypes, namely grid-like distributions of mes-
enchymal cells, collectively migrating epithelial monolayers,
and cellular spheroids. The soft character of the potential and
CIL interactions are key in producing structures and collective
behaviors observed in cell colonies. In particular, the former
enables the formation of 3D tissues via cell extrusion. In turn,
CIL gives rise to self-organized collective motion in continuous
cell monolayers. In line with [17], we find that this effective
repulsion induces tensile stresses in cell monolayers.
Model
We model a 2D colony of cells as a suspension of over-
damped self-propelled disks. Neglecting translational noise,
the equation of motion of cell i with position xi and polarity
pi = (cos θi, sin θi) reads
(cid:2)F cc
ij nij + γ ( xi − xj)(cid:3) ,
nn(cid:88)
j
Fmpi = γs xi +
[ 1 ]
for contacting nearest neighbor cells j, with nij = (xj−xi)/dij
and dij =xj −xi. Here, Fm is the magnitude of the cell self-
propulsion force, and γs and γ are cell-substrate and cell-cell
friction constants, respectively.
The central force F cc
ij
includes a soft repulsion F r
ij = 2Ws/R2 (2R − dij), with Ws =(cid:82) 2R
ij associ-
ated to the reduction of the cell-substrate adhesion area when
two cells are closer than their spread size 2R, and an attrac-
tive force F a
ij that accounts for active contractility transmit-
ted through cell-cell adhesions. F r
ij is assumed to increase
linearly with decreasing intercellular distance dij up to dij=R.
Hence, F r
ij ddij the
cell-substrate adhesion energy (gray in Fig. 1a). No further re-
duction of the cell-substrate contact area is allowed for dij < R.
As a result, cells can approach at smaller distances under com-
pression. In this regime cells do not exert any force on the sub-
strate and are considered to be extruded from the monolayer
(Fig. 1a-b). Cell extrusions may lead to 3D tissues, whose
structure and dynamics are not described by our 2D model.
F a
ij is assumed to increase linearly with distance up to dij = 2R.
Hence, F a
ij ddij the
cell-cell adhesion energy (red in Fig. 1a). Accordingly, the
ij = −2Wc/R2 (dij − R), with Wc =(cid:82) 2R
R F a
R F r
Fig. 2. Phase behavior of cell colonies as a function of cell-cell adhesion W c and
cell repolarization rate ψ associated to CIL. Colors indicate the predicted regions for
non-cohesive (green), cohesive (blue), and overlapped (red) organizations.
In addi-
tion to capturing these structural transitions, simulations allow to identify dynamically
distinct states such as an active gas, a cluster crystal, a gel-like percolated network,
dynamic clusters, and an active polar liquid, as illustrated in snapshots.
total interaction force (black in Fig. 1a) reads
(cid:40) 2
R [Ws − Ws+Wc
R
F cc
ij (dij) =
(dij − R)],
if R ≤ dij ≤ 2R
0,
else.
[ 2 ]
In turn, CIL tends to orient the cell polarity pi in the di-
rection pf
i pointing away from the weighted average position
of the contacting cells (Fig. 1c and SI Appendix). Similarly
to [18], we model this interaction via a harmonic potential for
the polarization angle θi that, in addition to rotational noise,
yields
θi =−fcil(θi − θf
i ) +
√
2Dr ξ.
[ 3 ]
(a)
(b)
(c)
Fig. 1. A model of self-propelled particles with cell-like interactions. (a) Central
cell-cell force F cc
ij (black) including a soft repulsion due to reduction of cell-substrate
adhesion area (gray) and attraction due to active contractility through cell-cell ad-
hesions (red). (b) Cell extrusion for intercellular distances dij < R, resulting in
vanishing cell-cell forces in the plane. (c) Cellular self-propulsion force Fm in the
direction of the cell polarity pi. CIL rotates the polarity towards the direction pf
i
pointing away from cell-cell contacts.
Here, fcil is the cellular repolarization rate upon cell-cell con-
tact, whereas ξ (t) is a typified Gaussian white noise, and Dr
is the rotational diffusion coefficient of cell motion.
The parameters of the model may be reduced to five di-
mensionless quantities: the packing fraction of cells φ, cell-cell
and cell-substrate adhesion energies W c := Wc/(2RFm) and
W s := Ws/(2RFm), cell-cell friction γ := γ/γs, and a param-
eter ψ := fcil/ (2Dr) that compares the timescale of cytoskele-
tal repolarization associated to CIL to the rotational diffusion.
Hereafter, we set φ = 0.85, W s = 1, γ = 0, and focus on the
effects of intercellular adhesion and CIL on the organization
of cell colonies. The results are summarized in the phase di-
agram of Fig. 2. Including cell-cell friction leads to jammed
configurations of cohesive tissues (SI Appendix), in line with
[16]. In turn, cell density does not affect the phase transitions
but modifies the dynamical behavior of the cell colony (SI Ap-
pendix). Thus, cell proliferation may drive the colony through
different dynamical states (SI Appendix).
2
www.pnas.org/cgi/doi/10.1073/pnas.0709640104
Footline Author
i
i
i
i
extrusion0.00.20.40.60.81.01.2Wc0.00.20.40.60.81.01.21.41.6ψPolar liquid3D aggregatesGasGelSlowly evolvingDynamic clustersCluster crystalβDpNPN›dijfic i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 3 - #3
i
i
(a)
(b)
Results
Non-cohesive phase. We first study the transition between a
cohesive phase in which cells remain in contact, dij < 2R,
and a non-cohesive phase in which they lose contact. Loss
of cell contact is only possible if the maximal attractive force
ij (2R) = −2Wc/R, is overcome by the compo-
at dij = 2R, F cc
nent of the cells' self-propulsion force along the interparticle
axis. Such component depends on the relative alignment of
self-propulsion forces, and hence on CIL. When averaged over
orientations, self-propulsion forces yield an effective central
repulsion F p
between cells that depends on their
repolarization rate ψ (SI Appendix).
In the relevant limit
ψ (cid:29) 1/ (2π) (Discussion), it reads
ij =(cid:104)Fmpi(cid:105)θi
(cid:18)
(cid:19)
ij = (cid:104)Fmpi(cid:105)θi
F p
≈ Fm exp
− 1
4ψ
nij.
[ 4 ]
t = 128
320
(c)
514
704
2000
Then, within this mean-field approximation, the condition
F p
ij + F cc
ij (2R) = 0 gives a prediction for the transition be-
tween the non-cohesive (green in Fig. 2) and cohesive (blue in
Fig. 2) phases. This sets a critical adhesion energy
(cid:18)
(cid:19)
W
coh
c =
1
4
exp
− 1
4ψ
,
[ 5 ]
above which cells are expected to be in contact or, alterna-
tively, a critical CIL rate above which cohesiveness is lost.
Therefore, at low cell-cell adhesion, CIL promotes cell disper-
sal, thereby hindering the formation of cohesive tissues.
In simulations, we quantify this transition in terms of parti-
cle number fluctuations. Phase separated self-propelled disks
feature giant number fluctuations [9, 26, 27]. There, the stan-
dard deviation of the number of particles N in a given region
scales as σN ∼ N β for large N , with β ≈ 1, whereas a system
at equilibrium would feature β = 1/2. Similarly, we compute
the exponent β (Fig. 3a) and identify the regions with β > 1/2
as phase-separated, and thus cohesive. Consequently, we iden-
tify the transition to the cohesive phase from the onset of giant
number fluctuations (triangles in Fig. 2), which qualitatively
agrees with the mean-field analytical prediction.
Within the non-cohesive phase (green in Fig. 2), the colony
forms an active gas state with equilibrium-like statistics (β ≈
(a)
(b)
Fig. 3. Number fluctuations and diffusion in cell colonies. (a) Exponent of
number fluctuations σN ∼ N β as a function of cell-cell adhesion W c and CIL repo-
larization rate ψ. Phase-separated states feature giant number fluctuations (β > 1/2)
whose onset identifies the transition to the cohesive phase (triangles in Fig. 2). In the
non-cohesive phase, colonies of slowly repolarizing cells (low ψ) feature equilibrium-like
fluctuations (β≈ 1/2), whereas faster repolarizations (higher ψ) induce a hyperuni-
form distribution of cells (β < 1/2). (b) Cell diffusion coefficient D as a function of
ψ for some values of W c. For increasing repolarization rate ψ, D initially increases
but then decreases as clusters form. The maximum of D (ψ) identifies the onset of
clustering (squares in Fig. 2). Dfree = Fm/(2γsDr) is the translational diffusion
coefficient of a persistent random walker with rotational diffusion [28].
Fig. 4. Dynamics and phase-separation kinetics in cell colonies. (a) MSD Ex-
ponent (∆x)2 ∼ tα as a function of cell-cell adhesion W c and CIL repolarization
rate ψ. The colony forms a gel-like network with subdiffusive dynamics (α < 1) at
low ψ. Faster CIL gives rise to collective cell motion as indicated by almost ballistic
dynamics (α = 2). (b) Evolution of the average domain size L (t), computed from
the structure factor (SI Appendix), for different W c at ψ = 1. Dimensionless time
reads t = Fm/(2Rγs)t. The colony phase separates for W c (cid:38) 0.4. CIL yields
faster phase-separation kinetics than the diffusive coarsening dynamics of passive sys-
tems, for which L (t)∼ t1/3 [33]. (c) Illustration of the phase separation from an
initial random configuration towards the active polar liquid at ψ = 1 and W c = 0.7.
1/2) at low CIL repolarization rates ψ (Movie S1). At larger
ψ, cells get hyperuniformly distributed, with β < 1/2 (Fig.
3a), forming a crystal of small cell clusters (Movie S2). This
state is reminiscent of the equilibrium cluster crystals formed
by purely repulsive soft spheres [29]. In our case, an effective
repulsion arises from anti-aligned propulsion forces via CIL
(Eq. 4). Similarly to [30], we set a dynamical criterion for
the clustering transition based on the cell diffusion coefficient
D obtained from the long-time mean-squared displacement
(MSD), limt→∞ ∆x2 = 4Dt. Increasing the repolarization
rate ψ initially enhances diffusion by promoting cluster evap-
oration. However, the stronger effective repulsion at larger ψ
progressively prevents cells from escaping the clusters, hence
reducing diffusion until it is eventually solely due to interclus-
ter hopping events [31] (see SI Appendix for a discussion on the
dependence of D on ψ). Consequently, we locate the clustering
transition (squares in Fig. 2) from the maximum of D (ψ) at
each W c (Fig. 3b). Increasing cell-cell adhesion favors cluster-
ing, thereby enabling the short-range CIL-associated repulsion
responsible for the crystalline order.
Cohesive phase. Increasing cell-cell adhesion beyond the tran-
sition to the cohesive phase (blue in Fig. 2), the colony initially
forms a percolating structure of clusters. At low CIL repolar-
ization rate ψ, cells arrange in a network with very slow, subd-
iffusive dynamics, as shown by the MSD ∆x2∼ tα with α < 1
(Fig. 4a). Thus, due to cell-cell adhesion, the colony forms a
near-equilibrium attractive gel [32] with few cell rearrange-
ments (Movie S3). At larger repolarization rates ψ (above
squares in Fig. 2, see Fig. 3b), the effective CIL-associated
repulsion yields smaller, dynamic, and locally crystalline clus-
ters (Movie S4). They arise from a kinetic balance between
the CIL-enhanced evaporation and the adhesion-induced con-
densation of clusters that prevents the completion of phase
separation into a continuous dense phase.
Footline Author
PNAS
December 20, 2016
vol. 113
no. 51
3
i
i
i
i
0.00.20.40.6Wc0.00.40.81.21.6ψ0.3500.4000.4500.5000.5000.6000.6000.7000.7000.8000.9000.40.50.60.70.80.9β0.00.51.01.5ψ0.00.10.20.30.4D/DfreeWc= 0Wc= 0.1Wc= 0.2Wc= 0.3Wc= 0.40.00.20.40.60.8Wc0.00.40.81.21.6ψ0.8001.0001.4001.2001.6000.81.01.21.41.61.82.0α100101102103104t100101102(t)ψ = 1.0~t1/3Wc = 0.1Wc = 0.2Wc = 0.3Wc = 0.4Wc = 0.5Wc = 0.6Wc = 0.7Wc = 0.8 i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 4 - #4
i
i
Complete phase separation occurs at larger cell-cell adhe-
sion, W c (cid:38) 0.4. The coarsening dynamics (Fig. 4b-c) are
much faster than in a passive system, for which particle do-
mains grow by diffusion as L (t)∼ t1/3 [33]. By orienting cell
motility towards free space, CIL induces an advective coars-
ening of the cell domains that enables a fast phase separation
of cell colonies.
Upon phase separation, the colony forms a continuous
cell monolayer that exhibits self-organized collective motion
(Movie S5). This is reflected in the MSD exponent, that
evolves from diffusive (α = 1) towards almost ballistic (α = 2)
above W c≈ 0.4 (Fig. 4a). CIL induces a coupling between cell
polarity and density fluctuations in the fluid phase that gives
rise to a macroscopic polarization via a spontaneous symme-
try breaking. The outward motion of cells at the boundary
of the monolayer creates free space behind them, which polar-
izes neighboring cells before the leading cell can reorient back.
Through this mechanism, self-organized collective cell motion
emerges from CIL, leading to an active polar liquid state.
The polar order is stable if the confinement imposed by
neighbors restores the position and orientation of a cell be-
fore its polarity turns towards a new free direction. The re-
polarization occurs within a time-scale 1/fcil, and the char-
acteristic time of position relaxation in a dense environment
is ∼ γs/k, with k = 4(Ws + Wc)/R2 the stiffness of a two-
neighbor confinement. Thus, an approximate stability crite-
rion reads γs/k (cid:46) f
−1
cil , which is satisfied for the whole param-
eter range in Fig. 2 (SI Appendix).
polarity of N cells scales as PN = (cid:80)N
As illustrated in Fig. 5a, isolated fluid monolayers may ac-
quire a global polarity, and consequently perform persistent
random walks with a persistence much larger than that of sin-
gle cells (Movie S6). For randomly oriented cells, the average
i=1 pi/N ∼ N−1/2.
√
If cell polarities align, the average polarity of a small region
of cells decreases slower with its size, so that
N PN > 1.
The larger the repolarization rate ψ, the faster the increase
of polarity with N (SI Appendix). At sufficiently large sizes,
multiple misaligned polarity domains appear that restore the
random scaling (Fig. 5b). Hence, we define the onset of macro-
√
scopic polarization (circles in Fig. 2) by the condition that
N PN has a maximum at N = 75, namely that connected
clusters consisting of up to 75 cells may form a single polar-
ity domain. The appropriate choice of N depends on system
size. However, for the sizes explored, the transition line (cir-
cles in Fig. 2) is hardly sensitive to values around N = 75 (SI
Appendix). In conclusion, by ensuring a complete phase sepa-
ration while still allowing for cell rearrangements, sufficiently
strong cell-cell adhesion and CIL are required to form a polar,
collectively moving cell monolayer.
Finally, the effective potential energy Ep of cell-cell interac-
tions gives information on the mechanics of the colony. Posi-
tive (negative) potential energies correspond to tensile (com-
pressive) intercellular stresses. Non-cohesive colonies at low
cell-cell adhesion feature average attractive interactions lead-
ing to the formation of clusters. In turn, by polarizing border
cells outwards, CIL induces tensile stresses in cell monolayers
(Fig. 5c), in agreement with [17].
Overlapped phase. We finally focus on the transition to 3D tis-
sues. When the average total cell-cell force is attractive, cells
eventually overcome the energy barrier associated to the soft
repulsive potential (Fig. 1a), which corresponds to cell extru-
sion events. Extruded cells are confined at distances smaller
than R, where they exert neither cell-cell nor traction forces.
Thus, our model can predict the onset of the transition to 3D
cell arrangements. Assuming a homogeneous distribution of
cells, and using Eq. 2, the average interaction force reads
(cid:10)F cc
ij
(cid:11) =
(cid:82) 2R
(cid:82) 2R
R 2πdijF cc
ij ddij
R 2πdij ddij
=
2
9R
(4Ws − 5Wc) .
[ 6 ]
This force adds to the effective repulsion F p
ij associated to anti-
aligned self-propulsion forces (Eq. 4), so that the transition
between monolayers (blue in Fig. 2) and 3D cell arrangements
ij = 0.
(red in Fig. 2) is predicted by the condition (cid:10)F cc
(cid:19)(cid:21)
This sets a critical cell-cell adhesion energy
− 1
4ψ
(cid:11) + F p
,
[ 7 ]
4W s +
3D
c =
(cid:18)
(cid:20)
9
4
exp
1
5
W
ij
above which cells are expected to fully overlap or, alterna-
tively, a critical CIL repolarization rate above which cell ex-
trusion is prevented. Therefore, by opposing cell extrusion,
CIL hinders the collapse of cell monolayers into 3D aggre-
gates. Indeed, a sufficiently fast repolarization of cell motility
may stabilize cell monolayers even when cell-cell adhesion is
stronger than cell-substrate adhesion, W c > W s = 1 (Fig. 2).
In simulations, we characterize the degree of cell overlap in
terms of the average distance between contacting cells (cid:104)dij(cid:105)c
(a)
(b)
(c)
(d)
√
√
N PN of a monolayer of N cells for different CIL repolarization rates ψ at a cell-cell adhesion W c = 0.7.
Fig. 5. Collective motion, mechanics, and dewetting of cell monolayers. (a) Snapshot of a globally polarized, collectively migrating cell monolayer. (b) Rescaled average
N PN = 1 corresponds to randomly oriented cells.
polarity
N PN > 1) that gives rise to collective motion. The appearance of several polarity domains reduces the average polarity of large cell groups.
CIL induces a global polarity (
N PN is at N = 75. (c) Average cell-cell potential energy
The transition to the active polar liquid state (circles in Fig. 2) is defined by the condition that the maximum of
Ep = Ep/(2RFm) as a function of cell-cell adhesion W c and CIL repolarization rate ψ. CIL-associated repulsion induces tensile stresses (Ep > 0) in cell monolayers.
(d) Average distance between contacting cells (cid:104)dij(cid:105)c =(cid:104)dij(cid:105)c /(2R) as a function of W c and ψ. The transition between cell monolayers and 3D aggregates is predicted
to occur at a vanishing average cell-cell force (dashed line), and identified by the condition (cid:104)dij(cid:105)c = 3R/2 (solid line, crosses in Fig. 2).
√
√
4
www.pnas.org/cgi/doi/10.1073/pnas.0709640104
Footline Author
i
i
i
i
100101102103N100101pNPNWc = 0.7ψ=0.00ψ=0.17ψ=0.34ψ=0.52ψ=0.69ψ=0.86ψ=1.03ψ=1.20ψ=1.37ψ=1.550.00.20.40.60.81.0Wc0.00.40.81.21.6ψ-2.000-1.0000.0001.0002.0003.0004.0005.000321012345Ep0.80.91.01.11.2Wc0.00.40.81.21.6ψ0.7500.550.600.650.700.750.80›dijfic i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 5 - #5
i
i
(Fig. 5d). We then identify the transition when half of the con-
tacting cells are at the critical distance for extrusion, dij = R,
while the other half are fully spread, dij = 2R. Hence, the
transition is defined by (cid:104)dij(cid:105)c = 1
2 2R = 3R/2 (crosses in
Fig. 2), in qualitative agreement with the mean-field analytical
prediction.
2 R + 1
Monolayer instability occurs through a dewetting process
whereby holes appear in the cell monolayer, which rapidly
evolves into a network structure, as observed in [34]. Sub-
sequently, different regions of the network slowly collapse into
separate aggregates (Movie S7). In general, the 3D aggregate
- monolayer transition can be viewed as a wetting transition
of the cell colony [35] enabled by cell insertion or extrusion
[36]. Thus, our results show how CIL favors tissue wetting by
orienting cell motility towards free space.
Discussion and perspectives
Based on experimental observations, we propose that the dif-
ferent organizations of cell colonies that emerge from our
generic model correspond to different well-known tissue phe-
notypes (Fig. 6). First, the non-cohesive phase, in which cells
are not in contact, might correspond to mesenchymal tissues.
Experiments show that CIL leads to regular distributions of
mesenchymal cells during development [24, 25]. This result is
consistent with the transition towards an ordered structure of
cell clusters by increasing CIL strength ψ (Fig. 2).
The cohesive phase, in which cells maintain contact, can cor-
respond to epithelial tissues. In the active polar liquid state,
CIL induces cells to spontaneously invade empty spaces within
the tissue, similarly to wound healing processes characteristic
of epithelia.
Indeed, simulations of prepared wounds repro-
duce the closure dynamics observed in experiments [37] (SI
Appendix).
In the absence of CIL, healing is severely im-
paired (SI Appendix), in agreement with experiments upon
inhibition of Rac1 [38], a key protein for CIL behavior [39].
m/D2
r
(cid:0)Drt + e−Dr t − 1(cid:1), to experimental data for MCF10a
In addition, the parameters of our phase diagram can be es-
timated from experiments for two epithelial cell lines. By fit-
ting the MSD of a SPP with rotational diffusion [28], ∆x2 =
2v2
cells (SI Appendix), we estimate a self-propulsion velocity
vm = Fm/γs ≈ 1 µm/min, and a diffusion coefficient Dr ≈ 0.05
min−1. This gives a P´eclet number Pe = 3vm/(2RDr)≈ 2, too
low to produce motility-induced phase separation [32, 40]. In
turn, the duration of cell-cell contact during CIL events allows
the estimation of the rate of repolarization of cell motility. For
two mesenchymal cell types, hemocytes [21] and fibroblasts
[41], this gives fcil ≈ 0.1 min−1. The same estimate is ob-
tained for epithelial MDCK cells from the time that a wound
needs to start closing [42]. Then, assuming these parameter
values are similar for MCF10a and MDCK cells, we estimate
ψ := fcil/(2Dr) ≈ 1 for both cell lines. Self-propulsion forces
can be estimated from traction force measurements, which
yield Fm ≈ 60 nN and Fm ≈ 25 nN in MCF10a and MDCK
tissues, respectively [43]. Finally, cell-cell and cell-substrate
adhesion energies can be related to an effective elastic mod-
ulus Γ of an expanding monolayer, and to the total cellular
strain tot at which the expansion stops [44]. From Eq. 2,
Γ ≈ (Ws + Wc) /R2. In turn, tot corresponds to the cell-cell
distance at mechanical equilibrium, deq
ij = (1 + tot) R, namely
at which the total cell-cell force F cc
ij = 0 vanishes:
ij + F p
(cid:18)
(cid:19)(cid:21)
(cid:20)
deq
ij =
R
Ws + Wc
2Ws + Wc +
RFm
2
exp
− 1
4ψ
.
[ 8 ]
Then, using R = 16 µm and the values of Γ and tot reported
in [44], we infer W s ≈ 1.1 and W c ≈ 0.8 for the MCF10a tis-
sue, and W s≈ 0.35 and W c≈ 0.42 for the MDCK tissue. The
Fig. 6. Proposal for the classification of different tissue phenotypes (bold) in
terms of the phases of the model (colors). The association is based on the indicated
features, and supported by the parameter estimates for two epithelial tissues [44],
and an EMT [43] (crosses, see text). Speculated trajectories in cellular interaction
parameters during cancer progression are also included (dashed arrows).
3D
c ≈ 1.3 for
transition to 3D structures would then occur at W
the MCF10a tissue. Thus, this tissue type falls well within the
polar liquid state, in which cells form a collectively migrating
continuous monolayer as experimentally observed. In contrast,
the MDCK tissue is closer to the wetting transition, which we
c ≈ 0.63. Thus, although the latter also falls
estimate at W
within the polar liquid state, it may form 3D structures more
easily, in line with experimental observations [12].
3D
Now, the transition from cohesive to non-cohesive phases
should correspond to the epithelial-mesenchymal transition
(EMT), which is associated to down-regulation of cell-cell ad-
hesion proteins [3, 4]. Our prediction sheds light on the role
of CIL in the EMT (Fig. 6). As above, we can estimate the
parameters for an EMT in an expanding MCF10a monolayer.
Upon a knockdown of cell-cell adhesion proteins, the epithe-
xx ≈ 300
lial tissue disaggregates at an intercellular stress σcoh
Pa [43]. This translates into the critical cell-cell adhesion for
xx hR2, with h ≈ 5 µm
the loss of cohesiveness by W
c ≈ 0.2,
the height of the monolayer. Hence, we estimate W
c ≈ 0.19 at ψ = 1.
consistent with the prediction W
c ≈ σcoh
coh
coh
coh
A tissue may also undergo an EMT by increasing cell trac-
tion forces, such as upon treatment with hepatocyte growth
factor [44, 45]. In our diagram, an increased self-propulsion
force Fm yields a lower dimensionless cell-cell adhesion energy
W c := Wc/(2RFm) whereas its critical value depends only on
CIL (Eq. 5) hence causing the EMT.
In conclusion, the estimates and observations support the
association of epithelial tissues to the cohesive phase. Never-
theless, some mesenchymal cells can also migrate collectively
as a consequence of CIL [6, 13, 23] or of increased cell-cell
adhesion [46]. Therefore, these specific phenotypes might also
correspond to the active polar liquid state. However, whether
the features of collective mesenchymal cell migration [47] fully
agree with our results deserves further exploration.
Finally, in our model, the overlapped phase corresponds to
3D tissues. Their structure is not captured by our 2D model,
which only predicts the onset of their appearance. In experi-
ments, the transition from a cell monolayer to a 3D aggregate
can be induced in many ways [48], such as by increasing the
density of cell-cell adhesion proteins [35, 49]. Alternatively,
Footline Author
PNAS
December 20, 2016
vol. 113
no. 51
5
i
i
i
i
0.00.20.40.60.81.01.21.4WcCell-cell adhesion energy0.00.20.40.60.81.01.21.41.6Epithelial tissuesCollective cell motionWound healing3D tissuesMDCKMCF10aEMTEMTWettingTumorigenesisTumor dispersalxCIL repolarization ratexxMesenchymeRegular cell distributions i
i
"CILphasediagram-final" - 2016/12/22 - 1:20 - page 6 - #6
i
i
one can reduce the density of cell-substrate proteins [49, 50]
which, in our diagram, entails a decrease of the critical cell-
cell adhesion for the wetting transition, Eq. 7. 3D aggregates
also form when the substrate is softened [51], which simul-
taneously decreases cell tractions Fm and cell-substrate ad-
hesion Ws. This increases W c := Wc/(2RFm) while keeping
W s := Ws/(2RFm), and hence W
3D
c , constant.
The monolayer-spheroid transition has been put forward as
an in-vitro model for tumor formation and spreading [48]. In
this context, our predictions may contribute to appreciate the
role of CIL in cancer progression [52] (Fig. 6). Indeed, down-
regulation of cell-cell adhesion and enhanced traction forces
promote metastasis, which may proceed through many steps
involving collective cell migration, dissemination of cell clus-
ters, and a final EMT [1, 2, 3, 4, 53].
Conclusions
In summary, we studied the organization of cell colonies by
means of self-propelled particle simulations. The interactions
capture specific cellular behaviors such as CIL, and give rise to
several structures and collective dynamics (Fig. 2). Our results
show how CIL leads to regular cell arrangements, and hinders
the formation of cohesive tissues, as well as their extrusion-
mediated collapse into 3D aggregates. Self-organized collec-
tive cell motion, with tensile intercellular stresses, also emerges
from CIL interactions.
In addition, we have analytically derived an effective CIL-
induced cellular repulsion force, which yields explicit predic-
tions for transitions between non-cohesive, cohesive, and 3D
colonies. Based on experimental observations and parameter
estimates, we associate these phases to mesenchymal, epithe-
lial, and 3D tissue phenotypes. Thus, our predictions may
have implications for processes in development and disease
that modify the tissue phenotype. In general, our active soft
matter approach paves the way towards a physical understand-
ing of multicellular organization and collective cell behavior.
Methods
We performed simulations of SPP in an overdamped system.
Velocities are computed by solving F = Γ · x. Positions are
updated using an explicit Euler Scheme and the orientations
using the Euler-Maruyama method, with ∆t =0.016. We sim-
ulate rectangular domains of 25×103 up to 105 cells, enclosed
by means of a stiff repulsive potential. To avoid boundary ef-
fects, cells close to the border are excluded from the analysis.
A full description of the methods is given in the SI Appendix.
ACKNOWLEDGMENTS. We thank E. Bazeli`eres and X. Trepat for experimental
help, and T. Odenthal and S. Vanmaercke for help with implementation and proof-
reading. H.R. and B.S. acknowledge support from the Agency for Innovation by
Science and Technology in Flanders (IWT nr. 111504) and KU Leuven Research Fund
(PF/2010/07). R.A. acknowledges support from Fundaci´o "La Caixa", MINECO
(FIS2013-41144-P), and DURSI (2014-SGR-878).
I.P. acknowledges support from
MINECO (FIS 2011-22603), DURSI (2014SGR-922), and Generalitat de Catalunya
under Program Icrea Acad`emia.
9. Henkes S, Fily J, Marchetti C. (2011) Active jamming: Self-propelled soft parti-
cles at high density Phys Rev E 84:040301.
10. Basan M, Elgeti J, Hannezo E, Rappel W-J, Levine H (2013) Alignment of cel-
lular motility forces with tissue flow as a mechanism for efficient wound healing.
Proc Natl Acad Sci USA 110:2452-2459.
11. Sep´ulveda N, et al. (2013) Collective Cell Motion in an Epithelial Sheet Can Be
Quantitatively Described by a Stochastic Interacting Particle Model PloS Comp
Biol 9:e1002944.
12. Deforet M, Hakim V, Yevick HG, Duclos G, Silberzan P (2014) Emergence of
collective modes and tridimensional structures from epithelial confinement. Nat
Commun 5:3747.
13. Woods ML et al. (2014) Directional collective cell migration emerges as a property
of cell interactions. PLoS One 9:e104969.
14. Tarle V, Ravasio A, Hakim V, Gov N (2015) Modeling the finger instability in an
expanding cell monolayer. Integr Biol 7:1218-1227.
15. Mones E, Czir´ok A, Vicsek T (2015) Anomalous segregation dynamics of self-
propelled particles. New J Phys 17:063013.
16. Garcia S, et al. (2015) Physics of active jamming during collective cellular motion
in a monolayer. Proc Natl Acad Sci USA 112(50):15314-15319.
17. Zimmermann J, Camley BA, Rappel W-J, Levine H (2016) Contact inhibition
of locomotion determines cell-cell and cell-substrate forces in tissues. Proc Natl
Acad Sci USA 113(10):2660-2665.
18. Camley BA, Zimmermann J, Levine H, Rappel W-J (2016) Emergent collective
chemotaxis without single-cell gradient sensing. Phys Rev Lett 116:098101.
19. Abercrombie M, Heaysman JEM (1954) Observations on the social behavior of
cells in tissue culture. Exp Cell Res 6:293-306.
20. Abercrombie M, Ambrose EJ (1958) Interference microscope studies of cell con-
tacts in tissue culture. Exp Cell Res 15:332-345.
21. Davis JR, et al. (2015) Inter-cellular forces orchestrate contact inhibition of lo-
comotion. Cell 161:361-373.
22. Desai RA, Gopal SB, Chen S, Chen CS (2013) Contact inhibition of locomo-
tion probabilities drive solitary versus collective cell migration. J R Soc Interface
10:20130717.
23. Carmona-Fontaine C, et al. (2008) Contact inhibition of locomotion in vivo con-
trols neural crest directional migration. Nature 456:957-961.
24. Villar-Cervino V et al. (2013) Contact repulsion controls the dispersion and final
distribution of Cajal-Retzius cells. Neuron 77:457-471.
25. Davis JR, et al. (2012) Emergence of embryonic pattern through contact inhibi-
tion of locomotion. Development 139:4555-4560.
26. Fily Y, Marchetti MC (2012) Athermal phase separation of self-propelled particles
with no alignment. Phys Rev Lett 108:235702.
27. Fily Y, Henkes S, Marchetti MC (2014) Freezing and phase separation of self-
propelled disks. Soft Matter 10:2132-2140.
28. Coffey WT, Kalmykov YP, Waldron JT (2004) The Langevin equation (2nd edi-
tion). World Scientific, Singapore.
29. Mladek B, Gottwald D, Kahl G, Neumann M, Likos CN (2006) Formation of
polymorphic cluster phases for a class of models of purely repulsive soft spheres.
Phys Rev Lett 96:045701.
30. Levis D, Berthier L (2014) Clustering and heterogeneous dynamics in a kinetic
Monte-Carlo model of self-propelled hard disks. Phys Rev E 89:062301.
31. Moreno AJ, Likos CN (2007) Diffusion and relaxation dynamics in cluster crystals.
Phys Rev Lett 99:107801.
32. Redner GS, Baskaran A, Hagan MF (2013) Reentrant phase behavior in active
colloids with attraction. Phys Rev E 88(1):012305.
33. Bray A (1994) Theory of phase-ordering kinetics. Adv Phys 43:357-459
34. Douezan S, Brochard-Wyart F (2012) Dewetting of cellular monolayers. Eur Phys
J E 35:34.
35. Douezan S, et al. (2011) Spreading dynamics and wetting transition of cellular
aggregates. Proc Natl Acad Sci USA 108(18):7315-7320.
36. Beaune G, et al. (2014) How cells flow in the spreading of cellular aggregates.
Proc Natl Acad Sci USA 111(22):8085-8060.
37. Cochet-Escartin O, et al. (2014) Border forces and friction control epithelial clo-
sure dynamics. Biophys J 106(1):65-73.
38. Anon E, et al. (2012) Cell crawling mediates collective cell migration to close
undamaged epithelial gaps. Proc Natl Acad Sci USA 109(27):10891-10896.
39. Roycroft A, Mayor R (2016) Molecular basis of contact inhibition of locomotion.
Cell Mol Life Sci 73(6):1119-1130.
40. Redner GS, Hagan MF, Baskaran A (2013) Structure and dynamics of a phase-
separating active colloidal fluid. Phys Rev Lett 110(5):055701.
41. Kadir S, et al. (2011) Microtubule remodelling is required for the front-rear po-
larity switch during contact inhibition of locomotion J Cell Sci 124:2642-2653.
42. Brugues A, et al. (2014) Forces driving epithelial wound healing Nat Phys 10:683-
690.
43. Bazelli`eres E, et al. (2015) Control of cell-cell forces and collective cell dynamics
by the intercellular adhesome. Nat Cell Biol 17:409-420.
44. Vincent R et al. (2015) Active tensile modulus of an epithelial monolayer. Phys
Rev Lett 115:248103.
References
1. Friedl P, Wolf K (2003) Tumour-cell invasion and migration: diversity and escape
mechanisms. Nat Rev Cancer 3(5):362-374.
45. Maruthamuthu V, Gardel ML (2014) Protrusive activity guides changes in cell-cell
tension during epithelial cell scattering. Biophys J 107(3):555-563.
46. Plutoni C, et al. (2016) P-cadherin promotes collective cell migration via a Cdc42-
mediated increase in mechanical forces. J Cell Biol 212(2):199-217.
2. Friedl P, Gilmour, D (2009) Collective cell migration in morphogenesis, regener-
47. Theveneau E, Mayor R (2013) Collective cell migration of epithelial and mes-
ation and cancer. Nat Rev Mol Cell Biol 10(7):445-457.
3. Thiery JP, Acloque H, Huang RY, Nieto MA (2009) Epithelial-mesenchymal tran-
sitions in development and disease. Cell 139(6):871-890.
4. Nieto MA (2013) Epithelial plasticity: A common theme in embryonic and cancer
5. M´ehes E, Vicsek T (2014) Collective motion of cells: From experiments to mod-
cells. Science 342:708.
els. Integr Biol 6:831-854.
Op Cell Biol 41:22-28.
enchymal cells. Cell Mol Life Sci 70:3481-3492.
48. Gonzalez-Rodriguez D, Guevorkian K, Douezan S, Brochard-Wyart F (2012) Soft
matter models of developing tissues and tumors. Science 388(6109):910-917.
49. Ryan PL, Foty RA, Kohn J, Steinberg MS (2001) Tissue spreading on implantable
substrates is a competitive outcome of cell-cell vs. cell-substratum adhesivity. Proc
Natl Acad Sci USA 98(8):4323-4327.
50. Ravasio A, et al. (2015) Regulation of epithelial cell organization by tuning cell-
51. Douezan S, Dumond J, Brochard-Wyart F (2012) Wetting transitions of cellular
aggregates induced by substrate rigidity. Soft Matter 8(17):4578-4583.
52. Abercrombie M (1979) Contact inhibition and malignancy. Nature 281:259-262.
53. Cheung KJ, Ewald AJ (2016) A collective route to metastasis: Seeding by tumor
cell clusters. Science 352(6282):167-169.
6. Szab´o A, Mayor R (2016) Modelling collective cell migration of neural crest. Curr
substrate adhesion Integr Biol 7:1228-1241.
7. Szab´o B, et al. (2006) Phase transition in the collective migration of tissue cells:
Experiment and model. Phys Rev E 74(6):061908.
8. Belmonte JM, Thomas JL, Brunnet LG, de Almeida TMC, Chat´e H (2008) Self-
propelled particle model for cell-sorting phenomena Phys Rev Lett 100:248702.
6
www.pnas.org/cgi/doi/10.1073/pnas.0709640104
Footline Author
i
i
i
i
|
1806.08767 | 3 | 1806 | 2018-12-19T09:19:24 | Filament flexibility enhances power transduction of F-actin bundles | [
"physics.bio-ph",
"cond-mat.stat-mech",
"physics.comp-ph"
] | The dynamic behavior of bundles of actin filaments growing against a loaded obstacle is investigated through a generalized version of the standard multi filaments Brownian Ratchet model in which the (de)polymerizing filaments are treated not as rigid rods but as semi-flexible discrete wormlike chains with a realistic value of the persistence length.
By stochastic dynamic simulations we study the relaxation of a bundle of $N_f$ filaments with staggered seed arrangement against a harmonic trap load in supercritical conditions. Thanks to the time scale separation between the wall motion and the filament size relaxation, mimiking realistic conditions, this set-up allows us to extract a full load-velocity curve from a single experiment over the trap force/size range explored. We observe a systematic evolution of steady non-equilibrium states over three regimes of bundle lengths $L$. A first threshold length $\Lambda$ marks the transition between the rigid dynamic regime ($L<\Lambda$), characterized by the usual rigid filament load-velocity relationship $V(F)$, and the flexible dynamic regime ($L>\Lambda$), where the velocity $V(F,L)$ is an increasing function of the bundle length $L$ at fixed load $F$, the enhancement being the result of an improved level of work sharing among the filaments induced by flexibility. A second critical length corresponds to the beginning of an unstable regime characterized by a high probability to develop escaping filaments which start growing laterally and thus do not participate anymore to the generation of the polymerization force. This phenomenon prevents the bundle from reaching at this critical length the limit behavior corresponding to Perfect Load Sharing. | physics.bio-ph | physics | Filament flexibility enhances power transduction of F-actin bundles.
Department of Physics, Sapienza University of Rome, P.le Aldo Moro 2, I-00185 Rome, Italy
Department of Chemistry, ´Ecole Normale Superi´eure, rue Lhomond 24, 75005 Paris, France
Alessia Perilli∗
Carlo Pierleoni†
DSFC, University of L'Aquila, 67100 L'Aquila, Italy
Maison de la Simulation, CEA, CNRS, Universit´e Paris-Sud,
UVSQ, Universit´e Paris-Saclay, 91191 Gif-sur-Yvette, France.
Jean-Paul Ryckaert‡
Physics Dept., Universit´e Libre de Brussels (ULB),
Campus Plaine, CP 223, B-1050 Brussels, Belgium
(Dated: December 20, 2018)
The dynamic behavior of bundles of actin filaments growing against a loaded obstacle is inves-
tigated through a generalized version of the standard multi filaments Brownian Ratchet model in
which the (de)polymerizing filaments are treated not as rigid rods but as semi-flexible discrete worm-
like chains with a realistic value of the persistence length. By stochastic dynamic simulations we
study the relaxation of a bundle of Nf filaments with staggered seed arrangement against a har-
monic trap load in supercritical conditions. Thanks to the time scale separation between the wall
motion and the filament size relaxation, mimiking realistic conditions, this set-up allows us to ex-
tract a full load-velocity curve from a single experiment over the trap force/size range explored. We
observe a systematic evolution of steady non-equilibrium states over three regimes of bundle lengths
L. A first threshold length Λ marks the transition between the rigid dynamic regime (L < Λ),
characterized by the usual rigid filament load-velocity relationship V (F ), and the flexible dynamic
regime (L > Λ), where the velocity V (F, L) is an increasing function of the bundle length L at fixed
load F , the enhancement being the result of an improved level of work sharing among the filaments
induced by flexibility. A second critical length corresponds to the beginning of an unstable regime
characterized by a high probability to develop escaping filaments which start growing laterally and
thus do not participate anymore to the generation of the polymerization force. This phenomenon
prevents the bundle from reaching at this critical length the limit behavior corresponding to Perfect
Load Sharing.
I.
INTRODUCTION
Cellular shape changes involve significant correlated deformations of the cell membrane overlying and embedding
the cytoskeleton.
In the particular case of lamellipodium and filopodium growth the nanoscale energy source for
such deformation is provided by actin polymerization: G-actin monomers assemble to the barbed end of filaments
in direct contact with the cell membrane, generating significant pushing forces. The speed of membrane deforma-
tion/displacement depends on the resisting load due to membrane deformation and to the crowded environment
around the cell [1]. In the specific case of filopodium growth, modelling has been developed over the last twenty years
in different directions. Mogilner and Rubinstein [2] pioneered the direct modeling of a filopodium protruding against
the membrane resistance, establishing conditions for its onset and general size limitations due to actin bundle buckling
or to the diffusion of G-actin monomers to the tip. A full stochastic model of filopodium growth based on a bundle of
a few rigid filaments taking into account retrograde flow, a redistribution of the total load on filament tips induced by
membrane fluctuations and a rather realistic treatment of monomer diffusion which couples to polymerization events
was used to unravel the different length and time scales associated with filopodium growth. In particular, the growth
rate of filaments and their stable sizes were related to various parameters, like the G-actin diffusivity or the amplitude
of membrane fluctuations[3]. In a subsequent refined model with explicit G actin particles interacting with the rigid
filaments via excluded volume interactions, it was found that the restricted diffusion of G-actin taking place in the
enclosing cylindric volume of the membrane tube with growing filaments as local obstacles, leads to an inhomogeneous
growth of the bundle which favors elongation of peripheral filaments with respect to filaments located near the centre
∗ [email protected]
† [email protected]
‡ [email protected]
8
1
0
2
c
e
D
9
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
3
v
7
6
7
8
0
.
6
0
8
1
:
v
i
X
r
a
of the tube [4]. Further, alternative ways to model membrane fluctuations and probe their effect on the bundle growth
velocity was specifically addressed by including in stochastic models membrane height fluctuations at the filament
tips [5, 6].
2
FIG. 1. Left panel: Bundle of mutually non-interacting flexible filaments grafted normally to the fixed wall on the left, pressing
on a mobile wall (blue plane) oriented perpendicular to the grafting direction. The latter wall, located at position x = L and
subject to an harmonic load F = −κT L (mechanical potential energy VOT = 1
2 kT L2), separates two chambers both in contact
with a reservoir of monomers (dark blue spheres) at fixed concentration. Light blue spheres depict monomers incorporated
into grafted F-actin filaments in the left chamber while the second chamber contains free monomers only. This illustrates the
underlying "microscopic" model, treated recently by equilibrium statistical mechanics in the grand canonical ensemble [7], from
which our coarse-grained dynamical stochastic model derives.
Right panel: Illustrations of (a) a grafted filament with size z(L) = 1 + int(L/d) (where int(x) represents the integer part
of x) corresponding to the longer size for which no interactions with the wall at L are possible; (b) a slightly bent grafted
filament with net polymerization arrested by the wall; (c) an escaping filament polymerizing along the obstacle while subject to
a bending mechanical force by the wall. A filament is considered as escaping when its size in monomers exceeds the threshold
z∗(L) = int( πL
2d )) for which the filament forms a quarter of a circle of radius equal to the trap width L. In the figure, the
monomers in excess to z∗ are shown in red.
Going one step further in dissecting different effects, focus is often brought on the basic mechanisms and the collective
strategies put in place by bundles of actin filaments to optimize the transduction of chemical into mechanical energy[8].
Grafted actin filaments in a solution of free G-actin monomers at density ρ1 polymerize, with a rate U0 = konρ1, and
depolymerize, with a rate W0 = kof f , at their free ends, where kon and kof f are kinetic constants. At the bulk critical
density ρ1c = kof f /kon the filaments neither grow nor shrink on average. Supercritical conditions ρ1 > ρ1c correspond
to a net filament growth while subcritical conditions (ρ1 < ρ1c) to a net filament shrinkage. The velocity of elongation
of the filaments in absence of obstacles is V0 = d(U0 − W0) where d = 2.7nm is the filament length increment per
added monomer. When growing against a loaded wall in supercritical conditions, the net velocity V < V0 of the wall
times the external load F is a measure of the power transduction of chemical energy into mechanical work provided
by growing pushing filaments. When filaments push on the wall at normal incidence, the wall velocity decreases for
increasing loads from V0 at zero load 1. For strong enough loads the filaments will be forced to depolymerize on
average and the net wall velocity will become negative even in supercritical conditions. The stalling force Fs is the
specific value of the load which stops the obstacle (V = 0) and hence equals the bundle polymerization force at rest.
On the experimental side, measurements of the polymerization force exerted by a bundle of actin filaments on an
obstacle and the corresponding load-velocity relation are difficult to extract in in-vivo conditions given the uncertain-
ties on various possible additional effects e.g. those linked to interferences of auxiliary proteins (see however ref. [10]).
In-vitro experiments targeting dynamic bundle properties exploit actin bundles glued on colloidal particles growing
1Taking into account viscous forces on the wall motion leads to a value slightly smaller than V0[9].
3
against a load, the obstacle being another colloidal particle or a fixed wall [11, 12]. The theoretical framework used
to interpret such experiments is usually provided by some form of multi filament Brownian Ratchet (BR) models of
fully rigid filaments polymerizing against a loaded rigid wall [13 -- 18]. The rigid filament hypothesis is assumed to
be valid because typical F-actin (uncrosslinked) filaments rarely reach contour lengths Lc longer than a few 0.1µm
which is much smaller that the persistence length (cid:96)p ≈ 15µm. To be able to grow normally against a loaded wall
while pushing, straight rigid filaments need to exploit temporary gaps of size ≥ d created by thermal fluctuations of
the wall, allowing monomer insertion and thus the targeted rectification of the biased random motion of the obstacle.
The rigid character of the filaments makes these models unidimensional while in reality F-actin filaments are semi-
flexible and this leads to two distinct contour length dependent features for their compressional behavior. First, a new
mechanism of generation of voids of size d as a result of tip bending fluctuations becomes possible, as proposed by the
Elastic Brownian Rachet (EBR) model [19]. Using this EBR model, only a moderate velocity increase in the load-
velocity relation of a single filament pushing a hard wall has been evidenced by numerical simulations [9, 20]. Next, in
absence of the lateral envelop around a bundle of uncrosslinked polymerizing filaments (as provided by the membrane
tube in in-vivo filopodia [21] or in similar finger-like structures reconstructed in-vitro [22]), filaments in a long enough
bundle may start growing in lateral directions rather than pushing the obstacle. Such diverging filaments are said
to be escaping as they do not participate anymore to the direct transformation of chemical energy into mechanical
work against the load, even if a mechanical force is still exerted by the wall to maintain the permanent filament
bending. The escaping filament phenomenon, also called the pushing catastrophe [9, 20], is minimal, but nevertheless
present, when semi-flexible filaments hit the wall at normal incidence, that is in the geometry illustrated in figure 1.
Simulations at constant load with the EBR model[9] found for escaping filaments a probability of occurrence per unit
of time increasing as the square of the contour length of the grafted filament. The presence of escaping filaments at
normal incidence was detected in in-vitro experiments when following actin bundle growth against an optical trap
load [11] within the geometry illustrated in figure 12. Measurements of the load-velocity of actin bundles obtained
by monitoring elongation velocity against a constant load [12] have been interpreted with the multi-filament BR
model with staggered filament seeds disposition: whether the flexibility features of filaments are negligible because
the filaments remain sufficiently short or whether these effects renormalize parameters (like the number of filaments
in the bundle) in the fitting procedure is difficult to be established.
As we develop below and exploit in the present paper, a refined theoretical analysis of the filament flexibility
effects is facilitated by considering the actin bundle in an optical trap, following the setup exploited in [11]. Figure 1
represents the situation of a thought experiment of a bundle of polymerizing semi-flexible filaments in an optical trap
apparatus, there represented by a harmonic resisting load (see the caption of the figure). The moving wall (blue in
the figure) will have its rest position calibrated at the left yellow wall (L = 0), in absence of filaments. The left yellow
wall is also the location where filaments seeds are grafted. In presence of free monomers in the solution in supercritical
conditions, the growing filaments will push the blue wall in the right direction but the optical trap will increase its
reaction linearly with the position of the blue wall (F = −κT L) until stalling is reached, which is the condition at
which the velocity of the wall vanishes. The intensity of the trap κT tunes the trap amplitude at stationarity Ls,
hence the stalling force Fs = κT Ls.
For bundles of rigid non interacting filaments in ideal solution conditions, equilibrium statistical mechanics predicts
[7, 8]
Fs = Nf
kbT
d
ln (ρ1).
(1)
For bundles of semi-flexible filaments facing an optical trap load (as depicted in figure 1), equilibrium statistical
mechanics applied to discrete wormlike chains (d-WLC) with step size d predicts slightly larger values of the stalling
force (1-2 percent)[7]. In order to avoid unbounded escaping filaments and remain in strict equilibrium conditions, a
maximum value must be imposed to the grafted d-WLC contour length: the upper limit z∗(L)d (precisely defined in
the caption of figure 1) corresponds to a grafted filament with uniform curvature 1/L squeezed at normal incidence
between two parallel planes separated by a distance L. Equilibrium properties in presence of this constraint are
pertinent to an actin bundle in the optical trap as long as the probability of the onset of escaping is negligible over
the observation time. In ref. [7] we have shown that the non escaping region is limited by
(cid:115)
L < Ll =
(cid:96)pd
ln ρ1
(2)
2In this reference the escaping phenomenon is confused with macroscopic buckling.
4
as this condition ensures a negligible population of filament sizes at the boundary of the escaping regime, z∗(L) (see
caption of figure 1). In these conditions, the force exerted by a bundle of compressed flexible filaments in an optical
trap set up at stalling results from the action of individual filaments which are continuously swapping between two
states illustrated in the right panel of figure 1: either they do not touch the wall hence not contributing to the force
(filament a in right panel of the figure) or they touch the wall and exert a L-dependent force f (L) (filament b in
the figure) which, assuming adiabatic conditions3, and the weak compression behavior of the WLC model [23], has
the form f (L) ∝ kBT (cid:96)p/L2. This implies in turn that the bundle at equilibrium needs to recruit an average fraction
x ∝ (L2/(cid:96)p) of touching filaments to face the L-independent stalling force eq. (1), at the equilibrium trap amplitude
Ls corresponding to the chosen values of κT . We observe that the rigid filament case is recovered in the limit (cid:96)p → ∞
where the wall will get impulsive forces from the most advanced filament with fb(L) → ∞ and x → 0 as (cid:96)p diverges.
This L-dependent load repartition for flexible filaments at equilibrium is expected to persist in non-equilibrium
relaxation against arbitrary loads, and hence affect the load-velocity relationship. The aim of the present paper is to
investigate the impact of filament flexibility on the bundle non-equilibrium dynamics by exploiting stochastic dynamics
(SD) simulations. The model of living d-WLC used in this work is similar to the one in ref. [7] except that the contour
length restriction is replaced by a realistic compressional response of the d-WLC in the strong compression regime,
obtained by Monte Carlo calculations, which allows to make excursions into the escaping regime within a unique
model. We follow the relaxation of a flexible bundle against a harmonic force mimicking the experimental conditions
of ref.
[11], as illustrated in the left panel of figure 1. The trap strength at given ρ1 and Nf controls the range of
the bundle length investigated, therefore the occurrence of escaping filaments. Similar to the rigid filaments case[24],
the wall relaxation is expected to be mostly adiabatic which allows us to consider the trap relaxation as a sequence
of stationary non-equilibrium states of the bundle at different loads. This implies that the celebrated "velocity-load"
relationship V (F ), used to characterize the system behavior in rigid filament models, can be generalized to a "velocity-
load-bundle length" relationship V (F, L) without the need to consider complicated history-dependent behaviors. By
tuning κT in the optical trap set up, we can generate non-equilibrium steady-states for independent values of the load
F and the bundle length L and characterize the relationship in the entire relevant range of the parameters.
The bulk of the paper is organized as follows. Our dynamical model of living d-WLC [7] in an optical trap apparatus
is described in section 2. In the following section, Stochastic Dynamics (SD) simulations are exploited to follow the
relaxation of the bundle against an optical trap load, for a range of trap strengths. Peculiar features due to flexibility
are emphasized by a direct comparison with the rigid filament predictions. A useful concept of chemical friction linked
to the trap relaxation time is introduced in section IV. Some results on the dynamical evolution towards the escaping
regime are illustrated in section C, while discussions and conclusions are provided in section V. Details of results and
analytical derivations are collected in six appendices. To help the reader following the developments of our paper, we
have regrouped in a separate table most physical quantities (mesoscopic or microscopic) appearing in our paper (see
appendix A).
II. THE DYNAMICAL MODEL
We consider a grafted bundle of Nf linear flexible filaments in supercritical conditions (ρ1 > 1) hitting a mobile
hard wall which is subject to an external harmonic load as illustrated schematically in figure 1. The filaments are
d-WLC with actin persistence length (cid:96)p = 14.5µm and discrete step d = 2.7nm equal to the incremental contour
length associated to the incorporation of a single actin monomer in the F-actin protofilament. Filaments are grafted
normally to the wall located at x = 0 by fixing permanently the position of the two first articulation points of the
n − th d-WLC filament (n = 1, Nf ) at hn and hn + d respectively along the x−axis (where hn is a longitudinal shift
with hn < d/2). In multi-filament bundles, two longitudinal arrangements are usually considered. In the unstaggered
bundle case, the filament seeds are all located at the origin, hn = 0, while in the staggered bundle case, the filament
seed positions are uniformly distributed within an amplitude d according to
(cid:20) n − 0.5
Nf
(cid:21)
− 0.5
hn =
d
n = 1 . . . Nf .
(3)
The sizes of the filaments ¯j = {jn}n=1,Nf are defined as the number of articulation points in each d-WLC filament.
The minimum size jn = 2 represents the permanent oriented seed from which a filament can (re)grow.
We now specify the stochastic model for the coupled dynamics of the continuous variable L, the position of the
moving obstacle, and of the Nf discrete filament sizes. The wall dynamics is described by the overdamped Langevin
3fast microscopic intra-filament relaxation with respect to wall diffusion and F-actin self assembly kinetics, see appendix D for details
5
FIG. 2. Reduced force exerted by a passive d-WLC against a rigid wall located at L as a function of(cid:101)γ = (Lc − L)/Lc for three
small range of(cid:101)γ depending on Lc and departs from the unitary plateau at(cid:101)γ (cid:39) 0.1. The orange solid line is the fitting function
eq. (F5). In the inset, the same reduced force as a function of η =(cid:101)γ(cid:96)p/Lc. The orange solid line in inset is the weak bending
different values of the contour length Lc. The force grows from zero to its buckling value (represented by the plateau) within a
universal curve eqs.(F1-F4).
equation
ξ
dL(t)
dt
= Fbun(t) − κT L(t) + R(t)
(4)
where ξ is the solvent friction coefficient on the obstacle related to its diffusion coefficient D = kBT /ξ, where T is the
absolute temperature and kB the Boltzmann constant. The use of the overdamped limit is motivated by experimental
conditions [11, 12], as detailed in appendix D. The second term in the rhs of eq. (4) represents the action of the
optical trap with a linear restoring force with trap force constant κT . The fluctuation dissipation theorem requires
that the white noise random force R(t) satisfies
(cid:104)R(t)(cid:105) = 0 (cid:104)R(t + t(cid:48))R(t(cid:48))(cid:105) = 2ξkBT δ(t).
(5)
On a coarse-grained time scale, Fbun(t) represents the evolution of the total force exerted by the filaments on the
wall. It depends on time through the wall position L(t) and the filament sizes ¯j(t) by expressing the bundle force as
the sum of single filament forces
Nf(cid:88)
Fbun(¯j, L) =
¯fjn (Ln).
(6)
n=1
where Ln = L − hn is the distance between the seed location of filament n and the position of the obstacle L.
¯fjn (Ln) is the equilibrium average force on a planar hard wall located at L, exerted by a passive (i.e. non-reacting)
d-WLC of contour length Lcn = (jn − 1)d with its two first articulation points clamped at hn and hn + d. Such
an adiabatic treatment of the individual filament forces on the moving wall is justified by the time scale hierarchy
τintra (cid:28) τD (cid:28) τchem existing between the intramolecular relaxation time of the filament (τintra), the local wall
diffusion time τD = d2/D and the characteristic time of the chemical reactions taken as τchem = W −1
(see appendix
D for a detailed comparison of these time scales) Direct filament-filament mutual interactions are disregarded.
0
0.00010.0010.010.11γ~=(Lc-L)/Lc0123456f~~(γ)0.010.1110100η00.511.52 f~(η)Lc=40dLc=76dLc=157d 6
For living filaments, i.e. filaments that undergo de/polymerization reactions, the total free energy of a clamped
filament of size jn with seed-wall distance Ln is
wjn (Ln, ρ1) = −kBT ln αjn (Ln) − (jn − 2)kBT ln ρ1 + C
(7)
The first term, −kBT ln αjn(Ln), is the free energy penalty due to compression with αjn (Ln) the ratio of the partition
functions of the clamped filament in presence and in absence of the wall [7]. The second term is the free energy
gain due to the self-assembly of (jn − 2) monomers to the seed to form the filament [8]. Finally, C is an irrelevant,
j -- independent constant (the free energy of the permanent seed). The force ¯fjn (Ln) in eq. (6) is thus
¯fjn (Ln) = − ∂wjn(Ln, ρ1)
∂Ln
= − kBT
αjn (Ln)
∂αjn (Ln)
∂Ln
(8)
The form of ¯fj(L) for a passive d-WLC of persistence length (cid:96)p, contour length Lc = (j − 1)d, clamped at h = 0
and hitting a hard wall located at L and normal to the grafting direction, is given by
(cid:101)γ < 0.1
(cid:101)γ ≥ 0.1
(cid:19)
(cid:96)p
Lc
(cid:18)(cid:101)γ
f
f ((cid:101)γ)
π2
4
kBT (cid:96)p
L2
c
.
¯fj(LLc, (cid:96)p) = fb(Lc, (cid:96)p) ×
fb(Lc, (cid:96)p) =
f ((cid:101)γ), where η =(cid:101)γ (cid:96)p
Lc
where(cid:101)γ = (Lc − L)/Lc is the relative compression and fb is the Euler buckling force for a clamped rod [25]
(9)
(10)
The form of the functions f (η) and
, is illustrated in figure 2 and discussed in details in appendix
c/lp (which
is typically less than d for actin filaments up to j ≈ 150) (see the inset in figure 2), and remains constant over a large
F. At weak bending ((cid:101)γ < 0.1), f (η) rises from zero to one over a short compression range (Lc − L) ≈ 0.25L2
range of compressions up to(cid:101)γ (cid:39) 0.1. For larger compression a universal behavior appears in terms of(cid:101)γ and has been
characterized numerically as detailed in appendix F. The compression law at weak bending, derived analytically in
ref. [23], leads to a linear decay of the free energy wj(L, ρ1) with L, which extends almost up to L = Lc beyond which
the filament force vanishes as illustrated in figure 3.4
Like in Brownian Ratchet models [24], the stochastic time evolution of the filament sizes ¯j(t) = {j1(t), . . . , jn(t), . . . , jNf (t)}
is specified by a Poisson distribution of waiting times for the next instantaneous single monomer polymeriza-
tion/depolymerization. For each filament n (n = 1, Nf ), we adopt rates
Un(jn, L) = U0
αjn+1(Ln)
αjn (Ln)
exp{−βwjn+1(Ln, ρ1)}
exp{−βwjn (Ln, ρ1)}
= W0
(cid:40)
Wn(jn, L) =
W0
0
if jn > 2
if jn = 2
if jn ≥ 2
(11)
(12)
similar to the ones for the rigid filaments case given in ref. [24]. For rigid filaments ((cid:96)p → ∞) the α factors are either 1,
if the filament does not interact with the wall, or 0 in case the filament touches the wall. For the flexible case, we still
have αjn (Ln) = 1 for jn < z(Ln), where z(Ln) = int(Ln/d) + 1 (see figure 1) is the largest size of a filament such that,
for a given seed-wall distance Ln, none of its microscopic configurations overlaps with the wall. Given the conditional
equilibrium size distribution Pj(L, ρ1) ∝ exp (−βwj(L, ρ1)) [7], it is readily verified that the adopted rates given by
eqs.(11),(12) satisfy micro-reversibility at equilibrium. Our choice assumes that the depolymerization rate W0 of the
filament is independent of the G-actin monomer concentration and is insensitive to any form of filament-wall contact.
0 = konρ1c = W0 at the critical monomer concentration ρ1c in the bulk,
The effective polymerization rate, equal to U c
increases linearly with the bulk monomer concentration but decreases with the work needed to compress the filament
as its length is increased by d at fixed L. This is illustrated in figure 3 where the L dependence of two free energy
4The change in slope is just a little before L = Lc because some small compression is needed to reach the buckling regime, see figure 2.
7
FIG. 3. Wall position (L) dependence of the reduced free energy [βwj(L)] of fixed-length grafted d-WLC filaments having
neighbouring sizes j(red) and j + 1(blue) for j = 30 (left) and j = 60 (right) (Lc = (j − 1)d). The potential remains constant
for the L regime where there is no direct contact between the wall and the fluctuating grafted d-WLC filament of size j, that is
as long as j ≤ z(L), as defined and illustrated in figure 1. The vertical difference between blue and red profiles is the free energy
gap in the exponential factor for the polymerization rate U (j, L) at fixed wall position. The green-dotted profiles indicates the
threshold of 2.5 between the significant and negligible polymerization rate regions. Polymerization transitions from 30 → 31
are seen to be rare (less than 0.1W0) if the wall is closer than 0.75d from the tip of an uncompressed filament of size j = 30,
hence for L < 29.75d. Conversely, polymerization transitions from 60 → 61 are possible ((cid:46) 0.1W0) even if the filament of size
j = 60 is already under compression.
expressions wj(L, ρ1) and wj+1(L, ρ1) for successive filament sizes is shown respectively for j = 30 and j = 60. The
observed linear increase of wj(L, ρ1) as the wall position decreases from L = Lcj = (j − 1)d is a direct consequence of
the constant compression force at weak bending, see figure 2. The difference between the horizontal plateaus for two
successive sizes is the free energy change when adding/removing one monomer, −kBT ln ρ1. If we arbitrarily consider
e−(β∆w)W0 = e−2.5W0 ≈ 0.1W0 as the crossover rate between "negligible" and "significant" polymerization rates,
figure 3 shows that the polymerization j → j + 1 transition rates are significant for a filament j = 60 even when it
is already under compression. By contrast, for a shorter filament of size j = 30, polymerization rates of compressed
filaments are negligible and become significant only when the wall lies beyond 0.75d from the filament tip, i.e. in the
region where the blue profile lays below the green dotted profile.
The time evolution of our model of coupled wall-filaments dynamics has been generated by the Explicit-Wall
algorithm (EWA) proposed in [24, 26 -- 28] which treats both the filaments sizes and the wall position as discrete
random variables to reformulate the time evolution as a purely Markov chain dynamics. This step requires to replace
the Langevin equation for the wall dynamics by an equivalent Fokker-Planck equation (see appendix E). To make
our simulations feasible, we adopted the value = τD/τchem = W0d2/D = 0.0469 (where D is the wall diffusion
constant) which in ref. [24] was shown to accurately represent the experimentally relevant situation of quasi infinite
wall diffusion (exp = 4.69 × 10−5).
Before presenting the result of our non-equilibrium simulations, we would like to discuss the nature of the steady
and the equilibrium states of our model. We follow the relaxation of an initially short bundle under the action of
polymerization and trap restoring forces. In supercritical conditions the trap amplitude grows until the reach of a
steady state. For rigid filaments this is a genuine equilibrium state where the pushing polymerization force is on
average balanced by the trap force and the trap amplitude is inversely proportional to the trap strength [24]. Also for
semiflexible filaments we observe the reach of a steady trap amplitude at large times. However, this corresponds to
a metastable local equilibrium state, where polymerization force is balanced by the trap load, only for trap strength
large enough to have Ls < Ll as given in eq.(2). If this condition does not hold, we incur in the escaping regime where
the steady trap amplitude does not correspond to a force balancing the polymerization force, as detailed in section
V. In the latter case we don't have an equilibrium state since filament lengths grow indefinitely. In ref. [7] we have
carefully characterized the stability of the non-escaping local equilibrium state. Being a metastable state, there is
always the possibility that even in the non-escaping regime, one filament becomes escaping by running trajectories for
long enough time. However, the probability of filaments with contour length longer than L decreases exponentially
fast with Lc − L at fixed κT , i.e. the probability per unit time to observe the occurence of an escaping filament
28.52929.53030.5L/d-28-26-24-22βwj(L)58.55959.56060.5 L/d-56-54-52-50βwj(L)j=30j+1=31j=60j+1=61 8
becomes exponentially small[9].
We point out that while the control of the L fluctuations is particularly easy in an optical trap device, specifically
by adjusting the trap force constant, the L fluctuations and the associated onset of escaping filaments are, on the
contrary, much more difficult to predict in constant force experiments where no local equilibrium exists [9, 20].
III. RELAXATION OF A STAGGERED SEMI-FLEXIBLE BUNDLE IN AN OPTICAL TRAP
We have studied systems at several values of Nf and ρ1 with both staggered and unstaggered seed dispositions, but
here we first concentrate on the specific case of a staggered bundle of Nf = 32 filaments with seeds disposition as in
Eq.(3) at ρ1 = 2.5. In the following for convenience, the staggered rigid bundle model (in the infinite diffusion limit)
will be specifically referred to as SR (for staggered rigid). In order to highlight the effects of flexibility we compare
FIG. 4. Panel (a): Relaxation of a staggered bundle of Nf = 32 filaments subject to optical trap load with trap strength
κT = 0.4511kBT /d2 at ρ1 = 2.5. Results for rigid (blue line) and flexible (red line) models are reported. The horizontal line
represents the value Ls = Fs/κT = 65d from eq. (1), while the flexible model has the slightly larger asymptotic value, in
agreement with the theoretical prediction Ls = 66.1877d [7]. The vertical bars on the data indicate the standard deviation of
L(t) which is seen to evolve rapidly towards its predicted equilibrium values σ = (kBT /κT )1/2 = 1.489d and σ = 1.5178d for
the rigid and the flexible models respectively. Exponential fits to the long time behavior for both models are shown by dashed
and dot-dashed lines. Numerical values of the parameters are reported in the text. The inset is an enlargement of the region
where flexible and rigid model behaviors start diverging from each other.
Panel (b): V (t) = d(cid:104)L(t)(cid:105)/dt versus F (t) = κT(cid:104)L(t)(cid:105) for both rigid (green line) and flexible (red squares) models. The green
line is an accurate theoretical prediction of the stationary velocity for the SR model [12]. The purple curve represents the PLS
behavior from eq. (15). Inset: velocity relative to the SR prediction versus the fractional load: PLS (purple line), flexible
model (red points).
results for flexible and rigid bundles at the same thermodynamic conditions in figure 4. The top panel shows the
average wall relaxation (cid:104)L(t)(cid:105) of both a rigid bundle ((cid:96)p = ∞) and a bundle with the F-actin value (cid:96)p = 5370d in an
optical trap at κT = 0.4511kBT /d2, a value large enough to avoid escaping filaments for the flexible case [7, 29], but
0400800120016002000 t (W0-1)010203040506070 <L(t)> (d)0100200 t(W0-1)01020304050<L(t)> (d)051015202530 F (kBT/d)00.511.5 V (dW0)0.20.40.60.81 F/Fs0246810 V/VSR00.20.40.60.81F/Fs(a)(b) 9
weak enough to allow flexibility effects to manifest. Results for the rigid model have been already discussed in [24].
For the flexible bundle, we followed, for a time interval of T = 700W −1
, the relaxation of a statistical sample of 103
replica with initial configurations sampled by the equilibrium size distribution with the wall fixed at L(0) = 5d [7].
The observed asymptotic values of the wall position and its variance are in full agreement with the equilibrium theory
predictions (see the caption of figure 4) [7]. The short vertical bars on the data represent the standard deviation σ,
not the statistical error. Its rather limited value at all times is a unique characteristic of the optical trap apparatus
which allows to obtain rather precise measurements already with a limited number of relaxations (just one in real
experiments! [11]).
After a fast initial relaxation, (cid:104)L(t)(cid:105) approaches exponentially its asymptotic value Ls
0
(cid:104)L(t)(cid:105) − Ls ∼ −Ae−t/τ OT
(13)
where τ OT is the final relaxation time and A the amplitude of the slowest relaxation mode which depends on the
initial conditions. Fits, represented in the figure by dashes lines, provide A = 65.9d, τ OT = 128W −1
and A = 21.8d,
τ OT = 653W −1
for flexible and rigid bundles respectively. Velocity and load both follow analogous long time behaviors
with the same relaxation time which can thus be expressed as
0
0
τ OT =
1
κT
lim
t→∞
Fs − F (t)
V (t)
(14)
where V (t) = d(cid:104)L(t)(cid:105)/dt and F (t) = κT(cid:104)L(t)(cid:105) is the time dependent load. Note that τ OT (cid:29) τmicro ∼ τchem = W −1
in both models, where τmicro is the relaxation time of the distribution of the filament tips with respect to the wall
position. More precisely, it was shown in the rigid bundle case [24] that τmicro is of the order of few W −1
and the
distribution of filament tips is peaked at the wall position with a characteristic width of the order of unity, as long as
F/Fs > 0.15. In the long time relaxation, for both the rigid and the present flexible bundles, the load changes very
slowly on the time interval over which the chemical events take place and the bundle tips relax. This separation of
time scales implies that during the slow variation of the trap amplitude, and thus of the load, the distribution of the
filament lengths (with respect to the position of the obstacle) remains equivalent to the distribution in a hypothetical
constant load experiment with load value set to κT(cid:104)L(t)(cid:105) at the current stage of the relaxation process.5 This finding
holds for any value of Nf and κT in the non-escaping regime which emphasizes the general character of our results.
The bottom panel of figure 4 reports the wall velocity V (t), obtained by numerical differentiation, and plotted as
a function of the time dependent load F (t) for both rigid and flexible bundles. Rigid model behavior is represented
by the approximate solution of ref. [12] which turns out to be very accurate (see also ref. [24] for the accuracy of this
result). On the same plot we also report for reference the so called "Perfect Load Sharing" (PLS) behavior given by
0
0
(cid:20)
(cid:18)
(cid:19)
(cid:21)
V = d
U0 exp
− dF
Nf kBT
− W0
(15)
which implies an equal repartition of the work among all filaments[14, 30]. Equivalently, eq. (15) expresses that the
power of transduction of chemical energy into mechanical work, P = V F , results to be Nf times the power developed
by a single filament against a force f = F
. Such equal repartition is interesting as it was shown to maximize the
Nf
power of transduction P = V F [30, 31]. The V (F ) data for the flexible model at the chosen κT value closely follow
the rigid bundle curve up to F ≈ (12− 13)kBT /d (i.e. F/Fs ≈ 0.35, indicated by the vertical bar in the bottom panel
of the figure), corresponding to L ≈ 25d at that κT value (see also the inset of the top panel, where the velocity starts
deviating significantly to larger and larger values). As stalling is approached, the V (F ) data get closer to the PLS
curve (see the inset in the bottom panel). This behavior shows that the velocity for flexible bundles is function not
only of the external load but also of the bundle length.
The observed strong reduction of τ OT for flexible bundles is a manifestation of a new polymerization mechanism
linked to flexibility, a mechanism first suggested by Mogilner and Oster in a series of seminal papers [19, 32, 33] where
the Elastic Brownian Ratchet (EBR) model was introduced. This mechanism improves the work repartition over the
filaments as the fraction of filaments simultaneously touching the wall correlates with the load sharing capacities of
the filaments in the bundle. When filaments are short, the buckling force fb(L) they develop when in contact with the
wall is rather large and few filaments (just a single one acting sporadically for very tight traps) are able to sustain the
entire load. Therefore in a very short trap the dynamic behavior of the flexible bundle is equivalent to the rigid model
behavior, as seen in the inset of figure 4(a) for L < (25− 30)d. When filaments get longer, the single filament buckling
5As F/Fs → 0, the separation of time scales brakes down as both τmicro and the width of the size distribution diverge [12, 24].
10
force becomes increasingly weaker and a significant fraction of filaments has to be recruited simultaneously to sustain
the increasingly large external load. As L further increases, all filaments would be recruited to act permanently and
the PLS picture would become valid. This situation however is never reached in the present model because thermal
bending fluctuations cause filaments to escape before the PLS is reached, as discussed in section V. In eq. (14) we
have defined the optical trap relaxation time, an experimentally accessible quantity, and found that it may shorten
significantly for flexible bundles as a result of improved load sharing capacities. We come back on this important and
original issue in the next section where τ OT will be related to the trap strength and to an effective chemical friction
term reflecting precisely the flexibility influence on the bundle dynamics close to stalling.
IV. ANALYSIS OF THE FILAMENT FLEXIBILITY ON THE LOAD-VELOCITY RELATIONSHIP.
The increase of velocity induced by flexibility originates from the additional polymerization mechanism in which
the monomer addition inside a tip-wall gap smaller than d is made possible by a bending fluctuation. To interpret
quantitatively the onset of this mechanism in the appropriate high wall diffusion limit, we first reconsider the SR
model. The fast wall motion is blocked by the most advanced tip in the bundle. Consider a second filament which,
because of the seeds shift, has its tip located at a distance nδ with n ∈ [1, Nf − 1] from the tip of the most advanced
filament. If the wall position fluctuates beyond (Nf − n)δ away from the most advanced tip, the gap between the tip
of the second filament and the wall is ≥ d and this filament can polymerize with a rate U0. This is the subsidiary
mechanism of the SR brownian ratchet bundle which enhances the wall velocity with respect to the unstaggered
rigid bundle case and hence improves the load sharing capacities of the bundle just as a result of a more favorable
disposition of the seeds of the rigid filaments [31].
In the same conditions, but now with flexible filaments, the above mechanism exploiting wall fluctuations will remain
operative. In addition, as the wall fluctuates around its most probable position directly against the most advanced tip,
the second filament in the above example could alternatively polymerize by undergoing a nδ compression, resulting
into another kind of subsidy effect originally suggested for the 13 protofilaments microtubule facing a non fluctuating
wall [16]. Such a transition towards a compressed grafted filament should be followed by a relaxation of the filament
by displacing the wall against the external force. A compression by nδ plus the addition of a monomer becomes
energetically possible when the difference in free energy between the uncompressed state and the compressed state
with one added monomer satisfies nδfb(L)− kBT ln ρ1 ∼ kBT , where nδfb(L) is the compression free energy difference
and −kBT ln ρ1 is the free energy gain per monomer addition.6 This relation, together with eqs. (9)-(10) at weak
compression in the buckling state, gives a n-dependent effective length
(cid:115)
(cid:115)
¯L(n) ≈
nπ2(cid:96)pd
4Nf (1 + ln ρ1)
(16)
for a bundle of Nf filaments at supercriticality ρ1.
In panel (a) of figure 4, the flexibility effects for a Nf = 32
bundle with (cid:96)p = 5370d at ρ1 = 2.5 are observed to start at 25d ≈ ¯L(3) (see the insets), corresponding, for the used
trap strength κT = 0.4511kBT /d2, to F (cid:39) 12.5kBT /d indicated by a vertical dashed bar in panel (b) of the same
figure. Further, a series of four additional similar relaxation experiments is presented in figure 5 for four different
bundle sizes Nf = 16, 32, 64, 128 in identical supercritical conditions ρ1 = 2.5. The trap strength has been set at
κT = 0.011453 Nf kBT /d2 in order to reach roughly the same stalling distance Ls ≈ 80d at large time and to cover
the same L window from the initial value ≈ 5d. We empirically observe that in all cases ¯L(3), indicated by a vertical
arrow in the figure, signals the beginning of the flexibility influence. Hence, we adopt
Λ = ¯L(3) =
3π2(cid:96)pd
4Nf (1 + ln ρ1)
(17)
as the crossover size between the rigid regime and the flexible regime for a staggered bundle of Nf filaments. We note
that the influence of ρ1 on Λ (not studied here as all experiments are performed at the same supercritical conditions)
is predicted to be marginal by comparison to the influence of Nf or (cid:96)p variables.
As the above analysis shows, the velocity enhancement due to filament flexibility depends on L. Therefore the
velocity-load relation V (F ; Nf , ρ1) characterizing the behavior of SR models should be generalized to staggered semi-
flexible bundles (in the experimentally relevant adiabatic trap regime) by including the bundle size L as a relevant
6Here we replace the filament contour length by the trap length since they differ at most by few d's, close enough for our qualitative
consideration [7, 23].
11
FIG. 5. Flexible bundles relaxation experiments against a harmonic load with trap strength κT = 0.011453 Nf kBT /d2 chosen
to cover a similar L window between the bundle initial size L0 ≈ 5d up to a common equilibrium value Ls ≈ 80d. The plots
show the instantaneous average wall velocity as a function of the trap width. The continuous lines show, for reference, the
SR behavior in identical conditions based on D´emoulin et al.
theory [12]. The data points are the average over 128 (for
Nf = 64, 128), 256 (for Nf = 16) or 992 (for Nf = 32) independent trajectories. The crossover to the escaping regime is
Ll = 76.6d at ρ1 = 2.5 from eq. (2) but trajectories are stopped before observing any escaping filament. The flexibility effect
is seen to start at a distance close to Λ given by eq. (17) and indicated by the vertical arrows. For sake of clarity the vertical
axis of different experiments are shifted upward by 0.5 dW0.
independent macroscopic variable, leading to the form V (F, L; Nf , ρ1). Flexibility being more adequately represented
by the reduced size λ = L/Λ, the alternative but equivalent form V (F, λ; Nf , ρ1) can be used to describe the dynamical
behavior of flexible bundles, the rigid case being recovered for λ ≤ 1.
T
In order to map the entire V (F, λ) relationship for the flexible bundle at a given (Nf = 32, ρ1 = 2.5) state point,
we performed 14 different optical trap experiments (averages over 992 individual trajectories in each case) at different
trap strengths in the range 0.2094kBT /d2 ≤ κT ≤ 1.10kBT /d2. A subset of those dynamical relaxations is reported in
figure 6. The data analysis requires a careful treatment of trajectories affected by escaping filaments. While Ll = 76.6d
obtained by eq. (2) at ρ1 = 2.5 is a good qualitative estimate of the crossover length to the escaping regime, a more
quantitative threshold to the escaping regime in an optical trap at fixed κT , as derived in ref. [7] at stalling conditions
is Ls ≤ 65d and ρ1 = 2.5. This condition remains valid for the SD sampling of the bundle growth relaxation towards
T (cid:39) 0.42kBT /d2 for the present system. For the few lowest
a final equilibrium state and it roughly corresponds to κmin
values of κT < κmin
employed, some filaments in all dynamical replica were found to escape at some stage and in
figure 6 we just show the portion of the average relaxation before the manifestation of the phenomenon. Approaching
the threshold values of κT from the small κT side, a larger and larger fraction of replica doesn't present escaping
filaments over the time window of the experiment. In these cases, this fraction is used for the statistical analysis.
The resulting dynamical behaviors are then used to extract values of the velocity V (F, λ). Similar to the analysis
performed to obtain the bottom panel of figure 4, we obtain the velocity V (t, κT ) for each relaxation experiment by
numerical differentiation and we eliminate the time in favor of the force F (t) = κT(cid:104)L(t)(cid:105). Since the bundle flexibility
is directly related to its size, to interpret and rationalize the results in terms of the flexibility we group together data
from the different experiments but at the same trap amplitude L. The justification of this procedure is again related
to the observed separation of time scales between the slow load variation and the faster relaxation of filament lengths.
The resulting picture is reported in figure 7 where we show the power of transduction of chemical into mechanical
energy. For short filaments (λ ≤ 1) the effect of flexibility is negligible and the power closely follows the SR behavior
(partial load sharing [30]). As the bundle length increases (λ > 1), the power progressively increases towards the
PLS mean field behavior of eq. (15) indicating that flexibility induces better work sharing capacities, as anticipated
01020304050607080L (d)00.511.522.53V (dW0)Nf=128Nf=64Nf=32Nf=16 by Schaus and Borizy [30]. The PLS behavior, however, cannot be reached without entering the escaping regime
occurring at λ = Ll/Λ (cid:39) 3 for the present system (see figure 9).
Data for the force dependence of the velocity or the power at fixed L (or fixed λ) can be quantitatively represented
12
by a linear combination of PLS and SR behaviors. We suggest the form
V (F, L) (cid:39) b(L)V P LS(F ) + [1 − b(L)]V SR(F )
(18)
with the simple ansatz b(L) = (L/Ll)2. Such a choice is suggested by the L2 growth of the number of filaments
recruited to produce a given force F at different trap widths, as mentioned earlier in this section. The crossover
length Ll, defined in eq (2) as the length where the bundle enters massively in the escaping regime, is here interpreted
as the upper limit of the non escaping regime where all filaments are permanently touching the wall, hence leading
to the PLS behavior. In figure 7, eq. (18) is tested for the power P (F, λ) = F V (F, λ) replacing L by λ, so
(19)
with b(λ) = λ2(Λ/Ll)2. These results and additional data for Nf = 64 and Nf = 128 provided in figure 10 of appendix
B show the pertinence of eq. (19) for a full characterization of the flexibility effects based on analytical expressions,
hence relatively straightforward to use in interpreting future in-vitro experimental data.
P (F, λ) (cid:39) b(λ)P P LS(F ) + [1 − b(λ)]P SR(F )
FIG. 6. Non-equilibrium dynamics of a staggered bundle of Nf = 32 flexible filaments at ρ1 = 2.5 growing against an optical
trap load. Time relaxation of the wall position (cid:104)L(t)(cid:105) and associated standard deviation σ(t) at the indicated value of the
trap strength κT . Data are obtained averaging the relaxation of 992 equivalent replica with initial conditions sampled by the
equilibrium distribution at L(0) = 5d. As long as the plateau remains below ≈ 65d (see text), escaping filaments influence is
absent and each relaxation curve goes asymptotically towards a plateau value Fs/κT in full agreement with the equilibrium
statistical mechanics prediction of the stalling force Fs for the same model [7], indicated by a horizontal dashed line. We observe
that such an agreement is still verified for κT = 0.4189kBT /d2 (Ls ≈ 70d) despite the presence of some escaping filaments. In
all curve portions shown, we checked that the bias in the mean (cid:104)L(t)(cid:105) resulting from including or not data from trajectories
with at least one escaping filament, did not exceed 0.03d. The points where the different (cid:104)L(t)(cid:105) curves intersect an horizontal
line (e.g. the continuous black line at L = 20d) lead to the fixed L series of (V, F ) data reported in figure 7. Along the constant
L horizontal line, the value of the force is given by κT L while the value of the velocity by the local slope indicated as black
dashed lines.
The long time exponential relaxation of the wall allows to define the concept of chemical friction γ
(cid:20) ∂V (F, κT )
(cid:21)−1
∂F
F =Fs
τ OT (κT ) = − 1
κT
=
γ(κT )
κT
(20)
0100200300400500600t (W0-1)020406080100<L(t)> (d) T=1.100 kBT/d2 T=0.7300 kBT/d2 T=0.5332 kBT/d2 T=0.4189 kBT/d2 T=0.3665 kBT/d2 T=0.2094 kBT/d2 13
FIG. 7. Power P (F, λ) = F V (F, λ) developed by the polymerization force of a homogeneous bundle of Nf = 32 actin semi-
flexible filaments, pressing against a mobile wall located at L and subject to a load F in super-critical conditions at ρ1 = 2.5.
λ = L/Λ measures the flexibility, where Λ is the threshold distance between the rigid and flexible bundle domains defined in
eq. (17). Two theoretical force velocity predictions are also shown, the PLS prediction eq. (15) and the SR model prediction
as represented by the D´emoulin et al. approximation [12, 24]. At given load, a systematic increase of the power with λ is
observed, the two theoretical curves representing upper and lower boundaries to flexible data points. Except at large λ > 2.4,
data at fixed λ can be fairly well represented by a linear combination of those boundaries with weights proportional to λ2 (see
eq. (19)) shown by dot-dashed lines.
as the inverse of the slope of the velocity-load relationship at stalling, in principle directly accessible in laboratory
experiments. As shown in appendix C the κT -dependence can be replaced by the Ls-dependence and the chemical
friction directly related to the F -derivative of V (F, L) at stalling, sensitive to the load-sharing feature of the bundle.
The ansatz eq. (18) provides
where the L-independent friction coefficients of the PLS and the SR Brownian Ratchet models are given by
(21)
(22)
(23)
1
γ(Ls)
γP LS = −
γSR = −
1
γSR
1
γP LS + [1 − b(Ls)]
(cid:21)
=
Nf kBT
d2W0
(cid:39) b(Ls)
(cid:32)(cid:20) ∂V P LS
(cid:32)(cid:20) ∂V SR
(cid:21)
∂F
(cid:33)−1
(cid:33)−1
F =Fs
∂F
F =Fs
= C(Nf , ρ1)γP LS
using the PLS velocity-load relation, eq.
(15), and the SR velocity-load relation in terms of an effective load
sharing coefficient C(Nf , ρ1)[12, 24]. For a bundle with Nf = 32 at ρ1 = 2.5, γP LS = 32kBT /(d2W0) and
γSR = 281.6kBT /(d2W0) predicted by eq. (23) using the approximate C expression in Eq. (C7) of ref. [24]. Values
of the the chemical friction from our numerical experiments, reported and discussed in appendix C, are in semi-
quantitative agreement with predictions from eq. (21) based on these values of γP LS and γSR. Moreover they confirm
a progressive decrease of the chemical friction with bundle length which is roughly given by γ ≈ γP LS(Ls/Ll)−2, for
b(Ls) = (Ls/Ll)2 > 0.3.
We stress again that probing the bundle relaxation in an optical trap near stalling, for a range of κT values
compatible with the non escaping regime, if possible, would provide an experimental measure of the chemical friction,
hence a direct measure of the Ls-dependent load-sharing capacities of a multi-filament actin bundle.
051015202530F(kBT/d)02468P (kBT W0)PLSSRλ=0.4λ=0.8λ=1.2λ=1.6λ=2.0λ=2.4λ=2.8 14
FIG. 8. Wall relaxation (cid:104)L(t)(cid:105) for a homogeneous bundle of Nf = 8 under conditions that mimic the experiments reported in
figure 4(a) (upper panel) and 4(b) (lower panel) of ref [11], respectively. In both cases an overshooting behavior is observed
and corresponds to the occurrence of the escaping filaments. The observed large time plateau values roughly correspond to the
mechanical equilibrium between the trap restoring force and the force to bend the eight filaments, while they keep increasing
their size unimpeded by the wall.
V. THE ESCAPING REGIME.
In supercritical conditions, the possibility exists for growing flexible filaments hitting a hard obstacle, to undergo a
large bending fluctuation and start growing laterally without restraint. The onset of such escaping filaments is allowed
by our dynamical model. Sections III-IV were focusing on flexibility effects in the non escaping regime defined by
L < Ll defined in eq. (2). In the present section we consider situations where the wall position can sample values
L > Ll, allowing the system to perform excursions into the escaping regime. We classify a filament as escaped (and
irremediably lost for the transduction of chemical free energy into mechanical work) when its size gets larger than
z∗(L) (see its definition and illustration in right panel of figure 1).
To our knowledge, a unique experimental attempt to follow actin bundle dynamics using the optical trap set-up
has been reported to date in ref.
[11]. More specifically, relaxation to stalling for a bundle of 8-10 actin filaments
have been monitored. The interpretation of these in-vitro experimental data [11] turned out to be delicate given
the possible bias coming from the presence of escaping filaments. It was claimed that in order to probe a genuine
polymerization force at stalling the observed final force must be lower than the buckling force of a single grafted actin
filament with contour length equal to the trap gap [11]. The relaxation towards a steady value of the trap width for
a bundle of 8-10 filaments at (ρ1 = 1.92, κT = 0.0028kBT /d2) and at (ρ1 = 1.69, κT = 0.01076kBT /d2) was reported
in figure 4a and 4b of ref. [11] respectively. Data of the first experiments were discarded while data from the second
0100200300400500600700t(W0-1)050100150200<L(t)> (d) T=0.01076, ✁^1=1.6900100100200200300300400400500500600600700700t(W0-1)050100150200250300<L(t)> (d) T=0.002811, ✁^1=1.920100200300400500600700t(W0-1)00.20.40.60.8<F(t)> (kBT/d)0100200300400500600700t(W0-1)00.511.52<F(t)> (kBT/d) experiments (and similar cases) were exploited on the basis of the above criteria. However, the final value of the trap
width measured in the second experiments (figure 4b of ref.
[11]), suggested that the steady value of the force is
close to the force needed to stalled a single filament and not a bundle of 8-10 filaments according to eq. (1). This
surprising result was interpreted as the result of "dynamical stalling" where only the longest filament has to be stall
as it quickly becomes unstable and depolymerizes rapidly, hence it is replaced by a new longest filament which is
stalled in turn by the loaded wall. This scenario has not been confirmed so far but it has been conjectured that for
many actin filaments to work together and for filaments to avoid buckling, their growing barbed ends at membrane
contact must form complexes with processive formin membrane proteins [21].
15
It is interesting to simulate the experiment within our model and characterize the response, i.e. the time relaxation
of the trap. We simulated the relaxation of a bundle of 8 filaments at the two experimental conditions described
above. For each system the dynamics of 32 equivalent replicas, with initial conditions sampled by the equilibrium
filament length distribution at L(0) = 5d, were followed up to the attainment of a steady value for the trap width
L(t). The results are reported in figure 8.
In both cases we observe a non-monotonous behavior of (cid:104)L(t)(cid:105) at variance with the results reported in figure 6.
The observed overshoot is, in both experiments, the signature of the occurrence of escaping filaments. Indeed, all
filaments of all replica became escaping in the time interval between the end of the first, almost linear, growth and
the beginning of the plateau observed at larger times. In this time interval we observe an increase of the variance
signaling the random character of the filament escaping times in different replica. The asymptotic value of the trap
width corresponds to the load exerted by the trap to equilibrate the elastic force provided by eight very long d-WLC
bent at 90 degrees within the width of the trap. Equation (F8) predicts Lesc (cid:39) 230d and Lesc (cid:39) 147d for the two
experiments respectively, in qualitative agreement with the observed values. The corresponding threshold values
(F9) are ¯κT = 0.064kBT /d2 and ¯κT = 0.046kBT /d2 both rather larger than the
for the trap strength from eq.
employed values. According to eq.(2), the crossover distance to the escaping regime in the two cases is Ll(1.92) (cid:39) 91d
and Ll(1.69) (cid:39) 101d rather shorter than the observed Lesc. From the analysis of the trajectories, in the upper
panel of figure 8 (κT = 0.002811kBT /d2) we detect the first escaping filaments after an average time t1 = 381W −1
(σt1 = 23W −1
(σt2 = 23W −1
). In
the lower panel (κT = 0.01076kBT /d2), we have corresponding times t1 = 437W −1
) and t2 = 498W −1
(σt2 = 51W −1
). The onset of escaping filaments thus takes place essentially during the backward displacement of
the wall, just after the overshoot at L values much higher than Ll. This is probably due to a kinetic effect linked to
the initial fast motion of the wall, V ≈ dW0 (higher panel) and V ≈ 0.5 dW0 (lower panel), close to the free growth
velocity at zero load (U0 − W0)d (which is 0.92 dW0 and 0.69 dW0 respectively). Such large wall velocity probably
prevents filaments to enter earlier in the escaping regime i > z∗(L).
) while all Nf = 8 filaments are found to escape after an average time t2 = 413.0W −1
(σt1 = 28W −1
0
0
0
0
0
0
0
0
In ref. [7] we have already pointed out that for a bundle of 8 filaments growing at (ρ1 = 1.69, κT = 0.01076kBT /d2)
(second experimental case) of the expected stalling force from eq. (1) would imply an average trap gap much larger
than the crossover Ll and hence the relaxation process could not avoid incurring in the escaping regime. Our new
results confirm the occurrence of dynamical instabilities when filaments become escaping, as the trap moves under
the combined action of the filaments force and the external load.
For the experimental data at κT d2/kBT = 0.01076 (figure 4b in [11]), the plateau is reached asymptotically after
a growth characterized by a relaxation time τ OT (cid:39) 55s (disregarding that the growth is affected by unexplained
oscillations). Hence, the experimental data are compatible with the relaxation of a single filament, not only for the
plateau value but also for the longest relaxation time as it is close to τ OT = kBT (W0κT d2)−1 = 66s predicted by
using the experimental κT value, W0 = 1.4s−1 and eqs. (C3)-(22) with Nf = 1.7
We want to emphasize that our present results concern a model based on a single species of actin monomer complexes
(no hydrolysis effects) with a depolymerizing rate supposed to be unaffected by the presence of the compressional
force due to the obstacle. Therefore it does not contain the "dynamical stalling" phenomenon invoked to explain the
puzzling experimental results. It would be interesting to perform a similar study with a model where a compression
dependent depolymerization rate is considered. However, at present, the experimental input about this dependence
is missing.
VI. DISCUSSION AND CONCLUSION
Using a realistic model of semi-flexible filaments, we have introduced, in a rather general fashion, the effects of
flexibility on the dynamics of actin bundles in supercritical conditions pressing against a loaded obstacle. Applying
7Note that for a single filament, the PLS and the SR models coincide.
16
FIG. 9. The three different regimes for a bundle of filaments with finite flexibility indicated by the filament persistence length,
illustrated here for Nf = 32 and ρ1 = 2.5. The case of actin (cid:96)p = 5370d is indicated by a vertical dashed line. The non escaping
regime for a homogeneous bundle is further divided into a rigid filament regime and a flexible filament regime by the line at
L = Λ given in the text. The escaping regime is met for L = Ll with Ll = ((cid:96)pd/ ln ρ1)1/2.
an external load that mimics the optical trap apparatus in realistic conditions we have followed the non-equilibrium
relaxation of the trap under the combined action of the polymerization force of a staggered actin bundle and the
external load. The relaxation of the trap is found to be mostly adiabatic as the reorganization of the filament sizes
against the loaded wall is one-two orders of magnitude faster than the time of the load. This fundamental property
allows to consider most states visited during the dynamical relaxations as non-equilibrium stationary states for the
bundle and, by eliminating the time from the analysis of the results, to generalize the well known velocity-load
relationship of rigid bundles to include flexibility. The velocity V (F, L) now depends in addition on the bundle length
L, which greatly enriches the theoretical scenario. The new fundamental features are: i) a new characteristic length
Λ ≈ (d(cid:96)p/Nf )1/2 (see eq. (17)), a bundle property that marks the transition between SR behavior and flexible bundle
behavior, so the system properties are more adequately described in terms of the reduced length λ = L/Λ; ii) the
enhancement of the power of transduction of chemical energy into mechanical work for λ > 1; iii) the occurrence of
the escaping regime beyond some threshold distance Ll related to single filaments characteristics. Figure 9 illustrates
these features and their dependence on the filament persistence length (cid:96)p. We show the three regimes: 1) a rigid
regime at L < Λ (λ < 1) where the flexibility effect is negligible; 2) the intermediate regime where flexibility effects
are present with an increase of the obstacle velocity with flexibility (i.e. bundle length at fixed (cid:96)p) at fixed external
load; 3) the escaping regime at large enough trap width (L > Ll). We see that the non escaping flexible regime gets
, which
wider for increasingly flexible filaments. 8 Let us consider the ratio Ll/Λ =(cid:2)4/3π2 Nf (1 + ln ρ1)/ ln ρ1
is independent of (cid:96)p. In supercritical conditions (ρ1 > 1) Ll/Λ ≥(cid:0)4/3π2(cid:1)1/2
, therefore a wide flexible regime
(cid:3)1/2
before escaping requires bundles with a large number of filaments.
Using the optical trap set up for different values of κT we characterized the three regimes and established their
8For an unstaggered bundle, flexibility would add the same new polymerization mechanism when a compression of d would become of the
order of kBT . But now Λ ∼ Ll and the new non escaping flexible regime is absent.
N 1/2
f
00.0050.010.0150.020.0250.03(lp/d)-1/200.020.040.060.08(L/d)-1escapingnon-escapingflexiblerigidLl-1=[log(U0/W0)]1/2lp-1/2 -1=[4Nf(1+log(U0/W0))/(3✁2d)] lp-1/2actin =25dLl=76d boundaries. We have shown that the power of the bundle is strongly increased by the flexibility in the intermediate
regime and can almost cover the entire gap between the rigid bundle behavior (SR) and the mean field Perfect Load
Sharing (PLS) behavior, which appears to be an upper bound for a many flexible filaments bundle as the filaments
in the bundle approaches the escaping regime. The spectacular power increase with L at a given load F is the result
of an improved work sharing capacity of the bundle due to the increasing fraction of filaments pressing on the wall
(∝ L2) when developing the polymerization force opposing the load F . A simple ansatz combining linearly the PLS
and SR behaviors provides a satisfactory description of the data. Our work shows how filament flexibility could be
considered in interpreting future experiments, in a way which enriches considerably the present dominant theoretical
model [2]. Despite a consistent treatment of filament flexibility including the escaping regime, our model is unable to
reproduce the in-vitro experimental results of ref. [11] which remain unexplained.
17
ACKNOWLEDGMENTS
C.P. and J.P.R. are grateful to J. Baudry, J.F. Joanny et D. Lacoste for useful discussions. We thank G. Destr´ee and
P. Pirotte for technical help. J.P.R. thanks the financial support and hospitality of the University of L'Aquila during
a three months visit. C.P. is supported by the Agence Nationale de la Recherche (ANR), France under the project
"HyLightExtreme". Computational resources have been partially provided by the "Consortium des `Equipements de
Calcul Intensif (C`ECI)", funded by the Fonds de la Recherche Scientifique de Belgique (F.R.S.-FNRS) under Grant
No. 2.5020.11.
Appendix A: Tables of symbols and physical properties
In this appendix we collect, in three tables, all symbols used in the present text. Table I is devoted to macroscopic
quantities, table II to microscopic quantities and table III to time scales.
Appendix B: Further support to the validity of the ansatz eq.(19) to interpret the power of a staggered
flexible bundle.
In this appendix we report results for bundles with different number of filaments at the same reduced trap strength,
to further support our analysis of the velocity-load-length relationship, in particular the weighted average between the
SR and the PLS behaviors given by the ansatz eq.(19). In figure 10, we compare data for the optical trap relaxation
of bundles of 64 and 128 flexible filaments to the prediction eq.(19) to stress its general validity, beyond the single
case Nf = 32 illustrated in the main text.
Appendix C: Relaxation time of a staggered flexible bundle in optical trap and chemical friction.
In this appendix we show how to formally go from the κT -dependent chemical friction defined in eq.(20) to the
L-dependent chemical friction. Applying the variable change y = g(x, z) to a function of two variables f (x, y), one has
. For the specific g(x, z) = x/z, identifying x = F , y = κT ,
the general relationship
z = L, one gets
(cid:104) ∂f
(cid:105)
=
∂x
z
(cid:18)(cid:104) ∂f
(cid:105)
∂x
y
+
(cid:104) ∂f
(cid:105)
(cid:21)
(cid:20) ∂V
∂y
x
(cid:19)
(cid:105)
(cid:20) ∂V
z
(cid:104) ∂g
∂x
=
(cid:21)
(cid:20) ∂V
(cid:21)
+
1
L
∂F
κT
∂F
L
∂κT
F
(cid:34)(cid:20) ∂V
(cid:21)
(cid:35)
(cid:20)(cid:20) ∂V
(cid:21)
(cid:21)
≈
∂F
κT
F =Fs
∂F
L
F =Fs
The second term in the rhs of eq. (C1), which is strictly zero in the rigid filament case, is negligible for a flexible
bundle close to stalling as it is linked to the very small change of Fs with equilibrium bundle length fixed by κT , and
we reduce to
(C1)
(C2)
TABLE I. Macroscopic/mesoscopic quantities
18
Meaning
Bulk free actin monomer density.
Number of filaments in the bundle.
Bulk polymerization rate per filament tip.
Bulk depolymerization rate per filament tip.
Critical free monomer density.
Reduced density.
Bundle length and/or optical trap displacement with respect to the grafting
wall position. eq. (2) and caption of figure 1.
Trap strength. Caption of figure 1.
Load force in the optical trap. Caption of figure 1.
Obstacle velocity.
Stalling force and stalling length. eq. (1).
Symbol
ρ1
Nf
U0 = konρ1
W0 = kof f
kof f
kon
ρ1c =
ρ1 =
ρ1
ρ1c
=
U0
W0
L
κT
F = κT L
dL
dt
Fs, Ls =
V =
Fs
κT
Ll
(cid:20) ∂V (F )
ξ
∂F
γ = −
P = V F
¯L(n)
Λ
λ =
L
Λ
(cid:21)−1
Fs
Crossover bundle length between non escaping and escaping regimes. eq. (2).
Friction coefficient of the wall in the solution. eq. (4).
Effective chemical friction coefficient near stalling. eq. (20).
Transduction power of the bundle.
Filament length at which a polymerization involving a wall compression of
nd/Nf becomes thermally activated. eq. (16).
Bundle size threshold between the rigid bundle regime and the flexible (non
escaping) regime. eq. (17).
Flexibility parameter: rigid regime λ < 1, flexible regime λ > 1. Caption of
figure 7.
Hence we can replace κT with Ls in eq. (20) to define
(cid:20) ∂V (F, L)
(cid:21)−1
∂F
F =Fs
τ OT (Ls) = − 1
κT
=
γ(Ls)
κT
(C3)
The chemical friction coefficients γ(κT ) from eq. (20) and γ(Ls) from eq. (C3) are equal when referring to a specific
equilibrium state linking the equilibrium wall position Ls and the trap strength κT , at a given (Nf , ρ1) state point.
We analyzed the flexibility influence on the chemical friction as a function of the bundle equilibrium length Ls.Data
for a bundle with Nf = 32 at various values of Ls in the non escaping regime (0.5 < κT < 1.1) from our SD
simulations are reported in figure 11 in terms of the weight b = (Ls/Ll)2 of the PLS contribution. The weighted
average ansatz of eq. (21) together with the approximation γ = (Ll/Ls)2γP LS derived from eq.(21) by neglecting
(1/γSR) versus (1/γP LS), are indicated to illustrate the γ ≈ 1/L2
s behavior. Extracting the relaxation time directly
from the numerical relaxations is not trivial since it requires small statistical errors in the long time region where
the deviations from the asymptotic trap length become vanishingly small. With the typical noise level of our data
the value of the relaxation time depends substantially on the time window of the fit. Two estimates of the chemical
friction extracted from SD results are shown as illustrated in the caption of figure 11. The results confirm a progressive
decrease of the friction as the bundle length increases, hence an increase of the load-sharing capacity of the bundle.
TABLE II. Microscopic parameters and intermediate quantities
19
F-actin incremental length per additional monomer in the double strand
Meaning
actin filament.
F-actin persistence length.
Seed longitudinal position of filament n with respect to grafting plane
(−d/2 ≤ hn ≤ d/2). eq.(3).
Number of monomers in filament n. Text below eq.(3).
n-th filament contour length Lcn = (jn − 1)d. Text below eq.(6).
n-th filament seed-wall distance Ln = L − hn. Text below eq.(6).
Largest monomer size of filament n for which bending fluctuations do not
involve any direct interactions with the wall. Caption of figure 1.
Assumed cross over filament size between the non-escaping (jn ≤ z∗) and the
escaping (jn > z∗) regime. Caption of figure 1.
Buckling force for a filament of length L. eq.(10).
Potential of mean compression force for a particular filament of fixed size j
and seed at h = 0 facing a wall at L. Eq.(7) for α definition.
Total free energy (chemical+compressional) of a particular filament of size j.
Eq.(7) and figure 3.
Force deriving from wj. Eq.(8).
Microscopic bundle force. Eq.(6)
TABLE III. Time scales
Meaning
Chemical characteristic time. Discussion justifying eqs.(6,7,8).
Local wall diffusion characteristic time. Discussion justifying eqs.(6,7,8).
Longest relaxation time of a bundle in supercritical conditions subject to an
optical trap compression load, related to the relaxation of the bundle in
optical trap towards equilibrium (κT Ls = Fs). Eqs.(13), (14).
Symbol
d = 2.7nm
(cid:96)p = 14.5µm
hn
jn
Lcn
Ln
z(L, hn)
z∗(L, hn)
kB T (cid:96)p
π2
4
fb =
−kBT ln αj(L)
L2
wj
¯fj
Fbun
Symbol
−1
τchem = W
0
ξd2
kBT
τD =
τ OT
Appendix D: Experimental time scales
The overdamped limit of the Langevin equation eq.(4) is justified by the very fast inertial relaxation time τ inertia =
3 R3ρ ≈
M/ξ = 2 × 10−7s estimated according to the experimental setup [11] for a micron radius bead M = 4π
4.2 × 10−15Kg and from Stokes law ξ = 1.9 × 10−8 Js/m2. Other relevant time scales are much longer: the trap
relaxation time in absence of bundle is τf ree = ξ/κT = 2.5 ms (using a typical value κT = 0.008 pN/nm [11]) and the
chemical reactions time scale is τchem = W −1
0 ≈ 0.7 s [25].
Another relevant time scale which characterize the dynamics is the diffusive one related to the diffusive motion of
the wall over the length corresponding to the size of the (de)polymerization unit and estimated by τD = d2/D where
D is the wall diffusion coefficient. As in reference [24], let's define = τchem/τD. Experimentally, = 4.7 × 10−5 for
actin, using ξ = 1.9 × 10−8 Js/m2, W0 = 1.4 s−1, d = 2.7 nm and kBT = 4.14 × 10−21J. Such a small value cannot
be adopted in the simulations for computational inefficiency reasons. Computer time would be essentially spent to
study the wall relaxation next to a bundle with quasi-fixed filament sizes while our main interest is the force-velocity
20
FIG. 10. Flexible bundles relaxation experiments against an harmonic load with trap strength κT = 0.011453Nf chosen to cover
the same L window between initial size L0 ≈ 5d up to a common equilibrium value Leq ≈ 80d. The data points correspond to
successive situations for increasing bundle size values during the fixed κT relaxations for Nf = 64 (upper panel) and Nf = 128
(lower panel). Each point represents P = F V versus F where V is the instantaneous velocity and F = κT L is the instantaneous
external force applied to the bundle by the loaded wall. The plot show the power predicted by eq.(19).
requiring both wall and filament size sampling. Hence, as in our recent investigation of the rigid bundle case [24], we
have adopted the value = 4.7 × 10−2, small enough to be representative of the experimental wall dynamics with
(cid:39) 5 × 10−5.
The longest intramolecular relaxation time of a grafted WLC with contour length Lc is[34]
(cid:18) 2
(cid:19)4 ξ(cid:48)L4
c
kBT (cid:96)p
π
τintra =
where ξ(cid:48) is a drag coefficient per unit length. ξ(cid:48) can be estimated by [34]
ξ(cid:48)
0 =
4πν
(ln (Lc/a) + 2 ln 2 − 0.5)
where ν is the solvent viscosity and a is the diameter of the rod modeling F-actin hydrodynamically (a = 5.4 nm= 2d
[35]). Writing Lc = (j − 1)d for the filament contour length with j articulation points, one gets
64(j − 1)4
νd3
kBT
(cid:32)
τintra (cid:39)
=
(cid:33)
2 + 2 ln 2 − 0.5]
((cid:96)p/d)π3[ln j−1
(j − 1)4
2 + 2 ln 2 − 0.5
ln j−1
· 1.83 × 10−12 s
(D3)
(D1)
(D2)
01020304050F (kBT/d)02468101214161820P (kBTW0)L=20dL=40dL=60dL=70dPLSSR020406080100120F (kBT/d)0481216202428323640P (kBTW0)Nf=64Nf=128 21
FIG. 11. Inverse of chemical friction (1/γ) versus b(Ls) = (Ls/Ll)2 from the long-time relaxation of a bundle of 32 flexible actin
filaments in optical trap relaxation data in the non escaping regime, illustrated in figures 4-6. The two parameters fit (p1, p2)
consists in extracting an effective relaxation time τ OT = p2 using the form (cid:104)L(t)(cid:105) = Ls − p1 exp (−t/p2) where Ls is the exact
theoretical prediction of the equilibrium position of the trap [7], in an adjusted time window where Tmin is selected empirically
to get an optimal single exponential fit for the long time behaviour. The three parameters fit (p1, p2, p3) consists in extracting
an effective relaxation time τ OT = p2 using the best fit with the form (cid:104)L(t)(cid:105) = p3 − p1 exp (−t/p2) in a fixed time window
. The continuous curves are the predictions of eq.(21) and its approximation γ = (Ll/Ls)2γP LS
[Tmin, Tmax] = [400, 600]W
(see text)
−1
0
.
using ν = 0.001, d = 2.7× 10−9, kBT = 4.14× 10−21 in MKS units. Thus one finally gets estimates τintra = 3× 10−6 s
for j = 51 and τintra = 4 × 10−5 s for j = 101, hence showing that the ratio τintra/τchem ≤ 5 × 10−5 as long as
j ≤ 100.
Appendix E: Explicit wall stochastic dynamics Algorithms (EWA)
To get a numerical solution of the model described above using a Langevin equation approach, we adopt an Explicit
Wall stochastic dynamics Algorithm (EWA) developed in ref. [24] to follow the dynamics of a bundle of rigid filaments
in optical trap, that we adapt below to the flexible case. We start from the equivalent Fokker-Planck formulation
of the time evolution of the joint distribution function Pj1,j2,....jNf
(L, t). Discretizing the wall position variable, the
dynamics reduces to a multi-variables continuous-time Markov chain for which standard algorithms are available to
generate representative stochastic trajectories.
The Fokker Planck equation describing the evolution of the joint probability distribution of the variables (L,{jn}n=1,Nf )
0.10.20.30.40.50.60.7b(Ls)=(Ls/Ll)200.0050.010.0150.020.0250.031/γEq.(21)Approximation of Eq.(21)3 parameter fit2 parameter fit in the case of flexible filaments reads:
n=1
Nf(cid:88)
Nf(cid:88)
Nf(cid:88)
Nf(cid:88)
n=1
n=1
+
−
+
−
∂Pj1,...,jNf
∂t
(L, t)
= − ∂
∂L
Jj1,...,jNf
(L, t)
Un(jn − 1, L)(1 − δ2,jn)Pj1,...,jn−1,...,jNf
(L, t)
Un(jn, L)Pj1,...,jn,...,jNf
(L, t)
Wn(jn + 1, L)Pj1,...,jn+1,...,jNf
(L, t)
Wn(jn, L)Pj1,...,jn,...,jNf
(L, t)
where Un and Wn are the (de)polymerization rates given in eqs.(11),(12), and the probability current density is
n=1
Jj1,...,jNf
(L, t) = −D
∂Pj1,...,jNf
∂L
− D
kBT
∂wjn(Ln, ρ1)
∂L
+ κT L
Nf(cid:88)
n=1
(L, t)
Pj1,...,jNf
(L, t).
22
(E1)
(E2)
The boundary conditions for these equations are different from the rigid case, where the most advanced tip's position
results in a reflecting boundary condition for the wall motion: here, conversely, filaments can bend and let the wall
move to positions L < (jn − 1)d. Two kinds of boundary conditions can be chosen:
a) filament sizes are limited to z∗(L) = int((πL)/(2d)), and hence Gholami's expression for the free energy
wjn(Ln, ρ1) [23], which is valid in the weak bending regime, can be used. This leads to the stationary state
described in reference [7] by equilibrium statistical mechanics.
b) filament sizes are not limited, and escaping filaments can appear. In this case, depending on κT , the evolution
of the system can lead to no true stationary state, with a finite average wall position (see figure 8 and appendix
F) and filament sizes keeping on growing unhindered.
These latter open boundary conditions (no real stationary state) have been used in this paper. We used for the mean
force potential wjn (Ln, ρ1) an expression which is valid also beyond the weak bending regime, as detailed in appendix
F.
To solve the set of Fokker-Planck eqs. (E1),(E2) numerically, we have discretized the wall position, allowing it to
move by forward or backward jumps on a grid of step δ(cid:48) = d/M , with M (cid:29) 1 integer. The SD trajectory follows the
time evolution of the ensemble of (Nf + 1) integer variables: the filament sizes, jn(t) (n = 1, Nf ), and the index k(t),
which gives the wall position L(t) = k(t)δ(cid:48). These variable can change by ±1 with elementary rates to be specified:
as for the filaments, the (de)polymerization rates are given by eqs.(11),(12), while the rates for the stochastic jumps
of the wall position are obtained through the discretization of the diffusive part of eq. (E1)[24], and read
W+1 =
W−1 =
D
δ(cid:48)2
D
δ(cid:48)2
β∆U+1
[exp (β∆U+1) − 1]
[exp (β∆U−1) − 1]
β∆U−1
(E3)
where the global potential energy U incorporates the optical trap term and the mean force potential eq.(7) of all
passive filaments in terms of an index kn defined as Ln = kδ(cid:48) − hn. We have
(cid:20)
(cid:20)
Nf(cid:88)
Nf(cid:88)
n=1
n=1
− ln
− ln
(cid:18) αjn(Ln + δ(cid:48))
(cid:18) αjn (Ln − δ(cid:48))
αjn (Ln)
(cid:19)(cid:21)
(cid:19)(cid:21)
αjn (Ln)
β∆U+1 =
β∆U−1 =
+
βκT
2
+
βκT
2
[(L + δ(cid:48))2 − L2]
[(L − δ(cid:48))2 − L2]
(E4)
(E5)
Given the state of the Nf + 1 discrete variables(cid:0){jn(tm)}n=1,...,Nf , k(tm)(cid:1) after m steps, the next state after m + 1
steps is given by changing one of these variables by ± one unit, according to stochastic rules based on all the above
elementary rates [27]. Random sampling fixes both the variable that is going to change and the associated time of
occurrence tm+1 − tm as follows. Let's denote by {Ti}i=1,...,2(Nf +1) the ordered rates of the 2(Nf + 1) possible single
variable changes:
23
depolymerizations: T2n = Wn(jn(m), k) using eq.(12)
polymerizations: T2n−1 = Un(jn(tm), k(tm)) using eq.(11)
wall jumps: T2Nf +1 = W+1({jn(tm)}, k(tm)) and T2Nf +2 = W−1({jn(tm)}, k(tm)) using eq.(E3)
and let's define the vector {S(cid:96) =(cid:80)(cid:96)
i=1 Ti}(cid:96)=1,2(Nf +1). Two independent random numbers (r1, r2) are sampled from
a uniform distribution between 0 and 1 to get the time to the next move τ = tm+1 − tm and the nature of the move,
defined by the index (cid:96) among the ordered rate vector T (1 ≤ (cid:96) ≤ 2(Nf + 1)):
τ =
1
S2(Nf +1)
ln (1/r1)
S(cid:96)−1
S2(Nf +1)
< r2 ≤
S(cid:96)
S2(Nf +1)
.
(E6)
(E7)
The set of Nf + 1 microscopic variables are stored at equally spaced times for analysis.
Appendix F: The mean force potential for a grafted d-WLC hitting a hard wall normal to the grafting
orientation
The very basic ingredient to introduce flexibility effects in the present modelling is the discrete-Wormlike model for
passive grafted filaments. As justified in section II and appendix D, the action of individual grafted filaments pressing
on a second obstacle can be treated adiabatically.
[23] a semiflexible WLC of contour length Lc tethered
normally to a rigid surface at x = 0 and impinging onto another rigid surface parallel to the first one at x = L was
investigated in the weak bending regime (L (cid:46) Lc). The average force exerted by the filament can be expressed as
In ref.
where the prefactor fb(Lc, (cid:96)p), the Euler buckling force, at fixed bending modulus kBT (cid:96)p, goes like L−2
c
¯f (L, Lc, (cid:96)p) = fb(Lc, (cid:96)p) f (η)
(F1)
(F2)
and f (η) is a universal function of the reduced compression η = (cid:2)(Lc − L)(cid:96)p/L2
fb(Lc, (cid:96)p) =
L2
c
(cid:3) ≥ 0 (L ≤ Lc). In ref.
[23] the
following expression of f (η) has been obtained for a continuous WLC in the small compression regime and tested
against Monte Carlo simulations of a d-WLC model
c
kBT (cid:96)p
π2
4
f (η) = − 4
π2
∂ ln Z(η)
∂η
Z(η) = 2
∞(cid:88)
k=1
(−1)k+1 e−λ2
kη
λk
λk = (2k − 1)
,
π
2
(F3)
(F4)
where as usual Lc = (j − 1)d with d the bond length and j the number of articulation points in the chain. It was
shown that f (η) rapidly grows from f (0) = 0 to the plateau value of 1 reached at η (cid:39) 0.25 and remains at this unitary
value in a wide range of η values. The behavior of continuous and discrete WLC models are in very good agreement
provided L (cid:29) d and Lc (cid:29) d in the d-WLC.
Note that η is proportional to (cid:96)p hence the more rigid the filament, the larger η at given Lc − L. In the case of
actin ((cid:96)p = 5730d), η (cid:39) 0.25 (or γ/Lc (cid:39) 5 × 10−5 d−1) corresponds to a compression Lc − L (cid:39) 0.1d for typical values
of Lc (cid:39) 50. Hence the force exerted by a compressed actin filament is almost always equal to the L−independent
Euler buckling force. Moreover for any polymerizing filament whose tip in its straight configuration is at a distance
less than 0.9d from the obstacle before the chemical step, the force will jump from 0 to fb as the result to the addition
of one bond.
In reference [29] the compressional force exerted by a d-WLC filament upon a rigid wall has been studied by Monte
Carlo simulations for values of (Lc − L) well beyond the weak bending limit. The dimensionless force f (η) was
observed to deviate from its plateau value of 1 at a η value inversely proportional to the chain contour length Lc. For
increasing η the force rapidly increases and it is expected to become independent of Lc because the chain is strongly
bent against the obstacle and only the fraction of the filament of length ∼ πL/2 (cid:28) Lc closer to its tethering point
remains under compression, while the remaining part is free to lay parallel to the obstacle wall (or better to undergo
equilibrium fluctuations without crossing the obstacle wall).
Introducing another reduced compression(cid:101)γ = (Lc − L)/Lc = ηLc/(cid:96)p plotting the large compression data for chains
of different Lc (but same (cid:96)p = 5370d) as a function of (cid:101)γ, they collapse on the same curve as shown in figure 2.
This large compression behavior departs from the unitary plateau value at(cid:101)γ (cid:39) 0.1. In the figure we also report the
following fitting function which reproduces reasonably well this unique behavior
f ((cid:101)γ) =
(a + b(cid:101)γ2)
(1 −(cid:101)γ)2 + 1
(cid:101)γ ≥ 0.1
with a = 0.044(5), b = 0.28(1). These values of the parameters have been obtained by fitting simultaneously the three
sets of data in the figure.
The full expression of the adiabatic force exerted by a d-WLC of contour length Lc upon a rigid wall at distance L
from its seed is given by:
¯f (L, Lc, (cid:96)p) = fb(Lc, (cid:96)p) ×
f
(cid:18)(cid:101)γ
(cid:19)
(a + b(cid:101)γ2)
(1 −(cid:101)γ)2 + 1
(cid:96)p
Lc
(cid:101)γ < 0.1
(cid:101)γ ≥ 0.1
The large contour length expression -- second line of eq. (F6) -- leads, in the infinite Lc limit (L/Lc → 0,(cid:101)γ → 1),
to the following asymptotic expression:
24
(F5)
(F6)
(F8)
lim
Lc→∞
¯f (L, Lc, (cid:96)p) =
π2
4
kBT (cid:96)p
a + b
L2
(F7)
For active filaments in supercritical conditions, this regime is attained rapidly after the filament escapes laterally.
Indeed the propensity to polymerize at the tip of the filament and the loss of contact between the filament tip and
the obstacle will rapidly result in an unlimited contour length growth to reach the asymptotic regime. In this regime
the force exerted by the filament is related essentially to the elastic response due to pure bending and not to chemical
activity at the tip of the filament. In an optical trap apparatus of strength κT , the mechanical equilibrium condition
( ¯f = κT L) in this asymptotic regime imposes
(cid:18) π2
(cid:19)1/3
Lesc =
kBT (cid:96)p
a + b
κT
4
For a bundle of Nf escaped filaments this value must be multiplied by N 1/3
f
.
Lesc can be compared to Ls = kBT ln(ρ1)/(dκT ) a rather good approximation for the average width of the trap in
the non escaping regime [7]. The condition Lesc = Ls for a single filament sets a threshold ¯κT for the trap strength
to switch from one regime to the other:
¯κ2
T =
4
π2
[ln(ρ1)]3
(a + b)
(kBT )2
(cid:96)pd3
(F9)
For κT > ¯κT the filament will be non-escaping while for κT < ¯κT the filament has high probability to escape and
to end up in the asymptotic regime. Using the values of the parameters a and b given above, the actin persistence
length and ρ1 = 2.5 with kBT = 1 and d = 1 we get: ¯κT = 0.01337. This value is indeed compatible with the value
used in ref. [7] in the non-escaping regime. For a bundle of Nf filaments the threshold value is multiplied by Nf . For
Nf = 32 we then obtain ¯κT = 0.4284 compatible with the results shown in figure 4 of the main paper.
Eq. (9) fully characterize the force-compression law of the d-WLC model. Using this relation we can study the
behavior of our "apparatus" (bundle+optical trap) for any number of filaments Nf and any trap strength κT , in
particular we can mimic the conditions of the experiments reported in ref [11].
25
[1] J. Zimmermann, C Brunner, M Enculescu, M Goegler, A Ehrlicher, J. Kas, and M. Falcke. Actin filament elasticity and
retrograde flow shape the force-velocity relation of motile cells. Biophysical Journal, 102:287, 2012.
[2] A. Mogilner and B. Rubinstein. The physics of filopodial protrusion. Bioph. J., 89:782, 2005.
[3] Y. Lan and G. A. Papoian. The stochastic dynamics of filopodial growth. Biophysical Journal, 94:3839, 2008.
[4] U. Dobramysl and R. Papoian, G. A.and Erban. Steric effects induce geometric remodeling of actin bundles in filopodia.
Biophysical Journal, 110(9):2066, 2016.
[5] E. Altigan, D. Wirtz, and S.X. Sun. Mechanics and dynamics of actin-driven thin membrane protusions. Bioph. Journal,
90:65, 2006.
[6] R. K. Sadhu and S. Chatterjee. Actin filaments growing against a barrier with fluctuating shape. Phys. Rev. E, 93:062414,
2016.
[7] A. Perilli, C. Pierleoni, G. Ciccotti, and J.-P. Ryckaert. On the properties of a bundle of flexible actin filaments in an
optical trap. Journal of Chemical Physics, 144(24):245102, 2016.
[8] T. L. Hill. Microfilament or microtubule assembly or disassembly against a force. Proceedings of the National Academy of
Sciences of the United States of America, 78(9):5613, 1981.
[9] N. J. Burroughs and D. Marenduzzo. Three-dimensional dynamic monte carlo simulations of elastic actin-like ratchets. J.
Chem. Phys., 123:174908, 2005.
[10] T. Bornschlogl, S. Romero, C. L. Vestergaard, J.F. Joanny, G. Tran Van Nhieu, and P. Bassereau. Filopodial retraction
force is generated by cortical actin dynamics and controlled by reversible tethering at the tip. PNAS, 110(47):18928 -- 18933,
2013.
[11] M. J. Footer, J. W. J. Kerssemakers, J. A. Theriot, and M. Dogterom. Direct measurement of force generation by actin
filament polymerization using an optical trap. Proc. Natl. Acad. Sci. USA, 104:2181, 2007.
[12] D. D´emoulin, M. F. Carlier, J. Bibette, and J. Baudry. Power transduction of actin filaments ratcheting in vitro against
a load. Proc. Natl. Acad. Sci., 111:17845, 2014.
[13] C. S. Peskin, G. M. Odell, and G. F. Oster. Cellular motions and thermal fluctuations: the brownian ratchet. Biophys. J,
65(1):316, 1993.
[14] G. Sander van Doorn, C. Tanase, B. M. Mulder, and M. Dogterom. On the stall force for growing microtubules. European
Biophysical Journal, 29(1):2, 2000.
[15] K. Tsekouras, D. Lacoste, K. Mallick, and Joanny J.-F. Condensation of actin filaments pushing against a barrier. New
J. Phys., 13:103032, 2011.
[16] A. Mogilner and G. Oster. The polymerization ratchet model explains the force-velocity relation for growing microtubules.
Eur. Biophys. J, 28(3):235, 1999.
[17] J. Krawczyk and J. Kierfeld. Stall force of polymerizing microtubules and filament bundles. Europhys. Lett., 93:28006,
2011.
[18] X. Li and A.B. Kolomeisky. The role of multifilament structures and lateral interactions in dynamics of cytoskeleton
proteins and assemblies. J. Phys. Chem. B, 119:4653, 2015.
[19] A. Mogilner and G. Oster. Cell motility driven by actin polymerization. Biophys. J, 71(6):3030, 1996.
[20] N. J. Burroughs and D. Marenduzzo. Growth of a semi-flexible polymer close to a fluctuating obstacle: application to
cytoskeletal actin fibres and testing of ratchet models. J. Phys.: Condens. Matter, 18(14):S357, 2006.
[21] L. Blanchoin, R. Boujemaa-Paterski, C. Sykes, and J. Plastino. Actin dynamics, architecture, and mechanics in cell
motility. Physiol Rev, 94:235, 2014.
[22] A. P. Liu, D. L. Richmond, L. Maibaum, S. Pronk, P.L. Geissler, and Fletcher D.A. Membrane-induced bundling of actin
filaments. Nature Physics, 4:789, 2008.
[23] A. Gholami, J. Wilhelm, and E. Frey. Entropic forces generated by grafted semi-flexible polymers. Phys.Rev.E, 74:041803,
2006.
[24] A. Perilli, C. Pierleoni, G. Ciccotti, and J.-P. Ryckaert. On the force-velocity relationship of a bundle of rigid bio-filaments.
Journal of Chemical Physics, 148:095101, 2018.
[25] J. Howard. Mechanics of Motor Proteins and the Cytoskeleton. Sinauer, Sunderland, MA, 2001.
[26] E. Gerritsma and P. Gaspard. Chemomechanical coupling and stochastic thermodynamics of the f1-atpase molecular motor
with an applied external torque. Biophys. Rev. Lett., 5:163, 2010.
[27] P. Gaspard and E. Gerritsma. The stochastic chemomechanics of the f1-atpase molecular motor. J. Theor. Biol, 247:672,
2007.
[28] A.E. Carlsson. Model of reduction of actin polymerization forces by atp hydrolysis. Phys. Biology, 5:036002, 2008.
[29] C. Pierleoni, G. Ciccotti, and J.-P. Ryckaert. A semi-flexible model prediction for the polymerization force exerted by a
living f-actin filament on a fixed wall. J. Chem. Phys., 143:145101, 2015.
[30] T. E. Schaus and G. G. Borisy. Performance of a population of independent filaments in lamellipodial protrusion. Bio-
physical Journal, 95(3):1393, 2008.
[31] R. Wang and A. E. Carlsson. Load sharing in the growth of bundled biopolymers. New Journal of Physics, 16:113047,
2014.
[32] A. Mogilner and G. Oster. The physics of lamellipodial protrusion. Eur. Biophys. J., 25:47 -- 53, 1996.
[33] A. Mogilner and G. Oster. Force generation by actin polymerization ii: The elastic ratchet and tethered filaments. Biophys.
J., 84:1591, 2003.
[34] K.M. Taute, F. Pampaloni, E. Frey, and E.L. Florin. Microtubule dynamics depart from the wormlike chain model. Phys.
Rev. Lett., 100(2):028102, 2008.
[35] R. Glotter, K Kroy, E. Frey, M. Barmann, and E. Sackmann. Dynamic light scattering from semidilute actin solutions: A
study of hydrodynamic screening, filament bending stiffness, and the effect of tropomyosin/troponin-binding. Macromol.,
29(1):30, 1996.
26
|
1807.08990 | 1 | 1807 | 2018-07-24T09:37:45 | Mechanical and Systems Biology of Cancer | [
"physics.bio-ph",
"q-bio.CB"
] | Mechanics and biochemical signaling are both often deregulated in cancer, leading to cancer cell phenotypes that exhibit increased invasiveness, proliferation, and survival. The dynamics and interactions of cytoskeletal components control basic mechanical properties, such as cell tension, stiffness, and engagement with the extracellular environment, which can lead to extracellular matrix remodeling. Intracellular mechanics can alter signaling and transcription factors, impacting cell decision making. Additionally, signaling from soluble and mechanical factors in the extracellular environment, such as substrate stiffness and ligand density, can modulate cytoskeletal dynamics. Computational models closely integrated with experimental support, incorporating cancer-specific parameters, can provide quantitative assessments and serve as predictive tools toward dissecting the feedback between signaling and mechanics and across multiple scales and domains in tumor progression. | physics.bio-ph | physics | Mechanical and Systems Biology of Cancer
Fabian Spill1+, Chris Bakal2, Michael Mak3+*
1 School of Mathematics, University of Birmingham, Birmingham, B15 2TT, UK
2 Division of Cancer Biology, The Institute of Cancer Research, London, SW3 6JB, UK
3 Department of Biomedical Engineering, Yale University, New Haven, USA
+ equal contribution
* correspondence: [email protected]
Key Words: mechanobiology; cancer; mathematical biology; computational modeling; signaling;
cytoskeleton; focal adhesions; mechanotransduction
Manuscript Information
Number of Figures: 3
Abstract
Mechanics and biochemical signaling are both often deregulated in cancer, leading to cancer
cell phenotypes that exhibit increased invasiveness, proliferation, and survival. The dynamics and
interactions of cytoskeletal components control basic mechanical properties, such as cell tension,
stiffness, and engagement with the extracellular environment, which can lead to extracellular matrix
remodeling. Intracellular mechanics can alter signaling and transcription factors, impacting cell
decision making. Additionally, signaling from soluble and mechanical factors in the extracellular
environment, such as substrate stiffness and ligand density, can modulate cytoskeletal dynamics.
Computational models closely integrated with experimental support, incorporating cancer-specific
parameters, can provide quantitative assessments and serve as predictive tools toward dissecting the
feedback between signaling and mechanics and across multiple scales and domains in tumor
progression.
Introduction
The mechanical microenvironment in cancer is vastly altered compared to healthy tissue.
Typically, the extracellular matrix (ECM) is stiffened in the tumor microenvironment (1-3), but
individual cancer cells may actually be softer (4). There is a bimodal distribution of nanomechanical
stiffness across advanced cancer tissues (5). Moreover, more complex mechanical and geometric
characteristics, including the fibrous matrix structure, porosity, or viscoelastic parameters may be
changed in tumors (6, 7). Similarly, solid and fluid stresses are greatly altered in cancers (8). It is well
known that cancers exhibit increased fluid pressures, in part due to remodeling of the vasculature and
lymphatics (9).
The altered ECM stiffness and geometry of the tumor microenvironment are sensed by tumor
cells via mechanosensing structures, which can activate intracellular signaling pathways that drive
behaviors such as unrestrained proliferation, increased survival, tissue invasion, stemness, and drug
resistance (10-12). While cancer has been traditionally considered a genetic disease, alterations in
ECM stiffness and geometry can force normal cells to adopt phenotypes characteristic of transformed
and/or metastatic cells in the absence of any genetic change (13, 14). Theoretical work suggests that
environmental cues, coupled with various possible oncogenic alterations (e.g. overexpression of c-Src
(15)), can drive cancer progression (16, 17). Cancer progression can be promoted by genetic changes
that alter how cells respond to ECM stiffness and geometry and that enable cancer cells to remodel
their environment in ways that promote disease.
To open new therapeutic avenues that seek to manipulate the response of cancer cells to their
environment as a way to treat cancer, predictive mathematical models are required to describe how
cell fate decisions are due to interactions between tumor cells and their ECM and how these
interactions differ between normal and cancer cells. The problem is inherently multiscale in nature
and involves diverse components such as biochemical reactions, cell-matrix and cell-cell interactions,
and tissue-level alterations. The field of mechanotransduction has long embraced modelling tools in
order to describe how cells respond to mechanical and geometric cues, and these models serve as key
starting points for more complex descriptions of how cancer cells interact with their ECM. For
example, models have been developed that provide insights into diverse aspects of mechanobiology
including: force-dependent molecular bonds (18-21), spatiotemporal organization of intracellular
molecules (22-24), impact of cell shape (25-29), and the dynamics of the cytoskeleton (30-32). Here
we review some of these models and supporting experimental findings with a look toward the future.
We first review recent work on cytoskeletal interactions that modulate intracellular mechanics and
the propagation of cytoskeletal forces inside and outside the cell. Next we focus on the cell-matrix
adhesion complexes that act as key signal transducers and mechanosensors. Finally, we review key
signaling networks implicated in mechanotransduction.
Generation and propagation of intracellular forces
The active actin cytoskeleton provides basic structure and force generation capabilities. The
key components include actin filaments, actin crosslinking proteins (ACPs) such as alpha-actinin and
filamin, and myosin II motors that generate contractility. Inside the cell, a large network of these
components undergoes dynamic and stochastic interactions, spontaneously resulting in pattern
formation -- including the actin cortex at the cell periphery, thick contractile bundles of actin (stress
fibers), and cell polarity (leading and trailing edges). Local interactions and kinetics can control overall,
global functionality of the cytoskeletal network. In particular, actin turnover rates can modulate
cytoskeletal network tension, and the interplay between actin turnover, actin crosslinking, and myosin
II walking activity can regulate the morphological state of the network, from homogeneous
morphologies to local clusters (Fig. 1a) (30). Computational simulations can isolate individual features
and determine their roles in cytoskeletal network behavior. For example, altering actin nucleation
rates can modulate the stress fluctuation magnitudes in the cytoskeleton, a phenotype observed in
intracellular microrheology experiments that modulate epidermal growth factor (EGF) signaling
(known to influence actin nucleation) in breast cancer cells (33). Additionally, spatial and temporal
profiles are important in regulating cell behavior. These can be precisely tuned in computational
models. For example, cell geometry and dimensionality influence the anisotropy and amplitude of
intracellular stress fluctuations (34). While overall cell tensions have an intuitive role of enabling cells
to apply forces onto their substrate (e.g. the ECM) and migrate, intracellular stress fluctuations can
facilitate the redistribution of organelles and molecular components inside the crowded cytoplasmic
space (35). Furthermore, malignant tumor cells appear to exhibit larger intracellular displacement
and stress fluctuations compared to benign counterparts, as shown by experiments measuring
intracellular stiffness and force fluctuations (35). Cytoskeletal mechanics and fluctuations are the
result of the interactions between many cytoskeletal components, each undergoing dynamic
processes (turnover, walking, binding, unbinding, etc.). Computational network models of the
cytoskeleton, based on physical principles (reaction kinetics, mechanics) and incorporating realistic,
experimentally tangible features, can help dissect the local, molecular-level contributions to
experimentally observable mechanical cellular phenotypes. High resolution experimental techniques,
e.g. super resolution imaging or atomic force microscopy, can help guide the development and
validation of models of fine and distinct cytoskeletal features (36). Furthermore, models coupling
cytoskeletal forces to critical intracellular and extracellular features, particularly the nucleus and the
ECM, can start to elucidate a more holistic picture of cell behavior.
Cytoskeletal forces can be transmitted to the cell nucleus via the LINC (Linker of
Nucleoskeleton and Cytoskeleton) complex (37). Substrate stiffness modulates cytoskeletal tension
and thus nuclear stress and shape, which interestingly also modulates the expression levels of a key
nucleoskeletal protein lamin A, nuclear stiffness, and stem cell differentiation (38). The mechanical
properties of the nucleus can also influence nuclear shape and dynamics during cell deformation and
invasion through confined spaces (e.g. ECM pores or endothelial junctions). Large nuclear
deformations can lead to rupture and DNA damage, as observed in experimental studies of cancer
cells invading through highly confined constrictions (39, 40). Computational models coupling cellular
forces to the nucleus can generate quantitative details of nuclear deformation and mechanical
remodeling during physiological processes and draw insights toward differences in nuclear behavior
due to biochemical or structural alterations. For example, experiments show that lamin A/C deficiency
leads to more plastic remodeling of the nucleus after larger strains, which can be captured in a
continuum model of the nucleus featuring a hyperelastic shell and a poroelasto-plastic core (Fig. 1b)
(41). Furthermore, the role of different types of lamins (A and B) in regulating nuclear shape and
geometry can be explored in continuum models through incorporating heterogeneous material
Figure 1: Computational models of cell mechanics. a) Brownian dynamics simulations of the active actin
cytoskeleton demonstrate cytoskeletal network evolution, a process dependent on the interplay between
actin turnover, network crosslinking, and myosin activity. The simulation domain is 3x3x3µm3 with periodic
boundary conditions. Actin filaments are teal, myosin II motors are red, and actin crosslinking proteins are
yellow. Adapted from (30). b) A finite elements model of the nucleus predicts stress profiles and plastic
remodeling after deformation through a confined barrier, e.g. endothelial junction. The model is composed
of a permeable hyperelastic shell surrounding a poroelastic-plastic core. The color bar indicates relative
stress levels. The green fluorescence experimental images (right) show nucleus morphologies at different
stages during the deformation process. Adapted from (41).
profiles. In particular, a preferred mesh size difference between lamin A and lamin B appears to
explain nuclear blebbing tendencies (42).
In many types of solid tumors, cancer cells are embedded in a dense fibrillar matrix.
Cytoskeletal forces are transmitted into the ECM via cell-matrix adhesions, which can lead to ECM
remodeling and propagate mechanical signals to surrounding cells (43). Stiffer substrates tend to
promote increased cell traction forces and lead to a more invasive phenotype (44, 45). Relaxation of
tension in the substrate in laser ablation experiments (46, 47) tends to revert cell invasiveness.
Moreover, ECM networks exhibit nonlinear strain stiffening (48), suggesting potential mechanical
feedback mechanisms. These phenomena have been demonstrated through a number of
experimental studies. Complementarily, computational models can provide quantitative, mechanistic
insights toward underlying driving factors of invasive behavior in 3D ECMs -- particularly to a level of
detail that may be unfeasible for experiments to achieve or parse out. Computationally intensive
models can capture a high degree of local details observed in high resolution experiments of cell-ECM
interactions. In a recent study, a model capturing an entire cell with dynamic protrusions inside a
surrounding ECM showed that dynamic filopodia can act as rigidity sensors that facilitate durotaxis in
HUVECs (Fig. 2a) (49). While stiffness sensing (and many other cell behaviors) is a phenomenon
exhibited by normal and cancer cells, cancer-related parameters can be tuned in generalizable models
to explore disease phenotypes. In particular, the above model showed that the number and length of
filopodia can modulate invasive behavior, supporting prior studies that showed that deregulation in
filopodia-related functions and pathways are implicated in cancer progression and metastasis (50). In
another model that incorporates dynamic local forces and force-sensitive ECM fiber-fiber crosslinks,
it is demonstrated that the coupling of mechanical forces and fiber-fiber biochemical kinetics can
result in ECM densification near the cell boundary, consistent with experiments in tumor and
endothelial cells (51). Furthermore, the fibrillar nature of the ECM and asymmetric contractility of
elongated cells can lead to long range anisotropic strain profiles in the environment due to fiber
realignment (Fig. 2b) (52-54), which can also generate spatial profiles of stiffness (48).
Tumors often grow as large multicellular masses in which cell-cell and tumor-ECM interactions as well
as environmental properties can dictate cancer progression. Computational models of collective
tumor invasion and evolution have been developed that aim to capture patterns observed in clinical
data (e.g. from histology or clinical databases). Mathematical, biophysical relationships (e.g. in the
form of partial differential equations for continuum features and/or rules and probabilities for discrete
features) can be used to govern the behavior and spatiotemporal profiles of tumor content, consisting
of a mixture of components (cells, ECM, fluid, concentration profiles of nutrients, chemokines, and
drugs, etc.) (55). For instance, the role of adhesions on invasion or growth was investigated in the
models (56-58) and the role of angiogenesis in tumor growth was modeled in (59). Some models are
able to capture overall tumor geometries seen in clinical data (60) as well as provide insights toward
complex factors influencing drug response (55). An important next step is the integration of multi-
physics tumor models with more realistic biophysical features in the tumor microenvironment and
Figure 2: Models of cell-matrix interactions. a) A 3D cell model with stress fibers, a nucleus, and filopodia
captures mechanical cell-matrix interactions and indicates a potential role for filopodia in stiffness sensing.
Filopodia can protrude, adhere to ECM fibers, and contract, pulling fibers and sensing their stiffness. The
cell body will tend to polarize more toward stiffer regions. Adapted from (49). b) An ECM fiber model
shows network strain distribution (bottom) as a function of cell contraction anisotropy (top). More spindle-
like cells, which tend to contract more along one axis, can generate farther reaching anisotropic strain fields
in the fibrillar ECM network. Adapted from (54). c) Multiple motor-clutch components are used to model a
cell migrating on a deformable substrate, predicting an optimal substrate stiffness to maximize cell migration
speed. Clutches bind and unbind in a force-dependent manner from the substrate, and molecular motors
retract F-actin which is connected to the clutches, thus pulling the substrate. Adapted from (79).
associated signal transduction networks and signaling mechanisms. We refer the reader to (61, 62) for
recent reviews specifically focusing on multicellularity and tumor modelling.
The physical environment surrounding solid tumor cells is dynamic and heterogeneous,
influenced by the presence of cancer and stromal cells (63). Precise spatial and temporal physical
profiles (of stiffness, architecture, ligand density, etc.) of this environment can be explored through
mechano-chemical models that interface active cells with a responsive, physiologically mimicking
ECM. These profiles in turn can act as signals that cells can sense through complex mechanisms
mediated by adhesion complexes. Elucidating detailed signal transduction effects then requires
models that couple mechanics with biochemical signaling networks.
Focal adhesion dynamics, mechanosensing, and signaling
Focal adhesions (FA) are multifunctional organelles that serve as primary points of sensing of
ECM stiffness and geometry by cells (64). FAs are much more than passive receptors, but rather are
dynamical systems comprised of complex interactions between the ECM, the cytoskeleton, and signal
transduction machinery across multiple spatiotemporal scales. The emergent behavior of these
systems underpins the generation, transmission, and coordination of diverse forces, changes in cell
shape, and cell fate determination, including the acquisition of malignant and therapy-resistant
phenotypes (65).
At the heart of FAs are membrane bound integrins (Fig. 3b), (66). As cells interact with their
ECM, for example via protrusions generated by actin polymerization at the leading edge, the binding
of individual integrin molecules to the ECM initiates -- integrin clustering; the activation of FAK; the
Figure 3: Mechanosensing and mechano-regulating pathways. a) Different cytoskeletal compartments are
regulated by different pathways, compete with each other, and affect different downstream effectors.
Cdc42 and Rac mediate different actin-driven protrusions, whereas Rho activates actin stress fibers.
Different cytoskeletal compartments (e.g. cortex, stress fibers, protrusions) may compete for the global
actin pool. Therefore, strengthening one compartment may weaken others. Moreover, crosstalk of
regulators, e.g. through PAKs, may directly lead to competition of these compartments. Focal adhesions
(FAs) are complex structures linking the cytoskeleton and the surrounding extracellular matrix. Some of the
complexity are shown in (b). b) Integrins provide a direct attachment from inside the cell to the outside
matrix. Inside, they link to the cytoskeleton through proteins including talin and vinculin that are force-
sensitive. Consequently, forces may affect chemical reactions and thus adhesion assembly and disassembly.
c) Focal adhesion mechanosensing will then lead to downstream effects on phenotypes, mediated, for
instance, through YAP/TAZ. Focal adhesion kinase (FAK) is shown to shift the stiffness response of YAP/TAZ.
Adapted from (110). d) Cell shape may also affect intracellular signaling directly. A model shows that the
round cell adapts cytoskeletal regulators (here Rac) in the direction of the external stimulus (e.g.
chemotactic gradient), whereas the ellipsoidal cell polarizes in a direction in between its longest axis and
the stimulus. Adapted from (28).
subsequent recruitment of the proteins such as paxillin, talin, and vinculin; and the formation of
nascent FAs. Talin and vinculin also bind actin in branched networks that are actively flowing over
adhesions, which results in the transmission of force to the ECM, allowing cells to probe the stiffness
of the ECM and generate traction (67-69). However, because engagement of actin by FAs prevents
polymerized actin from generating further protrusions, in order to maintain protrusiveness, FAs
transiently disengage from the actin network, allowing polymerized actin to slide by FAs and continue
to push on the leading edge (70). This ability to engage and disengage actin networks by FAs has been
termed the actin-FA "clutch". Increased clustering, further recruitment of molecules which couple FAs
to actin, and post-translational events can lead to the maturation of FAs at sites distal from the leading
edge. As adhesions mature, the nature of the actin organization at the FAs also differs, as stress fibers
predominate on more mature adhesions; which can propagate relatively large forces throughout the
cell body, leading to large morphological changes. During the formation of both nascent and mature
FAs, both mechanical and biochemical processes occur which will ultimately trigger their turnover,
and thus from a systems-perspective, FA dynamics involve both extensive positive feedback loops (i.e.
initial activating events such as integrin clustering become amplified), and negative feedback loops
(i.e. FA formation leads to force generation and upregulates signals that will ultimately induce FA
turnover).
Iteration between modelling and experimentation has yielded deep understanding of FA
dynamics. The FA clutch concept was first described theoretically (71), which was followed by
experimental observations of actin flow and FAs in living cells, (70, 72, 73). These works then led to
the development of a stochastic model of the clutch (74) that predicts the existence of a regime of
fast retrograde flow with low traction forces and one with slow retrograde flow and high traction
forces, with extracellular stiffness acting as the switch between the regimes. More recent models have
been developed which incorporate dynamic interactions between the individual FA components talin
and vinculin (75) and which incorporate information regarding the spatial distribution of individual
ECM-integrin-clutches (76). Furthermore, mechanochemical models that include feedback between
adhesion assembly and substrate rigidity demonstrate different possible regimes of focal adhesion
evolution, from nascent, unstable adhesions to stable adhesions with a steady-state size (77), and the
interplay of ECM stiffness, remodeling, and ligand density can influence focal adhesion size and growth
(78). A recent model with multiple motor-clutch modules in a cell-scaled geometry demonstrated the
number of motor-clutch constructs in the cell influences the optimal substrate stiffness for maximal
cell migration, supported by experiments in glioma cells (Fig. 2c) (79). Taken together, these and other
studies are providing the first types of multi-scale models that explain how cellular phenotypes
emerge from the dynamical interplay between FAs and the ECM.
Importantly, signaling events play both a key role in regulating short term FA dynamics
(seconds-minutes) and in regulating the organization of the actin cytoskeleton around FAs. A
simplified schematic linking important signaling pathways with cytoskeletal features is illustrated in
Fig. 3. In particular, signaling via Rho GTPases such as Rac1, Cdc42, and RhoA play essential roles
regulating the relationship between FAs and actin (Fig. 3a), (80). These roles have been particularly
well studied in the context of migrating cells. For instance, chemotactic signals may lead to the
activation of Rho GTPases that are known to regulate the cytoskeleton in different ways (81, 82).
Moreover, forces may also activate Rho, through ROCK and myosin, leading to further forces pulling
on focal adhesions and thus their reinforcement (83). Mechanical signals such as stiffness gradients
consequently activate mechanoregulatory pathways, resulting, for instance, in durotaxis, i.e. the
migration along stiffness gradients (84). Mathematical models were developed linking adhesion
dynamics and durotaxis and showed that cell velocity depends on stiffness in a non-monotonic way,
with a maximum at an intermediate stiffness (85). Collectively migrating cells may also durotax due to
cells deforming the substrate more in the low stiffness regions (86), in line with experimental
observations (87). Other models focus on the intricate details of the interplay of stress fibers and focal
adhesion dynamics. For instance, the interplay of stress fibers and adhesion bonds was investigated
(88, 89), and it was found that cyclic stretch may induce cell reorientation through reduction of the
catch bond lifetimes of focal adhesions (90).
Cell shape, which emerges from the spatiotemporal dynamics of FA generation and turnover,
can also impact both FA and cytoskeletal dynamics and signaling events (Fig. 3d), (25, 91). Cell shape
can be considered a geometric cue and, like the spatial organization of the ECM-integrin complexes
themselves (92), is important to consider in order to understand how cells respond to mechanical cues
in the environment. Because cell morphological dynamics are dysregulated in many cancer types,
mechanosensing by FAs may be affected, which can be explored through coupling cell geometries with
cytoskeletal and FA kinetics in models. Cell-scaled mechanochemical models that include spatial
profiles of tensional components inside the cell, particularly stress fibers, and substrate adhesions can
reproduce cell shapes and stress distributions comparable to experimental studies (89, 93, 94).
Signaling downstream of mechanical stimuli and feedback on mechanics
FAs serve as a platform for the assembly of large signaling complexes which regulate a host of
downstream processes, particularly transcription. Transcriptional changes can influence cell-wide
behaviors over much
longer terms (hours-days); or even have permanent consequences
(differentiation). These longer-term changes in cell state can modulate short term FA dynamics.
A classic example of how short term FA dynamics, in response to ECM properties (mechanical
cues) and geometric cues (ECM organization and cell shape), drives long term changes in cell fate is
how entry into the cell cycle by adherent cells is dependent on cell spreading. Observations by
Dulbecco and Folkman provided the first evidence of the link between mechanosensing and
proliferation (95-97). Following work demonstrated that spreading upregulated ERK and RhoA activity
which upregulates the transcription of pro-proliferative factors CyclinD1 and downregulates pro-
quiescence factors such as p21 and p27 (98, 99). Although the role of adherence in driving proliferation
is not fully understood in cancer cells, recent studies, demonstrating that FAK is a key mediator of
resistance to inhibitors of ERK activators, strongly hint FA mediated activation of ERK is an important
driver of tumorigenesis (100, 101).
More recent work has shown that the YAP and TAZ transcriptional co-activators are also
effectors of signaling complexes formed at FAs (75, 91, 102). Regulation of transcriptional events by
YAP/TAZ is involved in a broad number of cellular behaviors that are essential drivers of tumorigenesis
and metastasis, including proliferation, maintenance of stemness, and migration (103-105).
Intriguingly, the mechanisms by which YAP/TAZ is activated appear to be highly dependent on the
type of adhesion. At very early adhesions, YAP/TAZ is activated largely by focal adhesion kinase (FAK)
through PI3K and/or mTOR (106-108). At nascent adhesions, FAK remains important for YAP/TAZ
activation, and this activation appears to rely on signaling via the ARHFGEF7/beta-Pix Rho GTP
Exchange Factor (RhoGEF) which activates Rac1 and Cdc42 (91). As adhesions mature, FAK activity
becomes dispensable, but now adhesions rely on ARHGEF7 and RhoA GTPase (91). Importantly,
YAP/TAZ activity can regulate FA dynamics by altering the levels of different FA components (109).
Although it remains to be formally shown, these observations suggest that by having different types
of YAP dynamics downstream of different adhesion types, cells can tune transcriptional events to
match ECM stiffness and geometry. Critically, it has been shown that these systems that couple FA
dynamics to YAP activation are often highly altered in cancer cells, emphasizing that cancer cells have
evolved mechanisms such that fate determination decisions differ compared to normal cells in
response to the same mechanical and geometric cues (91).
Recent modelling work has provided insight into the pathways linking mechanical cues to
transcription, and how these may differ in cancer cells. A model of YAP/TAZ mechanosensing showed
that YAP/TAZ increases in a switch-like manner with stiffness, and the location and plateau value of
YAP/TAZ concentrations can be critically affected by the molecular state of the cell (Fig. 3c) (110). FAK
is predicted to shift the location of the switch in the YAP/TAZ stiffness-response curve, whereas mDia
is predicted to shift the YAP/TAZ plateau level at high stiffness. However, as discussed above, the role
of FAK on YAP/TAZ signaling is complex, and coupling a model of YAP/TAZ regulation to one
investigating the intricate details of FAK mechanosensing, as done in (111, 112) may provide insights
toward some of these complexities.
Some mechanical effects may also underlie the behavior of multiple pathways. For instance,
MRTF's are also sensitive to mechanosensing pathways, mainly through the effect of these pathways
on the actin pool (113). In this way, there is some indirect overlap with YAP/TAZ signaling, but there
are also direct crosstalks as discussed in (114). Further, matrix stiffness was shown to be sensed by
TWIST1-G3BP2, which subsequently initiates epithelial-mesenchymal transitions (115). Matrix
stiffness may also change the structure of the nucleus, in part through the coupling of the dynamics
of myosin motors and lamin A, as investigated in the model in (116). This model highlights that such
effects arise through general mechanisms: stable mechanosensitive gene expression may arise if a
structural protein positively regulates its own gene expression while stresses inhibit the degradation
of that protein. Lamin A is thus involved in mechanosensing in general and thus also affects YAP/TAZ.
The typical softness of nuclei in cancer (117) may thus play a role in the altered YAP/TAZ signaling that
contributes to increased malignancy (103). Recent work also showed that a direct coupling of forces
through the cytoskeleton from focal adhesions to the nucleus is involved in YAP/TAZ nuclear
translocation (118).
Mathematical models have also been used to predict how cell shape influences signaling
dynamics. For instance, Rho GTPases activate primarily on the plasma membrane, so that shape
changes will affect the effective activation rates of these GTPases as well as subsequent downstream
effects (28, 29). This, for instance, implies that cells in an identical chemical state but with different
shapes may react differently to chemotactic signals (Fig. 3d) (28). Moreover, since cell shape affects
cytoskeletal regulators, changing shape is expected to induce feedback on shape regulation. Similarly,
modeling revealed that the cAMP/PKA/B-Raf/MAPK1,2 network in neurons is controlled by cell shape
(27), making cell shape a physical variable used to store biological information (119). Given the
enormous heterogeneity of cellular shapes in tumors, it is thus likely that these shapes also directly
contribute to the dynamics of intracellular signaling pathways and thus the heterogeneity of cell
phenotypes in cancer.
Conclusions and Outlook
Mathematical models with realistic mechanical and biochemical features have revealed
underlying mechanisms and predictive insights toward how cytoskeletal components coordinate
dynamically to lead to physical behaviors (migration, shape, force generation) of interest in the field
of cancer biophysics. Moreover, models that directly incorporate experimentally observable or
controllable features, such as dynamic adhesions, actin turnover, motors activity, and signaling, can
facilitate the validation of model predictions. Further, mathematical models help provide insights
toward a variety of phenomena across multiple scales, from how forces affect molecular binding rates
to how tissue level stresses impact tumor progression. While many models are complex and may be
computationally expensive to simulate, advances in modeling techniques, computational algorithms,
and higher performance computing will enable the development of multiscale, multiphysics models
that can provide an integrated picture of the various scales and features of cancer.
A key area of opportunity is the integration of models consisting of complex mechanics and
biochemical signaling networks, including feedback mechanisms. Biochemical signaling networks
typically have many interacting components with feedback between many pathways. Quantifying the
mathematical nature of pathways that lead to cytoskeletal responses and pathways that respond to
mechanotransduction can facilitate their coupling to biophysical and biomechanical models. These
models can entail complex physical relationships (e.g. non-linear stiffening, viscoelasticity and
plasticity) that govern discrete cytoskeletal and extracellular components (e.g. protein fibers) or the
continuum representations of large quantities of these components. Novel mechanical features to
incorporate that play important roles include the turnover kinetics of cytoskeletal components and
the active remodeling of the cytoskeleton and ECM by molecular motors and cells. Different
timescales should also be considered, from short term (minutes to hours) mechanosensing responses
that lead to altered cell morphologies and cell migration directionality to long term (hours to days)
mechanotransduction that leads to altered gene expression and cell fates -- the mechanistic principles
underlying these phenomena are not fully understood. Moreover, macroscopic tumor growth and
remodeling of the ECM and metastasis may occur on even longer timescales (months to years), leading
to tissue level changes of mechanics. Mathematical models aimed at understanding the interplay of
mechanical processes at these vastly different time scales can help link information obtained from
experiments at the molecular or cellular scale with in vivo or clinical observations of the long-term
evolution of tumors. Coupled mechanical and systems biology models can ultimately facilitate the
design of therapeutic strategies aimed toward modulating cancer phenotypes with known biophysical
features, such as migratory plasticity, remodeled ECMs, and metastasis. The intersection between
signaling and mechanics can provide new treatment methods against cancer, such as inhibiting or
desensitizing the link between external mechanical cues (e.g. ECM stiffness) and the affected signals
that drive cancer invasion and transformation (e.g. YAP/TAZ-linked pathways, integrin-mediated
signals) or suppressing the pathways that lead to aggressive remodeling of the ECM. Less obvious
strategies, such as targeting actin turnover to modulate cell force generation, can also be elucidated
by models to guide more subtle methods toward reverting invasive phenotypes. While many drugs,
such as classical microtubule-targeting chemotherapeutics or targeted inhibitors of integrins or FAK,
have a clear impact on cell mechanics, little is known about the systems level effects of such drugs in
dependence on the physical microenvironment of cancer. The merger of mechanical and systems
biology will lead to predictive models that can help devise treatment strategies to overcome the
adverse effects of mechanics on tumor progression and therapeutic resistance.
Acknowledgements
M.M. acknowledges funding from Yale University. The content is solely the responsibility of the
authors.
References
Plodinec M, Loparic M, Monnier Ca, Obermann EC, Zanetti-Dallenbach R, Oertle P, et al. The
Spill F, Reynolds DS, Kamm RD, Zaman MH. Impact of the physical microenvironment on tumor
Nia HT, Liu H, Seano G, Datta M, Jones D, Rahbari N, et al. Solid stress and elastic energy as
1.
Schedin P, Keely PJ. Mammary gland ECM remodeling, stiffness, and mechanosignaling in
normal development and tumor progression. Cold Spring Harb Perspect Biol. 2011 Jan;3(1):a003228.
2.
Provenzano PP, Inman DR, Eliceiri KW, Knittel JG, Yan L, Rueden CT, et al. Collagen density
promotes mammary tumor initiation and progression. BMC medicine. 2008;6:11-.
3.
Provenzano PP, Eliceiri KW, Campbell JM, Inman DR, White JG, Keely PJ. Collagen
reorganization at the tumor-stromal interface facilitates local invasion. BMC medicine. 2006;4(1):38-.
4.
Guck J, Schinkinger S, Lincoln B, Wottawah F, Ebert S, Romeyke M, et al. Optical deformability
as an inherent cell marker for testing malignant transformation and metastatic competence.
Biophysical Journal. 2005;88(5):3689-98.
5.
nanomechanical signature of breast cancer. Nature Nanotechnology. 2012 Nov;7(11):757-65.
6.
Wolf K, Te Lindert M, Krause M, Alexander S, Te Riet J, Willis AL, et al. Physical limits of cell
migration: control by ECM space and nuclear deformation and tuning by proteolysis and traction force.
J Cell Biol. 2013;201(7):1069-84.
7.
Notbohm J, Lesman A, Tirrell DA, Ravichandran G. Quantifying cell-induced matrix
deformation in three dimensions based on imaging matrix fibers. Integrative biology : quantitative
biosciences from nano to macro. 2015;7:1186-95.
8.
measures of tumour mechanopathology. Nature biomedical engineering. 2017;1(1):0004.
9.
Stylianopoulos T, Martin JD, Snuderl M, Mpekris F, Jain SR, Jain RK. Coevolution of Solid Stress
and Interstitial Fluid Pressure in Tumors During Progression: Implications for Vascular Collapse. Cancer
Research. 2013;73(13):3833-41.
10.
progression and metastasis. Curr Opin Biotechnol. 2016 Aug;40:41-8.
11.
Delivery Reviews. 2016 2/1/;97:41-55.
12.
Nature Reviews Molecular Cell Biology. 2014;15(12):786-801.
13. Weaver VM, Lelie S, Lakins JN, Chrenek Ma, Jones JCR, Giancotti F, et al. Integrin-Dependent
Formation of Polarized Three- Dimensional Architecture Confers Resistance To Apoptosis in Normal
and Malignant Mammary Epithelium. Cancer Cell. 2002;2(September):205-16.
14. Weaver VM, Petersen OW, Wang F, Larabell Ca, Briand P, Damsky C, et al. Reversion of the
malignant penotype of human breast cells in three-dimensional cutlure and in vivo by integrin blocking
antibodies. JCell Biol. 1997;137(1):231-45.
15.
[Original Paper]. 2000;19:5636.
16.
2016;164(6):1105-9.
17.
microenvironmental cues. Nature Reviews Cancer. 2015;15(8):499-509.
18.
specific bonding. Biophysical Journal. 1984;45(6):1051-64.
19.
1978;200(4342):618-27.
20.
and with rebinding. The Journal of Chemical Physics. 2004;121(18):8997-9017.
21.
Letters. 2004;92(10):108102.
Bell GI. Models for the specific adhesion of cells to cells. Science (New York, NY).
Brock A, Krause S, Ingber DE. Control of cancer formation by intrinsic genetic noise and
Bell GI, Dembo M, Bongrand P. Cell adhesion. Competition between nonspecific repulsion and
Insua-Rodríguez J, Oskarsson T. The extracellular matrix in breast cancer. Advanced Drug
Bonnans C, Chou J, Werb Z. Remodelling the extracellular matrix in development and disease.
Irby RB, Yeatman TJ. Role of Src expression and activation in human cancer. Oncogene.
Ingber DE. Reverse Engineering Human Pathophysiology with Organs-on-Chips. Cell.
Erdmann T, Schwarz US. Stochastic dynamics of adhesion clusters under shared constant force
Erdmann T, Schwarz U. Stability of adhesion clusters under constant force. Physical Review
Spill F, Andasari V, Mak M, Kamm RD, Zaman MH. Effects of 3D geometries on cellular gradient
Meyers J, Craig J, Odde DJ. Potential for control of signaling pathways via cell size and shape.
Mak M, Zaman MH, Kamm RD, Kim T. Interplay of active processes modulates tension and
cytoskeletal networks. Nature
self-renewing, motor-driven
transition
Banerjee DS, Munjal A, Lecuit T, Rao M. Actomyosin pulsation and flows in an active elastomer
in 3D Environments. PLoS
Eliaš J, Dimitrio L, Clairambault J, Natalini R. The dynamics of p53 in single cells: physiologically
Ron A, Azeloglu EU, Calizo RC, Hu M, Bhattacharya S, Chen Y, et al. Cell shape information is
Goehring NW, Grill SW. Cell polarity: Mechanochemical patterning. Trends in Cell Biology.
Eliaš J, Clairambault J. Reaction -- diffusion systems for spatio-temporal intracellular protein
22.
networks: A beginner's guide with two examples. Computational and Structural Biotechnology
Journal. 2014;10(16):12-22.
23.
based ODE and reaction-diffusion PDE models. Physical Biology. 2014;11(4):045001-.
24.
2013;23(2):72-80.
25.
transduced through tension-independent mechanisms. Nat Commun. 2017 Dec 15;8(1):2145.
26.
Rangamani P, Fardin M-A, Xiong Y, Lipshtat A, Rossier O, Sheetz MP, et al. Signaling network
triggers and membrane physical properties control the actin cytoskeleton-driven isotropic phase of
cell spreading. Biophysical Journal. 2011;100(4):845-57.
Neves SR, Tsokas P, Sarkar A, Grace Ea, Rangamani P, Taubenfeld SM, et al. Cell Shape and
27.
Negative Links in Regulatory Motifs Together Control Spatial Information Flow in Signaling Networks.
Cell. 2008;133(4):666-80.
28.
sensing and polarization. Phys Biol. 2016;13(3):036008.
29.
Curr Biol. 2006;16(17):1685-93.
30.
in
drives phase
Communications. 2016;7:10323.
31.
with turnover and network remodeling. Nature Communications. 2017 2017/10/24;8(1):1121.
Müller KW, Birzle AM, Wall WA. Beam finite-element model of a molecular motor for the
32.
simulation of active fibre networks. Proceedings of the Royal Society A: Mathematical, Physical and
Engineering Science. [10.1098/rspa.2015.0555]. 2016;472(2185).
33.
Mak M, Anderson S, McDonough MC, Spill F, Kim JE, Boussommier-Calleja A, et al. Integrated
Analysis of Intracellular Dynamics of MenaINV Cancer Cells in a 3D Matrix. Biophysical Journal. 2017
2017/05/09/;112(9):1874-84.
34.
Microrheology of Cancer Cells
2014;10(11):e1003959.
35.
Guo M, Ehrlicher Allen J, Jensen Mikkel H, Renz M, Moore Jeffrey R, Goldman Robert D, et al.
Probing the Stochastic, Motor-Driven Properties of the Cytoplasm Using Force Spectrum Microscopy.
Cell. 2014 2014/08/14/;158(4):822-32.
Zhang Y, Abiraman K, Li H, Pierce DM, Tzingounis AV, Lykotrafitis G. Modeling of the axon
36.
membrane skeleton structure and implications for its mechanical properties. PLoS computational
biology. 2017;13(2):e1005407.
37. Wang N, Tytell JD, Ingber DE. Mechanotransduction at a distance: mechanically coupling the
extracellular matrix with the nucleus. Nature Reviews Molecular Cell Biology. [Perspective].
2009;10:75.
38.
with Tissue Stiffness and Enhances Matrix-Directed Differentiation. Science. 2013;341(6149).
39.
envelope rupture and repair during cancer cell migration. Science. 2016.
40.
envelope ruptures during cell migration to limit DNA damage and cell death. Science. 2016.
41.
Cao X, Moeendarbary E, Isermann P, Davidson Patricia M, Wang X, Chen Michelle B, et al. A
Chemomechanical Model for Nuclear Morphology and Stresses during Cell Transendothelial
Migration. Biophysical Journal. 2016 2016/10/04/;111(7):1541-52.
Mak M, Kamm RD, Zaman MH. Impact of Dimensionality and Network Disruption on
computational biology.
Denais CM, Gilbert RM, Isermann P, McGregor AL, te Lindert M, Weigelin B, et al. Nuclear
Swift J, Ivanovska IL, Buxboim A, Harada T, Dingal PCDP, Pinter J, et al. Nuclear Lamin-A Scales
Raab M, Gentili M, de Belly H, Thiam HR, Vargas P, Jimenez AJ, et al. ESCRT III repairs nuclear
Paszek MJ, Zahir N, Johnson KR, Lakins JN, Rozenberg GI, Gefen A, et al. Tensional homeostasis
Kraning-Rush CM, Califano JP, Reinhart-King CA. Cellular Traction Stresses Increase with
Kopanska KS, Alcheikh Y, Staneva R, Vignjevic D, Betz T. Tensile Forces Originating from Cancer
Storm C, Pastore JJ, MacKintosh FC, Lubensky TC, Janmey PA. Nonlinear elasticity in biological
Mak M, Kim T, Zaman MH, Kamm RD. Multiscale mechanobiology: computational models for
Integrative Biology. [10.1039/C5IB00043B].
Funkhouser CM, Sknepnek R, Shimi T, Goldman AE, Goldman RD, Olvera de la Cruz M.
42.
Mechanical model of blebbing in nuclear lamin meshworks. Proceedings of the National Academy of
Sciences. 2013;110(9):3248-53.
43.
integrating molecules to multicellular systems.
2015;7(10):1093-108.
44.
and the malignant phenotype. Cancer Cell. 2005 2005/09/01/;8(3):241-54.
45.
Increasing Metastatic Potential. PLoS One. 2012;7(2):e32572.
46.
Shi Q, Ghosh RP, Engelke H, Rycroft CH, Cassereau L, Sethian JA, et al. Rapid disorganization
of mechanically interacting systems of mammary acini. Proceedings of the National Academy of
Sciences. 2014;111(2):658-63.
47.
Spheroids Facilitate Tumor Invasion. PLoS One. 2016;11(6):e0156442.
48.
gels. Nature. 2005;435:191.
49.
Kim M-C, Silberberg YR, Abeyaratne R, Kamm RD, Asada HH. Computational modeling of
three-dimensional ECM-rigidity sensing to guide directed cell migration. Proceedings of the National
Academy of Sciences. 2018.
50.
& Migration. 2011;5(5):421-30.
51.
Matrix by Cell-Generated Forces. bioRxiv. 2017.
52. Wang H, Abhilash AS, Chen Christopher S, Wells Rebecca G, Shenoy Vivek B. Long-Range Force
Transmission in Fibrous Matrices Enabled by Tension-Driven Alignment of Fibers. Biophysical Journal.
2014;107(11):2592-603.
53.
Ma X, Schickel ME, Stevenson Mark D, Sarang-Sieminski Alisha L, Gooch Keith J, Ghadiali
Samir N, et al. Fibers in the Extracellular Matrix Enable Long-Range Stress Transmission between Cells.
Biophysical Journal. 2013 2013/04/02/;104(7):1410-8.
54.
Abhilash AS, Baker Brendon M, Trappmann B, Chen Christopher S, Shenoy Vivek B.
Remodeling of Fibrous Extracellular Matrices by Contractile Cells: Predictions from Discrete Fiber
Network Simulations. Biophysical Journal. 2014 2014/10/21/;107(8):1829-40.
55.
Nature Reviews Cancer. [Review Article]. 2015;15:730.
56.
adhesion. Mathematical medicine and biology : a journal of the IMA. 2005;22(2):163-86.
57.
Chauviere a, Preziosi L, Byrne H. A model of cell migration within the extracellular matrix based
on a phenotypic switching mechanism. Mathematical medicine and biology : a journal of the IMA.
2010;27(3):255-81.
58.
Modelling of Natural Phenomena. 2009;4(3):1-11.
59.
inhomogeneous environment. Journal of Theoretical Biology. 2003;225(2):257-74.
Bearer EL, Lowengrub JS, Frieboes HB, Chuang Y-L, Jin F, Wise SM, et al. Multiparameter
60.
Computational Modeling of Tumor Invasion. Cancer Research. [10.1158/0008-5472.CAN-08-3834].
2009;69(10):4493.
61.
Macklin P, Frieboes HB, Sparks JL, Ghaffarizadeh A, Friedman SH, Juarez EF, et al. Progress
Towards Computational 3-D Multicellular Systems Biology. In: Rejniak KA, editor. Systems Biology of
Tumor Microenvironment: Quantitative Modeling and Simulations. Cham: Springer International
Publishing; 2016. p. 225-46.
Preziosi L, Tosin A. Multiphase and Multiscale Trends in Cancer Modelling. Mathematical
Altrock PM, Liu LL, Michor F. The mathematics of cancer: integrating quantitative models.
Anderson ARA. A hybrid mathematical model of solid tumour invasion: the importance of cell
Arjonen A, Kaukonen R, Ivaska J. Filopodia and adhesion in cancer cell motility. Cell Adhesion
Malandrino A, Mak M, Trepat X, Kamm RD. Non-Elastic Remodeling of the 3D Extracellular
Alarcón T, Byrne HM, Maini PK. A cellular automaton model for tumour growth in
Mitchison T, Kirschner M. Cytoskeletal dynamics and nerve growth. Neuron. 1988
Jiang G, Giannone G, Critchley DR, Fukumoto E, Sheetz MP. Two-piconewton slip bond
Hu K, Ji L, Applegate KT, Danuser G, Waterman-Storer CM. Differential transmission of actin
Eke I, Cordes N. Focal adhesion signaling and therapy resistance in cancer. Seminars in Cancer
Hynes RO. Integrins: bidirectional, allosteric signaling machines. Cell. 2002 Sep 20;110(6):673-
Macklin P. Key challenges facing data-driven multicellular systems biology. arXiv preprint
Malandrino A, Mak M, Kamm RD, Moeendarbary E. Complex mechanics of the heterogeneous
Sun Z, Guo SS, Fassler R. Integrin-mediated mechanotransduction. J Cell Biol. 2016 Nov
62.
arXiv:180604736. 2018.
63.
extracellular matrix in cancer. Extreme Mechanics Letters. 2018 2018/05/01/;21:25-34.
64.
21;215(4):445-56.
65.
Biology. 2014;31:1-11.
66.
87.
67.
Ezzell RM, Goldmann WH, Wang N, Parashurama N, Ingber DE. Vinculin promotes cell
spreading by mechanically coupling integrins to the cytoskeleton. Exp Cell Res. 1997 Feb 25;231(1):14-
26.
68.
Giannone G, Jiang G, Sutton DH, Critchley DR, Sheetz MP. Talin1 is critical for force-dependent
reinforcement of initial integrin-cytoskeleton bonds but not tyrosine kinase activation. J Cell Biol. 2003
Oct 27;163(2):409-19.
69.
between fibronectin and the cytoskeleton depends on talin. Nature. 2003 Jul 17;424(6946):334-7.
70.
motion within focal adhesions. Science. 2007 Jan 5;315(5808):111-5.
71.
Nov;1(9):761-72.
72.
Jurado C, Haserick JR, Lee J. Slipping or gripping? Fluorescent speckle microscopy in fish
keratocytes reveals two different mechanisms for generating a retrograde flow of actin. Mol Biol Cell.
2005 Feb;16(2):507-18.
73.
revealed in stationary fibroblasts. Science. 1999 Nov 5;286(5442):1172-4.
74.
2008;322(5908):1687-91.
75.
Elosegui-Artola A, Oria R, Chen Y, Kosmalska A, Perez-Gonzalez C, Castro N, et al. Mechanical
regulation of a molecular clutch defines force transmission and transduction in response to matrix
rigidity. Nat Cell Biol. 2016 May;18(5):540-8.
76.
loading explains spatial sensing of ligands by cells. Nature. 2017;552:219.
77.
Cao X, Lin Y, Driscoll Tristian P, Franco-Barraza J, Cukierman E, Mauck Robert L, et al. A
Chemomechanical Model of Matrix and Nuclear Rigidity Regulation of Focal Adhesion Size. Biophysical
Journal. 2015 2015/11/03/;109(9):1807-17.
78.
Cao X, Ban E, Baker BM, Lin Y, Burdick JA, Chen CS, et al. Multiscale model predicts increasing
focal adhesion size with decreasing stiffness in fibrous matrices. Proceedings of the National Academy
of Sciences. 2017.
79.
optimal stiffness for cell migration. Nat Commun. 2017;8:15313.
80.
cellular tension. Nat Rev Mol Cell Biol. 2010 Sep;11(9):633-43.
81.
Reviews. 2013;93(1):269-309.
82.
Mechanism and Function. Antioxidants & Redox Signaling. 2012;18(3):120730082110002-.
83.
inhomogeneous stress fiber contraction. New Journal of Physics. 2007;9(11):425-.
Oria R, Wiegand T, Escribano J, Elosegui-Artola A, Uriarte JJ, Moreno-Pulido C, et al. Force
Chan CE, Odde DJ. Traction Dynamics of Filopodia on Compliant Substrates. Science.
Parsons JT, Horwitz AR, Schwartz MA. Cell adhesion: integrating cytoskeletal dynamics and
Cherfils J, Zeghouf M. Regulation of small GTPases by GEFs, GAPs, and GDIs. Physiological
Smilenov LB, Mikhailov A, Pelham RJ, Marcantonio EE, Gundersen GG. Focal adhesion motility
Bangasser BL, Shamsan GA, Chan CE, Opoku KN, Tüzel E, Schlichtmann BW, et al. Shifting the
Mitchell L, Hobbs GA, Aghajanian A, Campbell SL. Redox Regulation of Ras and Rho GTPases:
Besser A, Schwarz US. Coupling biochemistry and mechanics in cell adhesion: a model for
Oria R, Wiegand T, Escribano J, Elosegui-Artola A, Uriarte JJ, Moreno-Pulido C, et al. Force
Dulbecco R, Stoker MG. Conditions determining initiation of DNA synthesis in 3T3 cells. Proc
Folkman J, Moscona A. Role of cell shape in growth control. Nature. 1978 Jun 1;273(5661):345-
Plotnikov SV, Waterman CM. Guiding cell migration by tugging. Curr Opin Cell Biol.
Harland B, Walcott S, Sun SX. Adhesion dynamics and durotaxis in migrating cells. Physical
Chen B, Kemkemer R, Deibler M, Spatz J, Gao H. Cyclic Stretch Induces Cell Reorientation on
Sero JE, Bakal C. Multiparametric Analysis of Cell Shape Demonstrates that beta-PIX Directly
Russell RJ, Xia S-L, Dickinson RB, Lele TP. Sarcomere Mechanics in Capillary Endothelial Cells.
Qian J, Liu H, Lin Y, Chen W, Gao H. A mechanochemical model of cell reorientation on
84.
2013;25(5):619-26.
85.
Biology. 2011;8(1):015011-.
Escribano J, Sunyer R, Sánchez MT, Trepat X, Roca-Cusachs P, García-Aznar JM. A hybrid
86.
computational model for collective cell durotaxis. Biomechanics and Modeling in Mechanobiology.
2018 2018/03/02.
87.
Sunyer R, Conte V, Escribano J, Elosegui-Artola A, Labernadie A, Valon L, et al. Collective cell
durotaxis emerges from long-range intercellular force transmission. Science. 2016 2016-09-09
00:00:00;353:1157-61.
88.
substrates under cyclic stretch. PLoS One. 2013;8(6):e65864.
89.
Biophysical Journal. 2009 2009/09/16/;97(6):1578-85.
90.
Substrates by Destabilizing Catch Bonds in Focal Adhesions. PLoS One. 2012;7(11):e48346.
91.
Couples YAP Activation to Extracellular Matrix Adhesion. Cell Syst. 2017 Jan 25;4(1):84-96 e6.
92.
loading explains spatial sensing of ligands by cells. Nature. 2017 Dec 14;552(7684):219-24.
Kim M-C, Neal DM, Kamm RD, Asada HH. Dynamic Modeling of Cell Migration and Spreading
93.
Behaviors on Fibronectin Coated Planar Substrates and Micropatterned Geometries. PLOS
Computational Biology. 2013;9(2):e1002926.
94.
McMeeking RM, Deshpande VS. A Bio-chemo-mechanical Model for Cell Contractility,
Adhesion, Signaling, and Stress-Fiber Remodeling. In: Holzapfel GA, Ogden RW, editors. Biomechanics:
Trends in Modeling and Simulation. Cham: Springer International Publishing; 2017. p. 53-81.
95.
Natl Acad Sci U S A. 1970 May;66(1):204-10.
96.
9.
97.
Ingber DE, Folkman J. Mechanochemical switching between growth and differentiation during
fibroblast growth factor-stimulated angiogenesis in vitro: role of extracellular matrix. J Cell Biol. 1989
Jul;109(1):317-30.
98. Welsh CF, Roovers K, Villanueva J, Liu Y, Schwartz MA, Assoian RK. Timing of cyclin D1
expression within G1 phase is controlled by Rho. Nat Cell Biol. 2001 Nov;3(11):950-7.
99.
Mammoto A, Huang S, Moore K, Oh P, Ingber DE. Role of RhoA, mDia, and ROCK in cell shape-
dependent control of the Skp2-p27kip1 pathway and the G1/S transition. J Biol Chem. 2004 Jun
18;279(25):26323-30.
100.
Hirata E, Girotti MR, Viros A, Hooper S, Spencer-Dene B, Matsuda M, et al. Intravital imaging
reveals how BRAF inhibition generates drug-tolerant microenvironments with high integrin beta1/FAK
signaling. Cancer Cell. 2015 Apr 13;27(4):574-88.
101.
Fallahi-Sichani M, Becker V, Izar B, Baker GJ, Lin JR, Boswell SA, et al. Adaptive resistance of
melanoma cells to RAF inhibition via reversible induction of a slowly dividing de-differentiated state.
Mol Syst Biol. 2017 Jan 9;13(1):905.
Kuroda M, Wada H, Kimura Y, Ueda K, Kioka N. Vinculin promotes nuclear localization of TAZ
102.
to inhibit ECM stiffness-dependent differentiation into adipocytes. J Cell Sci. 2017 Mar 1;130(5):989-
1002.
103.
13;29(6):783-803.
104.
via cytoskeleton reorganization to induce anoikis. Genes and Development. 2012;26(1):54-68.
Zanconato F, Cordenonsi M, Piccolo S. YAP/TAZ at the Roots of Cancer. Cancer Cell. 2016 Jun
Zhao B, Li L, Wang L, Wang CY, Yu J, Guan KL. Cell detachment activates the Hippo pathway
Sun M, Spill F, Zaman Muhammad H. A Computational Model of YAP/TAZ Mechanosensing.
Hu JK, Du W, Shelton SJ, Oldham MC, DiPersio CM, Klein OD. An FAK-YAP-mTOR Signaling Axis
Bell S, Terentjev EM. Focal Adhesion Kinase: The Reversible Molecular Mechanosensor.
Lamar JM, Stern P, Liu H, Schindler JW, Jiang ZG, Hynes RO. The Hippo pathway target, YAP,
105.
promotes metastasis through its TEAD-interaction domain. Proceedings of the National Academy of
Sciences. 2012;109(37):E2441-E50.
106.
pathway. J Cell Biol. 2015 Aug 3;210(3):503-15.
107.
Yui S, Azzolin L, Maimets M, Pedersen MT, Fordham RP, Hansen SL, et al. YAP/TAZ-Dependent
Reprogramming of Colonic Epithelium Links ECM Remodeling to Tissue Regeneration. Cell Stem Cell.
2018 Jan 4;22(1):35-49 e7.
108.
Regulates Stem Cell-Based Tissue Renewal in Mice. Cell Stem Cell. 2017 Jul 6;21(1):91-106 e6.
109. Nardone G, Oliver-De La Cruz J, Vrbsky J, Martini C, Pribyl J, Skladal P, et al. YAP regulates cell
mechanics by controlling focal adhesion assembly. Nat Commun. 2017 May 15;8:15321.
110.
Biophysical Journal. 2016 Jun 7;110(11):2540-50.
111.
Biophysical Journal. 2017;112(11):2439-50.
112.
focal adhesion kinase mechanosensing. PLoS computational biology. 2015;11(11):e1004593.
113. Miralles F, Posern G, Zaromytidou AI, Treisman R. Actin dynamics control SRF activity by
regulation of its coactivator MAL. Cell. 2003;113(3):329-42.
114.
transcription. Cellular & Molecular Biology Letters. 2016 2016/12/07;21(1):28.
115. Wei SC, Fattet L, Tsai JH, Guo Y, Pai VH, Majeski HE, et al. Matrix stiffness drives epithelial --
mesenchymal transition and tumour metastasis through a TWIST1 -- G3BP2 mechanotransduction
pathway. Nature Cell Biology. 2015;17(5):678-88.
116.
Is Sufficient to Physically Control Gene Circuits. Biophysical Journal. 2014;107(11):2734-43.
117.
Nature Nanotechnology. 2007;2(12):780.
118.
YAP nuclear entry by regulating transport across nuclear pores. Cell. 2017;171(6):1397-410. e14.
119.
information in cell shape. Cell. 2013;154(6):1356-69.
Finch-Edmondson M, Sudol M. Framework to function: mechanosensitive regulators of gene
Zhou J, Aponte-Santamaría C, Sturm S, Bullerjahn JT, Bronowska A, Gräter F. Mechanism of
Kim NG, Gumbiner BM. Adhesion to fibronectin regulates Hippo signaling via the FAK-Src-PI3K
Dingal PCDP, Discher Dennis E. Systems Mechanobiology: Tension-Inhibited Protein Turnover
Cross SE, Jin Y-S, Rao J, Gimzewski JK. Nanomechanical analysis of cells from cancer patients.
Elosegui-Artola A, Andreu I, Beedle AE, Lezamiz A, Uroz M, Kosmalska AJ, et al. Force triggers
Rangamani P, Lipshtat A, Azeloglu EU, Calizo RC, Hu M, Ghassemi S, et al. Decoding
|
1210.7019 | 1 | 1210 | 2012-10-25T23:27:29 | Labyrinthine clustering in a spatial rock-paper-scissors ecosystem | [
"physics.bio-ph",
"physics.soc-ph",
"q-bio.PE"
] | The spatial rock-paper-scissors ecosystem, where three species interact cyclically, is a model example of how spatial structure can maintain biodiversity. We here consider such a system for a broad range of interaction rates. When one species grows very slowly, this species and its prey dominate the system by self-organizing into a labyrinthine configuration in which the third species propagates. The cluster size distributions of the two dominating species have heavy tails and the configuration is stabilized through a complex, spatial feedback loop. We introduce a new statistical measure that quantifies the amount of clustering in the spatial system by comparison with its mean field approximation. Hereby, we are able to quantitatively explain how the labyrinthine configuration slows down the dynamics and stabilizes the system. | physics.bio-ph | physics | Labyrinthine clustering in a spatial rock-paper-scissors ecosystem
University of Copenhagen, Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen, Denmark
Jeppe Juul, Kim Sneppen, and Joachim Mathiesen
(Dated: September 14, 2018)
The spatial rock-paper-scissors ecosystem, where three species interact cyclically, is a model ex-
ample of how spatial structure can maintain biodiversity. We here consider such a system for a
broad range of interaction rates. When one species grows very slowly, this species and its prey
dominate the system by self-organizing into a labyrinthine configuration in which the third species
propagates. The cluster size distributions of the two dominating species have heavy tails and the
configuration is stabilized through a complex, spatial feedback loop. We introduce a new statis-
tical measure that quantifies the amount of clustering in the spatial system by comparison with
its mean field approximation. Hereby, we are able to quantitatively explain how the labyrinthine
configuration slows down the dynamics and stabilizes the system.
PACS numbers: 87.23.Kg, 87.23.Cc, 87.18.Hf
Introduction -- Spatial migration of species is crucial
for the viability of many ecological systems. As a striking
example, crickets are known to locally deplete their nutri-
tional resources to an extend where mass-migration is the
only alternative to cannibalism [1, 2]. Once the crickets
have left an area, they can not return until the natu-
ral resources have been reestablished. Likewise, deadly
viruses and bacteria depend on constantly infecting new
hosts to survive [3 -- 5].
The rock-paper scissors game has emerged as a
paradigm to describe the impact of spatial structure on
biodiversity [6 -- 11]. In this system, three species interact
cyclically such that species 1 can invade species 2, which
can invade species 3, which, in turn, can invade species
1 (see Fig. 1a). Such intransitive interaction pattern is
very similar to the important genetic regulatory network
the repressilator [12, 13] and has been identified in many
ecological system, among others in marine benthic sys-
tems [14, 15], plant systems [16 -- 18], terrestial systems
[19, 20], and microbial systems [21 -- 25]. In such systems,
all species constantly need to migrate spatially to sur-
vive. Investigating three strands of E. coli bacteria with
cyclic interactions, it has been shown that biodiversity
can not be preserved unless spacial structure is imposed
by arranging the bacteria on a petri dish [7, 8, 26]. These
results have been reproduced in Monte Carlo simulations
[6, 27 -- 29], but even though many different analytical ap-
proaches have been applied, exactly how spatial structure
stabilizes the system is still an open problem [30 -- 33].
Model -- We study the rock-paper-scissors game on
a square lattice of L × L nodes and periodic boundary
conditions. Each node is occupied by one of the three
species 1, 2, or 3 growing at rates v1, v2, and v3, respec-
tively. In each update a random node i and a random of
its neighbors j are selected. If i can invade j according
to the cyclic interacting pattern illustrated in Fig. 1a, it
will do so with a probability equal to vi.
Results -- When the three species are initiated from a
random configuration and with equal growth rates, they
quickly organize into a steady state where all species are
equally abundant and form small clusters (see Fig. 1b).
If the growth rate of species 3 is increased compared to
species 1 and 2, species 2 becomes more abundant on the
lattice and all three species form larger clusters (see Fig.
1c). This paradoxical behavior, that the biomass of one
species increases proportional to the growth rate of its
prey, is characteristic for the rock-paper-scissors system
[6, 27].
Similarly, if the growth rate of species 1 is decreased,
species 3 slowly becomes scarcer. Approaching the limit
v1 → 0 a very large lattice is required in order for species
3 to be viable.
In this limit a new, interesting spatial
organization is observed. Species 3 propagates through
the lattice in thin and broken wave fronts in constant
flight from species 2. In the rest of the system the slowly
growing species 1 and its prey, species 2, is tangled in a
complex configuration with an enormous mutual perime-
ter. This spatial organization forms an ever-changing
labyrinth of narrow pathways in which species 3 propa-
gates (see Fig. 1d-f). The more narrow and twisted the
labyrinth becomes, the longer it will take for species 3 to
return to a particular location, which gives species 1 more
time to grow, forming broader pathways. This complex,
spatial feedback loop stabilizes the configuration.
In order to describe this spatial self-organization math-
ematically, we study the probabilities p1, p2, and p3 of a
random node to be occupied by species 1, 2, or 3, respec-
tively. Furthermore, we are interested in the perimeter
pij between species i and j. That is, the probability of a
random node and a random of its neighbors to be occu-
pied by species i and j, respectively.
Given these perimeters the time evolution of species
abundances is given by [30]
p1 = v2p12 − v3p31,
(1)
where the equations for p2 and p3 follow by cyclic per-
mutation of the indices 1, 2, and 3. This symmetry also
holds for all subsequent equations of this article.
2
1
0
2
t
c
O
5
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
9
1
0
7
.
0
1
2
1
:
v
i
X
r
a
2
(Color online) Species abundancies and
Figure 2:
perimeters for small v1. a) When the growth rate of
species 1 is decreased, species 3 becomes less abundant. The
mean field theory correctly predicts the abundancies. b) The
perimeters between the species are much lower than predicted
by the mean field theory due to clustering. c) The ratio be-
tween the predicted mean field perimeters and the observed
perimeters are equal for all species. This ratio defines χ. d)
When v3 (cid:29) v1, v2, the ratio χ diverges corresponding to the
large clustering in Fig. 1c. When one species grows much
slower than the others χ approaches 5, which gives rise to the
labyrinthine clustering in Fig. 1d.
inserted into (1) and the time derivatives are set to zero,
one obtains the steady state solution
p1 =
v2
v1 + v2 + v3
(v1 + v2 + v3)2 .
v2v3
p12 =
(2)
(3)
Figure 1: (Color online) Spatial self-organization in the
rock-paper-scissors game. a) The three species interact
cyclically. Species i invades its prey at rate vi. b-d) Snap-
shots of the steady state spatial organization of the three
species when b) All species grow at same rate and L = 300.
c) Species 3 grows 5 times faster than 1 and 2 and L = 300.
d) Species 1 grows 1250 times slower than 2 and 3 and
L = 12800. e-f ) zooms of the system in panel d.
To explain this behavior, one can adopt a mean field
approximation, where all nodes are linked and spatial
structure does not exist. Then, the perimeter between
two species is simply given by the product of species
abundances p12 = p1 p2, where tilde (∼) denotes that
the mean field approximation has been applied. If this is
In Fig. 2a-b the steady state abundances and perime-
ters are shown at constant v2 = v3 = 1 and varying
v1 ≤ 1. It is seen that a slow growth rate of species 1 leads
to a decline in the abundance of species 3 as expected.
The mean field approximation correctly predicts how the
abundances of the three species depend on the growth
rates. However, the mean field approach can not capture
the spatial organization of the species, and thus it pre-
dicts perimeters far longer than what is observed in sim-
ulations (see Fig. 2b). The fact that the abundances are
correctly predicted indicates that the mean field perime-
ters are proportional to the true, spatial perimeters. In-
deed, if (1) is set to zero for both the spatial and mean
field system one can derive the relations.
p12
p12
=
p23
p23
=
p31
p31
≡ χ.
(4)
Here, we have introduced the ratio χ, defined by how
much the perimeter between two species is longer in the
2000400060008000100001200020004000600080001000012000123v1v2v3bcdefa00.20.40.60.81051015χ0 0.10.20.3 0.20.40.60.81 00.210.40.60.81v00.20.40.60.81χ2.5500100.20.40.60.81v100.20.40.60.81v2v3v1v3v3p1p3p~1p2p~=~2p12p~12p31p~=23p~31p23p=12p~12p23p~23p31p~31pabcd steady state of the mean field approximation compared
to the spatial system (see Fig. 2c). This new statistical
measure describes the spatial and dynamical organization
of the rock-paper-scissors game for varying growth rates.
The intuition behind χ is the following:
3
The average time before a node of species 2 is invaded
for the spatial system
in mean field. Therefore, χ ≈ T1
T1
by species 1 is given by T1 = p1
v1p12
and T1 = p1
provides
v1 p12
a measure for how much longer each species on average
lives on a node before being invaded, compared to the
mean field system, i.e. how much the spatial organization
slows down the dynamics. Furthermore, when χ is large
the perimeters of the spatial system is much smaller than
in the mean field system, according to (4), so the species
must have a high degree of clustering. Hence, χ gives a
measure for the clustering of the spatial system. These
two interpretations are, of course, tightly connected. If
the average cluster diameters are doubled, each node will
live for twice as long before being invaded, corresponding
to increasing χ by a factor two.
and v2
v3
How does χ depend on the growth rates of the three
species? In Fig. 2d this dependency is shown as a func-
tion of the relative growth rates v1
, with v3 chosen
v3
to be the fastest growing species. When all growth rates
are equal we have χ ≈ 2.5, corresponding to the mod-
erate amount of clustering observed in Fig. 1b. When
species 3 grows much faster than the two other species,
such that both growth ratios go to zero, χ becomes very
large. This agrees well with the large amount of cluster-
ing observed in Fig. 1c.
When only v1 → 0 we see from Fig. 1d and 2d that
χ approaches a finite value close to 5. In this limit, we
expect from (2) that p3 → 0 while p1 ≈ p2 → 1
2 . In this
case, it is limited how much species 3 can cluster. The
observation of χ suggests that clustering only reduces the
perimeter between species 2 and 3 by a factor 5 compared
to the mean field system. This sets an upper bound for
how much species 1 and 2 can cluster. The mean field
4 , so with χ ≈ 5
approach predicts a perimeter p12 = 1
equation (4) dictates the perimeter in the spatial system
20 . This agrees well with the 12800 × 12800
to be p12 ≈ 1
system in Fig. 1d, where v1 = 0.0008, v2 = v3 = 1, and
p12 ≈ 0.05, which is also evident from Fig. 2b.
While the value of χ quantifies the average amount of
clustering, it does not provide information on the clus-
ter size distribution. In the case where all species grow
with the same rate, Fig. 1b suggests that clusters have a
characteristic size. Indeed, Fig. 3 shows that the cluster
size distribution in this case sharply decreases for clus-
ters larger than 1000 nodes. When the growth rate of
species 1 goes to zero, however, species 3 continues to be
organized in small clusters, but large clusters of species 1
and 2 become much more likely. The cluster size distri-
butions of these become exceedingly broad culminating
in a heavy tail distribution with a cut-off that is set by
Figure 3:
(Color online) Cluster size distributions. a
When all species grow at the same rate, all clusters consist of
less than 5000 nodes. Here v1 = v2 = v3 = 1. b When the
growth rate of species 1 is decreased, species 3 becomes less
abundant and large clusters of species 1 and 2 become more
likely. Here v1 ≈ 0.5, and v2 = v3 = 1. c In the limit v1 → 0
the cluster size distributions of species 1 and 2 become heavy
tailed with a cut-off set by the system size. Here v1 ≤ 0.007,
v2 = v3 = 1. For all plots L = 2048.
the system size.
An alternative approach that has been applied to de-
scribe the spatial organisation of the rock-paper-scissors
game is the pair approximation [31, 34]. Here the time
evolution of the perimeters pij are expressed through the
probabilities pi
jk of a random node belonging to species i
and a random pair of its neighbors belinging to species j
and k, respectively. When all growth rates are equal, the
pair approximation gives the steady state probabilities
[34]
p1 =
,
(5)
2
27
.
5
27
,
p12 =
p11 =
1
3
Since (3) p12 = 1
9 , the pair approximation predicts χ =
1.5, which is far from the observed value of χ = 2.5. This
further illustrates that the incapability of the pair ap-
proximation to describe the behavior of the rock-paper-
scissors game.
cluster sizeprobability density function10010210410610−1210−1010−810−610−410−2100species 1species 2species 3n−2.4probability density functioncluster size10010210410-10species 1species 2species 3species 1species 2species 310 010-5cluster size100102104abc Discussion -- Our results quantitatively describe how
spatial clustering slows down the dynamics of the rock-
paper-scissors game, and why this leads to a labyrinthine
spatial organization in the limit where one species grows
slowly compared to the two others. An organization that
includes a new type of excitable fronts that propagate on
self-organized labyrinthine clusters distributed over many
length scales.
In this limit of one slow species, the largest clusters of
both the slow species and its prey cover a large fraction
of the system, as seen in Fig. 2d. This consequence of the
labyrinthine configuration would not be possible in site
percolation, where each of the large species would need
to occupy close to 60% of the nodes to percolate [35].
Interestingly, the extreme version of the rock-paper-
scissors ecology with one slow species bear resemblance to
the forest fire model in a fire-tree-ashes analogy [36 -- 38].
The slow species would then be forest, which is burned
by fire, which is replaced by ashes, from which trees can
again slowly grow. The main differences from existing
forest fire models are that in the present system trees
can only grow in the neighborhood of other trees and fire
can only be extinguished in the neighborhood of ashes.
The method of quantifying how much clustering slows
down the dynamics of a spatial system, compared to the
mean field approximation, is quite general, and we ex-
pect it to be applicable on a broad range of dynamical
systems. In particular, it may be useful predicting the
spatial organization in predator-prey models, which con-
tinues to attract much attention within the field of com-
plex systems [39, 40].
[1] S. Simpson, G. Sword, P. Lorch, and I. Couzin, Proceed-
ings of the National Academy of Sciences of the United
States of America 103, 4152 (2006).
[2] G. Sword, P. Lorch, and D. Gwynne, Nature 433, 703
(2005).
[3] D. Morens, G. Folkers, and A. Fauci, Nature 430, 242
(2004).
[4] K. Sneppen, A. Trusina, M. H. Jensen, and S. Bornholdt,
PloS one 5, e13326 (2010).
[5] J. Juul and K. Sneppen, Physical Review E 84, 036119
(2011).
[6] M. Frean and E. R. Abraham, Proceedings of the Royal
Society of London. Series B: Biological Sciences 268,
1323 (2001).
[7] B. Kerr, M. A. Riley, M. W. Feldman, and B. J. M.
Bohannan, Nature 418, 171 (2002).
[8] T. Reichenbach, M. Mobilia, and E. Frey, Nature 448,
1046 (2007).
[9] G. Szab´o and G. F´ath, Physics Reports 446, 97 (2007).
4
[10] E. Frey, Physica A: Statistical Mechanics and its Appli-
cations 389, 4265 (2010).
[11] P. P. Avelino, D. Bazeia, L. Losano, J. Menezes, and
B. F. Oliveira, Physical Review E 86, 036112 (2012).
[12] M. Elowitz and S. Leibler, Nature 403, 335 (2000).
[13] L. A. S. E. Elowitz, M.B. and P. Swain, Science's STKE
297, 1183 (2002).
[14] J. Jackson and L. Buss, Proceedings of the National
Academy of Sciences 72, 5160 (1975).
[15] K. Sebens, Ecological Monographs pp. 73 -- 96 (1986).
[16] D. D. Cameron, A. White, and J. Antonovics, Journal of
Ecology 97, 1311 (2009).
[17] R. A. Lankau and S. Y. Strauss, Science 317, 1561
(2007).
[18] D. R. Taylor and L. W. Aarssen, The American Natural-
ist 136, 305 (1990).
[19] B. Sinervo and C. Lively, Nature 380, 240 (1996).
[20] T. R. Birkhead, N. Chaline, J. D. Biggins, T. Burke, and
T. Pizzari, Evolution 58, 416 (2004).
[21] R. Durrett and S. Levin, Journal of Theoretical Biology
185, 165 (1997).
[22] J. Nahum, B. Harding, and B. Kerr, Proceedings of the
National Academy of Sciences 108, 10831 (2011).
[23] B. Kirkup and M. Riley, Nature 428, 412 (2004).
[24] M. Hibbing, C. Fuqua, M. Parsek, S. Peterson, M. Hi-
bbing, C. Fuqua, M. Parsek, S. Peterson, et al., Nature
reviews. Microbiology 8, 15 (2010).
[25] P. Trosvik, K. Rudi, K. Straetkvern, K. Jakobsen, T. Naes,
and N. Stenseth, Environmental Microbiology 12, 2677
(2010).
[26] B. Kerr, C. Neuhauser, B. J. M. Bohannan, and A. M.
Dean, Nature 442, 75 (2006).
[27] C. R. Johnson and I. Seinen, Proc Biol Sci. 269, 655
(2002).
[28] J. Mathiesen, N. Mitarai, K. Sneppen, and A. Trusina,
Phys. Rev. Lett. 107, 188101 (2011).
[29] Q. He, M. Mobilia, and U. C. Tauber, Phys. Rev. E 82,
051909 (2010).
[30] T. Reichenbach, M. Mobilia, and E. Frey, Phys. Rev. E
74, 051907 (2006).
[31] G. Szab´o, A. Szolnoki, and R. Izs´ak, Journal of Physics
A: Mathematical and General 37, 2599 (2004).
[32] A. Dobrinevski and E. Frey, Phys. Rev. E 85, 051903
(2012).
[33] J. Juul, K. Sneppen, and J. Mathiesen, Phys. Rev. E 85,
061924 (2012).
[34] K. Tainaka, Physics Letters A 176, 303 (1993).
[35] D. Stauffer and A. Aharony, Introduction to percolation
theory (Taylor & Francis, 1992).
[36] P. Bak, K. Chen, and C. Tang, Physics letters A 147,
297 (1990).
[37] B. Drossel and F. Schwabl, Physical Review Letters 69,
1629 (1992).
[38] K. Christensen, H. Flyvbjerg, and Z. Olami, Physical
review letters 71, 2737 (1993).
[39] C. A. Lugo and A. J. McKane, Physical Review E 78,
051911 (2008).
[40] D. Vasseur and J. Fox, Nature 460, 1007 (2009).
|
1810.03013 | 1 | 1810 | 2018-10-06T16:05:19 | Orbiting of bacteria around micrometer-sized particles entrapping shallow tents of fluids | [
"physics.bio-ph",
"cond-mat.soft",
"physics.flu-dyn"
] | Hydrodynamics and confinement dominate bacterial mobility near solid or air-water boundaries, causing flagellated bacteria to move in circular trajectories. This phenomenon results from the counter-rotation between the bacterial body and flagella and lateral drags on them in opposite directions due to their proximity to the boundaries. Numerous experimental techniques have been developed to confine and maneuver motile bacteria. Here, we report observations on Escherichia coli and Enterobacter sp. when they are confined within a thin layer of water around dispersed micrometer-sized particles sprinkled over a semi-solid agar gel. In this setting, the flagellated bacteria orbit around the dispersed particles akin to planetary systems. The liquid layer is shaped like a shallow tent with its height at the center set by the seeding particle and the meniscus profile set by the strong surface tension of water. The tent-shaped constraint and the left handedness of the flagellar filaments result in exclusively clockwise circular trajectories. The thin fluid layer is resilient due to a balance between evaporation and reinforcing fluid pumped out of the agar. The latter is driven by the Laplace pressure caused by the curved meniscus. This novel mechanism to entrap bacteria within a minimal volume of fluid is relevant to near surface bacterial accumulation, adhesion, biofilm growth, development of bio-microdevices, and cleansing hygiene. | physics.bio-ph | physics | Orbiting of bacteria around micrometer-sized particles entrapping shallow tents of fluids
George Araujo1, Weijie Chen1,2, Sridhar Mani2, Jay X. Tang1*
1 Department of Physics, Brown University, 182 Hope st, Providence, RI, 02912, USA
2 Department of Genetics, Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, NY
10461, USA
*Corresponding author: [email protected]
Abstract
Hydrodynamics and confinement dominate bacterial mobility near solid or air-water boundaries,
causing flagellated bacteria to move in circular trajectories. This phenomenon results from the
counter-rotation between the bacterial body and flagella and lateral drags on them in opposite
directions due to their proximity to the boundaries. Numerous experimental techniques have been
developed to confine and maneuver motile bacteria. Here, we report observations on Escherichia
coli and Enterobacter sp. when they are confined within a thin layer of water around dispersed
micrometer-sized particles sprinkled over a semi-solid agar gel. In this setting, the flagellated
bacteria orbit around the dispersed particles akin to planetary systems. The liquid layer is shaped
like a shallow tent with its height at the center set by the seeding particle and the meniscus profile
set by the strong surface tension of water. The tent-shaped constraint and the left handedness of
the flagellar filaments result in exclusively clockwise circular trajectories. The thin fluid layer is
resilient due to a balance between evaporation and reinforcing fluid pumped out of the agar. The
latter is driven by the Laplace pressure caused by the curved meniscus. This novel mechanism to
entrap bacteria within a minimal volume of fluid is relevant to near surface bacterial accumulation,
adhesion, biofilm growth, development of bio-microdevices, and cleansing hygiene.
keywords: biophysics, bacterial motility, confinement, hydrodynamics, microfluidics
Significance
1
We report a novel observation of thin but robust tent-like layer of liquid around micro-sized particles
sprinkled over a semisolid agar surface. The layer is resilient to evaporation and provides a strong
confinement that allows orbital motion of flagellated bacteria around the particles. The experimental
setup is simple and it is surprising that such small particles each retain a water layer, sustaining circular
bacterial trajectories. Our discovery is explained by applying a hydrodynamic model on bacterial
swimming. The mechanism is relevant to practical situations involving bacterial behavior in
environments with scarcity of water, biofilm growth, infection control, etc.
Introduction
A flagellated bacterium propels itself by rotating its flagella relative to its body, which counterrotates
[1]. Freely swimming bacterial cells tend to accumulate close to solid surfaces [2-4], with their residence
times within a few micrometers (up to 1-3 µm) from the boundary lasting several seconds due to
hydrodynamic effects [5]. The coupling of body rotation and the enhancement in the drag force exerted
on the part of the cell closer to the solid surface causes the bacterium to move in circular trajectories [6].
The handedness of the flagella determines whether the circular trajectory is clockwise or
counterclockwise. Propelled by its flagellar filaments with left-handed helical structure, E. coli moves
forward in clockwise paths when looked from above [7]; it swims on the right-hand side when prevented
to turn by a solid wall on its side [8].
Numerous experimental techniques, including microfluidic devices [9,10], droplets [11-14], or vortices
[15], have been developed to confine and maneuver motile bacteria. The mode of bacterial motility can
also be shaped by confinement. For example, whereas flagellated bacteria swim in liquid, on a solid
surface they may move collectively in the form of swarming motility [16-18]. The swarming motility
triggers a variety of adhesive mechanisms, including those preceding to the growth of biofilms [19-21].
Results
Our experiment starts by allowing swarming bacteria to break away into swimming by adding a tiny
droplet of water (2-5 µl) next to a swarming colony (see Method; illustrated in Fig 1a). A number of
bacteria leave their pack at the swarm front and disperse over the agar surface (Fig. 1b). As the droplet
2
of water evaporates, most bacteria get stuck to the surface and cease the swimming motion within a short
time interval (~1 minute). However, graphite particles or silica beads contained in the water droplet
entrap a layer of water in their surrounding regions, allowing a few bacteria to swim around the particles,
exclusively in clockwise trajectories (Fig. 1c&1d). Such orbiting motion lasts over 10 min during
observation, much longer time than expected for a water film a few micrometers in thickness to disappear
due to evaporation.
We observed common orbiting motion among a wild type strain of E. coli (HCB33) and three
recently isolated strains of bacteria, including two novel enterobacter isolates, SM1 and SM3,
and one E. coli isolate, H5 (details in the Methods). For the two E. coli strains, we used agar
containing added surfactant Triton X-100 (Sigma-Aldrich) of 0.1% vol/vol in order to facilitate
the colony spread. We also experimented with Pseudomonas aeruginosa (PA01) and Bacillus
subtilis (3610). In these two latter cases, we also noticed the long-lasting water layers around
the micrometer-sized particles where bacteria were motile. For P. aeruginosa, we observed
orbiting around the micro-sized particle, but the motion was less organized, i.e. bacteria tended
to go off track from their orbits within a few seconds, possibly due to this species of bacteria
frequently reversing flagella motor rotation direction [22]. B. subtilis, which are longer cells (4-
10 μm in length) remained motile within the water meniscus, but they did not orbit well. These
results suggest universal motion of bacteria entrapped around microsized particles on moist
surface, although the exact trajectories are species dependent. From here on, we focus the
remaining report on Enterobacter sp. and E. coli, which reliably form circular trajectories.
Small graphite particles (< 2.5 µm in size) are usually not surrounded by swimming cells, possibly
because they could not retain the water reservoir large enough to entrap bacteria. Large graphite particles
(> 12 µm in size) retain liquid in a region over 40 µm in radius and thus keep many bacteria swimming
in their proximity. Intermediate graphite particles were frequently seen to be surrounded by a few active
swimmers and were the ones around which the orbiting of bacteria was most often observed and
measured (Fig. 1d). In cases where the number of swimmers around a dispersed particle is big (over a
dozen, as is commonly seen around large graphite particles or particles immersed in a dense bacterial
population), the cells frequently bump into each other, making it difficult to discern their trajectories. We
chose silica beads of uniform size (4.5 µm diameter; Bangs Laboratories, Inc.) to perform the same
experiment and observed equivalent results between the beads (Fig. 2a) and the graphite particles of
3
comparable size, suggesting that the key parameter is the particle size, not its chemical nature. The
common role of the graphite particles or beads is to sustain a tent-shaped water reservoir around them.
Around the particles where bacteria were observed to orbit, the radii of meniscus of water and bacterial
trajectories correlate to the particle size. For the silica beads, which are uniform in dimensions, the
meniscus and trajectory radii vary within narrow ranges dependent on the extent of evaporation. Graphite
particles, which were more variable in size and had irregular shapes, gave rise to larger variation in
meniscus and trajectory radii (Fig. 2b&2c).
Discussion
Hydrodynamic cause for the circular trajectories.
The circular trajectories are caused primarily by the solid confinement imposed by the agar surface. The
orbiting bacteria were observed to be in the same focal plane as the bacterial cells stuck in the nearby dry
regions. The surface contour of graphite particles had minimal effects on the bacterial trajectories with
exceptions of close encounters. In some cases bacteria were seen to swim within a body length from the
center particle, but often orbiting bacteria swam over 8 µm away from it, a distance by which the
hydrodynamic effect of the graphite boundary becomes negligible [5]. These observations reinforce the
argument that the circular trajectories are caused by the proximity of the bacteria to the bottom surface.
The observation that Enterobacter sp. move in the same clockwise direction as E. coli allows us to
conclude that both species of bacteria have left-handed helical flagella.
Measurements on all four strains show a relationship between the speed of the orbiting bacteria and the
radii of curvature of their trajectories. Bacteria swam slower in trajectories of smaller radii, but beyond
a certain threshold (~4 µm) the speed approached an upper limit, between 30-50 µm/s, depending on the
bacterial strain (Fig. 3a). These results indicate that graphite particles or beads only affect the swimming
speed by restricting the bacteria to their close proximity via the tent-shaped liquid film around them.
The experimental speed profile agrees with a hydrodynamic model described previously [23] (details on
its implementation discussed in Methods). Although it is possible to shift the theoretical line (Fig. 3a),
by adjusting the parameters for the dimensions of the model bacterium and the flagellar motor rotation
speed, the key feature remains the same: the speed initially increases with the radius of curvature, but
quickly attains an upper value. The consistent trend of the data provides further support that the circular
4
motion observed in the orbiting bacteria is mainly caused by the hydrodynamic boundary effect of the
bottom surface (Fig 3b).
Endurance of the water layer around the seeding particle.
One key finding of our study is that graphite or bead particles on the agar surface retain a layer of water
around them much longer than on a non-permeable surface. The evaporation of a water layer from a
borosilicate glass surface containing microspheres on top has been investigated experimentally using x-
ray transmission microscopy [24] (polystyrene beads of 6 µm diameter). The full process lasts about 12
seconds, with 96% of the time on a pinning stage when the water film surrounding the bead thins
gradually, and the last 4% on depinning of the water meniscus around the microsphere in the form of a
sudden collapse. In our case, however, the fact that the bottom surface is made by agar gel (>97.5%
water) allows for water to constantly permeate the upper surface. Thus, the meniscus lasts much longer
(Figure 4). This observation is explained as a balance between evaporation and additional water pumped
from the agar gel to the layer. Given the tiny thickness (a few µm) of the water layer, its internal pressure
P' is roughly constant everywhere. The internal pressure within the water layer (P') is lower than the
atmospheric pressure Po [25]. The difference between these two pressures is given by the Young-Laplace
equation, Po - P' = σ/R, where σ is the water surface tension and 1/R is the meniscus total curvature. Near
the top of the agar gel, the pressure is also close to the atmospheric pressure. Hence, a pressure gradient
is established between the permeable agar gel and the water layer, giving rise to water being pumped, as
expected from Darcy's law [26]. Consequently, even though water is constantly evaporating from the
thin layer, it is being replenished by that from the agar gel.
Conclusion
In summary, we report a novel mechanism by which micro-sized particles on a moist surface entrap a
tent-shaped layer of liquid within which flagellated bacteria swim in circular trajectories. The
phenomenon occurs due to strong surface tension at the liquid-air interface and the proximity of the cells
to a semi-permeable agar gel. Our finding implicates the potential for persistence of bacteria in
environments with restricted water supply. For instance, certain bacteria cooperate with plants to
maintain their growth and longevity through droughts [27]. The observation that small particles in a wet
environment can entrap bacteria and allow them to remain viable and motile, even after most of the
region has dried, raises awareness on the importance of proper hygiene habits to prevent bacterial
infection [28,29].
5
Fig. 1. Circular motion of bacteria around particles on agar surface. (a) Schematics representing a
bacterial swarm (in purple) with a water droplet (containing graphite particles or silica beads) added to a
location overlapping the swarm front. A region in the droplet is amplified to illustrate bacterial
trajectories around the dispersed particles. (b) Microphotograph showing different features following the
water droplet evaporation. The swarm front is marked by the red dashed line. Most of the region beyond
has dried and is populated by bacteria stuck to the surface, except for those around graphite particles. (c)
Artistic representation of bacterial cells moving around a dispersed particle within a tent-like layer of
water around it. (d) Time lapse following the trajectories of bacterial cells orbiting a graphite particle.
Three cells in particular are highlighted with different colors and geometrical contours to aid
visualization. The sequence goes from left to right and the pictures shown were taken approximately 0.47
s apart. These images were extracted from Supplementary Movie 1.
6
Fig. 2. Image of meniscus of water layer and the measured dependence of both meniscus and
trajectory radii on particle size. (a) Microphotograph highlighting the meniscus edge of water (radius
Rm) around silica beads. Bacteria outside these circles are stuck in dried regions, whereas a few bacteria
within these circles orbit around the beads (Supplementary Movie 2). (b) Relationship between the size
of dispersed particles and the cutoff radius of their surrounding water meniscus. Each data point
represents an individual observation. (c) Relationship between the size of the dispersed particles (graphite
or beads) and the radius of trajectories of the orbiting bacteria.
7
Fig. 3. Measured speed versus radius of trajectory in comparison with numeric prediction by
applying a simple hydrodynamic model. (a) Speed of bacteria with respect to the radius of curvature
of trajectories for the four strains tested. The red line is the prediction by applying the Resistive Force
Theory (with parameters specified in the Methods). The lower end of the line is set by the 5 nm minimal
gap chosen between the cell body and the bottom surface in the model. Each data point represents a
measurement for an individual bacterium. (b) Schematics showing the curvilinear trajectory of a
bacterium when it swims close to a solid boundary. The cell body and its flagellar bundle rotate in
opposite directions. The larger drag on the cell body at the location facing the surface exerts a lateral
force in the opposite direction to that on the flagella (forces indicated in the figure). The net torque leads
to a circular trajectory.
8
Fig. 4. Schematics illustrating a resilient water layer around a micron-sized particle on an agar gel.
The internal pressure (P') within the layer of water is lower than the atmospheric pressure (Po), which is
also roughly the pressure inside the agar gel close to the top surface. This pressure difference pumps
water up from the semisolid agar gel, counterbalancing constant evaporation. Hence, the water layer is
maintained much longer than on typical non-permeable solid surfaces. The thickness d corresponds to
the width of a bacterial cell, with its location defining a meniscus boundary outside which bacteria get
stuck to the agar surface.
9
Materials and Methods
Bacterial strains
Six strains of bacteria were used. Strain HCB33 (wild type E. coli) was obtained from Howard Berg
(Harvard University). P. Aeruginosa PA01 was obtained from Keiko Tarquinio (Emory University
Medical School). B. subtillis 3610 was obtained from Daniel B. Kearns (Indiana University). The
remaining three strains were isolated at the Albert Einstein College of Medicine (NYC). H5, a swarming
strain of E. coli from human feces (Committee on Clinical Investigation CCI# 2009-446-006; IRB# 2015-
4465). SM1 and SM3 are novel strains of swarming Enterobacter sp. isolated from mouse feces (presently
undergoing species characterization).
Bacterial culture
Bacteria were grown in suspensions of Lysogeny Broth (LB) medium overnight (~ 16 hours) and then a
5 µl droplet was taken to inoculate the center of a petri dish containing LB agar gel (LB + 0.5% agar).
Triton-X100 (Sigma-Aldrich) was added (0.1% in volume) to the LB agar as external surfactant for E.
coli inoculation in order to help with its colony spread. In particular, it was observed for the H5 strain
that the colony spreads in similar rate to what is seen in typical swarming colonies: it populates the whole
petri dish (~10 cm in diameter) within a few hours. For the HCB33, the effect was not as dramatic and
the colony grew only slightly faster compared to situations when no external surfactant is added to the
growth medium.
The samples were placed into an incubator with controlled temperature of 37 oC and 50% relative
humidity for 3 to 5 hours, depending on the growth rate. In the next step, a droplet containing micrometer-
sized particles (graphite or silica beads) was placed at a point overlapping the colony, as illustrated in
Figure 1a. Graphite particles were obtained by scrapping a pencil with a razor blade and dispersed in
deionized water (DI water). Silica beads (4.5 µm in diameter, Bangs Laboratories, Inc) were also diluted
in DI water. After the addition of the water droplet, the petri dish was gently tilted in different directions
to help the droplet to spread away from the bacterial colony. This motion created thin water layers that
evaporated within one minute on the agar surface. Most bacteria then got stuck on the dried surface,
leading a small number of swimming bacteria trapped in water reservoirs around the dispersed particles.
Imaging and measurements
10
The droplet region was imaged under an upright microscope (Nikon Eclipse E800) with a 40x objective
lens, coupled to a CCD camera (Photometrics Coolsnap EZ). The imaging was controlled by the software
Metamorph (Molecular Devices) from which the movies were recorded. The recording rate was in the
range between 10.72 frames per second (fps) to 33.88 fps, with lower frame rates chosen for bigger
imaging area.
Radius of meniscus was determined by defining the nearby immobilized bacteria as the cutoff perimeter,
as indicated on Figure 2a. Bacterial trajectories and their swimming speeds were calculated using ImageJ
from their captured positions in video clips. The numerical calculation applying the Resistive Force
model was implemented in a Python code.
Hydrodynamic model
E. Lauga et. al proposed a model based on Resistive Force theory to represent bacteria moving close to
solid boundaries [23]. It models a bacterium as formed by a sphere connected to a helical rod, to represent
the cell body and flagella bundle, respectively. The free swimming organism is subjected to zero net
force and torque and is propelled by the rotating flagella bundle. Body and flagella are characterized by
their respective mobility matrices which appear in the linear system of equations, equation (1), given the
velocities and rotation rates of the cell body in equations (2) and (3), respectively.
(A+B)X = BY
where,
(1)
X = (Ux, Uy, Uz, Ωx, Ωy, Ωz)T
Y = (0, 0, 0, 0, ω, 0)T
(2)
(3)
Matrix A is the mobility matrix of the cell body. B is the mobility matrix of the flagella. Ui gives the i-
th component of the velocity and Ωi the i-th component of the rotation rate of the cell body. The rotation
rate of the flagella is given by ω. The cell is set to initially align with the y axis, without loss of generality.
Reference [23] gives the details on how to build the mobility matrices.
11
The velocities and rotation rates in this model are dependent on the gap distances between the surface
and the material parts of the model bacterium. The speed U of the circular moving bacterium is given
by the components along the bottom surface: U = (Ux2 + Uy2)1/2 and the radius of curvature R of the cell
trajectory is also obtained from the solution of the linear system: R = U/Ωz.
Acknowledgements: We acknowledge funding from NSF DMR 1505878 (JXT), NIH R01 CA
CA161879 (SM), and CAPES BEX 13733-133 (GA). We thank Neha Mani for the initial idea
of her school project that led to the experimental design. We also thank Yinan Liu for her
assistance on graphic illustration.
Author Contributions: GA, WC & JXT conceived the study. GA and WC performed the
experiments and analysis. SM produced and provided the bacteria (SM1, SM3 and H5). GA
and JXT wrote the paper with contributions from all co-authors.
12
References
[1] Lauga E (2016) Bacterial hydrodynamics. Annu Rev Fluid Mech 48, 105 -- 130.
[2] Li G, et al. (2011) Accumulation of swimming bacteria near a solid surface. Phys Rev E 84
041932.
[3] Berke AP, Turner L, Berg HC, Lauga E (2008). Hydrodynamic attraction of swimming
microorganisms by surfaces. Phys Rev Lett 101 038102.
[4] Li G, Tang JX (2008) Accumulation of microswimmers near a surface mediated by collision and
rotational brownian motion. Phys Rev Lett 103 078101 .
[5] Drescher K, Dunkel J, Cisneros LH, Ganguly S, Goldstein RE (2011) Fluid dynamics and noise in
bacterial cell-cell and cell-surface scattering. Proc Natl Acad Sci 108 10940-10945 .
[6] Frymier PD, Ford RM, Berg HC, Cummings PT (1995) Three-dimensional tracking of motile
bacteria near a solid planar surface. Proc Natl Acad Sci 92 6195-6199 .
[7] Vigeant MA, Ford RM (1997) Interactions between motile Escherichia coli and glass in media
with various ionic strengths, as observed with a three-dimensional-tracking microscope. Appl
Environ Microbiol 63 3474 -- 3479.
[8] DiLuzio, WR et al. (2005) Escherichia coli swim on the right-hand side. Nature 435 1271 -- 1274 .
[9] Männik J, Driessen R, Galajda P, Keymer JE, Dekker C (2009) Bacterial growth and motility in sub-
micron constrictions. Proc Natl Acad Sci 106 14861-14866.
[10] Biondi SA, Quinn JA, Goldfine H (1998) Random motility of swimming bacteria in restricted
geometries. AICHE J. 44 1923-1929.
[11] Hennes M, Tailleur J, Charron G, Daerr A (2017) Active depinning of bacterial droplets: the
collective surfing of Bacillus subtilis. Proc Natl Acad Sci 114 5958-5963 .
[12] Sokolov A, Rubio LD, Brady JF, Aranson IS (2018) Instability of expanding bacterial
droplets. Nature Comm 9 1322.
[13] Wioland H, Woodhouse FG, Dunkel J, Kessler JO, Goldstein, RE (2013) Confinement
stabilizes a bacterial suspension into a spiral vortex. Phys Rev Lett 110 268102.
[14] Tuval I, et al. (2005) Bacterial swimming and oxygen transport near contact lines. Proc Natl
Acad Sci 102 2277 -- 2282.
[15] Sokolov A, Aranson IS (2016) Rapid expulsion of microswimmers by a vortical flow. Nature
Comm 7 11114.
13
[16] Turner L, Zhang R, Darnton NC, Berg HC (2010) Visualization of flagella during bacterial
swarming. Journal of Bact 192 3259-3267.
[17] Zhang HP, Be'er A, Florin EL, Swinney HL (2010) Collective motion and density fluctuations in
bacterial colonies. Proc Natl Acad Sci 107 13626-13630.
[18] Copeland MF, Weibel DB (2009) Bacterial swarming: a model system for studying dynamic
self-assembly. Soft Matter 5 1174-1187.
[19] Dunne Jr WM Bacterial adhesion: seen any good biofilms lately? (2002) Clin Microbio Rev
15 155-166.
[20] Jin F, Conrad JC, Gibiansky MS, Wong GCL (2011) Bacteria use type-IV pili to slingshot
on surfaces. Proc Natl Acad Sci 108 12617-12622.
[21] O'Toole GA, Kolter R (1998) Flagellar and twitching motility are necessary for
Pseudomonas aeruginosa biofilm development. Mol Microbiol 30:295-304.
[22] Cai Q, Li Z, Ouyang Q, Luo C, Gordon VD (2016) Singly flagellated Pseudomonas
aeruginosa chemotaxes efficiently by unbiased motor regulation. MBio. 7 e00013-1.
[23] Lauga E, DiLuzio WR, Whitesides GM, Stone HA (2006) Swimming in circles: motion of bacteria
near solid boundaries. Biophys Journal 90 400 -- 412.
[24] Cho K, et al. (2016) Low internal pressure in femtoliter water capillary bridges reduces evaporation
rates. Scientific Reports 6 22232.
[25] Honschoten JW, Brunets N, Tas NR (2010) Capillary at the nanoscale. Chem Soc Rev, 39 1096 --
1114.
[26] Cunningham R (1980). Diffusion in Gases and Porous Media. Springer US.
[27] Xu L, et al. (2018) Drought delays development of the sorghum root microbiome and enriches for
monoderm bacteria. Proc Natl Acad Sci 115 E4284-E4293.
[28] Grice EA, Segre JA (2011) The skin microbiome. Nat Rev Microbiol 9 244-253.
[29] Ottemann KM, Miller JF (1997) Roles for motility in bacterial host interactions. Mol Microbiol
24 1109-1117.
14
|
1201.2792 | 3 | 1201 | 2013-07-15T01:33:24 | Measurement of Photon Statistics with Live Photoreceptor Cells | [
"physics.bio-ph",
"physics.optics",
"quant-ph"
] | We analyzed the electrophysiological response of an isolated rod photoreceptor of Xenopus laevis under stimulation by coherent and pseudo-thermal light sources. Using the suction electrode technique for single cell recordings and a fiber optics setup for light delivery allowed measurements of the major statistical characteristics of the rod response. The results indicate differences in average responses of rod cells to coherent and pseudo-thermal light of the same intensity and also differences in signal-to-noise ratios and second order intensity correlation functions. These findings should be relevant for interdisciplinary studies seeking applications of quantum optics in biology. | physics.bio-ph | physics | Measurement of Photon Statistics with Live Photoreceptor Cells
Nigel Sim1, Mei Fun Cheng1, Dmitri Bessarab2, C. Michael Jones2, and Leonid A. Krivitsky1,∗
1 Data Storage Institute, Agency for Science Technology and Research (A-STAR), 117608 Singapore
2 Institute of Medical Biology, Agency for Science Technology and Research (A-STAR), 138648 Singapore
3
1
0
2
l
u
J
5
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
3
v
2
9
7
2
.
1
0
2
1
:
v
i
X
r
a
We analyzed the electrophysiological response of an isolated rod photoreceptor of Xenopus laevis under
stimulation by coherent and pseudo-thermal light sources. Using the suction electrode technique for
single cell recordings and a fiber optics setup for light delivery allowed measurements of the major
statistical characteristics of the rod response. The results indicate differences in average responses of
rod cells to coherent and pseudo-thermal light of the same intensity and also differences in signal-to-
noise ratios and second order intensity correlation functions. These findings should be relevant for
interdisciplinary studies seeking applications of quantum optics in biology.
PACS numbers: 42.50.Ar, 42.66.Lc, 87.80.Jg
Eyes of living organisms represent advanced light har-
vesting systems, developed through hundreds of millions
years of evolution. Some of their features are compa-
rable or even superior to existing man-made photode-
tection devices. For example, rod photoreceptor cells
of the retina, which are responsible for night vision and
form the focus of the present study, represent miniatur-
ized photodetectors containing a photosensitive element
(rhodopsin pigment) along with a "built-in" chemical
power supply (ATP produced by mitochondria) [1]. They
have sensitivity down to single-photon level, and demon-
strate a remarkable low-noise operation [2]. Under-
standing such properties of nature-given photodetectors
stimulates considerable interest in interfacing them with
sources of non-classical light, such as light with a "fixed"
number of photons, and "squeezed" light [3]. Implemen-
tation of such "bio-quantum" interfaces would contribute
to a better understanding of the performance of the visual
system near the threshold [4], allow a reference-free cali-
bration of its quantum efficiency [5], along with a direct
detection of entangled multi-photon states [6]. Study of
the statistics of rod responses to excitations by classical
light sources allows careful characterization of rods, and
open possibilities for future studies of non-classical light.
Vertebrate retinal photoreceptor rod cells, further re-
ferred to as rods, are densely packed in the retina laying
the inner surface of an eyeball [1]. Individual rods con-
sist of two distinctive functional regions: extended rod-
like outer segment (OS) which is filled with rhodopsin
photopigment molecules, and shorter rounded inner seg-
ment (IS) which contains various components of cell ma-
chinery. In the dark, the electrochemical gradient across
the rod cell membrane causes a continuous flux of ions
(N a+, K +, Ca2+) in and out of the cell through aque-
ous pores in the membrane, referred to as ion channels.
A photon, impinging on the OS, causes isomerization of
a rhodopsin molecule, which initiates a chain of intra-
cellular reactions, referred to as phototransduction cas-
cade. A fraction of ion channels closes, thus resulting
in hyperpolarisation of the cell. Illumination with a rel-
atively bright flash closes all light sensitive channels in
the OS, and bleaches the rod. A recovery process regu-
lates re-opening of the ion channels to the resting state.
Accurate measurement of the ion current flux through
the rod membrane is possible with the suction-electrode
technique [7]. In this case, the OS of the rod is drawn in
a tight fitting glass microelectrode filled with the phys-
iological solution. The rod functions in a way similar
to in-vivo system, however the ion current through the
membrane is now re-directed to a low noise amplifier.
In earlier experiments isolated rods were mainly stim-
ulated by multi-mode thermal sources (lamps, LEDs),
whose statistics, however, was not within the scope of in-
terest of those works [8–11]. Influence of well controlled
light statistics on the response of the entire visual system
was studied only in behavioral experiments with human
subjects [4, 12]. However, noise properties of individual
rods are difficult to access in this case, as several hun-
dred of rods are simultaneously illuminated, and their
collective response undergoes several intermediate stages
of the visual processing. In the present study, influence
of controllable photon fluctuations on the response of an
isolated rod is analyzed for the first time. Using well
characterized coherent and pseudo-thermal light sources,
together with a fiber optics setup for light delivery al-
lows us to determine major statistical characteristics of
the rod in the whole dynamic range of its response.
The probability distribution of photon numbers is
given by [13]:
Pph(m) =( mme−m
m!
mm
(1+m)m+1 ,
,
for the coherent source
for the thermal source
(1)
where m is the average number of photons. Based on
earlier findings, it is further assumed that each photon
isomerizes only a single rhodopsin molecule [14]. In this
case the probability of isomerization of n molecules by
m impinging photons is given by the binomial distribu-
tion: P (nm) = (cid:0)m
as an overall quantum efficiency of the photodetection
process [13]. Then, the probability of isomerization of n
n(cid:1)ηn(1 − η)m−n, where η is defined
2
molecules by the fluctuating light source is given by:
PI (n) =
∞Xm=0
P (nm)Pph(m).
(2)
It is further assumed, that in the absence of saturation
subsequent electrophysiological responses to individual
isomerizations are additive, and have a standard Gaus-
sian shape [8]. In this case, the probability of observing
the amplitude of the photocurrent A in response to n
isomerizations is given by:
eP (An) =
p2π(σ2
1
D + nσ2
A)
exp(cid:18)−
(A − nA0)2
2(σ2
D + nσ2
A)(cid:19), (3)
where A0 is the average, and σA is the standard deviation
of the photocurrent amplitude in response to a single
isomerization, and σD is the standard deviation of the
photocurrent in darkness. Hence the final probability
of observing the photocurrent amplitude A is given by
averaging over PI (n):
P (A) =
∞Xn=0eP (An)PI (n).
(4)
In order to account for the saturation of the photocur-
rent, P (A) is truncated at the saturation amplitude
AS, so that the probability of observing the amplitude
A > AS just adds up and results in observing AS:
PS(A) =
P (A),
1 −R AS
0,
A=0 P (A)dA,
if A < AS
if A = AS
if A > AS
(5)
Eq.
(5) allows calculation of any statistical mo-
ment of the photocurrent amplitude, given by Ak =
A=0 AkPS(A)dA. The relation between the average
2
A and the variance VarA ≡ A2 − A
, can be conve-
niently characterized by a dimensionless signal-to-noise
R ∞
ratio SNR ≡ A/pVar(A).
Investigation of joint statistics of the rod response with
some reference detector, measured in a Hanbury-Brown
and Twiss (HBT) type experiment [15, 16], allows finding
the second order intensity correlation function of the light
field g(2). The value of g(2) may evidence non-classical
properties of light [13]. From the joint pulse-to-pulse data
of photocurrent amplitudes of the rod and photocounts
of a conventional avalanche photodiode (APD), g(2) is
found as:
g(2) ≡
AK
A K
= R ∞
A=0P∞
K=0 AKP (K)PS(A)dA
,
(6)
A K
where K = ηDm is the average number of APD photo-
counts, ηD is the APD quantum efficiency, and P (K) is
the probability distribution of APD photocounts, which
is similar to (1).
FIG. 1: (color online) (a) Optical layout. A laser beam is
chopped by a shutter, attenuated by a neutral density filter
(NDF), and split by a 50/50 beamsplitter (BS). Both of the
beams are coupled into single mode optical fibers (SMF) with
lenses (L). One of the fibers is connected to an avalanche
photodiode (APD) whilst the other stimulates the rod. Inset
shows the pseudo-thermal light source: L is a lens, GD is a
rotating ground glass disc, D is a diaphragm. (b) Microscope
image (top view) of the fiber taper and a suction micropipette
with the constrained rod cell.
In the experiment (see Fig.1a) the beam of a 50 mW
continuous wave frequency-doubled Nd:YAG laser at 532
nm (Photop, DPL-2050) was chopped by an optical shut-
ter (Melles Griot) in 30 ms duration pulses with a repe-
tition rate of 10-15 seconds. The pseudo-thermal source,
shown in the inset of Fig.1a, was realized by focusing
the beam of the same laser with a lens (f =100 mm) into
a rotating ground glass disc (GD, 1500 grit), followed
by a 0.5 mm iris diaphragm (D), placed at a distance
1200 mm [17]. The diameter of the diaphragm was set
to be much smaller than the size of a single scattered
speckle. Then, either the laser beam (for the experiment
with the coherent source) or a speckle (for the experi-
ment with the pseudo-thermal source) was attenuated by
a variable neutral density filter (NDF), and directed to a
HBT setup, consisting of a non-polarizing 50/50 beam-
splitter (BS) and two single mode optical fibers (SM 460-
HP) with equalized coupling efficiencies preceded by as-
pheric lenses (f =6.24 mm). One of the fibers was used
to stimulate the cell, whilst the other one was connected
to an actively quenched APD (Perkin Elmer, SPCM-
AQRH-15-FC). The number of photons used to stimu-
late the cell was calculated from the number of APD
photocounts in each pulse, assuming the APD quantum
efficiency ηD =50% [18]. The statistics of coherent and
pseudo-thermal sources are checked in independent mea-
surements by the APD (see supplementary materials).
Methods of cells preparation, electrophysiology record-
ings, and light coupling were similar to the ones we de-
scribed previously [19]. In brief, dark adapted adult male
frogs Xenopus laevis were used for the experiment. All
procedures with the animals were carried out in accor-
dance with IACUC regulations. Shortly after the sacri-
fice, the retinas were taken from the eyes under IR-light
(850 nm) and sliced with a blade to release single rods.
The rods were loaded into a Ringer solution filled cham-
ber of an inverted microscope (Leica, DMI 3000), which
was placed in a light tight Faraday cage. The micro-
scope was equipped with an IR LED (920 nm), and a
CCD camera (Leica, DFX 380) to allow observation of
the preparation without bleaching the rods. A glass mi-
cropipette, pulled to the diameter of ≈5µm and bended
by about 45◦ was used for conventional suction electrode
recordings [7]. OS of intact rods were captured in the
pipette by gentle air suction, see Fig.1b. The photocur-
rent pulses were amplified and digitized by a patch-clamp
amplifier (Heka, EPC 10), the data were stored by the
computer. The experiment was synchronized by an ex-
ternal pulse generator, which triggered data acquisition
by the rod and the APD, and conducted at room tem-
perature. Each cell was used for continuous recordings
during 90-130 min; 1-2 cells per preparation were used.
The consistency of the response during the experiment
was checked prior to acquisition of every point by excit-
ing the saturation amplitude AS with a dim torch [7].
A tapered single mode optical fiber (Nanonics, working
distance 25-30 µm, spot size 4 µm), mounted on a motor-
ized stage (Sutter, resolution 0.3 µm), was used for light
delivery to the rod (see Fig.1b). The fiber tip and the
rod were carefully positioned at a distance of 25-30 µm,
so that the stimulus were sent along the rod long axis.
Such an "axial configuration" provides a nearly optimal
match between the light spot and the rod profile, long
absorption path in the rod, and an invariance of the rod
response to light polarization [19].
Actual waveforms of the rod photocurrent in response
to stimulation by different numbers of impinging pho-
tons, produced by the coherent source, are shown in the
inset of Fig.2. In addition, a histogram of rod responses
to dim coherent light flashes is shown in Fig.6 of supple-
mentary materials. The dependence clearly indicates the
quantization of the response, which supports the notion
that rods function as photon number resolving detectors
[2], in accordance with the suggested model. The de-
pendencies of the normalized average photocurrent am-
plitudes, on the average numbers of photons for coherent
and pseudo-thermal sources are shown in Fig.2. Ampli-
tudes were normalized by the saturation value AS, and
the numbers of photons were normalized by m0, the value
which initiated a response of a half saturation amplitude,
which, for the majority of studied cells, was in the range
of 550-2,500 photons/pulse. Lines represent numerical
calculations of the averages, with the probability distri-
bution given by (5). The dependencies for the major-
ity of investigated cells yielded parameters in the range:
AS ∈[18,26] pA, A0 ∈[0.5,0.85] pA, σA ∈[0.25,0.6] pA,
S
A
A
/
1.0
_
,
e
d
u
t
i
l
p
m
a
d
e
z
i
l
a
m
r
o
n
0.8
0.6
0.4
0.2
0.0
Photons per pulse
16000
9500
2400
1550
980
370
30
A
p
,
t
n
e
r
r
u
c
20
16
12
8
4
0
0
1
2
3
time,s
4
5
6
0.01
normalized number of photons, m/m0
0.1
1
_
3
10
FIG. 2: (color online) Dependence of the normalized average
amplitude of the rod photocurrent on the average number of
impinging photons for coherent (red squares, solid line) and
pseudo-thermal (black circles, dashed line) sources. Each data
point is an average response to 80-100 light pulses. 6 different
cells, taken from 6 different animals were used. Lines show
numerical calculations, and standard deviation is shown by
error bar. Inset shows raw waveforms of the rod photocurrent
in response to coherent pulses of different average intensity.
σD ∈=[0.1-0.2] pA, and η ∈[0.007,0.01]. Note that η ac-
counts for an overall efficiency of the photodetection in a
given experiment.
The dependencies in Fig.2 demonstrate that satura-
tion of the average amplitude for the coherent source is
significantly steeper than for the pseudo-thermal one. A
clear explanation for the difference can be found from the
experimental histograms of the normalized amplitudes,
which are shown for both of the studied light sources in
Fig.3. At relatively low number of impinging photons
rod operates in a linear regime, and the statistics of its
responses displays the statistical features of the incoming
light (Fig.3 a,c). However, when the rod is stimulated by
bright flashes, its saturation causes damping of response
fluctuations and modifies the light statistics (Fig.3 b,d).
For the coherent source photon numbers are well local-
ized around their average value (see (1)), and the ma-
jority of response amplitudes to bright flashes are close
to the saturation amplitude AS (Fig.3b). In contrast, for
the pseudo-thermal source with the same average photon
number, the probability of emission of pulses with rela-
tively low photon numbers remains non-negligible (see
(1)). Hence the amplitude probability distribution gets
significantly broader than in the case of the coherent
source (Fig.3b), resulting in smoother saturation of the
average response, as shown above (see Fig.2).
Note that the capability of the entire visual system of
living human subjects to distinguish between the light
sources of different statistics was shown in [4]. Our ex-
periments suggest that responses of single rods are at
the basis of this phenomena. Our results also imply that
the light source characteristics should be carefully con-
y
t
i
l
i
b
a
b
o
r
p
0.3
0.2
0.1
0.0
0.5
0.4
0.3
0.2
0.1
0.0
0.0
0.2
(a)
_
A=0.13AS
SNR=1.85
(b)
_
A=0.98AS
SNR=14.5
(c)
_
A=0.108AS
SNR=0.997
(d)
_
A=0.8AS
SNR=2.6
0.6
1.0
0.8
0.8
0.4
normalized amplitude, A/As
0.0
0.2
0.4
0.6
0.8
0.6
0.4
0.2
0.0
0.3
0.2
0.1
0.0
1.0
FIG. 3: (color online) Probability histograms of normalized
amplitudes measured for coherent (a,b) and pseudo-thermal
(c,d) sources at different values of the average, shown with
the corresponding SNR values. Each histogram was acquired
for 80-100 light pulses. Standard deviation is shown by error
bar.
sidered for faithful comparison of experiments on visual
transduction in rods.
The effect of saturation is evident from the experimen-
tal dependence of the SNR on the number of impinging
photons, shown along with numerical calculations using
(5), in Fig.4a. The raw data for the measurements of
SNR, obtained for 6 different cells, are shown in sup-
plementary materials (see Fig.7).
In the linear regime
of the rod response, SNR grows linearly for the coherent
source, and remains constant for the pseudo-thermal one.
At the same time, it demonstrates a steeper growth in the
saturation region for both of the sources. Note that sat-
uration of rod should be considered in experiments with
non-classical light. In particular, saturation might lead
to experimental artifacts, such as observation of sub-shot
noise fluctuations ("squeezing") in experiments with clas-
sical light sources [5].
0
The experimental dependencies of the correlation func-
tion g(2)
on the number of photons for coherent and
pseudo-thermal light sources, along with numerical cal-
culations, are shown in Fig.4b. The result for the coher-
ent source yields g(2)
0 =1.01±0.01 and does not demon-
strate any dependence on the number of impinging pho-
In contrast, for the pseudo-thermal source, g(2)
tons.
decreases with the increasing number of photons. Thus
"excess" correlations of the thermal source are gradually
neglected due to the saturation of the rod response. As
an additional check, in the experiment with the pseudo-
thermal source and for the linear regime of the rod re-
sponse, g(2)
τ was calculated between the photocurrent am-
plitude and the number of APD counts for consecutive
light pulses. One observed a drop from g(2)
0 =1.94±0.07,
to g(2)
τ =0.99±0.02, which is expected as consecutive light
0
4
(a)
102
R
N
S
101
1
2.0
(b)
)
2
0
(
g
1.5
1.0
0.01
10
normalized number of photons, m/m0
0.1
1
FIG. 4: (color online) Dependence of the SNR (a) and g
(b) on the average number of impinging photons for coherent
(red squares, solid line) and pseudo-thermal source (black cir-
cles, dashed line). Results are averaged similarly as in Fig.2.
Standard deviation is shown by error bar, which can not be
seen for the coherent source in (b). Lines show numerical
calculations.
(2)
0
pulses are uncorrelated. Thus it is shown that rods al-
low adequate measurement of g(2)
similar to conventional
man-made detectors, and could be readily used for char-
acterization of quantum light.
0
In conclusion, being inspired by the ultimate charac-
teristics of rod photoreceptors, we carefully investigated
the impact of photon fluctuations of various classical
light sources on their response. The experimental re-
sults revealed capabilities of isolated rods in measure-
ment of photon statistics. It is of future interest to in-
vestigate rods interfaces with nonclassical light, in partic-
ular with correlated two-photon light and intense multi-
photon twin-beam states. The developed approach can
be also used in interfacing rods with realistic sources of
intensity fluctuations, for example blinking star lights,
when observed through turbulent atmosphere [20].
We would like to acknowledge M. Chekhova, M. Tsang,
K. Eason, S.-H. Tan, and V. Volkov for stimulating dis-
cussions. This work was supported by A-STAR Joint
Office Council grant and A-STAR Investigatorship grant.
∗ Electronic address: [email protected]
[1] J.E. Dowling, The Retina (Harvard University Press,
1987).
[2] F. Rieke, and D. A. Baylor, Rev. Mod. Phys. 70, 1027-
5
1036 (1998).
[3] D. N. Klyshko, Phys.-Usp. 39, 573-596 (1996).
[4] M.C. Teich, et al., Biol. Cybern. 44, 157-165 (1982).
[5] I. Agafonov, et al., Optics Letters 36, 1329 (2011).
[6] E. Pomarico et al., New J. Phys. 13, 063031 (2011).
[7] D.A. Baylor, et al., J. Physiol. 288, 589-611 (1979).
[8] D.A. Baylor, et al., J. Physiol. 288, 613-634 (1979).
[9] T.D. Lamb, JOSA A l4, 2295-2300 (1987).
[10] F. Rieke, and D.A. Baylor, Biophys.J. 75 1836-1857
(1998).
[11] G.D. Field et al., Annu. Rev. Physiol. 67, 491-514 (2005).
[12] M.C. Teich, et al., JOSA 72, 419-431 (1982).
[13] L. Mandel, E. Wolf, Optical coherence and quantum op-
tics (Cambridge University Press, 1995).
[14] D. Baylor Proc. Natl. Acad. Sci. USA 93, 560-565 (1996).
[15] R. Hanbury Brown and R. Q. Twiss, Nature 178, 1046
(1956).
[16] D.N.Klyshko Physical foundations of quantum electronics
(World Scientific, Singapore, 2011).
[17] F.T. Arecchi, Phys. Rev. Lett. 15, 912-916 (1966).
[18] The number of photons per pulse never exceeded 3 · 104,
thus it was perfectly resolved by the APD with a dead
time τd ≈ 35 nsec.
[19] N.Sim, et al., Biomed. Optics Express, 2, 2926 (2011).
[20] A. Consortini, and G. Conforti, JOSA A, 1, 1075 (1984).
|
1911.01367 | 1 | 1911 | 2019-11-04T17:51:25 | Stochastic model for the CheY-P molarity in the neighbourhood of E. coli flagella motors | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.CB"
] | E.coli serves as prototype for the study of peritrichous enteric bacteria that perform runs and tumbles alternately. Bacteria run forward as a result of the counterclockwise (CCW) rotation of their flagella bundle and perform tumbles when at least one of their flagella rotates clockwise (CW), moving away from the bundle. The flagella are hooked to molecular rotary motors of nanometric diameter able to make transitions between CCW and CW rotations that last up to one hundredth of a second. At the same time, flagella move or rotate the bacteria's body microscopically during lapses that range between a tenth and ten seconds. We assume that the transitions between CCW and CW rotations occur solely by fluctuations of CheY-P molarity in the presence of two threshold values, and that a veto rule selects the run or tumble motions. We present Langevin eqs for the CheY-P molarity in the vicinity of each molecular motor. This model allows to obtain the run- or tumble-time distribution as a linear combination of decreasing exponentials that is a function of the steady molarity of CheY-P in the neighbourhood of the molecular motor, which fits experimental data. In turn, if the internal signaling system is unstimulated, we show that the runtime distributions reach power-law behaviour, a characteristic of self-organized systems, in some time range and, afterwards, exponential cutoff. In addition, our model explains without any fitting parameters the ultrasensitivity of the flagella motors as a function of the steady state of CheY-P molarity. In addition, we show that the tumble bias for peritrichous bacterium has a similar sigmoid-shape to the CW bias, although shifted to lower concentrations when the flagella number increases. Thus, the increment in the flagella number allows lower operational values for each motor increasing amplification and robustness of the chemotatic pathway. | physics.bio-ph | physics |
Stochastic model for the CheY-P molarity in the
neighbourhood of E. coli flagella motors
G. Fier1, D. Hansmann∗1,2, and R. C. Buceta†1,2
1Instituto de Investigaciones F´ısicas de Mar del Plata, UNMdP and CONICET
2Departamento de F´ısica, FCEyN, Universidad Nacional de Mar del Plata
Funes 3350, B7602AYL Mar del Plata, Argentina
November 5, 2019
Abstract
Escherichia coli serves as prototype for the study of peritrichous enteric bacteria that perform
runs and tumbles alternately. Bacteria run forward as a result of the counterclockwise (CCW)
rotation of their flagella bundle, which is located rearward, and perform tumbles when at least one
of their flagella rotates clockwise (CW), moving away from the bundle. The flagella are hooked to
molecular rotary motors of nanometric diameter able to make transitions between CCW and CW
rotations that last up to one hundredth of a second. At the same time, flagella move or rotate
the bacteria's body microscopically during lapses that range between a tenth and ten seconds. We
assume that the transitions between CCW and CW rotations occur solely by fluctuations of CheY-
P molarity in the presence of two threshold values, and that a veto rule selects the run or tumble
motions. We present Langevin equations for the CheY-P molarity in the vicinity of each molecular
motor. This model allows to obtain the run- or tumble-time distribution as a linear combination of
decreasing exponentials that is a function of the steady molarity of CheY-P in the neighbourhood
of the molecular motor, which fits experimental data. In turn, if the internal signaling system is
unstimulated, we show that the runtime distributions reach power-law behaviour, a characteristic
of self-organized systems, in some time range and, afterwards, exponential cutoff.
In addition,
our model explains without any fitting parameters the ultrasensitivity of the flagella motors as
a function of the steady state of CheY-P molarity.
In addition, we show that the tumble bias
for peritrichous bacterium has a similar sigmoid-shape to the CW bias, although shifted to lower
concentrations when the flagella number increases. Thus, the increment in the flagella number
allows lower operational values for each motor increasing amplification and robustness of the
chemotatic signaling pathway.
1
Introduction
Flagellated bacteria, when swimming in three-dimensional isotropic liquid media, execute mo-
tion modes, basically based on translations and/or turns, characteristic of each species. These
movements may be limited by the interaction with their congeners or the surrounding material
∗[email protected]
†[email protected]
1
environment. The motion modes of flagellated bacteria are determined by the rotation of their
flagella, which are hooked to a cation rotary motor. In multiflagellated species, the motors have
the ability to be synchronized. The internal biochemical processes involved in the dynamics of
the flagellar rotary motor at nanoscopic scale cause counterclockwise (CCW) and clockwise (CW)
flagella rotations at microscopic scale. These internal processes are spatially located in the neigh-
borhood of rotary motors and temporally trigger, quasi-instantaneously, the switching of flagella
rotation direction.
Enteric peritrichous bacteria as Escherichia coli have two alternating motion modes: persistent
runs without setbacks (or recoils), and abrupt turns known as tumbles [1 -- 3]. An E. coli advances
running forward when its flagella bundle (directed rearward) rotates CCW as a whole, and performs
tumbles when as few as one flagellum rotates CW, detaching from the flagellar bundle [4].
In
contrast, marine uniflagellated bacteria such as Vibrio alginolyticus swim with cyclic motion modes
established by persistent runs, first forward (with its flagellum behind) and then backwards (with
its flagellum to the front), motion known as run reverse. This cycle is restarted by a sudden turn
known as flick [5]. V. alginolyticus advances running forward when its flagellum rotates CCW,
and performs reverse runs when it flagellum rotates CW [6]. Self-propelled microorganisms, and
in particular flagellated bacteria, are systems that remain out of equilibrium, transiting between
two or more metastable states, e.g. E. coli transits between states of run and tumble and V.
alginolyticus transits between states of run and run-reverse. Experimental and theoretical studies
on motion of E. coli have shown that tumbles cannot be considered instantaneous events [7 -- 10].
Tumble-time distributions are a consequence of the internal signaling system and are essential
in a complete description of the run-and-tumble motion. A conclusion drawn from fluorescence
microscopy studies on E. coli have concluded that the transition from run to tumble is the result
of a change of direction from CCW to CW of at least one flagellar motor, which is visualized
when one or more flagella move away from the flagella bundle [4]. Recently, some very elaborated
models have been introduced using a 'veto' hypothesis, which assumes that the bacterium runs if
all flagella rotate CCW and the bacterium tumbles if at least one flagellum rotates CW [11 -- 13].
This veto hypothesis is based on high-definition observations using slow-motion techniques and
optical tweezers [14]. However, until now it is unknown whether the veto hypothesis is sufficient
to explain swimming behaviour, since the experimental measurements have not been reproduced
theoretically from a model for the activity of flagellar motors at nanoscopic scale. Figure 1 shows
a scheme of the veto hypothesis for a bacterium with three flagella running and tumbling.
The motion-modes behaviour of E.coli is a consequence of the internal biochemical processes
ocurring inside each bacteria. The signal-transduction system of histidine-aspartate phosphorelay
consists, in a simplified fashion, of histidine-kinase transmembrane which activates response regu-
lators proteins CheY and CheB on the cytoplasm [15]. This signaling system responds to chemical
stimuli and induces changes in bacterial behaviour, which are known as taxis. In turn, the response
regulator CheY interacts with the subunits of the flagella producing transitions between runs or
tumbles modes. However, the run-and-tumble motion in isotropic media is taxis free, although
changes in the rotation direction of flagella occur without an external stimulus. Several results
show that concentration of the CheY phosphorylated signaling protein (CheY-P), in the vicinity of
the flagellar motor C-ring switch, determines the rotation direction of each flagellum [16]. In turn,
the switch is composed of change proteins (FliN, FliG and FliM), with which CheY-P interacts
to trigger the CCW-CW transition.
It is known that the amount of FliM proteins involved in
the transition CCW→ CW is not equal to the transition CW→ CCW [17]. Additionally, chemical
stimuli on the receptor cluster of the bacterium poles modify the levels of CheY-P on the cytoplasm
and change the transitions between motion modes accordingly [16]. This phenomenon, known as
chemotaxis, modifies only the frequency of the transitions between CCW and CW. Assuming that
these transitions are caused by fluctuations of the CheY-P concentration between two threshold
values in the neighborhood of the flagellar motors, we introduce a phenomenological stochastic
equation for the CheY-P molarity in Section 2. This model allows to obtain the time distributions
2
Figure 1: The flow block diagram shows the veto hypothesis for a bacterium with 3 flagella from
the data obtained from the numerical simulation of the internal signaling model introduced by
us in Section 2. The 3 upper schematic timelines show the rotation intervals CW (coloured) and
CCW (white). The veto hypothesis acts with the AND gate, which assumes that the bacterium
runs if all its flagella rotate CCW. Otherwise, the bacterium tumbles. The lower schematic
timeline shows the intervals of tumble (black) and run (white). Using the right-hand rule, the
CCW and CW rotations have incoming (i) and outgoing (o) directions, respectively, seen from
the flagellar motor.
for run and tumble as a linear combination of decreasing exponentials, which fits experimental
data. Also, the model shows that the runtime distributions follow a power-law behaviour when
the internal signaling system is unstimulated.
An astonishing property of E.coli is the gain of its chemotactic signaling system [18]. Small
variations in receptor occupancy lead to significant changes in the fraction of time in which the
bacteria make runs or tumbles. This amplification has different origins. On the one hand, there
is amplification due to receptor clustering in the cell poles. On the other hand, the ultrasensitive
flagellar motors contribute to increase the amplification. The high cooperative response of the
flagellar motors to variations in the molar concentration of CheY-P has been widely reported
throughout the past decades [16, 19 -- 22]. In Section 2 we show that our model allows obtaining
the CW bias as a function of steady CheY-P molarity in agreement with the experimental data.
Besides, we show that the tumble bias is also a sigmoid function of steady molarity with the same
slope as the CW bias, although it is displaced to its left. We conclude that another possible
amplification factor is the flagella number. We show that, with the same ligand concentration, the
tumble bias is approximately an order of magnitude greater than the CW bias. Additionally, a
higher flagella number increases the robustness of the chemotactic response.
2
Internal signaling system
When E. coli performs runs or tumbles in an isotropic media, its internal signaling system is
unstimulated. Nevertheless, theoretical and experimental results show that the receptor-kinase-
activity has a steady state in the abscence of stimuli [23, 24]. In consequence, the CheY-P molar
concentration has a steady state under these conditions [25]. Under the assumption that the flagella
motor has cilindrical geometry with a diameter of 45 nm and the bacterium has a cigar shape of a
3
Figure 2: The plot shows the molar concentration c(t) of Chey-P as a function of time t in the
vicinity of one of the flagellar motors of E. coli, obtained through the numerical integration of
the Langevin equation (1). Dotted lines show the molarity threshold-values cio and coi. If the
molarity increases at c(t) = cio, there is a transition from CCW to CW, or else, if the molarity
decreases at c(t) = coi, there is a transition from CW to CCW. The coloured (colourless) regions
correspond to the CW (CCW) regimes. The values of the integration parameters are β = 1 and
µ = 2 µM, while the chosen molarity threshold-values shown in the plot are coi = 2.59 µM and
cio = 3.01 µM.
length of 1 to 10 µm and a diameter of 1 µm, the quotient between the area of a bacterium and the
area of a motor is approximately in the interval [2, 20] × 103, which means that each E. coli has
enough surface for thousands of flagella but has only a few. In the light of this difference of scales
between cell surface and the surface of the flagella motor, we model the concentration of CheY-
P around the motor as a homogeneous stochastic process. We assume that transitions between
CCW and CW occur by fluctuations in the concentration of CheY-P [26] under the presence of
two threshold values [17]. Assuming that molarity evolves temporarily in a similar way around
each flagellum motor, we introduce the following phenomenological Langevin equation (LE) for
the molarity cj(t) of CheY-P in the surroundings of the j-th motor (1 ≤ j ≤ n integer with n the
flagella number):
cj = −(cj − µ) cβ
j + ζj(t) ,
(1)
where cj denotes time derivative of cj(t), µ is the steady molarity, and β is a positive ex-
ponent to be determined. The noise ζj is Gaussian white with zero mean and correlations
(cid:104)ζj(t)ζk(t(cid:48))(cid:105) = 2 Qj δjk δ(t − t(cid:48)), where Qj is the noise intensity and j, k = 1, . . . , n. In order to
numerically obtain a mean and variance close to that of the experimental results, it is sufficient to
use a positive integer exponent β. The associated Fokker-Planck equation (FPE) to equation (1)
is
∂
∂t
p(cj, tc(cid:48)
j, t(cid:48)) = − ∂
∂cj
with initial condition p(cj, t(cid:48)c(cid:48)
function (PDF) of the molarity cj at time t given the initial condition c(cid:48)
A(cj)
j, t(cid:48)) = δ(cj − c(cid:48)
= −(cj − µ) cβ
.
j is the drift of the process.
[A(cj) p(cj, tc(cid:48)
p(cj, tc(cid:48)
j, t(cid:48))] + Qj
∂2
(2)
∂c2
j
j), where p(cj, tc(cid:48)
j, t(cid:48)) is the probability density
j at time t = t(cid:48) and
j, t(cid:48)) ,
In the vicinity of the j-th motor, if the molarity increases at cj(t) = cio, the transition from
CCW to CW occurs; or else, if the molarity decreases at cj(t) = coi (with coi < cio), the transition
from CW to CCW happens as shown in Figure 2. The labels i (or o) denote that the flagellum
performs CCW (or CW) rotation, that is, it moves with incoming (or outgoing) direction of rotation
seen from the motor. During the CCW rotation, close to any motor, the probability [or survival
probability function (SPF)] that the molarity has values 0 < c(t) ≤ cio (with the initial condition
c(t(cid:48)) = coi) is
4
[CheY-P](µM)CioCoi0.00.51.01.52.02.53.03.50.00.51.01.52.02.53.03.5tc(t) Figure 3: Both plots show the tumble-time (TTD) and runtime (RTD) distributions as functions
of tumble-time tT and runtime tR respectively, obtained from the internal signaling system model
for a bacterium with three flagella (n = 3). The plots with symbols are the result of numerical
integration of Langevin equations (1) taking β = 1, µ = 2 µM, and Qj = 0.7 (j = 1, 2, 3) .
The steady molarity value corresponds to an unstimulated internal signaling system. The
chosen molarity threshold-values are coi = 2.59 µM and cio = 3.01 µM. Inside, we show the
semilog plots in order to see that the tumble-time distribution after the maximum is very well
approximated by a single decreasing exponential function, behaviour that does not happen for
the runtime distribution. In both plots, the maxima appear at times ≈ 0.04 s and the time
distributions are equal to zero at initial times. The solid line corresponds to the data fit using
a linear combination of exponential functions. The temporal mean values with their standard
errors obtained from our simulations are 0.86 ± 1.21 s for runs and 0.146 ± 0.136 s for tumbles,
which are close to the ones obtained for E. coli [1].
Si(τ ) =
p(c, tcoi, t(cid:48)) dc ,
where τ = t − t(cid:48) and
0
p(c, tc(cid:48), t(cid:48)) = p(c, tc(cid:48), t(cid:48)) − pst(c) ,
(3)
(4)
(cid:90) cio
(cid:90) +∞
being pst(c) the stationary PDF. In addition, during the CW rotation, close to any motor, the
probability that the molarity has values coi ≤ c(t) < +∞ (with the initial condition c(t(cid:48)) = cio) is
So(τ ) =
p(c, tcio, t(cid:48)) dc .
(5)
Both SPF, given by equations (3) and (5), must satisfy initial conditions Si(0) = So(0) = 1 and
asymptotic conditions Si(+∞) = So(+∞) = 0. For a system of n equivalent and autonomous
flagellar motors, the probability that a bacterium runs or tumbles is
coi
FR = S n
i ,
FT = (Si + So)n − S n
i ,
respectively. The runtimes and tumble-times distributions are
fX(τ ) = − dFX
dτ
,
(6)
(7)
where X = R or T, which can be calculated in different ways. The time distributions with n = 1
[equations (6) and (7)] are known as first passage time densities (FPTD). This is the survival
probability density where the concentration reaches a threshold value cio (cio) at time t, with
initial condition coi (coi), corresponding to the CCW (CW) rotation mode. In the case n = 1, it
is convenient to refer to equation (7) as rotation-time distributions of the modes CW and CCW,
since this case does not apply to the multi-flagellated E. coli, but does apply to each of its rotary
5
▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲Tumble0.00.20.40.60.81.00.00.51.01.52.02.53.03.5tT(s)TTD▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲0.00.20.40.60.81.010-40.0010.0100.1001▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼Run0.00.51.01.52.00.00.20.40.60.81.01.21.4tR(s)RTD▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼0.20.40.60.81.00.10.20.51 motors. Figure 2 shows the numerical results of the CheY-P molarity (around any flagellar motor)
as a function of time, obtained by integrating one of the Langevin equations (1) with parameters
β = 1 and µ = 2 µM. Alternatively, it is possible to give an analytical approach near the steady
state. Using the ansatz of variables separation, the solution of the Fokker-Planck equation (2) near
the steady probability density is
p(c, tc(cid:48), t(cid:48)) = pst(c) +
Tj(τ ) Cj(c) ,
(8)
+∞(cid:88)
j=1
Sy(τ ) =
j=1
(9)
syj e−αj τ
where it is straightforward to find that Tj(τ ) ∼ e−αj τ , with αj > 0. The SPFs [from equations (3)
and (5)] are
+∞(cid:88)
0 Cj(c) dc for the run and soj =(cid:82) +∞
(with y = i, o), where the coefficients are sij =(cid:82) cio
the sum rule for the coefficients(cid:80)+∞
Cj(c) dc
for the tumble. Note from the equation (9) that the initial condition Sy(0) = 1 allows to establish
j=1 syj = 1 and that the asymptotic condition Sy(+∞) = 0 is
verified. It is easy to see through equations (6) and (9) that the run- and tumble-time distributions
(given by equations (7)) near the steady state are linear combinations of decreasing exponential
functions, which is a conclusion that is consistent with the experimental observations [1, 27]. This
conclusion includes the rotation-time distributions of the CW and CCW modes. However, the
variable separation approach is inadequate to establish the early time distributions. Instead, by
integrating Langevin's equations (1), we show that time distributions are increasing functions from
zero, reaching an early maximum and then decaying as mentioned above, as we show in Figure 3.
The first experimentally measurements (circa 1972) on run- and tumble-time distributions for
E. coli showed exponential decays although they did not show either maxima at early times or
zero value at initial time [28]. This is reasonable because both features happen at very short
times (cid:47) 0.04 s. The maxima of the time distributions at early times for E. coli have been first
experimentally observed fifteen years ago [27]. The behaviour described here has also been observed
in time distributions of runs and reverse runs performed by V. alginolyticus with a single flagellum
[29].
coi
Figure 4: The log-log plot shows the runtime distribution (RTD) as a function of runtime tR
(with green triangles), using the same data as in the right plot of Figure 3.
In the interval
[0.1, 1] we observe a power-law behaviour, i.e. fR(τ ) ∼ τ−ν with exponent ν = 0.89 (shown by
the red dashed line). For τ (cid:29) 1, we observe an exponential cutoff.
Based on the experimental observation and our theoretical conclusions, the time distributions
of runs and tumbles can be described by
fX(τ ) =
BXk e−βXk τ .
(10)
+∞(cid:88)
k=1
6
▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼∼tR-ν0.010.050.100.50150.010.050.100.501tR(s)RTD Both plots of Figure 3 show numerical data fits by using equation (10), which is a good approxima-
tion even at early times. The left plot shows that, after the maximum, the tumble-time distribution
can be fitted by a single exponential function. In turn, Figure 4 shows that the numerical data of
the runtime distribution can be fitted into at least one decade with a power-law function. After
this behaviour, an exponential cutoff is observed. The runtime distribution
(cid:26) τ−ν
e−λτ
fR(τ ) ∼
τ ∈ [a, b]
τ (cid:29) b ,
(11)
Rk
where ν is the power-law coefficient, λ is the cutoff coefficient, and [a, b] is the time interval where
the power law is observed. The experimental observation and our model agree that the tumble-time
distributions are mono-exponential. With the same model we show that multi-exponential runtime
distributions fit the experimental measurements very well. On the other hand, the runtimes
distributions have been adjusted, by us and other authors, with power laws for one or more decades
[27]. This mismatch should not be surprising, as it is possible to fit in a certain range power-law
distributions with multi-exponential functions [30]. Conversely, a weighted sum of N -exponential
functions may result in a power law at some interval, with particular weights BRk and characteristic
times β−1
, with k = 1, . . . , N . This suggests that the CheY-P molarity fluctuations during the
runs, assuming that the internal signaling system is unstimulated, self-organize the flagellar system
reaching runtime distributions with power-law behaviour at time range. The sleep- and wake-
stage distributions show similar behaviours to those described for tumble and run respectively:
the disruptive-sleep duration follows exponential distributions, while the wake duration is self-
organized with power-laws distributions [31]. These behaviours resemble the dynamics seen in some
models of self-organized criticality (SOC): avalanche-time distributions follow power laws, while
quiessence-time distributions can be exponential [32]. However, our system is out of criticality,
since the mean runtime is finite. If, when changing the molarity parameters, the power-law range
increases, then the mean runtime also increases.
If the power-law range and mean runtime go
to infinity, the system reaches a steady critical state [33], which does not happen in our system.
From our simulations we have observed that the power-law behaviour of the runtime distribution is
lost when the internal signaling system begins to be stimulated, preserving the multi-exponential
behaviour.
The ultrasensitive property of flagella motors can be quantitatively described by a Hill equation
[34] for the response θ as a function of the steady CheY-P molarity µ. The response of an individual
flagellar motor is defined as the time fractions in which the motor rotates in one direction and the
other, referred to as CW bias or CCW bias and denoted here θCW or θCCW , respectively. The Hill
parameters are the steady molarity producing half response µA and the Hill coefficient nH, which
measures the steepness of the response. With our model, varying the steady molarity µ, we study
the CW-bias change of each motor. As in this work we focused on explaining run-and-tumble time
distributions for unstimulated scenarios, we do not take into account the adaptive methylation
pathway. This is equivalent to use CheR-CheB mutants in ultra-sensitivity experiments. Thus,
our changes in the steady molarity µ corresponds to changes in Chey-P induced by plasmids at
different IPTG concentrations in bead assays [20]. We average CW bias for each flagellar motor
after several minutes, time scale much bigger than motor adaptation through turnover of C-ring
proteins [21]. Under this assumption, the two threshold values represent the steady state number of
FliM subunits of each motor. The left plot of Figure 5 shows how our model reproduces accurately
the ultrasensitive property of a flagellar motor, with fit Hill-parameters of the experimental and
simulation data very close to each other.
For a multi-flagellated bacterium, the response of the flagellar system is defined as the fraction
of times in which the bacterium runs or tumbles, called run bias or tumble bias [13] and denoted
here θR or θT, respectively. In addition, as we see in the right plot of Figure 5, the tumble bias also
shows ultrasensitive response under variations in CheY-P molarity, similar to CW bias, where its
sigmoid curve is shifted left respect to the CW-bias one. Consequently, while µA is a function of
7
Figure 5: Left: CW bias as a function of the equilibrium molarity µ. Experimental and sim-
ulations data agree in the individual response of the flagellar motors vs variations in CheY-P
molarity. Using the Hill equation to fit our simulation data (with blue triangles down) we found
that the molarity producing half CW-bias is µA = 2.95 µM and the Hill coefficient is nH = 11.0
(see blue dashed line). On the other hand, experimental data (with green squares) have been
fitted with Hill parameters µA = 3.1 µM and nH = 10.3 ± 1.1 [20], which were confirmed with
data from other measurements (with red triangles up) [16]. Right: Tumble bias as a function of
the equilibrium molarity µ. Plot shows the simulation data (with triangles) for a system with
three flagella motors. The fit with the Hill equation has the same coefficient as that of the left
plot although here µA = 2.6 µM (see blue dashed line).
(cid:17)nH ,
(cid:16) µA
µ
1 − θ
θ
=
the flagella number n, the Hill coefficient nH is invariant. This implies that a higher number of
flagella motors allows each bacteria to decrease the operational steady molarity leaving invariant
the amplification of the flagella system. Using the alternative expression for the phenomenological
Hill equation
it is easy to see that the right-hand side of this equation is equal to (cid:104)tR(cid:105)/(cid:104)tT(cid:105) when θ = θT is the
tumble bias or is equal to (cid:104)tCCW(cid:105)/(cid:104)tCW(cid:105) when θ = θCW is the CW bias. Chemotactic response
modifies cell behaviour changing the tumble frequency. Under the presence of attractants, such
as amino acids like aspartate, E.coli extends the runtimes although tumble-times remain almost
constant [1]. Because of this response, E. coli has been described as an 'optimist' [28], since when
life gets better, it keeps swimming in the same direction. In the left plot of Figure 6 we show this
optimist behaviour. For short tumble-bias θT (cid:47) 0.15, we show how the mean run- and tumble-
times shift against constant chemotactic conditions. We found that runtimes increase with tumble
bias following a power-law behaviour (see right plot of Figure 6). In contrast, mean tumble-times
remain almost constant around 0.1 s as was observed [1]. A similar behaviour is found for the mean
CW- and CCW-times, which we do not show here because of its great similarity. For long tumble-
bias θT (cid:39) 0.85 we found the opposite effect, where mean tumble-time behaves like mean runtimes
and vice versa. To the right of Figure 6, the log-log plot shows that mean run- and tumble-times
as functions of tumble- and run-bias, respectively. These mean times have a power-law behaviour
for more than four decades. Surprisingly, the power-law exponent takes the same value regardless
of the flagella number and matches the exponent value of the runtime distribution.
Kinase activity have been measured via fluorescence resonance energy transfer (FRET) tech-
nique following the signal of CheY-P-CheZ complex [23 -- 25]. From this measurements, phenomeno-
logical equations for kinase activity and methylation level has been proposed in references [23, 24].
Thus, we can link the steady CheY-P molarity µ with the kinase activity function G through the
linear relationship µ = k G(m, [L]), where k is a constant, [L] is the ligand concentration, and m
is the average methylation level of the receptors. We show that the two-threshold hypothesis is
8
□□□□□□□□□□□□□□□□□□□△△△△△△△△△△△△△△△△△△△△△△△△△△▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼□Experimental[20]△Experimental[16]▼SimulationDataHillfunctionfit012345670.00.20.40.60.81.0μ(μM)CWbias▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼Hillfunctionfit▼SimulationData012345670.00.20.40.60.81.0μ(μM)Tumblebias Figure 6: Left: The semi-log plot shows the mean run- and tumble-times as functions of the
tumble bias, where J = R, T. The vertical red dashed line shows tumble-bias value θT ≈ 0.15 for
unstimulated run-and-tumble. Low (high) tumble-bias results in long (short) runs and short
(long) tumbles. Right: The log-log plots show the mean runtime (and mean tumble-time) as
functions of the tumble bias X = θT (and run bias X = θR = 1−θT), plotted with green triangles
down (and blue triangles up), respectively. The red dashed line shows the power-law behaviour
with exponent ζ = 0.89.
sufficient to reproduce the second step of amplification (see Figure 5). The addition of the kinase
activity allows the model to reproduce the first step of amplification. For CheR-CheB mutants, the
methylation remains always at the same steady state; therefore, we can simplify the dependence
of the kinase activity only to ligand concentration, i.e. G = G([L]). We choose the steady state
methylation so that the kinase activiy reaches the basal value for unstimulated conditions, which
we maintain fixed. Through the inverse function G−1 of the kinase activity, we can find the rela-
tionship [L] = G−1(µ/k). This allows us to relate the tumble bias and CW bias with attractants
concentration, as has been done experimentally [16]. Because one of our main goals is to reproduce
the run-and tumble-distributions, we fixed the unstimulated tumble bias θT = 0.15, which varies
from strain to strain [1]. Thus, we study the effect of attractans for tumble bias θT ≤ 0.15. This
also impacts on the CW bias, which will always be smaller than tumble bias, being θCW = 0.04
for unstimulated conditions.
In left plot of Figure 7 we show how the tumble bias is modified
against flagella number. In this parametric plot we analyze the tumble bias as a function of the
CW bias under increasing µ; that is, the enhancement in the response against flagella number. We
found that increasing the number of flagella allows each bacteria to perform chemotaxis with lower
concentrations of CheY-P. This same results have been found experimentally in reference [13] for
other E.coli strain. Combining with results of Figure 5, we conclude that more flagella allows lower
CheY-P concentrations where the shape of the response will be invariant. Another way to study
the impact of flagella number in chemoctactic response can be achieved by the linear relationship
between µ and [L] through the inverse function G−1 as we mentioned above.
In right plot of
Figure 7 we found that a higher flagella number increases the bias, allowing higher amplification
in the response, which is a consequence of the veto rule. Furthermore, more flagella increases the
robustness of the response; e.g. the CW bias of any flagellar motor drops to zero for [L] ≈ 5 µM
and the tumble bias of three-flagella system drops to zero for [L] (cid:39) 10 µM (see right plot of Fig-
ure 7). The range of response for the ligand concentration agrees with the mean values used for
chemotactic experiments [1]. Thus, our simulations confirm the 'gain paradox' [28] with another
possible amplification step and robustness enhancement of the chemotactic response induced by
an increase of flagella number.
Finally, we remark that our conclusions do not depend of the flagella number used in the
simulations. Recently, has been observed that flagella number thus not only vary from strain to
strian but also is sensitive to growing media conditions [35]. Thus, the parametrization used in
our model depends of the strain and environment conditions used in reference [1]; comparisons
9
▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▼Run▲Tumble0.00.20.40.60.81.00.11101001000Tumblebias<tJ>▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲▲∼X-ζ10-40.0010.0100.10010.11101001000X<tJ> Figure 7: Left: The plot of the tumble bias as a function of the CW bias shows the bias deviation
with the increase of the flagella number. The intersection of the dashed lines shows the bias
values θCW = 0.04 and θT = 0.15 for an unstimulated bacteria with three flagella. Right: Log-log
plot of the tumble- and CW-bias as functions of the ligand concentration [L]. The tumble-bias
data shown correspond to a three-flagella system of our simulation. When [L] → 0 we recovered
the motion biases from unstimulated bacteria. When the CW bias dropping to zero the tumble
bias is about an order of magnitude greater, which shows the robustness of the bias response.
with other experimental setups must be done qualitatively and not quantitatively. The latter one
requires reparametrization.
3 Conclusions
The last five decades of research on bacterial systems, whether as a single bacterium or a colony,
account for the challenge of explaining phenomena that are dependent on different spatial and
temporal scales. When we study the individual movement of a flagellated bacterium in a non-
chemotactic media free from interactions with other bacteria or boundaries, we observe that the
motion modes on a microscopic scale are determined by the rotation direction of the flagella that
are hooked to molecular rotary motors of nanometric diameter. Each motor makes transitions in
the rotational direction (CW or CCW) in a few hundredths of a second [28], while the time of
each motion mode is at least an order of magnitude greater (e.g. E. coli can have runtimes of
the order of a second and tumble-times of the order of a tenth of a second), and a motion-mode
sequence can last several tens of minutes. The statistical properties of microscopic movement does
not reveal information about the internal processes of the bacteria if we do not link phenomena at
different scales. The CW bias as a function of the equilibrium molar concentration of CheY-P (in
the vicinity of the rotating molecular motor) can be established experimentally, although it has not
been conveniently modeled until today. The run- and tumble-times distributions, and therefore
the mean run- and tumble-times, observed microscopically, must be closely linked to the CheY-P
molarity in the vicinity of each motor. This is one of the most relevant connections, although it
is not the only one. This paper presents a simple model for E. coli that tries to establish how
phenomena that occur at a nanometric scale in the cytoplasm give rise to motile behaviours on a
microscopic scale.
In Section 2 we present a system of Langevin equations for the molarity of CheY-P in the vicinity
of each rotating molecular motor. We assume that when E. coli runs or tumbles in an isotropic
environment, its internal signaling system is unstimulated, and as a result, the CheY-P molar
concentration is stable. We also assume that the transitions between CCW and CW rotations occur
due to fluctuations in CheY-P concentrations in the presence of two threshold values, a hypotheses
introduced about two decades ago [17, 26]. Inspired by the theory of first passage time density,
which we show is valid for a uniflagellated bacterium, we propose a model for E. coli, valid for
10
▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■▼n=3■n=10.00.20.40.60.81.00.00.20.40.60.81.0CWbiasTumblebias▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼▼■■■■■■■■■■■■■■■■▼Tumblebias■CWbias0.5151010-510-40.0010.0100.100[L](μM)Bias other peritrichous enteric bacteria, e.g. Salmonella enterica, assuming a veto hypothesis [11 -- 13].
The most relevant conclusion of our internal signaling model is that, near the steady state, the
run- and tumble-time distributions, as well as the rotation-time distributions of the CW and
CCW modes, are linear combinations of decreasing exponential functions. These results are in full
agreement with the pioneering experimental observations on time distributions made by Berg and
Brown [1]. The formalism presented cannot predict the maximum at short times, which has been
recently observed experimentally, as well as it cannot predict the behaviour at early times, which
we describe through the numerical integration of Langevin equations. This complex response of
the flagellar system can be attributed, on the one hand, to the two regulatory thresholds of the
CW-CCW transitions and, on the other hand, to the veto hypothesis. There has been a long debate
about whether these runtime distributions are multi-exponential or power law after the maximum.
Here we show that the tumble-time distribution is mono-exponential and the runtime distribution
is multi-exponential, which can be precisely fitted with a power law in certain time range only
if the internal signaling system is unstimulated. We conclude that, during runs without stimuli,
the CheY-P molarity fluctuations close to rotatory motors self-organize the flagellar system out of
criticality during some time range to reach runtimes that follow a power-law-like distribution. The
runtime distribution in later times, after those of power-law behaviour, presents an exponential
cutoff, which is a sign that the flagellar system loses its self-organization. Similar power-law
behaviours with exponential cutoff have been established by Tu and Grinstein [36] for CCW-
rotation-time distributions from a linear mean-field model for the concentration of CheY-P. The
mono-exponential behaviour of the tumble-times distribution and the power-law-like behaviour
of the runtimes distribution are lost when the system is stimulated, preserving only the multi-
exponential behaviour of both distributions. Nevertheless, the run- and tumble-time distributions
of E. coli swimming unstimulated gain importance in chemotactic scenarios. When the receptor
kinase activity is modified by external chemical stimuli, a slow adaptation process begins through
the response regulator CheB. After time scales in the range of minutes, the CheY-P molarity shifts
back to the unstimulated stable value. In consequence, after some minutes, the system returns to
the default run- and tumble-time distributions even under invariant chemotactic conditions. Our
model also allows to accurately describe the ultrasensitive response of flagellar motors, as well as
the entire flagellar system. The responses (CW bias and tumble biases) are sigmoid functions of
steady molarity similar to each other, which shift to the left when the flagella number increases.
A higher flagella number produces an increase in amplification and robustness of the chemotactic
response, being another possible amplification step in the signaling system pathway. Thus, the
two-threshold hypothesis [17, 21] combined with the veto rule [13] allow to explain a variety of
phenomena for stimulated and unstimulated conditions described through the past decades. We
finally conclude that the knowledge of the time distributions of each motion mode as a function
of nanoscopic parameters is essential when studying other observables usually measured in the
laboratory at a microscopic scale.
Acknowledgements
This work was partially supported by Consejo Nacional de Investigaciones Cient´ıficas y T´ecnicas
(CONICET), Argentina, PIP 2014/16 No 112-201301-00629. R.C.B. thanks to M. Semp´e for her
suggestions on the final manuscript.
References
[1] Berg HC, Brown DA (1972) Chemotaxis in Escherichia coli analysed by three-dimensional
tracking. Nature 239: 500.
11
[2] Macnab RM, Koshland DE (1972) The gradient-sensing mechanism in bacterial chemotaxis.
PNAS 69: 2509-12.
[3] Macnab RM, Ornston MK (1977) Normal-to-curly flagellar transitions and their role in bac-
terial tumbling. Stabilization of an alternative quaternary structure by mechanical force. J
Mol Biol 112: 1-30.
[4] Turner L, Ryu WS, Berg HC (2000) Real-time imaging of fluorescent flagellar filaments. J
Bacteriol 182: 2793-2801.
[5] Taylor BL, Koshland DE (1974) Reversal of flagellar rotation in monotrichous and peritrichous
bacteria: Generation of changes in direction. J Bacteriol 119: 640-642.
[6] Homma M, Oota H, Kojima S, Kawagishi I, Imae Y (1996) Chemotactic responses to an
attractant and a repellent by the polar and lateral flagellar systems of Vibrio alginolyticus.
Microbiology 142: 2777-2783.
[7] Kafri Y, da Silveira RA (2008) Steady-state chemotaxis in Escherichia coli. Phys Rev Lett
100: 238101.
[8] Saragosti J, Silberzan P, Buguin A (2012) Modeling E. coli tumbles by rotational diffusion.
implications for chemotaxis. PLoS ONE 7: e35412.
[9] Fier G, Hansmann D, Buceta RC (2017) A stochastic model for directional changes of swim-
ming bacteria. Soft Matter 13: 3385-3394.
[10] Fier G, Hansmann D, Buceta RC (2018) Langevin equations for the run-and-tumble of swim-
ming bacteria. Soft Matter 14: 3945-3954.
[11] Vladimirov N, Lebiedz D, Sourjik V (2010) Predicted auxiliary navigation mechanism of
peritrichously flagellated chemotactic bacteria. PLoS Comput Biol 6: e1000717.
[12] Sneddon MW, Pontius W, Emonet T (2012) Stochastic coordination of multiple actuators
reduces latency and improves chemotactic response in bacteria. PNAS 109: 805-809.
[13] Mears PJ, Koirala S, Rao CV, Golding I, Chemla YR (2014) Escherichia coli swimming is
robust against variations in flagellar number. eLife 3: e01916.
[14] Darnton NC, Turner L, Rojevsky S, Berg HC (2007) On torque and tumbling in swimming
Escherichia coli. J Bacteriol 189: 1756.
[15] Wadhams GH, Armitage JP (2004) Making sense of it all: bacterial chemotaxis. Nature Rev
Mol Cell Biol 5: 1024-1037.
[16] Sourjik V, Berg HC (2002) Binding of the Escherichia coli response regulator CheY to its
target measured in vivo by fluorescence resonance energy transfer. PNAS 99: 12669-74.
[17] Bren A, Eisenbach M (2001) Changing the direction of flagellar rotation in bacteria by mod-
ulating the ratio between the rotational states of the switch protein flim. J Mol Biol 312:
699-709.
[18] Segall JE, Block SM, Berg HC (1986) Temporal comparisons in bacterial chemotaxis. PNAS
83: 8987-8991.
[19] Kuo SC, Koshland DE (1989) Multiple kinetic states for the flagellar motor switch. J Bacteriol
171: 6279-87.
[20] Cluzel P, Surette M, Leibler S (2000) An ultrasensitive bacterial motor revealed by monitoring
signaling proteins in single cells. Science 287: 1652-1655.
[21] Yuan J, Branch RW, Hosu BG, Berg HC (2012) Adaptation at the output of the chemotaxis
signalling pathway. Nature 484: 233-236.
[22] Yuan J, Berg HC (2013) Ultrasensitivity of an adaptive bacterial motor. J Mol Biol 425:
1760-64.
12
[23] Tu Y, Shimizu TS, Berg HC (2008) Modeling the chemotactic response of Escherichia coli to
time-varying stimuli. PNAS 105: 14855-60.
[24] Shimizu TS, Tu Y, Berg HC (2010) A modular gradient-sensing network for chemotaxis in
Escherichia coli revealed by responses to time-varying stimuli. Mol Syst Biol 6: 382.
[25] Sourjik V, Berg HC (2002) Receptor sensitivity in bacterial chemotaxis. PNAS 99: 123-127.
[26] Morton-Firth CJ, Bray D (1998) Predicting temporal fluctuations in an intracellular signalling
pathway. J Theor Biol 192: 117-128.
[27] Korobkova E, Emonet T, Vilar JMG, Shimizu TS, Cluzel P (2004) From molecular noise to
behavioural variability in a single bacterium. Nature 428: 574-578.
[28] Berg HC (2004) E. coli in motion. Spinger-Verlag. doi:10.1007/b97370.
[29] Xie L, Altindal T, Chattopadhyay S, Wu XL (2011) Bacterial flagellum as a propeller and as
a rudder for efficient chemotaxis. PNAS 108: 2246-2251.
[30] Bochud T, Challet D (2007) Optimal approximations of power laws with exponentials: Ap-
plication to volatility models with long memory. Quantitative Finance 7: 585-589.
[31] Chu-Shore J, Westover MB, Bianchi MT (2010) Power law versus exponential state transition
dynamics: Application to sleep-wake architecture. PLoS ONE 5: e14204.
[32] Buceta RC, Muraca D (2011) Model for domain wall avalanches in ferromagnetic thin films.
Physica A 390: 4192-4197.
[33] Yang CB (2004) The origin of power-law distributions in self-organized criticality. J Phys A:
Math Gen 37: L523-L529.
[34] Weiss JN (1997) The Hill equation revisited: uses and misuses. The FASEB J 11: 835-841.
[35] Sim M, Koirala S, Picton D, Strahl H, Hoskisson PA, et al. (2017) Growth rate control of
flagellar assembly in Escherichia coli strain RP437. Sci Rep 7: 41189.
[36] Tu Y, Grinstein G (2005) How white noise generates power-law switching in bacterial flagellar
motors. Phys Rev Lett 94: 208101-104.
13
|
1702.08317 | 1 | 1702 | 2017-02-27T15:12:27 | Design of elastic networks with evolutionary optimised long-range communication as mechanical models of allosteric proteins | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.BM"
] | Allosteric effects are often underlying the activity of proteins and elucidating generic design aspects and functional principles which are unique to allosteric phenomena represents a major challenge. Here an approach which consists in the in silico design of synthetic structures which, as the principal element of allostery, encode dynamical long-range coupling among two sites is presented. The structures are represented by elastic networks, similar to coarse-grained models of real proteins. A strategy of evolutionary optimization was implemented to iteratively improve allosteric coupling. In the designed structures allosteric interactions were analyzed in terms of strain propagation and simple pathways which emerged during evolution were identified as signatures through which long-range communication was established. Moreover, robustness of allosteric performance with respect to mutations was demonstrated. As it turned out, the designed prototype structures reveal dynamical properties resembling those found in real allosteric proteins. Hence, they may serve as toy models of complex allosteric systems, such as proteins. | physics.bio-ph | physics | Design of elastic networks with evolutionary optimised long-range communication as
mechanical models of allosteric proteins.
Holger Flechsig∗
Department of Mathematical and Life Sciences, Graduate School of Science,
Hiroshima University, 1-3-1 Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan
(Dated: February 24, 2017)
Allosteric effects are often underlying the activity of proteins and elucidating generic design as-
pects and functional principles which are unique to allosteric phenomena represents a major chal-
lenge. Here an approach which consists in the in silico design of synthetic structures which, as the
principal element of allostery, encode dynamical long-range coupling among two sites is presented.
The structures are represented by elastic networks, similar to coarse-grained models of real proteins.
A strategy of evolutionary optimization was implemented to iteratively improve allosteric coupling.
In the designed structures allosteric interactions were analyzed in terms of strain propagation and
simple pathways which emerged during evolution were identified as signatures through which long-
range communication was established. Moreover, robustness of allosteric performance with respect
to mutations was demonstrated. As it turned out, the designed prototype structures reveal dy-
namical properties resembling those found in real allosteric proteins. Hence, they may serve as toy
models of complex allosteric systems, such as proteins.
Keywords: evolution, conformational motions, nonlinear dynamics, communication pathways, strain propa-
gation, robustness
INTRODUCTION
The functional activity of proteins and its precise regu-
lation often relies on allosteric coupling between different
functional regions within the macromolecular structure.
According to the mechanical perspective allosteric com-
munication originates from structural changes mediated
by a network of physically interacting residues [1]. Much
resembling the occurrence of a proteinquake, local con-
formational motions initiated, e.g., upon ligand binding
to one specific site propagate across the protein struc-
ture to spatially remote regions eventually generating a
functional change in the conformation of another site.
Experiments have indeed evidenced the existence of
communication networks in proteins, which are formed
by only a set of amino acids and constitute allosteric
pathways, physically linking remote binding sites [2, 3].
Various computational strategies aimed at predicting
such pathways have been developed, including structure-
based network analysis [4–6], stochastic Markov mod-
elling [7, 8], and sequence-based statistical methods [9–
12]. Understanding of allosteric communication at the
molecular level has also been widely addressed at atom-
istic resolution in molecular dynamics (MD) simulations
[13–19].
To circumvent
the heavy computational burden
present in MD simulations, collective conformational mo-
tions and allosteric transitions in proteins have been, to
a vast extend, investigated at the coarse-grained level
using elastic-network models and the analysis of normal
modes [20–26]. Despite their approximate nature such
simplified descriptions have significantly contributed in
understanding the mechanistic underpinnings of allostery
in proteins [27, 28].
While in the majority of cases the important question
is considered to be how allostery works for a particular
protein given its specific structure, there may also be
more fundamental and general questions to be addressed
such as what are generic design and functional principles,
requisite to make allostery work, and, which dynamical
properties are unique to allostery. One possibility to ap-
proach such aspects would consist in a systematic screen-
ing of the biophysical properties among allosteric proteins
with available structural data, which has already been
sporadically attempted earlier [6, 29].
Here, in an attempt to address this subject, we present
an alternative approach which is motivated by the idea
to engineer, in silico, artificial spatial structures with dy-
namical properties resembling those found in real pro-
teins. In previous works such an approach was employed
to design structures of elastic networks which can oper-
ate as a model molecular machine [30–32] or swimmer
[33]. Moreover, we recently used this machine to con-
struct a model motor which in its function mimics the
myosin protein motor [34].
In this paper we aim to establish a generalised struc-
tural model of an allosteric system. To this end we design
through evolutionary optimisation elastic-network struc-
tures which, as the principal element of allostery, encode
long-range coupling among two spatially remote local re-
gions. We first explain how the strategy of iterative evo-
lution was developed and applied to stepwise improve,
starting from a random elastic network, the allosteric
response in the emerging structures.
In the designed
structures allosteric communication was then analysed in
terms of the propagation of strain and its spatial distri-
bution was used to identify pathways through which re-
mote interactions are established. Moreover, the effect of
7
1
0
2
b
e
F
7
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
7
1
3
8
0
.
2
0
7
1
:
v
i
X
r
a
mutations was demonstrated and robustness of allosteric
performance in the designed structures examined. Fi-
nally we discuss the relevance of our model system in the
light of actual allosteric proteins.
RESULTS
Random elastic network and Evolutionary
optimisation
Our structural model of an allosteric system was based
on evolutionary optimisation starting from a random
elastic network. The initial random network was con-
structed as a two-domain structure.
It was obtained
by randomly folding two polymer chains independently,
each consisting of 100 elastically linked identical beads
and representing one domain, and then merging them
such that they form a common domain interface.
In
the model physical interactions between all beads, be-
yond those acting between neighbouring beads in each
chain, were introduced by connecting those two beads by
an elastic link which have a separation smaller than a
prescribed interaction radius. Thus the complete elastic
network was obtained. Details of the construction are
described in the Methods section.
When elastic networks are constructed based on the
structures of real proteins, the beads would corresponds
to atoms or typically represent entire amino acid residues,
and elastic links between them empirically mimic effec-
tive interactions between them [35, 36].
Here, we proceed with the constructed random network
of two compactly folded coupled domains shown in Fig.
1. In each network domain a pocket site, for simplicity
formed by two beads only, was chosen such that a suffi-
cient separation of the two sites was ensured (see Meth-
ods). One site represented the allosteric pocket whereas
the other mimicked the regulated pocket, respectively.
As shown in Fig. 1, the pockets were located opposite to
each other near the surface of the respective domain.
With this setup, the ability of the elastic network
to conduct allosteric communication between the pock-
ets was examined using a simple force-probe scheme, in
which conformational dynamics in the allosteric site was
initiated through local binding of an additional ligand
bead, and the subsequent response generated in the regu-
lated site has been detected. In particular, attractive pair
forces which were acting on the pocket beads of the al-
losteric site have been applied to mimic binding of the fic-
titious substrate bead to its centre. The dynamics of pro-
tein elastic-networks consists in processes of over-damped
relaxation motions (see Methods). Hence, as a result of
such forces, all beads underwent coupled relaxation mo-
tions bringing the elastic network from its original con-
formation (without the forces) to a deformed steady state
of the network, having the ligand bead tightly bound to
2
FIG. 1. Random elastic-network structure. The ini-
tial two-domain network consisting of 200 beads connected
by elastic springs. The randomly folded chains, each form-
ing one domain, are shown as blue, respectively red, traces
of thick bonds. Other springs are displayed as thin bonds in
grey. In each domain the two beads representing the remotely
placed pockets are highlighted as coloured beads.
the allosteric site (i.e, with the forces applied for a long
time). During this process the additional forces tend to
close this site, i.e. its two beads move towards each other,
and first the elastic links localised in their vicinity became
deformed. Eventually deformations propagated through
the entire network structure until a final steady network
conformation was reached. In the simulations the con-
formational motions inside the network were followed by
numerically integrating the equations of motion for all
network beads (see Methods). The response generated
inside the regulated pocket in the second domain was
quantified in terms of distance changes between its two
corresponding beads.
As we found, the random network structure did not re-
veal any allosteric coupling between the two sites. While
the allosteric site closed upon substrate binding, with
the distance between pocket beads changing by approx.
−4A, no response was detected at the regulated site with
a change of < 10−4A (for the choice of lengths units,
see Methods). Proceeding with this observation, an al-
gorithm of evolutionary optimisation aimed to design
networks with pronounced allosteric communication was
established. The optimisation process consisted in se-
quences of mutations followed by selection, which were
applied iteratively starting from the initial random net-
work. In each evolution step the following sequence was
carried out:
i) mutation; a single structural mutation
was performed by randomly selecting one network bead
(excluding one of the four pocket beads) and changing
its equilibrium position, which corresponded to an alter-
ation of elastic connections in the vicinity of the mutation
spot. ii) probing of allostery; the force-probe scheme was
applied in a simulation of the new mutant network struc-
ture and its conformation with the ligand bead bound
to the allosteric site was obtained.
iii) selection; the
3
FIG. 2. Evolutionary optimisation. Snapshots of network conformations taken at different steps of accepted mutations along
the processes of evolutionary optimisation towards designing the two prototype allosteric networks. The initial random network
structure is shown in the top row. Each depicted network conformation corresponds to the steady state with the allosteric
pocket being closed after ligand binding and the allosteric response in the regulated pocket becoming progressively improved. In
the design process of the network with symmetric cooperativity the regulated pocket increasingly closed as evolution proceeds,
whereas during the design of the network with asymmetric cooperativity its ability to open emerged (indicated by black arrows).
Pocket beads are highlighted in blue colour and the size of the regulated pocket in A is given for all snapshots. Mutation-induced
structural changes and concomitant remodelling of elastic interactions can be seen by comparing snapshots. Final structures
of designed networks are shown in the bottom row.
allosteric response in the regulated pocket of the net-
work before and after the mutation was compared and
the mutation was scored favourable and was accepted if
the mutant network showed improved allosteric coupling
between the two sites; otherwise it was rejected. The
optimisation procedure was iteratively applied and ter-
minated, once a network structure with a prescribed level
of sufficient allosteric coupling was obtained. Details of
the implementation are described in the Methods section.
In this work two prototype examples of elastic network
structures with allosteric coupling were designed and
analysed. In two independent simulations of evolutionary
optimisation, starting both times with the same initial
random network, two different network structures with
optimised allostery have been designed. The first one
was designed under the side condition that allosteric cou-
pling between the remote sites had the type of symmetric
cooperativity, i.e., binding of the substrate bead to the al-
losteric site and the concomitant closing of that pocket
resulted in consequent closing of the regulated site. In the
second case, the requirement was vice versa; allosteric
communication was optimised under the premise that
4
FIG. 3. Allosteric coupling in designed network structures. (A,a) For both designed networks the initial structure and
the final structure with the substrate ligand bound to the allosteric pocket are shown. Elastic connections between network
beads are shown as thin transparent lines and pocket beads are coloured blue (allosteric pocket) and red (regulated pocket),
In both designed networks the initial and respective final structure is very similar (see text), but the large-
respectively.
amplitude motions of pocket beads evidencing intra-structure allosteric communication are apparent.
In the network with
symmetric cooperativity substrate-induced closing of the allosteric pocket triggers allosteric closing of the regulated pocket,
whereas in the network with asymmetric cooperativity the allosteric response consisted in opening of the regulated site. (B,b)
Traces showing changes in the size of both pockets upon binding of the substrate bead to the allosteric pockets are also provided
(blue colour for allosteric the pocket, red for the regulated pocket).
substrate induced closing of the allosteric pocket triggers
opening in the regulated site, i.e. asymmetric coopera-
tive coupling was realised. In both cases, as a result of
several hundreds of mutations which were needed during
evolution to improve allosteric coupling, the networks un-
derwent significant structural remodelling and concomi-
tant rewiring of elastic interactions between their beads.
In Fig. 2 snapshots of network architectures along the
processes of evolution are displayed. For both designed
networks it could be observed that besides mutation-
induced changes taking place inside the individual do-
mains, in particular their common interface became sig-
nificantly remodelled. As evolution progressed the al-
losteric response generated inside the regulated pocket
upon ligand-binding to the allosteric pocket was gradu-
ally enhanced during both design processes, as can be
seen by comparing corresponding snapshots of network
conformations in Fig. 2. In the next paragraphs the two
prototype designed network structures are presented and
their dynamical properties analysed.
Designed prototype networks
The designed networks are shown in Fig. 2 (bottom
panel). Both networks retained the initial two-domain
architecture after the evolution, their structures, how-
ever, were clearly different from the initial random net-
work. In both designed networks a pronounced domain
interface with many inter-domain links had emerged;
such interface was rather sparse in the initial random
network (see Fig. 2). Remodelling of that region un-
der the process of evolution apparently points towards
the importance of the pattern of elastic connections in
the domain contact region for allosteric communication
in our model systems. Those aspects will be further dis-
cussed in the next section.
In Fig. 3 (A,a) the two designed networks are shown
each in their equilibrium conformation and in the respec-
tive steady conformations with the ligand bound to the
allosteric pocket and the allosteric response having been
triggered in the remote regulated sites. Traces showing
changes in the size of both pockets upon binding of the
ligand bead to the allosteric pockets are also provided
(see Fig. 3B,b). In both designed networks the allosteric
pocket closed rapidly when the substrate bead bound
there, with the size changing by −3.8A in both cases.
In striking contrast, motions inside the regulated pocket
were much slower, when changes in its size only gradually
set in and the pocket smoothly approached its closed con-
formation in the network with symmetric cooperativity
(change by −2.3A), or open conformation in the network
with asymmetric cooperativity, respectively (change by
+2.0A). Movies visualising conformational motions dur-
ing the allosteric transition in the designed networks are
provided as Supplementary Videos V2 and V3.
5
FIG. 4. Strain propagation. Propagation of strain subsequent to ligand-binding to the allosteric pocket is visualised in the
random network and in the two designed structures. Snapshot taken at different time moments during each simulation are
shown in similar perspectives. Network bonds are displayed bicolored, blue indicating stretching of the bond and red denoting
compression as compared to the corresponding natural bond length. The bond thickness encodes the absolute value of strain
the bond was subject to.
Comparing the ligand-bound network conformation
with the initial structure reveals that in both de-
signed networks the conformational changes underlying
allosteric dynamics did not involve any large-scale struc-
tural rearrangements but was rather governed by small-
amplitude subtle motions. In fact, initial and substrate-
bound structures compared by RMSD's of only 0.6A in
the symmetric case and 0.46A in the asymmetric case, re-
spectively. While local motions in the two pockets were
indeed pronounced, the communication between them
must therefore have been resulted from cascades of small-
amplitude displacements of network beads located in the
region connecting both pockets.
By the application of evolutionary optimisation to
the random network we could successfully design spe-
cial networks whose two-domain structure encodes en-
hanced allosteric interactions between two remote pock-
ets. All three network structures are provided as Sup-
porting Data. Next, to unravel signatures which may
underly allosteric communication in the two prototype
networks we have analysed the conformational dynamics
in terms of mechanical strain propagating through the
network structures.
Propagation of strain and communication pathways
Apparently, allosteric interactions in our model sys-
tems result from conformational changes propagating
from the allosteric site across the domain interface to
the regulated site. Therefore, in both designed struc-
tures the elastic strain inside the networks was computed
during the entire simulation following ligand binding to
the allosteric site until the steady state of the respec-
tive network was reached. The strain of a network was
determined in terms of the deformation of elastic links
connecting the beads, i.e. deviations from their natu-
ral lengths in the initial ligand-free network. Details of
the computation are found in the Methods section and
SI text. To conveniently visualise strain propagation,
the network links are displayed bicoloured, to distinguish
stretching or compression, and their width is propor-
tional to the magnitude of the respective deformation.
In order to compare to the designed networks the strain
propagation was also followed in the initial random net-
work which was allosterically inactive. Snapshots of the
strain distribution in all three networks taken at different
time moments during the simulation are shown in Fig. 4.
Movies visualising conformational motions and the prop-
agation of strain are provided as Supplementary Movies
V1-V3.
In all three networks the strain was first localised in
the vicinity of the allosteric pocket where the ligand was
bound. After that, rapidly interactions between other
beads also set in and the majority of links in domain
1 became deformed. In the random network the distri-
bution of strain in domain 1 was rather homogeneous;
most links were stretched or compressed at similar lev-
els and there was no distinctive feature present. In this
network any significant deformations of links located in
the central domain interface region could be detected and
therefore spreading of strain into the second domain dur-
ing the simulation was practically absent. The structure
of this domain is apparently very stiff preventing any
internal conformational motions to occur. Hence, the
random elastic network cannot conduct allosteric com-
munication. The situation in both designed networks
was completely different. There, first, after ligand bind-
ing to the allosteric site conformational motions spread
across the domain interface and generated deformations
in the second domain and in the vicinity of the regu-
lated site. Second, not all elastic pair interactions in
the designed structures were equally involved in the pro-
cess of allosteric coupling. Rather, the temporal distri-
bution of strain shows that communication between the
remote pockets proceeded through specific sub-networks
which were apparently critical for allostery in the de-
signed structures. Those networks are formed by a sub-
set of beads connected by springs which undergo major
deformations and, hence, contribute essential pathways
6
FIG. 5. Communication chains in designed structures.
Elastic bonds which are significantly involved in the transport
of strain are found to form simple connected chains which me-
ander through the structure and connect the allosteric pocket
in domain 1 (green beads, left side) with the regulated pocket
in the remote second domain (green beads, right side).
for the spreading of conformational motions from the al-
losteric site across the structure to the regulated site.
In both designed network structures such remarkably
strained springs were found in domain 1 and, in contrast
to the random network, in the central domain interface
region and the second domain, where they form sparse
clusters of important pair interactions between beads.
Regions in which the elastic springs remained only
marginally populated by strain were rather stiff as com-
pared to the rest of the network. In that regard we ob-
served that the structure of domain 2 in the designed net-
works is special. In the network with symmetric allosteric
cooperativity the two lobes in the tweezer-like domain
structure, each harbouring one of the pocket beads of the
regulated site, were found to be quite flexible whereas the
other parts in domain 2 were stiff (see Fig. 4). In the
network with asymmetric allosteric coupling the hinge
region connecting both lobes appears to be very flexible
instead (see Fig. 4). The specific pattern of flexibility
7
FIG. 6. Communication pathways in mutant structures. Communication chains in the designed wildtype network struc-
tures (A,a) are compared with the communication pattern obtained from strain propagation in a corresponding selected mutant
structure (B,b). Communication chains in (A,a) are the same as shown in Fig. 5 but here another perspective is chosen for
better comparison with the mutant structures. B) Communication pathways in the 52− 197− 196 double mutant of the network
with symmetric allosteric coupling. Beads with index 52,197 and 196 are highlighted in yellow and the two bonds removed
between them are shown in transparent. Links which in this mutant became important to redirect strain over the mutation site
are indicated (by an ellipse). b) Communication pathways in the 52 − 185 mutant of the designed network with asymmetric
allosteric coupling. Beads 52 and 185 are displayed in yellow and the deleted link between them is transparent. Allosteric
communication in this mutant was maintained via a complex strain transportation network and a simple chain as found in the
wildtype was not available (compare a,b).
in the designed networks has apparently emerged dur-
ing the process of evolutionary optimisation in order to
make opening/closing motions of the regulated site possi-
ble and hence enable the designed network architectures
to conduct allosteric communication.
The analysis of strain propagation in the designed
networks clearly reveals functionally important signa-
tures underlying the investigated types of allosteric cou-
pling. Nonetheless the strain distribution was rather
complex and we therefore aimed to deduce simpler path-
ways which could characterise communication between
the pockets, by focusing only on those network springs
which carry major strain (see Methods). Results are
shown in Fig. 5. In both networks simple chains formed
by significantly deformed adjacent network springs were
found to meander from the allosteric pocket in domain 1
via the domain interface to the regulated pocket in do-
main 2. Such chains are composed of a series of stretched
springs in domain 1, potentially caused by the closing
motion of the allosteric site, connected to a linkage of
compressed springs located in domain 2 and triggering
the respective allosteric response there.
The identified communication chains are obviously
critical for the transmission of conformational changes
between the two domains and therefore may represent a
functional skeleton, critical for allosteric activity in the
designed network architectures.
Immediately one may
pose the question whether, and to what extend, allosteric
coupling can be maintained if structural changes would
be applied to such motifs.
Robustness of allosteric communication
With regard to the robustness of allosteric communi-
cation in the two designed networks the intention was
not to perform a systematic screening of the effect of
structural perturbations. Instead, the aim was rather to
present a demonstration. Therefore the analysis of ro-
bustness was limited to a few exemplary structural mu-
tations, those applied to beads belonging to the identified
communication pathways, and on examining their influ-
ence on allosteric interactions. In a first set of simula-
tions mutations which consisted in the deletion of only a
single elastic link between two network beads were con-
sidered.
In the realm of proteins such kind of pertur-
bation would roughly correspond to the mutation of a
single amino acid residue, resulting in specific local inter-
actions to disappear. For each specific mutant network a
single simulation starting with ligand binding to the al-
losteric pocket was carried out and after completion the
response in the regulated pocket was quantified and the
level of allosteric communication determined. The results
are listed in Supplementary Table T1. It is found that
allostery in both designed networks is generally robust
with respect to the removal of single interactions. To un-
derstand how long-range coupling between both pockets
is maintained in the mutant networks we have visualised
the propagation of strain and determined communication
pathways for some exemplary cases, similar to what has
been performed for the designed wildtype networks (see
previous section).
For the network with symmetric cooperativity three
cases were focused on in more detail, all of which main-
tained allosteric activity (see Supplementary Table T1).
They corresponded to a mutant with one major inter-
action at the central domain interface having been re-
moved and another mutant with a link deleted in do-
main 2, closer to the regulated pocket. The computed
communication chains in those two mutants are shown
in Supplementary Fig. S3. Movies of strain propagation
in those two mutant networks are provided as Supple-
mentary Movies V4 and V5. To highlight robustness in
the first designed network, a third example consisting in a
double mutant network in which both selected links were
deleted, was considered too (see Supplementary Movie
V6). For all three mutant networks we find that the
strain propagation networks as well as the extracted com-
munication chains are very similar to that of the wild-
type networks, except in the vicinity of the respective
mutation site (see Fig. 6A,B and Fig. S3). There it is
found that in the neighbourhood of the deleted link a few
bead interactions, which in the wildtype network have
not played a major role in the transmission of strain, be-
came important, forming a bridge through which strain
was able to side-track the mutation site (see Figs. 6B
and S3). This result demonstrates that in the mutant
structures similar communication pathways are available
which, together with the bridge motifs that compensate
the mutation defect, are employed for the propagation of
conformational changes and hence may provide the foun-
dation for the robustness of allosteric coupling in this
designed network.
For the designed network with asymmetric allosteric
coupling one mutant structure with a single deleted link
located at the domain interface has been selected for il-
8
lustration. Despite the mutation it revealed full allosteric
activity. Propagation of strain subsequent to ligand bind-
ing to the allosteric pocket is shown in Supplementary
Movie V7 and the extracted communication pathways
are depicted in Fig. 6b.
It is found that the network
of bead interactions through which communication be-
tween both pockets was transmitted is rather complex,
with the strain being populated in large parts of the mu-
tant structure. Hence, in contrast to the wildtype net-
work, allosteric coupling in this mutant does not proceed
via simple communication chains.
While generally allosteric communication in both de-
signed networks was found to be robust with respect to
the deletion of a single bead interaction, there were still
few critical mutations which involved a drastic decrease
in the coupling between both pockets. They can corre-
spond to perturbations in the vicinity of either pocket
or located at the domain interface (see Supporting Figs.
S2A,B). In the final step of investigations the effect of
stronger mutations was considered, where an entire bead
was deleted in a simulation of each network. Those per-
turbations typically led to a remarkable drop or even
complete knockout of allosteric communication (see Sup-
porting Table T1).
Summary and Discussion
The phenomenon of allosteric coupling between differ-
ent functional regions within a macromolecular structure
is ubiquitously present in proteins and therefore raises
important questions of the fundamental nature of the un-
derlying mechanisms. In contrast to typically employed
structure-based modelling of protein dynamics, a dif-
ferent methodology towards approaching such essential
problems is presented in this paper. Instead of consid-
ering real protein structures, artificial analog structures
which encode pronounced long-range allosteric coupling
between two spatially remote pockets were designed. The
structures were represented by elastic networks, similar
to coarse-grained models widely used to describe confor-
mational dynamics of real proteins.
A force-probe scheme consisting in ligand-binding to
the allosteric pocket, following conformational motions
spreading over the structure, and detecting the response
generated in the regulated pocket was implemented to
evaluate the ability of the network to conduct allosteric
communication. Initially starting with a random elastic-
network, which did not reveal any allostery, a scheme of
evolutionary optimisation was iteratively applied to de-
sign two prototype elastic-network structures with per-
fected allosteric coupling, one with symmetric and the
second with asymmetric cooperativity.
In the designed networks well-defined pathways and
simple chains of important interactions, established by
only a few network beads, were identified to constitute
the signatures which underly allosteric communication.
Remote interactions were robust even if minor structural
perturbations were applied. However, a single critical
mutation could knockout completely allosteric communi-
cation in the designed networks.
While the first descriptions of allosteric systems,
namely the MWC and the KNF models [37, 38], were
formulated more than 50 years ago in the absence of any
structural data and were thus of phenomenological na-
ture, the development of very sophisticated experimen-
tal techniques and the vast amount of protein structures
becoming available in the last decades have ever since
allowed to investigate allosteric communication in pro-
teins on the molecular level and led to a decent under-
standing of the mechanism underlying allostery in several
role model proteins. In the recent past structure-based
computational modelling, aiming to follow the conforma-
tional motions as the underpinning of allosteric effects
in proteins, has gained a lot of attention.
In particu-
lar, due to the time-scale gap present in atomistic-level
molecular dynamics simulations, coarse-grained elastic
network models which are limited to the mechanical as-
pect of protein operation became very popular to inves-
tigate slow allosteric transitions with timescales beyond
the microsecond range. In those studies the analysis of
conformational changes is typically based on the com-
putation of normal modes and allosteric effects are dis-
cussed in terms of short- and long-ranged correlations of
amino-acid residue displacements.
The aim in the present study was to present a model
which emphasises the mechanical picture of allosteric sys-
tems. Therefore, also the elastic network model was
used. Here, however, the full elastic dynamics of the net-
work was considered by always numerically integrating
the non-linear equations of motion and monitoring pro-
cesses of conformational relaxation; no linearisation was
performed and the conformational dynamics beyond the
limiting normal mode approximation could be followed.
As a consequence, this model has the obvious advantage
that it can resolve the temporal order of events which
eventually establish allosteric communication in the net-
work structures, starting from forces and strains which
were first generated locally at the allosteric site as a con-
sequence of ligand binding, followed by the subsequent
propagation of conformational motions via the domain
interface, to finally induce structural changes in the re-
mote regulated pocket. The current model therefore nat-
urally includes causality as the guiding principle to mani-
fest allosteric communication and therefore is richer in its
explanatory power compared to correlation-based analy-
ses.
The presented model emphasises the structural view-
point of allostery (as described in [1]) in which allostery
is regarded as a consequence of optimised communi-
cation between the remote functional sites, established
through the propagation of strain along pathways which
9
are formed by a set of interacting residues. The em-
ployed elastic network description clearly implies limita-
tions. All network particles are of the same kind and
physical interactions between them are incorporated by
empirical effective potentials; thermal fluctuations were
also neglected for simplicity. Despite such gross simpli-
fications the dynamical properties of the designed pro-
totype allosteric structures reveal remarkable similarities
with those found in real allosteric proteins:
i) specific
parts of the structure are flexible whereas other regions
form stiff clusters; ii) the propagation of conformational
changes which results in the long-range coupling of the
remote sites proceeds through communication pathways
involving only a few of the many intra-structural inter-
actions;
iii) single critical mutations can knockout al-
losteric coupling. In summary, the designed elastic net-
work structures can provide a general physical model for
the mechanics of complex allosteric biomolecules, such as
proteins.
It should be remarked that the evolutionary pressure
imposed in the design process consisted only in magni-
fying changes in the regulatory pocket in response to
ligand-induced changes in the allosteric pocket; no other
requirements were imposed and the dynamical properties
which actually improved allosteric communication in the
evolving structures emerged autonomously. However, in
future studies design algorithms which involve coevolu-
tion can also be implemented and multiple optimisation
criteria can be imposed. Moreover, in future studies the
design and analysis of a larger number of allosteric elastic
networks can be undertaken. That would allow to com-
pare properties such as communication pathways among
them and possibly draw conclusions on the generality of
such dynamical motifs.
Previously the relaxational elastic-network approach
employed in this study was already applied to investigate
allosteric coupling in helicase motor proteins [39, 40] and
in the myosin-V molecular motor [41]. In those studies,
however, allosteric coupling was only qualitatively dis-
cussed. The analysis of strain propagation performed in
this paper for the designed artificial protein structures,
and the methods to quantify and visualise communica-
tion pathways, can easily be applied in the structure-
based modelling of real allosteric proteins and allow to
investigate intramolecular communication in dynamical
simulations.
Added note After finishing the manuscript I became
aware of two quite recently appeared works in which
also the design of mechanical networks with allosteric
coupling was undertaken (published in PNAS [42, 43]).
There, however, different computational algorithms were
used and optimisation of allosteric coupling was per-
formed within the linear response limitation; the un-
derlying network architectures were also very different
(e.g. 2D on-lattice models [42]). Both works principally
demonstrate the applicability of design processes to ob-
10
tain desired responses.
The design process introduced in my manuscript was
developed completely independent. Besides other dif-
ferences, its principal distinction is that during the de-
sign process the allosteric response was optimised by al-
ways considering the full nonlinear elastic dynamics of
the networks. The importance of nonlinearities for pro-
tein function involving allostery has previously been evi-
denced (e.g. in [44, 45]). Secondly, the network architec-
tures used in the present study resemble more closely the
three-dimensional compact fold of real proteins; in fact
the designed networks can be regarded as coarse-grained
representations of fictitious protein structures. Most im-
portantly this work goes beyond developing the design
process. In fact, the main emphasis here was put on un-
derstanding the mechanistic principles underlying long-
range communication in the designed networks, which
was achieved by visualising and analysing propagation of
strain, extracting communication pathways and chains
which were critical for allosteric coupling, and, investi-
gating robustness of the designed functional properties.
spring connecting beads i and j in some deformed network con-
formation, with (cid:126)Ri being the actual position vector of bead i, and
d(0)
is the corresponding natural spring length. Coefficients Aij
ij
have value 1 if beads i and j are connected by a spring (i.e. when
d(0)
ij < rint), and equal 0, otherwise.
The dynamics of the elastic network is governed by Newton's
equation of motion in the over-damped limit, where the velocity of
each network bead is proportional to the forces applied to it. The
equation for bead i was
(cid:126)Ri = − ∂
∂ (cid:126)Ri
U + (cid:126)fi
d
dt
= − N(cid:88)
Aij
j
dij − d(0)
ij
dij
( (cid:126)Ri − (cid:126)Rj ) + (cid:126)fi.
(1)
On the right side are the elastic forces exerted by network springs
which are connected to bead i, they only depend on the change in
the distance between two connected beads. Additionally, an exter-
nal force (cid:126)fi could be applied to bead i, which was used in probing
allostery (see next section). In the above equations a rescaled time
was used to remove dependencies of the beads' friction coefficient
(equal for all beads) and κ. To obtain the positions of network
beads, and hence the conformation of the network at any time mo-
ment, the set of equations of motion was numerically integrated.
In the simulations a first order scheme with a time step of 0.1 was
employed.
METHODS
Pocket sites and probing of allostery
Construction of random elastic network
The initial two-domain elastic network was constructed by first
randomly folding two chains of linked beads, then bringing them
into contact and finally determine the network connectivity. One
chain consisted of 100 identical beads each and its folded form was
constructed as follows. After fixing the position of the first bead,
each next bead was placed at random around the position of the
previous bead, with the following restrictions:
i) the distance to
the preceding bead had to lie within the interval between lmin and
lmax, ii) the new bead had to be separated from each previous
bead by at least the distance lmin, and, iii) the distance from the
new bead to the geometric centre of all previous beads should not
exceed the threshold rmax.
In the simulations prescribed values
lmin = 4.0A, lmax = 5.0A, and rmax = 20.0A were used in order
to generate a compactly folded backbone chain. After constructing
two such chains they were merged in such a way that they came into
tight contact but still did not overlap. Positions of the 200 beads
in the initial two-domain random structure are denoted by (cid:126)R(0)
and their spatial coordinates are provided as Supporting Data. To
complete the network construction we have checked all distances
between beads i and j and introduced an
ij = (cid:126)R(0)
d(0)
elastic spring between those pairs of beads for which the distance
d(0)
ij was below a prescribed interaction radius of rint = 9A. With
this choice the initial two-domain elastic network had 1467 links. It
should be noted that in this model all length units are in principle
arbitrary. When data from real protein structures are used, the
distances between amino acids have the scale of Angstroms. Hence,
throughout the paper this notion is adopted.
i − (cid:126)R(0)
j
i
Elastic conformational dynamics
tions of all elastic links, i.e. U = (cid:80)N
The total elastic energy of the network is the sum of contribu-
ij )2. Here,
N = 200 is the number of beads, κ is the spring stiffness constant
(equal for all springs), dij = (cid:126)Ri − (cid:126)Rj is the actual length of a
2 (dij − d(0)
i<j κ
Aij
In the network model the allosteric site and the regulates site
were, for simplicity, each represented by two beads. Those beads
have been selected in such a way that the two pockets were suffi-
ciently remote from each other. At the same time the two beads
forming one pocket should be adjacent, but not directly connected
by an elastic spring. In the constructed two-domain network the
allosteric pocket was defined by beads with indices 38 and 81, be-
longing to the first domain, and for the regulated pocket beads
from the second domain, with indices 149 and 189, were chosen.
To probe allosteric communication between the two pockets a
simple force-probe scheme in which conformational dynamics in the
allosteric site was initiated through the application of additional
forces and the subsequent response generated in the regulated site
was probed by evaluating structural changes there. In the simu-
lations pair forces have been applied to the beads of the allosteric
pocket. The forces were always acting along the direction given by
the actual positions of the two pocket beads. Specifically, the force
applied to pocket bead 38 was (cid:126)f38 = 0.5 · (cid:126)u, with the unit vector
(cid:126)u = ( (cid:126)R81− (cid:126)R38)/ (cid:126)R81− (cid:126)R38. The same force, but with the different
sign, was acting on the second pocket bead, i.e. (cid:126)f81 = −0.5 · (cid:126)f38.
Such pair forces would induce only internal network deformations
and result in a decrease in the distance between the two pocket
beads, thus leading to closing of the allosteric pocket. Since this
situation is apparently equivalent to assuming that an additional
network bead becomes bound to the centre of the two pocket beads,
where it generates attractive forces between itself and each pocket
bead, we can also say that the chosen force scheme mimics binding
of a fictitious ligand bead to the allosteric pocket. The action of
the additional forces generated deformations of the network which
were first localised in the vicinity of the substrate pocket but grad-
ually spread over the entire network structure. The corresponding
process of conformational relaxation was followed by integrating
the equations of motion until a steady state of the elastic network,
in which all motions were terminated, was reached (at final time
T , see SI text). In the resulting conformation of the network, the
effect of allosteric coupling was examined by evaluating the dis-
tance between the beads corresponding to the regulated site. This
distance d149,189(T ) = (cid:126)R149(T ) − (cid:126)R189(T ) =: A is termed the
allosteric parameter.
Evolutionary Optimisation
To design networks with perfected allosteric communication a
process of evolutionary optimisation, consisting of mutations fol-
lowed by selection, was applied iteratively starting from the ran-
dom network. In particular the following sequence was carried out.
First the allosteric response of the elastic network before the muta-
tion was determined and the allosteric parameter A stored. Then
a single structural mutation was performed by randomly select-
ing one network bead (except for one of the four pocket beads)
and changing its equilibrium position. The new equilibrium posi-
tion was chosen to be randomly oriented within a sphere of radius
2.0A around the old equilibrium position. To preserve the back-
bone chain of each domain, it was additionally required that, after
the mutation, the distance between the mutated bead and its left
and right neighbour in the chain still lie within the interval be-
tween lmin and lmax. Furthermore, distances from the mutated
bead to all other network beads should not be smaller than lmin.
After the mutation the network connectivity Aij was updated by
reexamining distances between all beads and the mutated bead;
only those pairs which were separated by a distance less than rint
were linked by a spring. After a mutation the elastic network may
posses internal free rotations originating from loosely coupled net-
work parts. They can lead to local movements free of energetic
cost which was to be prevented. Therefore, when the number of
non-zero eigenvalues in the spectrum of the elastic network was
smaller than 3N − 6 (indicating the occurrence of internal rotation
zero modes), the mutation was rejected. Once a mutation which
fulfilled all criteria was found, probing of allostery in the new elas-
tic network was proceeded as described in the previous section, the
allosteric parameter after the mutation Amut was determined, and
the mutation was evaluated by comparing the allosteric parameter
before the mutation A with that after the mutation Amut. Only
mutations which were favourable, i.e. those which improved the al-
losteric response in the network, were selected. Two situations were
distinguished. In the evolution process where symmetric coupling
between the allosteric and regulated pocket was to be optimised, a
mutation was accepted only if Amut < A, and rejected otherwise.
In the second independent evolutionary process corresponding to
the asymmetric situation, the acceptance criteria for the mutations
was Amut > A. As a termination condition for the two evolution
processes we imposed (A− Arandom) < −2.0A for the design of the
network with symmetric coupling and (A − Arandom) > 2.0A for
the design of the network with asymmetric coupling. Arandom de-
notes the allosteric parameter of the initial random network. Dur-
ing both design processes the improvement of allosteric response in
the evolving networks was recorded and it shown in Supplementary
Fig. S1.
Strain propagation and pathways
In the initial random, in the two designed networks, and in the
selected mutant networks, the propagation of strain after ligand
binding to the allosteric pocket was monitored. The strain of
an elastic link connecting beads i and j was defined as sij (t) =
dij (t) − d(0)
ij . In the employed model conformational changes cor-
responded to relaxation processes of the elastic network structure
(see Equ.'s (1)). Therefore, the energy injected locally at the al-
losteric site as a consequence of ligand-binding there would be not
only converted into deformations of elastic bead connections but
also dissipate. In particular deformations of elastic springs become
significantly damped the farther away they are located from the
pocket. Since we still wanted to discuss the anisotropy of strain
distribution in the network, a method in which the strain was nor-
malised with respect to the distance from the allosteric site (in
terms of the minimal path) was employed. Details are described
in the Supplementary Information. For the visualisation of strain
11
a link was coloured blue (if sij > 0) or red (sij < 0) and the
width corresponded to sij (superscript ∼ refers to the normalised
strain). To determine the communication chains shown in Fig. 5
the maximum absolute strain of each link during the simulation
was stored and only those links whose normalised strain exceeded
a prescribed threshold st were considered (see SI text). For both
designed networks a threshold value of st = 60% was imposed. To
obtain the communication skeleton in the 52−196−197 double mu-
tant of the designed network with symmetric cooperativity (shown
in Fig. 6B) a threshold of st = 45% was used. For the 52− 185 mu-
tant of the designed network with asymmetric cooperativity (Fig.
6b) a threshold of st = 25% was used.
Robustness
All performed mutations are listed in SI Table T1. For each
mutant network a robustness coefficient was computed as the ra-
tio of the change in the pocket size of the regulated pocket in the
considered mutant network and the pocket size change in the wild-
type network, i.e. ( d(0)
149,189). Here,
superscript ∼ denotes distances in the mutant network and super-
script f inal denotes distances in the corresponding steady state of
the elastic network after ligand binding to the allosteric site (i.e.
at time T ).
149,189 − df inal
149,189)/(d(0)
149,189 − df inal
ACKNOWLEDGEMENTS
The author is grateful to Alexander S. Mikhailov and
Yuichi Togashi for helpful discussions. This work is sup-
ported by JSPS KAKENHI Grant Number 16K05518.
All figures of networks and the movies have been pre-
pares using the VMD software [46].
∗ [email protected]
[1] Tsai C-J, Nussinov R (2014) A unified view of "how al-
lostery works". PLoS Comput Biol 10: e1003394.
[2] Brueschweiler S, Schanda P, Kloiber K, Brutscher B,
Kontaxis G, et al. (2009) Direct observation of the dy-
namic process underlying allosteric signal transmission.
J Am Chem Soc 131: 3063-3068.
[3] Grutsch S, Bruschweiler S, Tollinger M (2016) NMR
methods to study dynamic allostery. PLoS Comput Biol
12(3): e1004620.
[4] del Sol A, Fujihashi H, Amoros D, Nussinov R (2006)
Residues crucial for maintaining short paths in network
communication mediate signaling in proteins. Mol Syst
Biol 2
[5] del Sol A, Arauzo-Bravo MJ, Amoros D, Nussinov R
(2007) Modular architecture of protein structures and
allosteric communications: potential implications for sig-
naling proteins and regulatory linkages. Genome Biol 8:
R92.
[6] Daily MD, Upadhyaya TJ, Gray JJ (2008) Contact rear-
rangements form coupled networks from local motions in
allosteric proteins. Proteins 71: 455-466.
[7] Chennubhotla C, Bahar I (2006) Markov propagation of
allosteric effects in biomolecular systems: application to
GroEL-GroES. Mol Syst Biol 2: 36.
[8] Chennubhotla C, Bahar I (2007) Signal propagation in
proteins and relation to equilibrium fluctuations. PLoS
Comput Biol 3(9): e172.
[9] Lockless SW, Ranganathan R (1999) Evolutionarily con-
served pathways of energetic connectivity in protein fam-
ilies. Science 286: 295-299.
[10] Suel GM, Lockless SW, Wall MA, Ranganathan R (2003)
Evolutionarily conserved networks of residues mediate al-
losteric communication in proteins. Nat Struct Biol 10:
59-69.
[11] Dima RI, Thirumalai (2006) Determination of network of
residues that regulate allostery in protein families using
sequence analysis. Protein Sci 15: 258-268.
[12] Tang S, Liao JC, Dunn AR, Altman RB, Spudich JA, et
al. (2007) Predicting allosteric communication in myosin
via a pathway of conserved residues. J Mol Biol 373:
1361-1373.
[13] Ma J, Sigler PB, Xu Z, Karplus M (2000) A dynamic
model for the allosteric mechanism of GroEL. J Mol Biol
302: 303-313.
[14] Ghosh A, Vishveshwara S (2007) A study of communica-
tion pathways in methionyl-tRNA synthetase by molecu-
lar dynamics simulations and structure network analysis.
Proc Natl Acad Sci USA 104: 15711-15716.
[15] Cui Q, Karplus M (2008) Allostery and cooperativity re-
visited. Protein Sci 17: 1295-1307.
[16] Dixit A, Verkhivker GM (2011) Computational model-
ing of allosteric communication reveals organizing prin-
ciples of mutation-induced signaling in ABL and EGFR
kinases. PLoS Comput Biol 7(10): e1002179.
[17] Laine E, Auclair C, Tchertanov L (2012) Allosteric com-
munication across the native and mutated KIT receptor
tyrosine kinase. PLoS Comput Biol 8(8): e1002661.
[18] Naithani A, Taylor P, Erman B, Walkinshaw MD (2015)
A molecular dynamics study of allosteric transitions in
leishmania mexicana pyruvate kinase. Biophys J 109:
1149-1156.
[19] Hertig S, Latorraca N, Dror R (2016) Revealing atomic-
level mechanisms of protein allostery with molecular dy-
namics simulations. PLoS Comput Biol 12(6): e1004746.
[20] Xu C, Tobi D, Bahar I (2003) Allosteric changes in pro-
tein structure computed by a simple mechanical model:
hemoglobin T-R2 transition. J Mol Biol 333: 153-168.
[21] Zheng W, Brooks BR, Thirumalai D (2006) Low-
frequency normal modes that describe allosteric transi-
tions in biological nanomachines are robust to sequence
variations. Proc Natl Acad Sci USA 103: 7664-7669.
[22] Chennubhotla C, Yang Z, Bahar I (2008) Coupling be-
tween global dynamics and signal transduction pathways:
a mechanism of allostery for chaperonin GroEL. Mol
BioSyst 4: 287-292.
[23] Yang Z, Majek P, Bahar I (2009) Allosteric transitions
of supramolecular systems explored by network models:
application to chaperonin GroEL. PLoS Comput Biol 5:
e1000360.
[24] Morra G, Verkhivker G, Colombo G (2009) Modeling sig-
nal propagation mechanisms and ligand-based conforma-
tional dynamics of the Hsp90 molecular chaperone full-
length dimer. PLoS Comput Biol 5(3): e1000323.
[25] Erman B (2013) A fast approximate method of identify-
ing paths of allosteric communication in proteins. Pro-
teins 81: 1097-1101.
[26] Krieger J, Bahar I, Greger IH (2015) Structure, dynam-
ics, and allosteric potential of ionotropic glutamate re-
12
ceptor N-terminal domains. Biophys J 109: 1136-1148.
[27] Bahar I, Chennubhotla C, Tobi D (2007) Intrinsic dy-
namics of enzymes in the unbound state and relation to
allosteric regulation. Curr Opin Struct Biol 17: 633-640.
[28] Bahar I, Lezon TR, Yang L-W, Eyal E (2010) Global
dynamics of proteins: bridging between structure and
function. Annu Rev Biophys 39: 23-42.
[29] Daily MD, Gray JJ (2007) Local motions in a benchmark
of allosteric proteins. Proteins 67: 385-399.
[30] Togashi Y, Mikhailov AS (2007) Nonlinear relaxation dy-
namics in elastic networks and design principles of molec-
ular machines. Proc Natl Acad Sci USA 104: 8697-8702.
[31] Cressman A, Togashi Y, Mikhailov AS, Kapral R (2008)
Mesoscale modeling of molecular machines: cyclic dy-
namics and hydrodynamical fluctuations. Phys Rev E 77:
050901(R).
[32] Huang M-J, Kapral R, Mikhailov AS, Chen H-Y (2013)
Coarse-grain simulations of active molecular machines in
lipid bilayers. J Chem Phys 138: 195101.
[33] Sakaue T, Kapral R, Mikhailov AS (2010) Nanoscale
swimmers: hydrodynamic interactions and propulsion of
molecular machines. Eur Phys J B 75: 381-287.
[34] Sarkar A, Flechsig H, Mikhailov AS (2016) Towards syn-
thetic molecular motors: a model elastic-network study.
New J Phys 18: 043006.
[35] Tirion MM (1996) Large amplitude elastic motions in
proteins from a single-parameter, atomic analysis. Phys
Rev Lett 77: 1905-1908.
[36] Haliloglu T, Bahar I, Erman B (1997) Gaussian dynamics
of folded proteins. Phys Rev Lett 79: 3090-3093.
[37] Monod J, Wyman J, Changeux JP (1965) On the nature
of allosteric transitions: a plausible model. J Mol Biol 12:
88-118.
[38] Koshland DE, Nemethy G, Filmer D (1966) Comparison
of experimental binding data and theoretical models in
proteins containing subunits. Biochemistry 5: 365-385.
[39] Flechsig H, Mikhailov AS (2010) Tracing entire operation
cycles of molecular motor hepatitis C virus helicase in
structurally resolved dynamical simulations. Proc Natl
Acad Sci USA 107: 20875-20880.
[40] Flechsig H, Popp D, Mikhailov AS (2011) In silico in-
vestigation of conformational motions in superfamily 2
helicase proteins. PLoS ONE 6(7): e21809.
[41] Duettmann M, Togashi Y, Yanagida T, Mikhailov AS
(2012) Myosin-V as a mechanical sensor: An elastic net-
work study. Biophys J 102: 542-551.
[42] Yan L, Ravasio R. Brito C, Wyart M (2017) Architecture
and coevolution of allosteric materials. Proc Natl Acad
Sci USA (early edition)
[43] Rocks JW, Pashine N, Bischofberger I, Goodrich CP, Liu
AJ, et al. (2017) Designing allostery-inspired response
in mechanical networks. Proc Natl Acad Sci USA (early
edition).
[44] Miyashita O, Onuchic JN, Wolynes PG (2003) Nonlin-
ear elasticity, proteinquakes, and the energy landscapes
of functional transitions in proteins. Proc Natl Acad Sci
USA 100: 12570-12575.
[45] Togashi Y, Yanagida T, Mikhailov AS (2010) Nonlinear-
ity of mechanochemical motions in motor proteins. PLoS
Comput Biol 6: e 1000814.
[46] Humphrey W, Dalke A, Schulten K (1996) VMD - Visual
molecular dynamics. J Molec Graphics 14: 33-38.
SUPPLEMENTARY INFORMATION
PROBING OF ALLOSTERY AND
EVOLUTIONARY OPTIMISATION
The force-probe employed scheme to score the al-
losteric response in the networks during the design
processes was explained in the main text. To follow
conformational motions in a network subsequent to
ligand-binding to the allosteric site, the set of equations
(Equ.'s (1), main text) was numerically integrated,
always until a steady state of that network in which mo-
tions were sufficiently terminated was reached (the time
needed was referred to as T ). As a termination condition
for the process of numerical integration a requirement
for the instantaneous bead velocity (averaged over all
beads) to drop below a prescribed threshold was imposed
vi < 10−6.
in the simulations. The condition was 1
N
(cid:80)N
i
During the process of evolutionary optimisation in the
two designed networks the improvement of of the al-
losteric response was recorded. It is shown in Fig. S1.
Details are given in the figure caption.
STRAIN PROPAGATION AND
COMMUNICATION PATHWAYS
To discuss the anisotropy of the strain distribution in
the random and designed networks we have introduced
a method to normalise the strain of elastic links with
respect to the distance from the allosteric site. There-
fore, we introduced the shortest graph distance between
a network bead with index i and a bead of the allosteric
pocket with index p as the minimal number of links of
a path connecting the two beads, denoted by Dip. The
shortest graph distance of an elastic spring connecting
beads i and j to the allosteric pocket was then defined
as Dij := min{Dip1, Dip2, Djp1, Djp2}, where p1 and p2
were the indices of beads of the allosteric pocket and
p3 and p4 were those of the regulated pocket. Next
we defined shells Sn of links, each of which contained
all those elastic links (ij) that had the same shortest
graph distance n (n = 1, 2, . . . ) to the allosteric pocket,
i.e. Sn := {all links (ij) for which Dij = n}. Hence,
each elastic network link (ij) was uniquely assigned to
one shell Sn and its elastic strain at time t (defined in
the Methods section of the main text) was denoted by
sij(t, n). After this procedure, obviously all elastic links
had been sorted into their respective shells; all links with
shortest graph distance 1 were in shell S1, those with
shortest graph distance 2 were in shell S2, etc. Now,
during a first simulation of the allosteric operation of a
designed network - following ligand binding to the al-
losteric site until the steady conformation was reached
- for each shell Sn a maximum absolute strain value
13
mn = max{sij(t, n)} of all links belonging to the same
shell Sn was determined and stored. Then, in a repeated
simulation of the same network, the strain of each elas-
tic link (ij) was normalised by dividing its value by the
maximum absolute strain value mn of the shell n, the
particular link belonged to, i.e. sij(t) = sij(t, n)/mn.
The normalised strain sij(t) was used to visualise the
strain propagation in both designed networks in a time-
resolved fashion, presented in Supplementary Movies V2
and V3. Corresponding snapshots are shown in the main
text Fig. 4.
In the same simulation we have memo-
rised for each link (ij) the maximum absolute value mij
of its normalised strain, i.e. mij = maxsij(t). Those
values were employed to determine the communication
pathways (shown in the main text Fig. 5), which were
constructed by considering only those links that were sig-
nificantly involved in the strain propagation, imposed by
the condition mij > st, where st was a prescribed thresh-
old value for the normalised strain (with its values given
in the main text Methods section).
In the random elastic networks a similar procedure of
strain normalisation was undertaken. However, to com-
pare the propagation of strain to that in the two designed
networks, the link strain was normalised with respect to
the coefficients mn from the designed network with sym-
metric allosteric coupling.
For the selected mutant networks the procedure of strain
normalisation was also applied to visualise strain prop-
agation during the allosteric transition (Movies V4-V7)
and to construct communication pathways (Fig. 6 main
text and Fig. S3).
It should be noted that for the normalisation procedure
the shortest graph distance between a network bead and
the a bead of the allosteric pocket was determined with
a standard algorithm using the powers of the adjacency
matrix.
ROBUSTNESS OF ALLOSTERIC
COMMUNICATION
Robustness of allosteric communication with respect
to exemplary single structural mutations applied to each
of the two designed prototype networks was analysed. To
determine the communication chains for the two mutants
of the designed network with symmetric allosteric com-
munication shown in Fig. S3, a threshold of st = 55%
was used for the 52−197 mutant, and st = 60% was used
for the 196 − 197 mutant.
SUPPORTING MOVIES AND DATA
Time-dependent propagation of strain in the random
network, the two designed networks, and the selected mu-
tant networks are provided as Supporting Movies V1-V7.
In each of the movies the frame rate is not constant and
has been adapted such the fast dynamics inside the al-
losteric pocket, the inter-domain propagation, as well as
the slow motions inside the regulated pocket, can be con-
veniently followed. The actual time during the simulation
is always given at the bottom right in the movies.
• Supporting Movie V1 Conformational motions
and strain propagation in the allosterically inactive
initial random elastic network.
• Supporting Movie V2 Conformational motions
and strain propagation in the designed network
with symmetric allosteric coupling.
• Supporting Movie V3 Conformational motions
and strain propagation in the designed network
with asymmetric allosteric coupling.
• Supporting Movie V4 Conformational motions
and strain propagation in the 52−197 mutant of the
designed network with symmetric allosteric cou-
pling.
• Supporting Movie V5 Conformational motions
and strain propagation in the 196 − 197 mutant
of the designed network with symmetric allosteric
coupling.
14
• Supporting Movie V6 Conformational motions
and strain propagation in the 52−197−196 double-
mutant of the designed network with symmetric al-
losteric coupling.
• Supporting Movie V7 Conformational motions
and strain propagation in the 52−185 mutant of the
designed network with asymmetric allosteric cou-
pling.
i
As Supplementary Data the set of spatial coordinates
(cid:126)R(0)
of the initial random network and the two designed
networks are provided in the respective equilibrium con-
formation.
• random network.dat:
text file containing spatial
coordinates of the initial random network. 1st col-
umn: bead index (0 to 199); 2nd,3rd,4th columns:
x,y,z coordinate of the bead position.
• designed network1.dat: text file containing spatial
coordinates of the designed network with symmet-
ric allosteric coupling. 1st column: bead index (0
to 199); 2nd,3rd,4th columns: x,y,z coordinate of
the bead position.
• designed network2.dat: text file containing spatial
coordinates of the designed network with asymmet-
ric allosteric coupling. 1st column: bead index (0
to 199); 2nd,3rd,4th columns: x,y,z coordinate of
the bead position.
15
FIG. S 1. Improvement of the allosteric response in the evolving networks during the two independent processes of evolutionary
optimisation; d149,189(T ) is the allosteric parameter of the elastic network at the current stage of evolution and drandom
149,189 (T ) is
the value in the initial random elastic network. A successful evolution step corresponded to an accepted mutation, i.e. one
after which the allosteric response of the network had improved. The total number of successful evolution steps was 1, 138 in
the design of the network with symmetric cooperativity and 744 in the design of the network with asymmetric cooperativity,
respectively.
16
FIG. S 2. and Table T1. Mutations and robustness of allosteric communication. A,B): Communication chains in the
designed wildtype networks are shown in the same perspective as in Fig. 5 (main text), with the beads which were subjected
to mutations being indicated by the respective bead indices. Mutations performed in both designed networks are listed in the
table on the right side. A colour scale from blue to red shall illustrate neutral to fatal mutations, according to the respective
robustness coefficient (defined in the main text Methods section). For the 52 − 197 − 196 double mutant of the network with
symmetric coupling (which is not listed in the Table) the robustness coefficient was 0.99.
FIG. S 3. Communication chains in selected mutant structures. For the designed network with symmetric cooperativity
the communication chains in two representative mutants, each in which a single elastic link was deleted, are shown. The 52−197
mutant (A) as well as the 196 − 197 mutant (B) both retained full allosteric activity, with robustness coefficients of 1.00 and
0.98, respectively (see Table T1). In each mutant the deleted link is shown in transparent and the corresponding beads are
highlighted in yellow.
|
1703.05188 | 1 | 1703 | 2017-03-15T14:45:54 | Negative membrane capacitance of outer hair cells: electromechanical coupling near resonance | [
"physics.bio-ph"
] | The ability of the mammalian ear in processing high frequency sounds, up to $\sim$100 kHz, is based on the capability of outer hair cells (OHCs) responding to stimulation at high frequencies. These cells show a unique motility in their cell body coupled with charge movement. With this motile element, voltage changes generated by stimuli at their hair bundles drives the cell body and that, in turn, amplifies the stimuli. In vitro experiments show that the movement of these charges significantly increases the membrane capacitance, limiting the motile activity by additionally attenuating voltage changes. It was found, however, that such an effect is due to the absence of mechanical load. In the presence of mechanical resonance, such as in vivo conditions, the movement of motile charges is expected to create negative capacitance near the resonance frequency. Therefore this motile mechanism is effective at high frequencies. | physics.bio-ph | physics |
Negative membrane capacitance of outer hair cells:
electromechanical coupling near resonance
Kuni H. Iwasa
35A Convent Dr., Rm 1F242A
NIDCD, National Institutes of Health
Bethesda, Maryland 20892
∗
Abstract
The ability of the mammalian ear in processing high frequency sounds, up to ∼100 kHz,
is based on the capability of outer hair cells (OHCs) responding to stimulations at high
frequencies. These cells show a unique motility in their cell body coupled with charge
movement. With this motile element, voltage changes generated by stimuli at their hair
bundles drives the cell body and that, in turn, amplifies the stimuli. In vitro experiments
show that the movement of these charges significantly increases the membrane capacitance,
limiting the motile activity by additionally attenuating voltage changes.
It was found,
however, that such an effect is due to the absence of mechanical load.
In the presence
of mechanical resonance, such as in vivo conditions, the movement of motile charges is
expected to create negative capacitance near the resonance frequency. Therefore this motile
mechanism is effective at high frequencies.
key words: mammalian ear, cochlear amplifier, piezoelectricity
Introduction
The remarkable sensitivity and frequency bandwidth as high as 100 kHz, depending on
animal species [1], of the mammalian hearing is based on the ability of its ear to function
a frequency analyzer [2]. The frequency components are then transferred to the brain in
parallel by a bundle of neurons. Thus a key question is how a cell-based biological system
can be sensitive and capable of operating at high frequencies.
It has been found that an amplifier that counteracts viscous drag is essential for a
sensitive mechanoeletrical analyzer such as the mammalian ear [3, 4] and that outer hair
cells (OHCs) play such a key role [5, 6]. These cells have a motile mechanism in their
cell body based on piezoelectricity, called "somatic motility" or "electromotility," utilizing
electrical energy [7 -- 11]. The electric potential that is used by the motile mechanism is
generated by mechanotransducer current of the sensory hair bundle of these cells, in response
to mechanical stimuli. This process is assisted by the endocochlear potential, the unusual
positive potential in the K+-rich endolymphatic space, generated by the stria vascularis.
∗
email: [email protected]
1
Indeed, the electrical energy and the ionic environment provided to OHCs are exceptional.
However, a question remains as to how the OHCs can be effective at high frequencies
because viscous drag increases with the frequency while the capacitive conductance of the
basolateral membrane increases with frequency and significantly attenuates the receptor
potential, which drives the motile mechanism in the cell body of OHCs [12].
This puzzle has been called the "RC time constant" problem, leading to a dispute
regarding the basis for the amplifying role of OHCs: active process in the hair bundle alone
[13], or somatic motility coupled with hair bundle transduction [14, 15], or a combination
of both [16]. The second point of view was examined by considering various mechanisms
that could improve the effectiveness of somatic motility [17 -- 23].
Despite their differences, all these analyses assume that the membrane capacitance,
which consists of two components, linear and nonlinear, is unaffected by the mechanical load
on OHCs. Of the two components, the linear component is structural, primarily based on the
capacitance of the plasma membrane. Nonlinear component is due to the movement of the
motile charge, which is associated with the motile function of the cell. This component has
a bell-shaped membrane potential dependence in the load-free condition and the magnitude
of this component at its peak can be larger than the linear capacitance [7, 8]. For this
reason, the motor charge appeares to enhance "RC attenuation," and thereby to decrease
the effectiveness of OHCs as an amplifier [12].
A recent analysis, however, showed that nonlinear capacitance should depend on me-
chanical load, leading to a prediction that viscous drag increases mechanical energy output
of OHCs by reducing the attenuation by motile charges [24]. Here it is shown by using a
simple model system that the effect of mechanical resonance is more substantial. It can fully
nullify the membrane capacitance, increasing energy output of OHCs. The implications of
the finding to the cochlea are discussed by connecting the input and output by additional
assumptions and by examining the resulting inequality that describes the upper bound of
the effectiveness of OHCs.
Model System
Here we consider a system with an OHC, which is connected to a spring with stiffness K, a
dashpot with friction coefficient η, and a mass m. We assume here that the cell has n motile
elements, which has two discrete states, compact and extended, and during a transition from
the compact state to the extended state, the cell length increases by a and the electric charge
q flips across the plasma membrane. The axial stiffness of the cell is k (Fig. 1). Since a
set of the equations that govern this system has been derived previously [24], only a brief
description is given below.
Basic Equations
Let P be the fraction of the motile units in the extended state. Its equilibrium value P∞
follows the Boltzmann distribution,
P∞ =
exp[−β∆G]
1 + exp[−β∆G]
,
2
(1)
Figure 1: Mechanical connectivity and the equivalent elec-
tric circuit of the model system. The system is driven by
changes in hair bundle conductance Ra. Unlike in vivo con-
dition, movement of the cell body does not affect Ra. In the
mechanical schematics (A), K is stiffness of the external me-
chanical load, m the mass, and the drag coefficient is η. The
contribution of the motile element to cell length x is pro-
portional to anP , where P , a, and n respectively represent
the fraction of the motile elements in the elongated state,
unitary length change, and the number of such units, the
unitary change of charge of which is q. The stiffness of the
cell due to the material property is k. The broken line indi-
cates the border of the OHC. In the equivalent circuit (B),
the membrane potential is V , the basolateral resistance Rm,
and the total membrane capacitance of the basolateral mem-
brane Cm, consisting of the structural capacitance C0 and
the contribution of charge movements in the motile element,
which depends on the load. The endocochlear potential is
eec and the potential eK is due to K+ permeability of the
basolateral membrane. The apical capacitance is ignored in
this model.
with β = 1/(kBT ), where kB is Boltzmann's constant and T the temperature, and
∆G = q(V − V1/2) + Ka2n(P − P0),
(2)
where K = kK/(k + K). If the system has not reached equilibrium, P∞ can be regarded as
the target value of P for the given set of variables, i.e. V and force applied to the motile
unit, at a given moment, and changes take place toward that goal. The equation of motion
can then be expressed,
m
d2P
dt2 + η
dP
dt
= (k + K)(P∞ − P ),
which is intuitive for m = 0. The receptor potential V is described by,
eec − V
Ra
V − eK
Rm
=
+ C0
dV
dt
− nq
dP
dt
.
(3)
(4)
Here Ra is the apical membrane resistance, which is dominated by mechanotransducer
channels in the hair bundle. The basolateral membrane has the resistance Rm and the
linear capacitance C0, which is determined by the membrane area.
Response to Small Oscillatory Stimuli
Here we assume small periodic changes with an angular frequency ω from a resting resistance
¯Ra of the hair bundle resistance,
Ra(t) = ¯Ra(1 + r exp[iωt]).
3
AKmηkanPBeecRaRmeKCmV The response of the system should be described by small periodic changes of the variables
from their steady state values:
V (t) = ¯V + v exp[iωt],
P∞(t) = ¯P∞ + p∞ exp[iωt],
P (t) = ¯P + p exp[iωt],
where the variables expressed in lower case letters are small and those marked with bars on
top are time-independent. Namely, ¯V = (eecRm + eK ¯Ra)/(Rm + ¯Ra) and ¯P = ¯P∞. Thus,
¯P is expressed by Eq. 1 with ∆G, in which P is replaced by ¯P .
The equations for the small amplitudes are,
p∞ = −γ(qv + a2n Kp),
(cid:18) 1
(cid:19)
[− (ω/ωr)2 + i ω/ωη]p = p∞ − p,
− eec − V
Ra
r =
+
1
Rm
Ra
v + iω(C0 − nq · p),
(5)
(6)
(7)
where the resonance frequency ωr, viscous roll-off frequency ωη, and definitions of parame-
ters K and γ are given by,
ω2
r = (k + K)/m,
ωη = (k + K)/η,
γ = βP (1 − P ).
Eq. 7 for the receptor potential can be made simpler by introducing two parameters,
i0 = (eec − eK)/( ¯Ra + Rm), and σ = 1/Ra + 1/Rm into
−i0r = (σ + iωC0)v − iωnqp.
The combination of Eqs. 5 and 6 leads to,
[− (ω/ωr)2 + i ω/ωη + (1 + γa2n K)]p = −γqv.
(8)
(9)
Here let us introduce a parameter α2 = 1 + γa2n K for brevity. Note here that α = 1 in the
absence of external elastic load and otherwise α > 1. The contribution Cnl of the motor
charge to the membrane capacitance Cm is given by Cnl = (qn/v)Re[p]. This leads to,
Cm = C0 + Cnl,
Cnl =
γnq2[α2 − ¯ω2]
[α2 − ¯ω2]2 + (¯ω/¯ωη)2 ,
(10)
where ¯ω = ω/ωr, ¯ωη = ωη/ωr, and C0 is the regular membrane capacitance proportional
= γq2n in the absence of
to the membrane area of the cell. Eq. 10 indicates that C
mechanical load, consistent with earlier studies [24 -- 26].
(max)
nl
Voltage oscillation v exp[iωt] generates current iωp exp[iωt]. The admittance is given by
Y (ω) = iωp/v. Since the fluctuation-dissipation theorem [27, 28] relates the admittance to
the power spectrum of current noise with the formula SI (ω) = 4kBT Im[Y (ω)], we have
SI (ω) =
4 ¯P (1 − ¯P )q · ωr/¯ωη · ¯ω2
[α2 − ¯ω2]2 + (¯ω/¯ωη)2 .
(11)
4
A
B
Figure 2: Nonlinear capacitance Cnl and power spectral density SI (ω) of current noise. A:
Nonlinear capacitance plotted against ¯ω(= ω/ωr). Nonlinear capacitance Cnl is normalized by
γnq2. B: Power spectral density of current noise is plotted against ¯ω. SI (ω) is normalized by
S0(= 4 ¯P (1 − ¯P )qωr). Traces respectively correspond to the values of ¯ωη, 1 (black), 2 (blue), and 5
(red).
It has a peak 4 ¯P (1 − ¯P )qωη at ¯ω2 = α2 (Fig. 2B). This spectral shape is quite different
from that without mechanical resonance, which has high-pass characteristics [29, 30].
Now let us examine power output elicited by hair bundle stimulation. Since the voltage
change v is the result of a change r in the hair bundle resistance as described by Eq. 8, it
is expressed by,
By combining Eqs. 9 and 12, we obtain,
−i0r + iωnqp
σ + iωC0
.
v =
(cid:19)
(cid:35)
(cid:34)
−
(cid:18) ω
(cid:19)2
ωr
(cid:18) 1
ωη
+ iω
+
γnq2
σ + iωC0
+ α2
p =
(12)
(13)
γqi0
σ + iωC0
r.
High Frequency Asymptote
Since we are interested is in high frequency range, we may assume σ + iωC0 → iωC0. Then
Eq. 13 can be simplified into(cid:34)
(cid:18) ω
(cid:19)2
(cid:35)
−
ωr
+ i
ω
ωη
+ α2 + ζ
p = −i
γqi0
ωC0
r,
(14)
with a parameter ζ = γq2n/C0, which is the ratio of maximal nonlinear capacitance to the
regular capacitance.
The work against drag per half cycle is Ed = (1/2)ηω( K/K)2nap2. Power output Wd,
which is 2ω/(2π)Ed, can be expressed by
Wd =
(γanqi0)2
[α2 + ζ − ¯ω2]2 + (¯ω/¯ωη)2 ·
ηk2r2
2π(k + K)2C2
0
,
(15)
5
0.20.5125-2-1123=5210.20.512512345=521 A
B
Figure 3: Power output per unit resistance change (r = 1). A: Frequency dependence of power
output, assuming α2 + ζ = 1. Power output W (¯ω) is normalized by W0 = (γanqi0)2ηk2/[2π(k +
K)2C 2
0 ]. Traces correspond to the values of ¯ωη: 1, (black); 2, (blue); and 3 (red). B: Maximum
power output plotted against ¯ωη. The scale of power output is the same as in A. Traces correspond
to the values of α2 + ζ: 1, (black); 1.5, (blue); and 2 (red).
using a reduced frequency ¯ω = ω/ωr, and ¯ωη = ωη/ωr.
Eq. 15 is maximized at ¯ω2 = α2 + ζ − 1/(2¯ω2
η) and the maximal value is,
W
(max)
d
=
4(γanqi0)2 ¯ω4
η
η − 1
4(α2 + ζ)¯ω2
·
ηk2r2
2π(k + K)2C2
0
.
(16)
Notice that the maximum W
α2 + ζ − 1/(2¯ω2
approximated by
η) > 0.
(max)
d
If ¯ωη is sufficiently large to satisfy 4(α2 + ζ)¯ω2
is a monotonically increasing function of ¯ωη because
η (cid:29) 1, it can be
W
(max)
d
Negative Capacitance
≈ (γanqi0)2 ¯ω2
η
α2 + ζ
·
ηk2r2
2π(k + K)2C2
0
.
(17)
To understand the condition for maximizing power output, it would be instructive to ex-
amine the value of the membrane capacitance Cm.
η (cid:29) 1 makes
The condition for maximizing power output, ¯ω2 = α2 + ζ − 1/(2¯ω2
η) with ¯ω2
nonlinear capacitance negative:
C0 + Cnl
C0
≈ − 2α2 − ζ
2ζ 2
1
¯ω2
η
.
(18)
This means that, under this condition, nonlinear capacitance Cnl cancels out the regular
capacitance C0. The net membrane capacitance Cm is small in magnitude because of the
factor 1/¯ω2
η and likely negative because α2 > 1 by definition and experimental data usually
show ζ (cid:47) 2.
6
0.20.512524683211234501020125 Receptor Potential
The amplitude of the receptor potential v can be obtained from Eqs. 9 and 13 by eliminating
p. At high frequencies, this combination leads to
· α2 − ¯ω2 + i¯ω/¯ωη
α2 + ζ − ¯ω2 + i¯ω/¯ωη
i0r
ωC0
v = i
(19)
which consists of two factors. The first factor i0/(ωC0) can be expressed as (eec−eK)/[ω( ¯Ra+
Rm)C0], recalling the definition of i0. It indicates low-pass attenuation with a time constant
τRC = ( ¯Ra + Rm)C0. The second factor represents enhancement near ¯ω2 ≈ α2 + ζ. The
magnitude of v at the peak frequency can be expressed,
,
vmax ≈ ¯ωη
ωrτRC
·
(20)
assuming ¯ωη (cid:29) 1. The first factor on the right-hand-side can be expressed in a more sym-
metric form: ωηωRC/ω2
r . This expression illustrates the presence of attenuation, which is
still determined by C0, even though the net membrane capacitance Cm is virtually elimi-
nated by the motile charge.
α2 + ζ
ζ
· (eec − eK)r,
Frequency Limit
Here the results obtained for our simple model system (Fig. 1) are examined for implications
to the mammalian cochlea, a complex system, specifically with regard to the limit of the
effectiveness of OHCs for amplifying the oscillation in the cochlea.
This examination is based on two major additional assumptions as in a previous treat-
ment [18]: that the output of OHC feeds back to hair bundle displacement and that the
major source of the drag is the shear in the gap between the tectorial membrane and the
reticular lamina, which is essential for hair bundle stimulation.
Hair bundle stimulation gives rise to changes r in hair bundle resistance, which leads
to the amplitude x of cell displacement, which is expressed as x = anp · k/(k + K) and
p is described by Eq. 13. If the resulting cell displacement brings about the mechanical
stimulation same as the initial one, and their phases match [18], the movement of the
system is self-sustaining. The amplitude is determined by the nonlinearity of the system
[18], which is not described here.
Let us assume that hair bundle displacement z and OHC displacement x are proportional
and described by z = λx. The dependence of the change r in hair bundle resistance on
hair bundle displacement z has been experimentally studied. Let g the sensitivity of the
hair bundle transducer. Although the relationship between z and r is nonlinear, let g the
mechanosensitivity at the operating point. Then a condition for an effective amplifier is
given by
gλx(max) ≥ r,
(cid:19)2
where x2 is expressed by Eq. 13 for high frequencies,
(cid:18) k
x2 =
(γaqni0)2
(ωrC0)2
k + K
7
H(¯ω) · r2,
(21)
(22)
with
H(¯ω) =
¯ω2[(α2 + ζ − ¯ω2)2 + (λ¯ω/¯ωη)2]
1
,
(23)
where λ appears in the denominator because it changes the amplitude of length and in
effect changes the drag coefficient in the subtectorial space, where the dominant drag loss is
expected. Notice here that the regular capacitance C0 remains as an important factor that
determines the effectiveness of OHC, even though Cm is very small under this condition.
tion (2 − √
Function H(¯ω) is a monotonically decreasing function of ¯ω2 except for where the condi-
3)(α2 + ζ) is satisfied. A narrow band of parameter
values within this condition, the maximum of H exceeds 40 (Fig. 4). This condition enables
amplifying function at high frequencies.
3)(α2 + ζ) < 1/¯ω2
√
η < (2 +
If the transfer function g(z) is linearized to r = gz in the immediate neighborhood of
the operating point, the frequency limit ωb is expressed by,
(cid:18)
(cid:19)2
b < (α2 + ζ)ζ 2
ω2
λgi0 · a
q
·
k
k + K
Hmax(α2 + ζ, ¯ωη/λ),
(24)
where the best frequency ωb is related to the mechanical resonance frequency ωr by ω2
b =
(α2 + ζ)ω2
r and the definition of ζ is used to replace C0. The local maximum of H(¯ω) is
expressed by Hmax(α2 + ζ, ¯ωη/λ). The dependence of this function on the two parameters
is plotted as a contour graph (Fig. 4).
A
B
Figure 4: Contour plots of Hmax and ζ 2Hmax for λ = 1. A: Contour plot of Hmax Ordinate axis:
¯ω2
η = (¯ωη/¯ωr)2; abscissa: α2 + ζ. The values of Hmax are indicated in the plot. Brighter shades
indicate higher values. B: Contour plot of (α2 + ζ)ζ 2Hmax assuming α2 = 1.1, corresponding to a
η; abscissa: ζ. The values of (α2 + ζ)ζ 2Hmax are indicated
10 kHz cell (see text). Ordinate axis: ¯ω2
in the plot. Brighter shades indicate higher values.
The inequality indicates the importance of the ratio k/(k + K). The optimum condition
is K = 0 and ω2
r = k/m. While a larger value of K elevates the mechanical resonance
frequency ωr, it reduces ωb, making the effectiveness of higher frequency unfavorable. This
issue will be discussed later.
8
7.414.822.229.63744.451.859.20204060801001.01.52.02.53.0(ωη/ωr)2α2+ζ28568411214016819622425228002040608010001234(ωη/ωr)2ζ Values of the Parameters for a 10 kHz cell
Here parameter values are examined using a set of data available for 10 kHz cells. The first
to be examined is the validity of an inequality ωη/ωr (cid:29) 1, the optimizing condition for
H(¯ω). Assume that the source of the major drag is shear in the gap between the reticular
lamina and the tectorial membrane. Then the drag coefficient η is given by η = µS2/d,
where µ is the viscosity of the fluid, S the surface area, and d the gap. If S = 10µm× 20µm,
and d = 1µm, η = 1.6 × 10−7N/m, using the viscosity of water (µ = 8 × 10−4 Pa).
Given the experimentally determined axial elastic modulus of 510 nm/unit strain [31], a
20 µm long OHC has stiffness k of 2.6 × 10−2 N/m (510 nm/20µm). Even if we let K = 0,
i.e. without an external elastic load, we obtain ωη = (k + K)/η ≈ 1.5 × 106, much higher
than the auditory frequency. This value would be even larger for shorter cells of higher
frequency region. Thus the condition ωη/ωr (cid:29) 1 holds.
Now let us examine the frequency limit. For a 20 µm long cell, typical of the 10 kHz,
the linear capacitance is C0 = 8 pF and an = 1 µm, which is 5 % of the cell length. Most
in vitro experiments show the unitary motile charge of q = 0.8 e, where e is the electronic
charge. The maximal value of γ is 1/(4kBT ) when the transducer channel is half open. Here
kB is the Boltzmann constant and T the temperature. The resting basolateral resistance is
7 MΩ and the resting membrane potential of −50 mV requires the resting apical resistance
of 30 MΩ [21]. These values lead to i0 = 4 nA.
It has been pointed out that values for the sensitivity g of hair bundles determined by
in vitro experiments tend to be underestimates due to the matching of the force probe with
hair bundles [32]. For this reason, g=1/(25 nm) [33] is taken.
If we assume k/(k + K) = 1/10 together with λ = 1 and H = 20, an underestimate (See
Fig. 4A), we obtain fb = ωb/2π < 1.1 × 103, consistent with the location of 10 kHz. Power
output can be evaluated using this set of parameters. With this set of the parameter values,
a typical value for maximal power output would be 0.1 fW for r = 0.1. An extrapolation to
the maximal output is 10 fW. These values are in a reasonable agreement with the expected
output range of a single 10 kHz cell estimated from cochlear mechanics [34].
It should be noticed, however, that these agreements do not mean that the given value for
k/(k + K) is reasonable as will be seen in the next section. It simply means k/(k + K) > 0.1
for the given set of parameters because we assumed that OHC output is at the phase optimal
for amplifying to derive the inequality.
Performance at Higher Frequencies
mechanical resonance frequency ωr(=(cid:112)
For an OHC effective at higher frequencies, two conditions should be met. One is that the
m/(k + K) ) must be high. The other is ωb, which
is proportional to k/(K + K) must be larger than ωr. For this reason if k/(K + K) = 0.1 for
a 10 kHz cell, an OHC cannot be effective at higher frequencies, as shown in the following.
The membrane resistance decreases about 3-fold for 10-times higher frequency [21]. A
3-fold reduction of membrane resistance alone would lead to a 3-fold increase in the limiting
frequency. Now a 10-fold increase of ωr requires 1 100-fold increase of the ratio (k + K)/m.
Considering that each OHC is held by stiff Deiters' cup in at the base around the nucleus,
we can expect a 10-fold difference in the stiffness k between a 5 µm cell and a 20 µm cell, the
elastic modulus of OHCs being approximately constant [31]. In addition, a 10-fold increase
9
√
in the frequency reduces the thickness of boundary layer by 1/
10-fold. This factor may
lead to factor ∼ 3 in reducing the mass m. Thus, a ∼ 30-fold increase in k/m could be
expected. If we assume the ratio k/(k + K) is 0.1 for mechanical resonance of 10 kHz, the
ratio turns into 0.03 and the limiting frequency cannot significantly exceed 10 kHz.
If we assume, however, that the resonance at 10 kHz is achieved without the external
elastic load, we require K > 2.3k to achieve 100-fold increase in (k + K)/m. This gives the
maximal value of 0.3 for the stiffness ratio k/(k + K), leading to a limiting frequency well
above 100 KHz, even after a decrease in Hmax due to a 10-fold increase in k, which makes
α2 ≈ 2.
Another important factor is the ratio ζ of the magnitude of nonlinear capacitance Cnl to
the linear capacitance C0. An increase in this factor can have a significant effect in elevating
the limiting frequency (Fig. 4B). A two-fold increase in ζ may lead to an additional 70 %
increase in the limiting frequency. Guinea pig data indeed shows a 4-fold increase of ζ from
low frequency cells (C0 =35 pF) to high frequency cells (5 pF) [35]. However, rat data show
no significant difference between 4 kHz cells (12.1 pF) and 30 kHz cells (5.4 pF) [36].
The inequality 24 shows that limiting frequency is unlikely increased further by a higher
value of Hmax because this factor is large where ζ is small but limiting frequency depends
on (α2 + ζ)ζ 2Hmax, which is larger for large values of ζ and ¯ωη (Fig. 4).
It is possible that the limiting frequency could be raised by other factors, including the
amplitude ratio λ, hair bundle sensitivity g, or the quantity a/q, which may somewhat de-
pend on the cellular property even though it can be regarded as the molecular characteristic
of the motile element.
If these factors do not significantly increase their contributions at higher frequencies, the
ratio k/(k + K) must remain relatively large. Since OHCs should be involved in a relative
motion between the basilar membrane and the reticular lamina [37, 38], the effectiveness
of OHC requires that the resonance frequency of this relative motion must be close to that
of the local basilar membrane. Since the cell bodies of OHCs would be much less stiff
than the basilar membrane, the associated mass must be much smaller.
In this regard,
experimental observations, which reveal the modes of motion in the cochlea, are of great
interest to understand the detailed mechanism of the cochlear amplifier [39].
References
1. Long, G. R., 1994. Psychoacoustics. In R. R. Fay, and A. N. Popper, editors, Comparative
Hearing: Mammals, Springer, New York, NY, 18 -- 56.
2. B´ek´esy, G. v., 1960. Experiments in Hearing. MacGraw-Hill, New York, NY.
3. Gold, T., 1948. Hearing. II. The physical basis of the action of the cochlea. Proc. Roy. Soc., B
135:492 -- 498.
4. Zweig, G., 1991. Finding the impedance of the organ of Corti. J. Acoust. Soc. Am. 89:1229 -- 1254.
5. Liberman, M. C., and L. W. Dodds, 1984. Single neuron labeling and chronic cochlear pathology.
III. stereocilia damage and alterations of threshold tuning curves. Hearing Res. 16:55 -- 74.
10
6. Dallos, P., X. Wu, M. A. Cheatham, J. Gao, J. Zheng, C. T. Anderson, S. Jia, X. Wang, W. H. Y.
Cheng, S. Sengupta, D. Z. Z. He, and J. Zuo, 2008. Prestin-based outer hair cell motility is
necessary for mammalian cochlear amplification. Neuron 58:333 -- 339.
7. Ashmore, J. F., 1987. A fast motile response in guinea-pig outer hair cells: the molecular basis
of the cochlear amplifier. J. Physiol. 388:323 -- 347.
8. Santos-Sacchi, J., and J. P. Dilger, 1988. Whole cell currents and mechanical responses of
isolated outer hair cells. Hearing Res. 65:143 -- 150.
9. Iwasa, K. H., 1993. Effect of stress on the membrane capacitance of the auditory outer hair cell.
Biophys. J. 65:492 -- 498.
10. Gale, J. E., and J. F. Ashmore, 1994. Charge displacement induced by rapid stretch in the
basolateral membrane of the guinea-pig outer hair cell. Proc. Roy. Soc. (Lond.) B Biol. Sci.
255:233 -- 249.
11. Dong, X. X., M. Ospeck, and K. H. Iwasa, 2002. Piezoelectric reciprocal relationship of the
membrane motor in the cochlear outer hair cell. Biophys. J. 82:1254 -- 1259.
12. Housley, G. D., and J. F. Ashmore, 1992. Ionic currents of outer hair cells isolated from the
guinea-pig cochlea. J. Physiol. 448:73 -- 98.
13. Hudspeth, A. J., 2008. Making an effort to listen: mechanical amplification in the ear. Neuron
59:530 -- 545.
14. Liberman, M. C., J. Gao, D. Z. He, X. Wu, S. Jia, and J. Zuo, 2002. Prestin is required for
electromotility of the outer hair cell and for the cochlear amplifier. Nature 419:300 -- 304.
15. Dallos, P., J. Zheng, and M. A. Cheatham, 2006. Prestin and the cochlear amplifier. J Physiol
576:37 -- 42.
16. O Maoil´eidigh, D., and F. Julicher, 2010. The interplay between active hair bundle motility and
electromotility in the cochlea. J. Acoust. Soc. Amer. 128:1175 -- 1190.
17. Dallos, P., and B. Evans, 1995. High-frequency motility of outer hair cells and the cochlear
amplifier. Science 267:2006 -- 2009.
18. Ospeck, M., X.-X. Dong, and K. H. Iwasa, 2003. Limiting frequency of the cochlear amplifier
based on electromotility of outer hair cells. Biophys. J. 84:739 -- 749.
19. Iwasa, K. H., and B. Sul, 2008. Effect of the cochlear microphonic on the limiting frequency of
the mammalian ear. J Acoust Soc Am 124:1607 -- 1612.
20. Mistr´ık, P., C. Mullaley, F. Mammano, and J. Ashmore, 2009. Three-dimensional current flow
in a large-scale model of the cochlea and the mechanism of amplification of sound. J R Soc
Interface 6:279 -- 291.
21. Johnson, S. L., M. Beurg, W. Marcotti, and R. Fettiplace, 2011. Prestin-driven cochlear ampli-
fication is not limited by the outer hair cell membrane time constant. Neuron 70:1143 -- 1154.
22. Meaud, J., and K. Grosh, 2011. Coupling active hair bundle mechanics, fast adaptation, and
somatic motility in a cochlear model. Biophys J 100:2576 -- 2585.
23. Ospeck, M., and K. H. Iwasa, 2012. How close should the outer hair cell RC roll-off frequency
be to the characteristic frequency? [Correction: 103:846-847]. Biophys. J 102:1767 -- 1774.
11
24. Iwasa, K. H., 2016. Energy Output from a Single Outer Hair Cell. Biophys. J. 111:2500 -- 2511.
25. Santos-Sacchi, J., 1991. Reversible inhibition of voltage-dependent outer hair cell motility and
capacitance. J. Neurophysiol. 11:3096 -- 3110.
26. Iwasa, K. H., 2010. Chapter 6. Electromotility of outer hair cells. In P. A. Fuchs, editor, The
Oxford Handbook of Auditory Science volume 1: The Ear, Oxford University Press, Oxford,
UK, 179 -- 212.
27. Nyquist, H., 1928. Thermal agitation of electric charge in conductors. Phys. Rev. 32:110.
28. Kubo, R., 1957. Statistical-mechanical theory of irreversible processes. I. General theory and
simple applications to magnetic and conduction problems. J. Phys. Soc. Jpn. 12:570 -- 586.
29. Iwasa, K. H., 1997. Current noise spectrum and capacitance due to the membrane motor of the
outer hair cell: theory. Biophys. J. 73:2965 -- 2971.
30. Dong, X., D. Ehrenstein, and K. H. Iwasa, 2000. Fluctuation of motor charge in the lateral
membrane of the cochlear outer hair cell. Biophys J 79:1876 -- 1882.
31. Iwasa, K. H., and M. Adachi, 1997. Force generation in the outer hair cell of the cochlea.
Biophys. J. 73:546 -- 555.
32. Nam, J.-H., A. W. Peng, and A. J. Ricci, 2015. Underestimated sensitivity of mammalian
cochlear hair cells due to splay between stereociliary columns. Biophys. J. 108:2633 -- 2647.
33. Russell, I. J., G. P. Richardson, and A. R. Cody, 1986. Mechanosensitivity of mammalian
auditory hair cells in vitro. Nature 321:517 -- 519.
34. Wang, Y., C. R. Steele, and S. Puria, 2016. Cochlear Outer-Hair-Cell Power Generation and
Viscous Fluid Loss. Sci. Rep. 6:19475.
35. Santos-Sacchi, J., S. Kakehata, T. Kikuchi, Y. Katori, and T. Takasaka, 1998. Density of
motility-related charge in the outer hair cell of the guinea pig is inversely related to best fre-
quency. Neurosci Lett. 256:155 -- 158.
36. Mahendrasingam, S., M. Beurg, R. Fettiplace, and C. M. Hackney, 2010. The ultrastructural
distribution of prestin in outer hair cells: a post-embedding immunogold investigation of low-
frequency and high-frequency regions of the rat cochlea. Eur. J. Neurosci. 31:1595 -- 1605.
37. Gao, S. S., R. Wang, P. D. Raphael, Y. Moayedi, A. K. Groves, J. Zuo, B. E. Applegate, and
J. S. Oghalai, 2014. Vibration of the organ of Corti within the cochlear apex in mice. J.
Neurophysiol. 112:1192 -- 1204.
38. Ren, T., W. He, and P. G. Barr-Gillespie, 2016. Reverse transduction measured in the living
cochlea by low-coherence heterodyne interferometry. Nature commun. 7:10282.
39. Dong, W., and E. S. Olson, 2013. Detection of cochlear amplification and its activation. Biophys.
J. 105:1067 -- 78.
12
|
1512.02015 | 1 | 1512 | 2015-12-07T12:36:08 | Role of the Number of Microtubules in Chromosome Segregation during Cell Division | [
"physics.bio-ph",
"q-bio.SC"
] | Faithful segregation of genetic material during cell division requires alignment of chromosomes between two spindle poles and attachment of their kinetochores to each of the poles. Failure of these complex dynamical processes leads to chromosomal instability (CIN), a characteristic feature of several diseases including cancer. While a multitude of biological factors regulating chromosome congression and bi-orientation have been identified, it is still unclear how they are integrated so that coherent chromosome motion emerges from a large collection of random and deterministic processes. Here we address this issue by a three dimensional computational model of motor-driven chromosome congression and bi-orientation during mitosis. Our model reveals that successful cell division requires control of the total number of microtubules: if this number is too small bi-orientation fails, while if it is too large not all the chromosomes are able to congress. The optimal number of microtubules predicted by our model compares well with early observations in mammalian cell spindles. Our results shed new light on the origin of several pathological conditions related to chromosomal instability. | physics.bio-ph | physics | Role of the Number of Microtubules in Chromosome Segregation
during Cell Division
Zsolt Bertalan and Zoe Budrikis
ISI Foundation, Via Alassio 11/c, 10126 Torino, Italy
Caterina A. M. La Porta∗
Center for Complexity and Biosystems,
Department of Bioscience, University of Milan,
via Celoria 26, 20133 Milano, Italy
Stefano Zapperi†
ISI Foundation, Via Alassio 11/c, 10126 Torino, Italy
Center for Complexity and Biosystems,
Department of Physics, University of Milan,
via Celoria 16, 20133 Milano, Italy
CNR-IENI, Via R. Cozzi 53, 20125 Milano, Italy and
Department of Applied Physics, Aalto University,
P.O. Box 14100, FIN-00076 Aalto, Espoo, Finland
(Dated:)
5
1
0
2
c
e
D
7
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
1
0
2
0
.
2
1
5
1
:
v
i
X
r
a
1
Abstract
Faithful segregation of genetic material during cell division requires alignment of chromosomes
between two spindle poles and attachment of their kinetochores to each of the poles. Failure
of these complex dynamical processes leads to chromosomal instability (CIN), a characteristic
feature of several diseases including cancer. While a multitude of biological factors regulating
chromosome congression and bi-orientation have been identified, it is still unclear how they are
integrated so that coherent chromosome motion emerges from a large collection of random and
deterministic processes. Here we address this issue by a three dimensional computational model of
motor-driven chromosome congression and bi-orientation during mitosis. Our model reveals that
successful cell division requires control of the total number of microtubules: if this number is too
small bi-orientation fails, while if it is too large not all the chromosomes are able to congress. The
optimal number of microtubules predicted by our model compares well with early observations in
mammalian cell spindles. Our results shed new light on the origin of several pathological conditions
related to chromosomal instability.
∗ [email protected]
† [email protected]
2
INTRODUCTION
Cell division is a complex biological process whose success crucially depends on the cor-
rect segregation of the genetic material enclosed in chromosomes into the two daughter
cells. Successful division requires that chromosomes should align on a central plate between
the two poles of an extensive microtubule (MT) structure, called the mitotic spindle, in a
process known as congression [1]. Furthermore, the central region of each chromosome, the
kinetochore, should attach to MTs emanating from each of the two poles, a condition known
as bi-orientation [2]. Only when this arrangement is reached, do chromosomes split into two
chromatid sisters that are then synchronously transported towards the poles [3]. Failure for
chromosomes to congress or bi-orient can induce mitotic errors which lead to chromosomal
instability (CIN), a state of altered chromosome number, also known as aneuploidy. CIN is
a characteristic feature of human solid tumors and of many hematological malignancies [4],
a principal contributor to genetic heterogeneity in cancer [5] and an important determinant
of clinical prognosis and therapeutic resistance [6, 7].
Chromosome congression occurs in a rapidly fluctuating environment since the mitotic
spindle is constantly changing due to random MT polymerization and depolymerization
events. This process, known as dynamic instability, is thought to provide a simple mecha-
nism for MTs to search-and-capture all the chromosomes scattered throughout the cell after
nuclear envelope breakdown (NEB)[8]. Once chromosomes are captured, they are trans-
ported to the central plate by molecular motors that use MTs as tracks. The main motor
proteins implicated in this process are kinetochore dynein, which moves towards the spin-
dle pole (i.e. the MT minus end) [9 -- 12] and centromere protein E (CENP-E or kinesin-7)
[13 -- 15] and polar ejection forces (PEFs) [16], both moving away from the pole (i.e. they are
directed towards the MT plus end). PEFs mainly originate from kinesin-10 (Kid) and are
antagonized by kinesin-4 (Kif4A) motors [17], sitting on chromosome arms[18]. While PEFs
are not necessary for chromosome congression, they are vital for cell division [15] since they
orient chromosome arms [18], indirectly stabilize end-on attached MTs [19] and are even able
to align chromosomes in the absence of kinetochores [14]. Recent experimental results show
that chromosome transport is first driven towards the poles by dynein and later towards the
center of the cell by CENP-E and PEF [15] (see Fig 1).
A quantitative understanding of chromosome congression has been the goal of intense
3
theoretical research focusing on the mechanisms for chromosome search-and-capture [20 --
22], motor driven dynamics [23 -- 27] and attachments with MTs [28, 29]. A mathematical
study of search-and-capture was performed by Holy and Leibler who computed the rate for
a single MT to find a chromosome by randomly exploring a spherical region around the pole
[20]. Later, however, Wollman et al. [21] showed numerically that a few hundred MTs would
take about an hour to search and capture a chromosome, instead of few minutes as observed
experimentally. It was therefore argued that MTs should be chemically biased towards the
chromosomes [21]. An alternative mechanism proposed to resolve this discrepancy is the nu-
cleation of MTs directly from kinetochores [30], which was incorporated in a computational
model treating chromosomal movement as random fluctuations in three dimensions [22].
Describing motor driven chromosome dynamics and MT attachment [28, 29] has also
been the object of several computational studies mainly focusing on chromosome oscillations
[23, 24]. These one-dimensional models do not account for congression, because they do not
consider peripheral chromosomes, not lying between the spindle poles at NEB, which are,
however, experimentally observed in mammalian cells [15] Three dimensional numerical
models have been extensively introduced to study cell division in yeast [25 -- 27] but in that
case motor proteins are not essential for congression and there is no NEB. It is not therefore
not clear to which extent these models can be applied to mammalian cells.
Despite the number of insightful experimental and theoretical results, it is still unclear
how a collection of deterministic active motor forces interact with a multitude of randomly
changing MTs to drive a reliable and coherent congression process in a relatively short time.
A key factor that has been completely overlooked in previous studies is the role of the
number of MTs composing the spindle. This is because, on the one hand, it is very difficult
to measure this number experimentally in a dividing cell: The only measurement to our
knowledge is reported in an early paper estimating the number of MTs in the mitotic spindle
of kangaroo-rat kidney (PtK) cells as larger than 104 [31]. On the other hand, computational
limitations have restricted the number of simulated MTs to justs few hundred [1, 21, 22].
Yet the misregulation of several biochemical factors controlling MT nucleation (e.g.
the
centrosomal protein 4.1-associated protein CPAP [32]) or MT depolymerization (e.g. the
mitotic centromere-associated kinase or kinesin family member 2C MCAK/Kif2C [33 -- 35])
are known to affect congression, suggesting that the number of MTs should indeed play an
important, but as yet unexplored, role in the process.
4
Here we tackle this issue by introducing a three dimensional model of motor driven chro-
mosome congression and bi-orientation during mitosis involving a large number of randomly
evolving MTs. Our model describes accurately the processes of stochastic search-and-capture
by MTs and deterministic motor-driven transport, reproducing accurately experimental
observations obtained when individual motor proteins were knocked down[13, 15, 36 -- 38].
Furthermore, the model allows us to explore ground that is extremely difficult to cover
experimentally and vividly demonstrate the crucial role played by the number of MTs to
achieve successful chromosome congression and bi-orientation.
Increasing the number of
MTs enhances the probability of bi-orientation but slows down congression of peripheral
chromosomes due to the increase of PEFs with the number of MTs. Conversely when the
number of MTs is too low, congression probability is increased but bi-orientation is im-
paired. Most importantly, the numerical value of the optimal number of MTs is around 104,
which agrees with experimental estimates [31] but is two orders of magnitude larger than
the numbers employed in previous computational studies [1, 21, 22].
MATERIALS AND METHODS
We consider a three-dimensional model for chromosome congression and bi-orientation in
mammalian cells based on the coordinated action of three motor proteins and a large number
of MTs emanating from two spindle poles. Chromosomes and MTs follow a combination of
deterministic and stochastic rules. Attached chromosomes obey a deterministic overdamped
equation driven by motor forces and use MTs as rails, but attachments and detachments
occur stochastically. Similarly, MTs grow at constant velocity but can randomly switch
between growing and shrinking phases. The dynamics is confined within the cell cortex,
modelled as a hard envelope that repels MTs and chromosomes. We set the cortex major
principal axis a parallel to the x axis, and the minor axes as b = 0.9a and c = 0.7a parallel
to the y and z axes, respectively. This results in a slightly flattened but almost circular
cell. nC = 46 chromosomes are initially uniformly distributed in a sphere of radius 0.65a
representing the nuclear envelope.
5
Microtubules
We assume that spindle poles are already separated and kept at a constant distance
throughout the congression/bi-orientation process [39], in positions (±a/2, 0, 0). MTs em-
anate from each pole radially as straight lines in random spatial directions. A fraction psc
of interpolar MTs forms a stable scaffold, and the remainder grow or shrink with veloc-
ities vg and vs, following the dynamical instability paradigm [40].
In this paradigm, the
transition from growing to shrinking, known as catastrophe, occurs with rate pcat and the
reverse process, known as rescue, occurs with rate pres. Following Ref.
[41], the rate of
MT catastrophe and rescue both depend on the force F acting on the tip of the MT as
cat exp(−F/Fcat) and pres = p0
pcat = p0
res exp(F/Fres), where Fcat and Fres are the sensitiv-
ities of the processes. In our simulations, the only forces on the MTs are due to end-on
attachments with kinetochores, which we describe in detail below. In most simulations, we
consider a constant number of NMT, but we also study the case of in which MTs nucleate
at rate knucl from each pole.
Chromosomes
Chromosomes consist of two large cylindrical objects, the chromatid sisters, joined at
approximately their centers. Chromosome arms are floppy, with an elastic modulus around
500 Pa [42 -- 44] but they tend to be aligned on a plane by PEFs [18]. We therefore treat
chromosome arms as a two dimensional disk of radius rC, representing the cross-section
for their interaction with MTs (see Fig 2a). At the centre of each chromosome sit two
kinetochores, highly intricate protein complexes fulfilling a wide variety of tasks, chief of
which is interacting with MTs. In the model, the two kinetochores are treated as a sphere
of radius rk defining the interaction range with MTs (see Fig 2a).
Chromosomes can interact with MTs in three distinct ways: PEFs (Fig 2b), lateral at-
tachments (Fig 2c) and end-on attachments (Fig 2d). Each of these interactions is associated
with a specific motor force, as illustrated in the schematic in Fig 2 and described below.
Time is discretized and at each time step ∆t we first implement stochastic events in
parallel, then perform MT growth/shrinking and update chromosome positions ri according
6
to the discretized overdamped equations of motion
ri(t + ∆t) = ri(t) + Fi∆t/η
(1)
where η is the drag coefficient and Fi is the total motor force acting on chromosome i. The
total force is the sum of PEFs, FPEF, lateral attachment forces due to dynein, Fdynein and
CENP-E, FCENPE, and end-on-attachment spring forces Fk. The precise form of these forces
is described in detail below.
Polar ejection forces
For every MT crossing the chromosome within a distance rC of its geometrical center (Fig
2b), the chromosome acquires a PEF FPEF due to motors sitting at the chromosome arms
[43], in direction of the plus end of the MT.
Lateral attachments
In our model, lateral kinetochore-MT attachments form when a MT crosses the kine-
tochore interaction sphere of radius rk. Then the MT serves as a track along which the
chromosome is slid by one of two groups of motor proteins, CENP-E or dynein. CENP-E
applies a force FCENPE towards the plus end of the MT, away from the spindle pole, while
dynein applies a force FDYN towards the minus end of the MT, thus pointing in the direc-
tion of the spindle pole, as illustrated in Fig 2c. Since we use overdamped dynamics, a
constant force corresponds to a constant velocity with which the group of motor proteins
moves the chromosome. To determine which type of motor is active, we take a determinis-
tic approach motivated by experimental results [15]: we initially set CENP-E as the active
motor for chromosomes that are inside a shell of radius 0.45a and dynein for the rest of
peripheral chromosomes. Experiments show that dynein brings peripheral chromosomes to
the poles [9 -- 12] and is then inactivated by the action of the kinase Aurora A, while CENP-E
is activated [45]. We simulate this by switching off dynein at the pole and replacing it by
CENP-E.
The CENP-E motor prefers to walk on long-lived MTs [45], giving the chromosome a
necessary bias to congress at the cell center. The biochemical factor underlying this process
7
has been recently identified with the detyrosination of spindle microtubules pointing towards
center of the cell [46]. In the model, we form lateral attachments when CENP-E is active
only if the MT has a lifetime larger than τM T = 60s.
End-on attachments
The two kinetochores in our model are represented as half-spheres and each has Nk slots
for end-on attachments with MTs. In general, when the tip of an itinerant MT is within
distance rk of a kinetochore with available slots, the MT and the kinetochore form an end-on
attachment. However, after NEB the kinetochores of peripheral chromosomes are covered
by dynein, inhibiting end-on attachments [12]. Hence, we allow for end-on attachments only
when CENP-E is active. The force on the chromosome from an end-on attached MTs is
translated via a harmonic coupling with zero rest length and spring constant kk.
MTs can detach stochastically from kinetochores with a rate that depends on the applied
force and on the stability of the attachment[47]. Biochemical factors, such as Aurora B
kinase, ensure that faulty attachments are de-stabilized [48, 49] and correct attachments
stabilized. In particular, intra-kinetochore tension in bi-oriented chromosomes inhibits the
de-stabilizing effect of Aurora B kinase on end-on attachments [49]. Furthermore, stabiliza-
tion of chromosomes at the central plate is also due to action of kinesin-8 motors [17, 50, 51].
In the present model, we simply stabilize attachments if both kinetochores have end-on at-
tached MTs stemming from both poles, while we treat as unstable the cases in which only
a single kinetochore has end-on attachments or in which two kinetochores have end-on at-
tached MTs all stemming from a single pole.
Unstable attachment detach with a probability that decreases exponentially with applied
detach = p(u),0
detach exp(F/F (u)
force p(u)
the kinetochore and F (u)
detach), where F is the force on the MT tip due to coupling with
detach is the sensitivity [41]. When the attachment is stable, we assume
that the growth/shrinkage velocity of the attached MTs is slowed exponentially (see Table
I and Ref. [41]), and that attachment is -- contrary to intuition -- stabilized by an applied
load p(s)
detach). This peculiar behavior, known as a catch-bond, has
detach exp(−F/F (s)
detach = p(s),0
been revealed experimentally [41] and explained theoretically [29].
8
Implementation
The numerical solution is implemented in a custom made C++ code. Images and videos
are rendered in 3D using Povray. Simulation and rendering codes are available at https:
//github.com/ComplexityBiosystems/chromosome-congression All parameters used in
the model are summarized in Table I. Where experimentally-measured parameters are not
available, we have used estimated values. We have tested these to ensure simulation results
are robust against changes in parameter values.
RESULTS
Control of MT number by MT nucleation rate
In most of our simulations, the number of MTs is fixed. To justify this, we have performed
simulations in which MTs nucleate from the two spindle poles with a rate knucl. At the
beginning of the simulation, we assume that the mitotic spindle is already formed, the
nuclear envelope is broken, and 46 chromosomes are randomly distributed in a spherical
region enclosing the poles. We then integrate the equations of motion for each chromosome
and monitor the number of MTs NMT as a function of the nucleation rate knucl. We find
that after a transition time (approximately 50s), that is much shorter than the congression
time (Fig 3a), the number of MTs fluctuates around a constant value (cid:104)NMT(cid:105) that is linearly
dependent on knucl (Fig 3b).
The result shown in Fig 3 can be understood from a simple kinetic equation for the
number of MTs
dNMT
dt
= knucl − koutNMT,
(2)
where the second term on the right-hand side is the total rate of MT collapse. The rate of
collapse per MT, kout, is the inverse of the MT lifetime, proportional to the MT half-life.
The solution of Eq. 2
NMT = knucl/kout(1 − exp(−koutt))
(3)
provides an excellent fit to the data with kout = 0.09s−1 (Fig 3a). The theory also shows
that for long times, t (cid:29) 1/kout, the number of MTs approaches NMT = knucl/kout. Hence
the number of MTs is essentially constant during the congression process, depending only
9
on the rate of nucleation and collapse, which are controlled by several biochemical factors.
Based on this result, we ignore the transient and keep NMT constant during each simulation.
Incorrect chromosome congression due to knock-down of motor proteins
After nuclear envelope breakdown, there are two possible scenarios for congression. In
the first case, all chromosomes already lie between the poles, and have access to stable MTs.
Hence, CENP-E overcomes dynein, moving the chromosome directly towards the center
of the cell. The second scenario involves chromosomes not having access to stable MTs,
because their initial position does not lie between the poles. Those chromosomes are first
driven by dynein to the nearest pole and remain there until they find a stable MT to which
they attach laterally. At this point, they slide towards the central plate using CENP-E
motor on the stable MT. We show the evolution of these two scenarios in S1 and S2 Videos.
In all simulations we ran with the present parameters (n > 30 instances per scenario) all
chromosomes congress and bi-orient.
Next, we switch off motor proteins individually (dynein, CENP-E or PEF) to show that
the model successfully reproduces what happens in cells, where all these motors are essential.
The results are summarized in Fig 4 (see also S3,S4 and S5 Videos) and show that the sup-
pression of each of the motors leads to incorrect congression or bi-orientation. Suppressing
kinetochore dynein does not allow peripheral chromosomes to congress, as shown in row 2
of Fig 4. Deletion of CENP-E traps chromosomes at the poles, as shown in row 3, and PEF
knockdown severely reduces the cohesion of the central plate where chromosomes can not
bi-orient, as shown in row 4.
These knock-downs have also been studied experimentally, yielding results in line with
ours. In Refs. [13, 36] principal contributors to PEF are knocked down, and it is shown that
in cases where there are no peripheral chromosomes, the chromosomes can congress but are
not stable at the central plate. Furthermore, other experiments show that chromosomes are
also stabilized at the central plate due to the effect of the kinesin-8 Kif18A on MT plus ends
[17, 50, 51]. These observations fold neatly into our model and yield a possible explanation
of the above mentioned slowing down of MT plus ends at kinetochores. It should also be
noted that when the effect of Kif18A is removed and MT plus ends follow fast dynamics
again, the effective PEFs in the vicinity of the central plate are reduced, further destabilizing
10
chromosome alignment [15]. In Refs. [37, 38], on the other hand, CENP-E is knocked down or
suppressed, and results in chromosomes being trapped at spindle poles. Finally. in Ref. [15]
all three motors are suppressed individually, with exactly the same results as presented here
from our simulations.
Optimal number of MTs for chromosome congression and bi-orientation
We find that the ratio of the number of total MTs in the system divided by twice the total
number of chromosomes that is, the total number of kinetochores, affects the congression
process in a non-trivial manner, as illustrated in Fig 5 and S6 and S7 Videos. In particular,
chromosome congression and bi-orientation are influenced by the number of MT in opposite
ways: While a large number of MTs enhances the chances of bi-orientation, it slows down
congression. This is due to the fact that PEFs increase with the number of MTs, thus acting
against kinetochore dynein and possibly hindering the motion of peripheral chromosome
towards the poles. In the wild-type case, kinetochore dynein in usually strong enough to
overcome these PEFs [15]. Overexpression of motors giving rise to PEFs can have adverse
effects, such as the over-stabilization of kinetochore-MT attachments [19]. On the other
hand, stabilizing MTs by disrupting various MT-depolymerase chains results in much slowed
down congression and bi-orientation [58]. We show the effect of too strong PEFs on our
model in Fig 6a, where the distribution of congressed chromosomes is plotted versus time
for different MT densities. On the other hand, PEFs stabilize congressed chromosomes at the
central plate, and in a simple search and capture scenario [20], like the one implemented in
our model, the more MTs there are the faster chromosomes become bi-oriented, as indicated
in Fig 6b. In Fig 6c, we plot the median of the congression/bi-orientation time distribution
defined as the time for which the probability of congression (black) and bi-orientation (red)
is one half. At very low MT densities (blue shaded area), reported in the left-hand-side of Fig
6c, not all samples congress within the limit of 103 seconds. At slightly higher MT densities
(red shaded area), not all samples bi-orient within the limit of 105 seconds. Finally, at very
high MT densities, PEFs become so strong that they reduce the congression probability.
These observations indicate the existence of a sweet spot for the MT density suggesting that
successful congression and bi-orientation can only happen only if the total number of MTs
in the spindle lies in the range of 7·103 − 1.8 · 104.
11
Overexpressing MT depolymerases reduces the congression probability
An experimentally testable prediction of our model is the effect on congression of the
overexpression of factors affecting MT depolymerization [59]. The catastrophe/rescue rate
ratio determines the MT length distribution during cell division. Shorter MTs would signifi-
cantly hamper the search and capture process: Chromosomes lying at the extreme periphery
would be harder to reach, decreasing the chances for congression. To quantify this effect, we
performed n = 10 simulations for each MT density and increasing the value of the catastro-
phe rate, as illustrated in Fig 7 and in S8 Video. The corresponding congression probability
is reported in Fig 8. For low MT densities the effect is very drastic and even partial con-
gression is suppressed. For the sweet-spot densities, MT depolymerases overexpression has
only a small effect, until the catastrophe rate becomes too large and congression disappears.
DISCUSSION
Understanding cell division and its possible failures is a key problem that is relevant for
many pathological conditions including cancer. While many biochemical factors controlling
several aspects of the division process have been identified, how these factors work together
in a coherent fashion is still an open issue. We have introduced a comprehensive three
dimensional computational model for chromosome congression in mammalian cells, using
stochastic MT dynamics as well as motor-protein interplay. The model incorporates move-
ment of the peripheral chromosomes to the poles and their escape from there towards the
central plate. Contrary to previous models that only used a limited number of MTs (e.g. a
few hundred in Ref. [22]), we are able to simulate up to 3·104 MTs. McIntosh et al. reported
already in 1975 that the number of MTs in the mitotic spindle of kangaroo-rat kidney (PtK)
cells during metaphase is larger than 104 [31], in good agreement with our predictions. Also,
to put this number in perspective, we notice that each human chromosome has up to 50
end-on attachment slots per kinetochore, and on average 25 MTs attached [52]. Since there
are 46 chromosomes in human cells, this corresponds to 2300 attached MTs on average. The
total number of MTs in the spindle should be much larger than the number of attached MT
and therefore 104 MTs appears to be a reasonable number. It is interesting to remark that
with this number of MTs, congression and bi-orientation of chromosomes is quick enough
12
that the assumption of biased search [21] is not needed.
With our model we show that the total number of MTs in the spindle is per se a crucial
controlling factor for successful cell division. When this number is too low or too high, con-
gression and/or bi-orientation fail. This explains apparent paradoxes where the same factors
can lead to different pathological conditions when up or down regulated. For instance, the
centrosomal protein 4.1-associated protein (CPAP), belonging to the microcephalin (MCPH)
family [60], is known inhibit MT nucleation [32]. CPAP overexpression leads to abnormal
cell division [61, 62], whereas mutations in CPAP can cause autosomal recessive primary
microcephaly, characterized by a marked reduction in brain size [63]. In the model, we can
account for CPAP overexpression by inhibiting MT nucleation, while its mutation can be
simulated by increasing knucl. The two processes push the number of MTs out of its sweet
spot, along different directions and therefore explain the different pathological conditions
with a single mechanism.
A similar reasoning explains the role of mitotic centromere-associated kinase or kinesin
family member 2C (MCAK/Kif2C) that is localized at MT plus ends [35] and functions
as a key regulator of mitotic spindle assembly and dynamics [64, 65] by controlling MT
length [35]. Higher expression of MCAK level has been found in gastric cancer tissue [66],
colorectal and other epithelial cancers [67] and breast cancer [68]. In fact, both depletion
[33, 34] and overexpression [58, 59] of MCAK lead to cell division errors. From the point of
view of our model, we can understand that MCAK overexpression increases the rate of MT
depolymerization reducing their length and number to a level in which bi-orientation is not
possible. Finally our model explains the recent results linking CIN to the overexpression
of AURKA or the loss of CHK2, both enhancing MT assembly rate [69]. Increasing MT
velocity effectively reduces the amount of tubulin units available for MT nucleation, thus
decreasing the number of MTs and imparing bi-orientation.
In conclusion, our model represents a general computational tool to predict the effect of
biological factors on cell division making it a valid tool for in silico investigation of related
pathological conditions. The main strength of our computational approach is that can it
help answer questions that are extremely difficult to address experimentally, such as the role
of the number of microtubules in driving successful cell division.
13
ACKNOWLEDGMENTS
We thank M. Barisic and H. Maiato for useful suggestions and for sharing the results of
Ref.
[15] before publication. We thank J. R. McIntosh for pointing out Ref.
[31] and M.
Zaiser for critical reading of the manuscript.
[1] Magidson V, O'Connell CB, Loncarek J, Paul R, Mogilner A, Khodjakov A. The spatial
arrangement of chromosomes during prometaphase facilitates spindle assembly. Cell. 2011
Aug;146(4):555 -- 67.
[2] Walczak CE, Cai S, Khodjakov A. Mechanisms of chromosome behaviour during mitosis. Nat
Rev Mol Cell Biol. 2010 Feb;11(2):91 -- 102.
[3] Matos I, Pereira AJ, Lince-Faria M, Cameron LA, Salmon ED, Maiato H. Synchronizing
chromosome segregation by flux-dependent force equalization at kinetochores. J Cell Biol.
2009;186:11 -- 26.
[4] Boveri T. Ubermehrpolige Mitosenals Mittelzur Analysedes Zellkerns. Stuber; 1903.
[5] Burrell RA, McGranahan N, Bartek J, Swanton C. The causes and consequences of genetic
heterogeneity in cancer evolution. Nature. 2013 Sep;501(7467):338 -- 45.
[6] Lee AJX, Endesfelder D, Rowan AJ, Walther A, Birkbak NJ, Futreal PA, et al. Chromosomal
instability confers intrinsic multidrug resistance. Cancer Res. 2011 Mar;71(5):1858 -- 70.
[7] Bakhoum SF, Compton DA. Chromosomal instability and cancer: a complex relationship
with therapeutic potential. J Clin Invest. 2012 Apr;122(4):1138 -- 43.
[8] Kirschner M, Mitchison TJ. Beyond self-assembly: from microtubules to morphogenesis. Cell.
1986;45:329 -- 342.
[9] Rieder CL, Alexander SP. Kinetochores are transported poleward along a single astral mi-
crotubule during chromosome attachment to the spindle in newt lung cells. J Cell Biol.
1990;110:81 -- 95.
[10] Li Y, Yu W, Liang Y, Zhu X. Kinetochore dynein generates a poleward pulling force to
facilitate congression and full chromosome alignment. Cell Res. 2007;17:701 -- 712.
[11] Yang Z, Tulu US, Wadsworth P, Rieder CL. Kinetochore dynein is required for chro-
mosome motion and congression independent of the spindle checkpoint. Curr Biol. 2007
14
Jun;17(11):973 -- 80.
[12] Vorozhko VV, Emanuele MJ, Kallio MJ, Gorbsky PTSGJ. Multiple mechanisms of chromo-
some movement in vertebrate cells mediated through the Ndc80 complex and dynein/dynactin.
Chromosoma. 2008;117:169 -- 179.
[13] Kapoor TM, Lampson MA, Hergert P, Cameron L, Cimini D, Salmon ED, et al. Chromosomes
can congress to the metaphase plate before biorientation. Science. 2006;311:388 -- 91.
[14] Cai S, O'Connell CB, Khodjakov A, Walczak CE. Chromosome congression in the absence of
kinetochore fibres. Nat Cell Biol. 2009;11:832 -- 838.
[15] Barisic M, Aguiar P, Geley S, Maiato H. Kinetochore motors drive congression of periph-
eral polar chromosomes by overcoming random arm-ejection forces. Nat Cell Biol. 2014
Dec;16(12):1249 -- 56.
[16] Rieder CL, Davison EA, Jensen LC, Cassimeris L, Salmon ED. Oscillatory movements of
monooriented chromosomes and their position relative to the spindle pole result from the
ejection properties of the aster and half-spindle. J Cell Biol. 1986 Aug;103(2):581 -- 91.
[17] Stumpff J, Wagenbach M, Franck A, Asbury CL, Wordeman L. Kif18A and chromokinesins
confine centromere movements via microtubule growth suppression and spatial control of kine-
tochore tension. Dev Cell. 2012 May;22(5):1017 -- 29.
[18] Wandke C, Barisic M, Sigl R, Rauch V, Wolf F, Amaro AC, et al. Human chromokinesins
promote chromosome congression and spindle microtubule dynamics during mitosis. J Cell
Biol. 2012;198:847 -- 863.
[19] Cane S, Ye AA, Luks-Morgan SJ, Maresca TJ. Elevated polar ejection forces stabilize
kinetochore-microtubule attachments. J Cell Biol. 2013;200:203.
[20] Holy TE, Leiber S. Dynamic instability of microtubules as an efficient way to search in space.
PNAS. 1995;91:5682 -- 5685.
[21] Wollman R, Cytrynbaum EN, Jones JT, Meyer T, Scholey JM, Mogilne A. Efficient Chro-
mosome Capture Requires a Bias in the 'Search-and-Capture' Process during Mitotic-Spindle
Assembly. Curr Biol. 2005;15:828 -- 832.
[22] Paul R, Wollman R, Silkworth WT, Nardi IK, Cimini D, Mogilner A. Computer simulations
predict that chromosome movements and rotations accelerate mitotic spindle assembly without
compromising accuracy. PNAS. 2009;106:15708 -- 15713.
15
[23] Joglekar AP, Hunt AJ. A Simple, Mechanistic Model for Directional Instability during Mitotic
Chromosome Movement. Biophys J. 2002;83:42 -- 58.
[24] Civelekoglu-Scholey G, Sharp DJ, Mogilner A, Scholey JM. Model of Chromosome Motility
in Drosophila Embryos: Adaptation of a General Mechanism for Rapid Mitosis. Biophys J.
2006;90:3966.
[25] Gardner MK, Pearson CG, Sprague BL, Zarzar TR, Bloom K, Salmon ED, et al. Tension-
dependent Regulation of Microtubule Dynamics at Kinetochores Can Explain Metaphase Con-
gression in Yeast. Mol Biol Cell. 2005;16:3764 -- 3775.
[26] Chac´on JM, Gardner MK. Analysis and Modeling of Chromosome Congression During Mitosis
in the Chemotherapy Drug Cisplatin. Cell Mol Bioeng. 2013 Dec;6(4):406 -- 417.
[27] Glunci´c M, Maghelli N, Krull A, Krsti´c V, Ramunno-Johnson D, Pavin N, et al. Kinesin-8
motors improve nuclear centering by promoting microtubule catastrophe. Phys Rev Lett. 2015
Feb;114(7):078103.
[28] Hill TL. Theoretical problems related to the attachment of microtubules to kinetochores. Proc
Natl Acad Sci U S A. 1985 Jul;82(13):4404 -- 8.
[29] Bertalan Z, Porta CAML, Maiato H, Zapperi S. Conformational Mechanism for the Stability
of Microtubule-Kinetochore Attachments. Biophys J. 2014;107:289 -- 300.
[30] O'Connell CB, Khodjakov AL. Cooperative mechanisms of mitotic spindle formation. J Cell
Sci. 2007 May;120(Pt 10):1717 -- 22.
[31] McIntosh JR, Cande WZ, Snyder JA. Structure and physiology of the mammalian mitotic
spindle. Soc Gen Physiol Ser. 1975;30:31 -- 76.
[32] Hung LY, Chen HL, Chang CW, Li BR, Tang TK.
Identification of a novel microtubule-
destabilizing motif in CPAP that binds to tubulin heterodimers and inhibits microtubule
assembly. Mol Biol Cell. 2004 Jun;15(6):2697 -- 706.
[33] Bakhoum SF, Genovese G, Compton DA. Deviant kinetochore microtubule dynamics underlie
chromosomal instability. Curr Biol. 2009 Dec;19(22):1937 -- 42.
[34] Stout JR, Rizk RS, Kline SL, Walczak CE. Deciphering protein function during mitosis in
PtK cells using RNAi. BMC Cell Biol. 2006;7:26.
[35] Domnitz SB, Wagenbach M, Decarreau J, Wordeman L. MCAK activity at microtubule tips
regulates spindle microtubule length to promote robust kinetochore attachment. J Cell Biol.
2012 Apr;197(2):231 -- 7.
16
[36] Antonio C, Ferby I, Wilhelm H, Jones M, Karsenti E. Xkid, a chromokinesin required for
chromosome alignment on the metaphase plate. Cell. 2000;102:425 -- 435.
[37] Putkey FR, Cramer T, Morphew MK, Silk AD, Johnson RS, McIntosh JR, et al. Unstable
Kinetochore-Microtubule Capture and Chromosomal Instability Following Deletion of CENP-
E. Dev Cell. 2002;3:351 -- 365.
[38] Silk AD, Zasadil LM, Holland AJ, Vitre B, Cleveland DW, Weaver BA. Chromosome misseg-
regation rate predicts whether aneuploidy will promote or suppress tumors. Proc Natl Acad
Sci U S A. 2013 Oct;110(44):E4134 -- 41.
[39] Sharp DJ, Rogers GC, Scholey JM. Microtubule motors in mitosis. Nature. 2000;407:41 -- 47.
[40] Mitchison TJ, Kirschner M. Dynamic instability of microtubule growth. Nature. 1984;312:237 --
242.
[41] Akiyoshi B, Sarangapani KK, Powers AF, Nelson CR, Reichow SL, Arellano-Santoyo H, et al.
Tension directly stabilizes reconstituted kinetochore -- microtubule attachments. Nature.
2010;468:576 -- 579.
[42] Nicklas RB. Measurements of the force produced by the mitotic spindle in anaphase. J Cell
Biol. 1983;97:542 -- 548.
[43] Marshall WF, Marko JF, Agard DA, Sedat JW. Chromosome elasticity and mitotic polar
ejection force measured in living Drosophila embryos by four-dimensional microscopy-based
motion analysis. Curr Biol. 2001;11:569 -- 578.
[44] Marko JF, Poirier MG. Micromechanics of chromatin and chromosomes. Biochem Cell Biol.
2003;81:209 -- 220.
[45] Kim Y, Holland AJ, Lan W, Cleveland DW. Aurora kinases and protein phosphatase 1
mediate chromosome congression through regulation of CENP-E. Cell. 2010;142:444 -- 455.
[46] Barisic M, Silva e Sousa R, Tripathy SK, Magiera MM, Zaytsev AV, Pereira AL, et al. Micro-
tubule detyrosination guides chromosomes during mitosis. Science. 2015;348(6236):799 -- 803.
Available from: http://www.sciencemag.org/content/348/6236/799.abstract.
[47] Maresca TJ, Salmon ED. Welcome to a new kind of tension: translating kinetochore mechanics
into a wait-anaphase signal. J Cell Sci. 2010;123:825 -- 834.
[48] Cimini D, Wan X, Hirel CB, Salmon ED. Aurora Kinase Promotes Turnover of Kinetochore
Microtubules to Reduce Chromosome Segregation Errors. Curr Biol. 2006;16:1711 -- 1718.
17
[49] Lampson MA, Cheeseman IM. Sensing centromere tension: Aurora B and the regulation of
kinetochore function. Trends Cell Biol. 2011;21:133 -- 128.
[50] Stumpff J, von Dassow G, Wagenbach M, Asbury C, Wordeman L. The kinesin-8 motor
Kif18A suppresses kinetochore movements to control mitotic chromosome alignment. Dev
Cell. 2008 Feb;14(2):252 -- 62.
[51] Stumpff J, Du Y, English CA, Maliga Z, Wagenbach M, Asbury CL, et al. A tethering
mechanism controls the processivity and kinetochore-microtubule plus-end enrichment of the
kinesin-8 Kif18A. Mol Cell. 2011 Sep;43(5):764 -- 75.
[52] McDonald KL, O'Toole ET, Mastronarde DN, McIntosh JR. Kinetochore microtubules in Ptk
cells. J Cell Biol. 1992;118:369 -- 383.
[53] Salmon ED, Saxton WM, Leslie RJ, Karow ML, McIntosh JR. Spindle microtubule dynam-
ics in sea urchin embryos: analysis using a fluorescein-labeled tubulin and measurements of
fluorescence redistribution after laser photobleaching. J Cell Biol. 1984;99:2157.
[54] Brouhard GJ, Hunt AJ. Microtubule movements on the arms of mitotic chromosomes: Polar
ejection forces quantified in vitro. PNAS. 2005;102:13903 -- 13908.
[55] Kim Y, Heuser JE, Waterman CM, Cleveland DW. CENP-E combines a slow, processive
motor and a flexible coiled coil to produce an essential motile kinetochore tether. JCell Biol.
2008;181:411 -- 419.
[56] Mallik R, Carter BC, Lex SA, King SJ, Gross SP. Cytoplasmic dynein functions as a gear in
response to load. Nature. 2004;427:649.
[57] Rusan NM, Fagerstrom CJ, Yvon AC, Wadsworth P. Cell cycle-dependent changes in micro-
tubule dynamics in living cells expressing green fluorescent protein-alpha tubulin. Mol Biol
Cell. 2001;12:971.
[58] Tanenbaum ME, Macurek L, van der Vaart B, Galli M, Akhmanova A, Medema RH. A com-
plex of Kif18b and MCAK promotes microtubule depolymerization and is negatively regulated
by Aurora kinases. Curr Biol. 2011 Aug;21(16):1356 -- 65.
[59] Ganguly A, Yang H, Cabral F. Overexpression of mitotic centromere-associated Kinesin
stimulates microtubule detachment and confers resistance to paclitaxel. Mol Cancer Ther.
2011 Jun;10(6):929 -- 37.
[60] Bond J, Roberts E, Springell K, Lizarraga SB, Lizarraga S, Scott S, et al. A centroso-
mal mechanism involving CDK5RAP2 and CENPJ controls brain size. Nat Genet. 2005
18
Apr;37(4):353 -- 5.
[61] Kohlmaier G, Loncarek J, Meng X, McEwen BF, Mogensen MM, Spektor A, et al. Overly
long centrioles and defective cell division upon excess of the SAS-4-related protein CPAP.
Curr Biol. 2009 Jun;19(12):1012 -- 8.
[62] Schmidt TI, Kleylein-Sohn J, Westendorf J, Le Clech M, Lavoie SB, Stierhof YD, et al. Control
of centriole length by CPAP and CP110. Curr Biol. 2009 Jun;19(12):1005 -- 11.
[63] Thornton GK, Woods CG. Primary microcephaly: do all roads lead to Rome? Trends Genet.
2009 Nov;25(11):501 -- 10.
[64] Hunter AW, Caplow M, Coy DL, Hancock WO, Diez S, Wordeman L, et al. The kinesin-related
protein MCAK is a microtubule depolymerase that forms an ATP-hydrolyzing complex at
microtubule ends. Mol Cell. 2003 Feb;11(2):445 -- 57.
[65] Zhu C, Zhao J, Bibikova M, Leverson JD, Bossy-Wetzel E, Fan JB, et al. Functional analysis
of human microtubule-based motor proteins, the kinesins and dyneins, in mitosis/cytokinesis
using RNA interference. Mol Biol Cell. 2005 Jul;16(7):3187 -- 99.
[66] Nakamura Y, Tanaka F, Haraguchi N, Mimori K, Matsumoto T, Inoue H, et al. Clinicopatho-
logical and biological significance of mitotic centromere-associated kinesin overexpression in
human gastric cancer. Br J Cancer. 2007 Aug;97(4):543 -- 9.
[67] Gnjatic S, Cao Y, Reichelt U, Yekebas EF, Nolker C, Marx AH, et al. NY-CO-58/KIF2C is
overexpressed in a variety of solid tumors and induces frequent T cell responses in patients
with colorectal cancer. Int J Cancer. 2010 Jul;127(2):381 -- 93.
[68] Shimo A, Tanikawa C, Nishidate T, Lin ML, Matsuda K, Park JH, et al.
Involvement of
kinesin family member 2C/mitotic centromere-associated kinesin overexpression in mammary
carcinogenesis. Cancer Sci. 2008 Jan;99(1):62 -- 70.
[69] Ertych N, Stolz A, Stenzinger A, Weichert W, Kaulfuss S, Burfeind P, et al. Increased micro-
tubule assembly rates influence chromosomal instability in colorectal cancer cells. Nat Cell
Biol. 2014 08;16(8):779 -- 791. Available from: http://dx.doi.org/10.1038/ncb2994.
SUPPORTING INFORMATION
: S1 Video. Congression of scattered chromosomes. Representative example of the
congression process in the case in which some of the chromosomes are initially scattered
19
beyond the poles.
: S2 Video. Congression of interpolar chromosomes. Representative example of the
congression process in the case in which all of the chromosomes initially lie between
the poles.
: S3 Video. Congression with PEF knockdown. Representative example of the con-
gression process when PEF is suppressed.
: S4 Video. Congression with Dynein knockdown. Representative example of the
congression process when Dynein is suppressed.
: S5 Video. Congression with CENP-E knockdown. Representative example of the
congression process when CENP-E is suppressed.
: S6 Video. Congression with a small number of MTs. Representative example of
the congression process with 10MTs per kinetochore.
: S7 Video. Cpngression with a large number of MTs. Representative example of
the congression process with 300MTs per kinetochore.
: S8 Video. Congression with MT depolymerases overexpression. Representative
example of the congression process overexpressing MT depolymerases. In these simu-
lations p0
cat = 0.348s−1.
20
Fig 1. Schematic of the dynamics of a single chromosome. a) Peripheral chromosomes, not lying
between the spindle poles, are driven to the nearest pole by dynein. b) Chromosomes are driven
from the pole to the central plate by the combined action of CENP-E and PEF. c) At the central
plate, chromosomes attached to both poles are called bi-oriented.
21
FDYNFCENPECentral plate+FPEFbi-orientationSpindlepolea)b)c) Fig 2. Schematic of the chromosome model and forces acting on it. a) The chromosome consists
of freely rotating arms and of a sphere of radius rk, representing the kinetochore. In the model
the arms is represented by a disk of radius rC, corresponding to the chromosome cross-section, and
the kinetochore by a sphere of radius rk. Microtubules (red) interact with the chromosome and
exert forces on it. b) A MT passing through a chromosome arm, adds a force FPEF in the direction
of the plus-end of the MT. c) Lateral attachments add constant forces originating from groups of
motor proteins at the kinetochore. Which group, dynein or CENP-E is active, is determined by
the simulation and described in detail in the main body of the text. d) MT tips can form end-on
attachments with the kinetochore, which is represented by a harmonic spring with stiffness kk and
zero rest length.
22
Name
Symbol Values used Comment/Reference
TABLE I. Model Parameters
Cell major axis
Effective kinetochore radius
Kinetochore slots
Kinetochore -- MT spring
Unstable detach rate
a
rk
Nk
kk
p(u),0
detach
Unstable detach sensitivity F (u)
Stable detach rate
Stable detach sensitivity
Chromatid radius
Number of chromosomes
PEF
CENP-E force
15µm
estimate
0.3 µm
estimate
25-50
based on PtK1 cells[52]
100.0 pN/µm magnitude similar to [23, 24]
0.1/s
4 pN
estimate, unloaded [41]
estimate
0.001/s
estimate, unloaded
4 pN
estimate
detach
p(s),0
detach
F (s)
detach
rC
nC
1.1-1.5 µm estimate [53]
46
human cell
FPEF
FCENPE
0.5 pN
per MT [54]
5×10 pN total group
Dynein force
FDYN
1.0×50 pN per group
based on stall force [55]
based on stall force [56]
MT growth velocity
MT growth sensitivity
MT shrinking velocity
MT shrinking sensitivity
Rescue rate
Rescue sensitivity
Catastrophe rate
Catastrophe sensitivity
vg
Fg
vs
Fs
p0
res
Fres
p0
cat
Fcat
12µm/min [57], unloaded
6pN
[41]
14µm/min [57], unloaded
4pN
[41]
0.045/s
[57], unloaded
2.3pN
[41]
0.058 - 0.58/s [57], unloaded and overexpression
2.4pN
[41]
Tot. number of MTs
NMT
900 -- 30000
Fraction of linked MTs
Drag coefficient
psc
η
0.1
estimate
10−7 Kg/s
estimate based
on cytoplasmic viscosity[53]
List of parameter values employed in the simulations.
23
Fig 3. The rate of microtubule nucleation controls their number. (a) The number of MTs reaches
a constant value in a time that is much shorter than the typical congression time. Different curves
refer to different values of knucl. (b) The number of MTs is proportional to the rate of nucleation
knucl. The numerical results here refer to a single pole. Lines are fits with the theory discussed
in the text. The curves have been obtained by averaging over n = 1000 independent runs of the
simulations. Error bars are smaller than the plotted symbols.
24
Fig 4. Time-lapse snapshots of the simulated congression process when motors are suppressed.
Chromosomes are shown as having chromatid arms (green) for viewing purposes, while the kine-
tochores are shown as yellow spheres. Not all MTs are shown, only those that serve as rails for
kinetochore motor-proteins (orange) and end-on attached MTs (red). The nuclear envelope is
shown for reference in each of the first panels as a white sphere. The cortex is represented in dark
grey. The wild type (WT) case, in which all motor proteins are active, is shown for comparison
in row 1. When dynein is suppressed (row 2), PEFs push peripheral chromosomes to the cortex.
However, when all chromosomes start between the poles, congression takes place normally. When
CENP-E is depleted (row 3), peripheral chromosomes or other chromosomes that are transported
to the poles get trapped there. Depleting PEFs (row 4) delays congression significantly and desta-
bilizes the coherence of the central plate. It makes no difference whether chromosomes start all
between poles or there are peripheral chromosomes.
25
Fig 5. Time-lapse snapshots of the simulated congression process for different values of the number
of MTs per kinetochore. Congression fails if this number is too small or too large. Chromosomes
are shown as having chromatid arms (green) for viewing purposes, while the kinetochores are shown
as yellow spheres. Not all MTs are shown, only those that serve as rails for kinetochore motor-
proteins (orange) and end-on attached MTs (red). The nuclear envelope is shown for reference in
each of the first panels as a white sphere. The cortex is represented in dark grey.
26
Fig 6. The distribution of (a) congression and (b) bi-orientation times for various MT densities.
The arrows indicate the trends for increasing MT densities. Congression is faster for a lower
number of MTs per kinetochore, because PEFs are directly proportional to the number of MTs.
However, bi-orientation is much slower for low MT densities, because the time needed to find every
kinetochore is strongly influenced by the number of MTs. This is summarized in (c) showing the
time tp=1/2 for which the congression/bi-orientation probability is one half. The maximum waiting
time for congression is 103s and for bi-orientation 105s.
If the MT density is too low, not all
samples bi-orient, as indicated by the red shaded area. Decreasing the MT density even further
severely reduces the congression probability, indicated by the blue shaded area. On the other hand,
increasing the MT density too much also impairs congression since kinetochore dynein will not be
strong enough to overcome PEFs. These results show that there is a sweet spot for congression/bi-
orientation as a function of the number of MT, lying between 7 × 103 and 1.8 × 104 MTs. All
27
curves have been obtained by averaging over n = 100 independent runs of the simulations. Error
bars are smaller than the plotted curves.
✁✂✄☎✆✝✞✟✝✞✠✝✞✡✝✞☛☞✌✍✞✎✞✞✞✝✞✞✞✞✝✎✞✞✞✏✞✞✞✞✏✎✞✞✞✑✁✒✓✔✕✖✗✘☎☎✘✙✓✚✗✔✛✔☎✔✛✘☎✏✞✜✞✢✞✎✞✣✤✥✦✞✝✞✟✏✧✝✞✟✜✧✝✞✟✢✧✝✞✟✎✧✝✞✟★✁✒✩✪✫✔✗✪✘✕✬✘✙✓✚✗✔✛✔☎✔✛✘☎✞✝✞✏✞✜✞✢✞✎✞✣✤✥✦✝✞✭✮✝✞✯✝✞✮✝✞✟✝✞✠✝✞✡✝✞☛✰✱✲✳✴✵✵✶✷✵✵✷✸✵✵✸✳✵✵✹✴✺✵✵✹✺✶✵✵✳✴✵✵✵✰✱✲✳✴✵✵✶✷✵✵✷✸✵✵✸✳✵✵✹✴✺✵✵✹✺✶✵✵✳✴✵✵✵ Fig 7. Time-lapse snapshots of the simulated congression process for different values of the rate
of MT catastrophes p0
cat. Large values of p0
cat, that is, overexpression of MT depolymerases, lead
to unsuccessful congression. The nuclear envelope is shown for reference in each of the first panels
as a white sphere. The cortex is represented in dark grey.
28
Fig 8.
Congression probability plotted against catastrophe rate. Overexpressing catastrophe
inducing factors can severely limit the congression probability. Each point represents the fraction
of n = 10 independent runs of the simulations that have reached congression during a waiting time
of 103s. Congression is stable over a wide range of catastrophe rates, but breaks down completely
at approximately at p0
cat = 0.046s−1.
29
|
1810.00701 | 1 | 1810 | 2018-10-01T13:39:25 | Effect Of Site Selective Ion Channel Blocking on Action Potential | [
"physics.bio-ph",
"q-bio.NC",
"q-bio.SC"
] | In this work we have theoretically investigated how the action potential generation and its associated intrinsic properties are affected in presence of ion channel blockers by adapting Gillepie's stochastic simulation technique on a very basic neuron of Hodgkin-Huxley type. With a simple extension of the Hodgkin-Huxley Markov model we have mainly investigated three types of drug blocking mechanisms and showed that the major experimental and physiological observations such as ionic currents, spiking frequency trends, change in action potential shape and duration, altered gating dynamics etc due to the presence of ion channel blockers can be well reproduced. The nature of action potential termination process in presence of sodium and potassium channel blockers are distinct and physiologically very different from each other. Channel blockers have distinct signatures on ionic currents. In presence of only sodium channel blockers the frequency of action potential generation falls off exponentially with increasing drug affinity, whereas in contrast, for only potassium channel blockers initially an enhanced spiking activity of action potential is found followed by a gradual decrease of the spiking frequency as the drug affinity increases. In case of dual type blockers with equal sodium and potassium channel binding affinity, the spiking frequency passes through maxima and minima due to the competition between channel number fluctuation and overall sodium and potassium conductances. We have found that sodium channel blockers shorten the duration of action potential while the potassium channel blockers delay it. We have also shown how the ion channel blockers alter the gating dynamics. Some experimental results of ion channel blocking in diverse systems have been validated through our site selected binding scheme. | physics.bio-ph | physics |
Effect Of Site Selective Ion Channel Blocking on Action Potential
Krishnendu Pal1,2,+ and Gautam Gangopadhyay2,∗
1Indian Association for the Cultivation of Science, Jadavpur, Kolkata-700032, India.
2S N Bose National Centre for Basic Sciences, Block-JD, Sector-III, Salt Lake, Kolkata-700106, India.
(Dated: October 2, 2018)
In this work we have theoretically investigated how the action potential generation and its as-
sociated intrinsic properties are affected in presence of ion channel blockers by adapting Gillepie's
stochastic simulation technique on a very basic neuron of Hodgkin-Huxley type. With a simple
extension of the Hodgkin-Huxley Markov model we have mainly investigated three types of drug
blocking mechanism such as (i) only sodium channel blocking, (ii) only potassium channel blocking
and (iii) dual type blocking and showed that the major experimental and physiological observations
such as ionic currents, spiking frequency trends, change in action potential shape and duration,
altered gating dynamics etc due to the presence of ion channel blockers can be well reproduced.
Our results show that the nature of action potential termination process in presence of sodium and
potassium channel blockers are distinct and physiologically very different from each other. We have
found that although the sodium and potassium ionic currents have interdependent relationship over
a course of an action potential but sodium and potassium channel blockers have distinct signatures
on ionic currents. In presence of only sodium channel blockers the frequency of action potential
generation falls off exponentially with increasing drug affinity, whereas in contrast, for only potas-
sium channel blockers initially an enhanced spiking activity of action potential is found followed
by a gradual decrease of the spiking frequency as the drug affinity increases. In case of dual type
blockers with equal sodium and potassium channel binding affinity, the spiking frequency passes
through maxima and minima due to the competition between channel number fluctuation and over-
all sodium and potassium conductances. We have found that sodium channel blockers shorten the
duration of action potential while the potassium channel blockers delay it which are of great phys-
iological and pharmacological importance. We have also shown how the ion channel blockers alter
the gating dynamics and such altered gating itself modulates the ion channel blocking which opens
the possibility of finding fundamental informations regarding probabilistic and dynamical features.
Some experimental results of ion channel blocking in diverse systems have been validated through
our site selected binding scheme. Many other types of blocking mechanisms such as closed state
blocking, inactive state blocking etc can also be explored using our method with desired level of
structural and functional details.
Keywords: Ion Channels; Sodium Channel Blockers; Potassium Channel Blockers; Action po-
tential, Local Anesthetics, Gillespie simulation.
I.
INTRODUCTION
Ion channels are typically very complex transmem-
brane proteins[1] exhibiting a high degree of both struc-
tural and functional diversity[2]. They have very distinct
electrical potential dependent gating mechanism where
the protein structures can adopt several conformational
states such as closed, open and inactive states in which
states they can either transport ions across the mem-
brane creating pores or restrict the ion permeability by
closing the pore when required[3]. The ion channels are
∗Electronic address: [email protected],*[email protected]
responsible for generating action potentials which are the
basic requirement of cell to cell communication and sig-
nal propagation in nervous system[2]. Among the var-
ious cation, anion and neutral ion channels in general
sodium and potassium channels are mainly responsible
for action potential generation and its termination in
most of the neurons. There exists certain compounds
or molecules, typically called as ion channel blockers
which selectively bind to specific protein conformations
of the ion channels and regulate their gating mechanism
by blocking the passage of ions across the membrane.
The high degree of specificity of these channel blockers
on certain channels make them a valuable tool to treat
numerous neural disorders[4 -- 6]. Channel blockers of dif-
ferent types such as cationic blockers, anionic blockers,
amino acids, and other chemicals regulate the functional
properties of the channel or prevent them to respond nor-
mally. The naturally occurring sodium channel selective
blocker TTX[7] was known since 1964, followed by Sax-
itoxin (STX), Neosaxitoxin (NSTX). Local anesthetics
such as Lidocain, Phenytoin, Amiloride, Bupivacaine and
Tetracaine etc are clinically used[8, 9] for sodium channel
blocking. On the other hand potassium channel blockers,
such as 4-aminopyridine and 3, 4-diaminopyridine etc are
used as local anaesthetics[10] and Tetraethylammonium
(TEA) is used only for experimental purpose[6, 11]. A
sodium blocker that blocks the open pore of the channel
are called open state blocker[8, 12, 13]. Also there exists
closed state blocker, inactive state stabilizing blockers
etc[8, 12, 13]. Potassium channel blockers either bind
with the selectivity filter or at the central cavity(open
state) of the channel[14]. From the discovery of TTX
toxicity[7] the present knowledge of ion channel block-
ers have been a very vast, complicated and rigorous
journey which was mostly developed for pharmacologi-
cal interests[4 -- 6, 9, 11].
Recently Markov modelling and computer simula-
tions prove to be promising techniques which provide
new insights into the fundamental principle and mech-
anism about how these drugs alter the normal biologi-
cal process[5, 15]. Single channel Markov models with
discrete protein conformational states can simulate state
specific channel properties and their alterations by mu-
tations, disease or drug binding[15]. As Markov mod-
els can be developed both at the level of single chan-
nel activity and at the macroscopic state, they pro-
vide an implicit relationship between the single channel
recording and the macroscopic current[16]. Now the fa-
mous neuronal action potential model of Hodgkin and
Huxley[17] is based on an assumption that the ion per-
meation process can be approximated as both continu-
ous and deterministic[17] as it considers infinite number
of ion channels inside a neuron. However, the perme-
ation process existing within active membrane is now
known to be neither continuous nor deterministic. Ac-
tive membrane consists of finite number of ion channels
which undergo random fluctuations between open and
closed states[2] which scale inversely proportional to the
number of channels present in a particular patch of a
neuron. Recent works also reveal that fluctuation in the
states of these channels are physiologically important in
small neuronal structures[18 -- 21]. When the number of
ion channels inside a neuron cell membrane is finite or
2
small, the effect of internal noise become more and more
important[18, 21 -- 23]. Channel number fluctuations can
itself cause spontaneous spiking activity[23] without any
stimulus and thereby the stochastic version of Hodgkin-
Huxley model is proposed[18, 20, 23]. Therefore, the ef-
fect of channel noise can not be ignored also in the study
of drug binding. Although innumerable number of exper-
iments have been performed to study the channel block-
ing phenomena, the theoretical investigation and com-
prehensive understanding of the state specific blocking
still requires detailed investigation from which the corre-
sponding physiological consequences of drug blocking can
be deduced that too from the level of action potential de-
scription itself. From the vast number of studies already
made it is very difficult to comprehend the mechanistic
link between the site specific blocking and the biophysical
consequences of that. So a microscopic reverse analysis
is necessary from theoretical perspective to validate the
known observations.
Inspired by the work done by Schmid, Goychuk and
Hanggi[24] on the effect of ion channel blockers using
Langevin type of stochastic description, we present here
a simple yet detailed and realistic approach for chan-
nel blocking kinetics using the standard Markovian squid
axon model of Hodgkin-Huxley[17] which represents a
basic neuron cell type. We have incorporated new drug
blocked states in the original Hodgkin-Huxley model to
theoretically extend it to a drug binding model. We then
have suitably adapted Gillespie[25] stochastic simulation
algorithm to study the effect of ion channel blockers on
action potential, ionic current, spiking frequency, action
potential duration and gating dynamics in moderately
stochastic limit of channel noise. We have proposed
three types of drug blocking mechanisms, e.g, sodium
channel only blocking, potassium channel only blocking
and a dual type blocking scheme with comparative bind-
ing affinity to these two channels. We have shown here
through a few experimental evidences[26 -- 38] that our ap-
proach successfully correlates to the known physiological
effects of ion channel blocking.
The Layout of the paper is as follows. In Section (II)
we have discussed the kinetic scheme of drug blocking and
the suitable modification of the simulation algorithm for
the present purpose. Effect of channel noise on the action
potential has been discussed also under this section. In
Section (III) the sodium and potassium channel blocking
effects on the trains of action potential have been dis-
cussed with fundamental biophysical details. In Section
(IV) comparative study on the effect of different types of
blockers on ionic current, spiking frequency, action po-
tential duration and gating dynamics has been discussed
under various subsections. In Section (V) we have dis-
cussed how our results are in good agreement with the
experimental and theoretical works done earlier. Finally
the paper is concluded in Section (VI).
II. PROPOSED KINETIC SCHEME AND
STOCHASTIC SIMULATION OF ACTION
POTENTIAL IN PRESENCE OF CHANNEL
BLOCKERS
The
such as
implemented in two ways
stochasticity in the Hodgkin-Huxley model
theoretical
is
approximation[39] method where it approximates the dy-
namics of the internal gating variables are governed by
Langevin[39] type of description of stochasticity, where
the Gaussian white noise terms are added to the internal
gating dynamical variables, described in the later part of
this paper. The other one is the computational stochas-
tic simulation based on kinetic Monte-Carlo simulation
of Hodgkin-Huxley Markov model[18 -- 20, 25, 40 -- 44]. In
this paper we have used the second one.
A. Markov Model of Hodgkin-Huxley Action
Potential Simulation
Here we briefly describe the Hodgkin-Huxley Markov
model and its kinetic scheme.
The Hodgkin and
Huxley(1952)[17] gating mechanism consists of four ac-
tivating n gates for potassium channel and its Markov
model consists of five states, such as, n0(resting state,
when all four gates are closed), n1(only one gate is open),
n2, n3 and n4(open state or the single ion conducting
state, when all four gates are open). The model of potas-
sium channel thus have 8 transition rates designated by
αns and βns between these 5 states. On the other hand,
the sodium channel has three activating m gates with
3
four distinct states and one inactivating h gate with
two distinct states. Thus the kinetic scheme based on
Markov process has 8 states, such as, m0h0(resting state),
m1h0, m2h0, m3h0, m0h1, m1h1, m2h1, m3h1(open state
or ion conducting state), with a total of 20 transition
rates designated by αm,hs and βm,hs. Thus as a whole
there are 28 transitions between 13 states to be consid-
ered for simulating action potential[43]. Each ion channel
randomly fluctuates between these discrete states.
kpon
FPB
α
2
n
β
3
α
n
β
4
n
α
n
β
α
β
n
4
n
3
n
2
5
n
1
n
0
3
2
2
n
1
n
n
4
n
3
4
D
P
(a) Markov Kinetic Scheme of Potassium Ion Channel
5
m h
10
α
h
β
h
1
m h
0 0
α
3
m
β
m
3
α
m
β
m
6
m h
1
1
2
α
m
β
m
2
7
m h
2
1
α
m
3
β
m
8
m h
3
1
ks on
FSB
DS
α
h
β
h
2
m h
1
0
α
m
β
m
2
2
α
h
β
h
3
m h
2
0
α
m
β
m
3
βh
α
h
4
m h
3
0
(b) Markov Kinetic Scheme of Sodium Ion Channel
FIG. 1: Markov model of Potassium and Sodium
ion channel: The potassium channel in figure (a) is a
five state model where we have added an extra
'drug-bound' state, DP . Similarly in figure (b) sodium
channel is an eight state model where we have added a
'drug-bound' state, DS.
The expressions of the voltage dependent rates[23]
are given as follows αm(V ) = (0.1(V + 40))(1 −
exp[−(V + 40)/10])−1,
βm(V ) = 4 exp[−(V +
65)/18], αh(V ) = 0.07 exp[−V + 65)/20], βh(V ) =
1 + exp[−(V = 35)/10]−1, αn(V ) = (0.01(V + 55))(1 −
exp[−(V + 55)/10])−1, βn(V ) = 0.125 exp[−(V + 65)/80].
The Hodgkin-Huxley[17] action potential or transmem-
brane voltage, V is given by the equation,
Cm
d
dt
V (t) + GK(t)(V (t) − EK) + GN a(t)(V (t) − EN a) + GL(V (t) − EL) = Iext(t).
(1)
Here V is in mV unit and the rates are in ms−1. Param-
eters and their descriptions are given in TABLE I [23].
4
Cm
EK
ρK
gmax
γK
K
EN a
ρN a
gmax
γN a
N a
EL
gL
Membrane capacitance
K+ reversal potential
K+ channel density
Maximal K+ channel conductance(all K+ channels are open)
Single K+ channel conductance
Na+ reversal potential
Na+ channle density
Maximal Na+ channel conductance(all Na+ channels are open)
Single Na+ channel conductance
Leak reversal potential
Leak conductance
TABLE I: Parameters of Hodgkin-Huxley equation[23].
1 µF/cm2
-77.0 mV
18 channels/ µm2
36.0 mS/cm2
20 pS
50.0 mV
60 channels/µm2
120.0 mS/cm2
20 pS
-54.4 mV
0.3 mS/cm2
For a discrete stochastic channel populations the potas-
sium and sodium membrane conductances across are ex-
pressed by the following equations (2),
GK(V, t) = gmax
GN a(V, t) = gmax
K [Nn4/NK] ,
N a [Nm3h1/NN a] ,
(2a)
(2b)
where Nn4 and Nm3h1 are the number of potassium and
sodium channels in open state, respectively and NK and
NN a are the total number of potassium and sodium chan-
nels present in the membrane patch considered, respec-
tively. Solving the stochastic simulation one obtains the
population of individual states from which the open state
populations for sodium and potassium channel are put
into conductance equation (2) and then solving equation
(1) one obtains the action potential.
B. Proposed Scheme of Channel Blocking
Now, in this work we have studied the effect of ion
channel blockers on action potential. Ion channel block-
ers can bind to several conformational states of ion chan-
nels such as closed state, inactivated state and open-state
of ion channels[2]. However, here we have only considered
the case of those blockers which preferentially bind to the
open states of the channels. As these blockers block the
open-pore of the channel and hinder ion permeation, we
can think of a state, say,"drug bound" state which is ac-
cessible only if the channels enter the open state. Thus
the drug bound state is coupled to the open state of the
sodium or potassium channel. The drug bound state ba-
sically represents the state of channels which are blocked
by the blockers. To implement this on potassium chan-
nel we designate this drug bound state as DP and for
sodium channel it is DS as seen from figure (1). The
forward transition rate for sodium or potassium blocking
are given by 'kxon', where x=s for sodium blockers and
x=p for potassium blockers. The affinity of drug bind-
ing is usually defined as kxon= kxb*[D], where kxb is the
binding constant of the drug and [D] is the concentra-
tion of the drug[12]. Thus any change of the value of
'kxon' means the change of concentration of the block-
ers or drugs of a particular type having specific binding
constant.
Now the backward transition for reopening is a very
slow process compared to the forward. Channel block-
ers or local anesthetics take around 1-4 hours(duration
of action) before they are completely removed1 from the
channel proteins[45]. In general, clinically used channel
blockers or local anesthetics can be divided into three
categories:
short acting (e.g., 2-chloroprocaine, 45-90
minutes), intermediate duration (e.g., lidocaine, mepi-
vacaine, 90-180 minutes), and long acting (e.g., bupi-
vacaine,
levobupivacaine, ropivacaine, 4-18 hours)[45].
Thus it is seen that elimination of the drug from the
site or the recovery of the channel is a very slow pro-
cess compared to that of drug binding where the onset
of local anaesthetic actions take place within few min-
utes(as we shall also see that the simulation runs with
the transition rates mentioned earlier take few seconds
only to completely terminate the action potential gen-
eration for a patch area we have considered in this pa-
per ). Thus the backward flux can therefore reasonably
be considered as a very small valued constant during the
1 Amide blockers(lidocaine, bupivacaine, mepivacaine etc) are bio-
transformed in the liver, ester blockers(cocaine, benzocaine, pro-
caine etc ) are hydrolyzed in the bloodstream by plasma esterases
or the hydrolysis of the side chains of the blockers make them
inactive and then they are eliminated via blood circulation[45].
course of drug binding action. We designate the back-
ward flux as FxB=0.001, a constant throughout2. Now,
we have gradually increased the drug binding affinity,
kxon from 0.0001 to 1.0 or more, keeping FxB=0.001,
constant. When we consider only sodium blockers we
do not consider the potassium blocking state, DP and
vice verse except for dual type of blocking as we shall
discuss later. Thus with the attached single drug-bound
state Hodgkin-Huxley model now have 14 states and 30
transition rates.
C. Stochastic Ion Channel Simulation Using
Gillespie Algorithm
The computer based stochastic simulation algorithm
using Markovian model can be classified into two al-
gorithms, such as, i) Channel-State-Tracking(CST) al-
gorithm and ii) Channel-Number-Tracking(CNT) algo-
rithm. The CST algorithm tracks the specific states of
each channel and superimposes individual channel cur-
rents corresponding to the states. This algorithm al-
though simple but it is computationally very costly and
intensive[19 -- 21]. However an easy alternate is CST algo-
rithm using Gillespie method[25, 40, 42, 43] which keeps
track of the number of channels in each state with the
assumption that multi-channel systems are independent
and memoryless. This Channel Number Tracking(CNT)
algorithm provides much greater efficiency in cases where
many channels are simulated as the algorithm calculates
an effective transition rate associated with the multi-
channel system by allowing only one transition among
all states in a random time interval. Thus in this paper
we have utilized Gillespie's CST simulation algorithm to
study the effect of ion channel blockers. The simulation
method is very well known in the literature[25, 40, 42, 43].
The steps that we considered to develop the stochastic
1
1×3600×103
2
2 There is roughly 18,000 channels in total are present in 200 µm
patch area(area that we have chosen in this work). If we even
consider the blocking action prolongs for 1 hour, then the approx-
imate flux becomes FxB ≈
× 18000 ≈ 0.005. Now
in fact not exactly 18000 blocked channels will be there. Blood
circulation hinders the channel blockers to penetrate the lipid bi-
layers by washing them out and a certain fraction of the blockers
can enter to block channels, thus have lesser chance to block all
the channels, also density of the patch is not homogeneous ev-
erywhere that some patches can have lesser number of channels.
Besides some drug actions last more than 4 hours. Consider-
ing these facts together we have taken a lower value which is
0.001(easy to compare), which hardly impacts the nature of the
results as we have verified.
algorithm for the site selective binding of sodium and
potassium blockers for a particular patch size of a neu-
ron have been elaborately discussed in Appendix (A1).
5
D. Effect of Channel Number Fluctuation
Next we have studied the action potential and con-
ductance of sodium and potassium channels without the
presence of external stimulus or external current, Iext =
0.0µA/cm2. We start from very high stochastic limit of
channel noise with patch size as low as A = 1.0 µm2
and extend it to the deterministic limit such as A = 200
µm2. Here we do not consider the presence of any drug,
hence kxon = FxB = 0.0. Thus here we consider 13 states
and 28 reactions: the original Hodgkin-Huxley model. In
90
60
30
0
-30
-60
60
30
0
-30
-60
)
V
m
(
V
)
V
m
(
V
)
V
m
(
V
60
30
0
-30
-60
)
V
m
(
V
60
30
0
-30
-60
0
0
0
0
Iext=0.0
A=1.0
a
50
100
150
200
250
300
350
400
450
500
A=10.0
b
50
100
150
200
250
300
350
400
450
500
A=100.0
c
50
100
150
200
250
300
350
400
450
500
A=200.0
d
50
100
150
200
250
Time (mS)
300
350
400
450
500
FIG. 2: Effect of channel number fluctuation on
action potential: Action potentials of various patch
size are plotted here. From (a) to (d) the action
potentials are shown for A= 1.0, 10.0, 100.0, 200.0 µm2
at Iext = 0.0 µA/cm2. It shows how the effect of
channel noise and consequent spontaneous action
potential generation falls off as the deterministic limit is
approached.
figure (2) we have shown how the action potential vary
from patch size, A=1.0 to 200 µm2. As seen from figure
2(a) that at very low patch size channel noise plays very
important role. These channel number fluctuations can
alone originate spontaneous spiking activity[23] without
even the presence of external current. As soon as the
patch size is increased the spontaneous spiking rate de-
creases. As we can see at moderately high patch size,
for example, A = 200µm2, the spiking phenomenon van-
ishes away and system behaves similarly as it does in
deterministic limit at Iext = 0.0µA/cm2.
As a side note we would like to mention here that the
Gillespie method that we have used to simulate action
potential, although is a very popular method and often
claimed to be an exact simulation technique in litera-
ture, however, contrary to that popular belief it can not
be claimed as an exact method especially while consider-
ing very low patch sizes. It is basically an approximate
method which works fine for higher patch sizes. A brief
discussion about its applicability has been discussed in
Appendix (A2). Exact simulation for figure 2 (a) and
(b) is beyond the scope here. These two plots are not ex-
act but here their sole purpose is to convey the message
that at low patch sizes the channel fluctuations do play
very important role.
6
A. Only Sodium Channels Are Blocked
As mentioned earlier, here we have considered only the
open state sodium channel blockers in this study. Here we
do not add any potassium blockers together. Thus here
only DS state is present not DP . In figure (3) we have
Effect of Sodium Channel Only Blockers
-1
( I = 6.9 A/cm , A = 200 m, Fs = 0.001 ms )
µ 2
µ
2
ext
B
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
1000
2000
3000
4000
5000
6000
7000
8000
No Drugs
9000
ks = 0.001
on
10000
9000
ks = 0.01
on
10000
9000
ks = 0.05
on
10000
9000
10000
ks = 0.1
on
9000
ks = 0.5
on
10000
a
0
b
0
c
0
d
0
e
0
f
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
III. SODIUM AND POTASSIUM CHANNEL
BLOCKERS
Now we come to the main focus of this work, i.e. study-
ing stochastic drug binding kinetics in ion channels. To
see the effect on action potentials we have gradually in-
creased the external stimulus from zero to higher values
and seen that at Iext = 6.9 µA/cm2 they show consid-
erable amount of spiking activity. The values of ionic
current lesser than 6.9 µA/cm2 show very irregular spik-
ing pattern with higher intervals which requires very long
time simulations and become very clumsy to be presented
in plots. At this selected value of ionic current they show
considerable amount of spiking activity with lesser in-
tervals which can be plotted very nicely and helps us
to interpret the effect of drugs on action potential spik-
ing trend. However the results in general discussed here
show exactly similar behavior for all other choices of Iext
other than zero. We have kept the patch size, A=200
µm2 for which the effect of channel number fluctuation
is moderately close to the deterministic limit but not ex-
actly the deterministic limit, because with the value of
Iext = 6.9 µA/cm2, a considerably higher patch size is re-
quired to match deterministic result. On the other hand
at very low patch size the channel noise makes it almost
impossible to study the binding effects. At this value
of patch size, A=200 µm2, we are dealing with almost
12000 sodium channels and 3600 potassium channels.
0
100
200
300
400
500
600
700
800
900
1000
Time (ms)
FIG. 3: Effect of sodium channel only blockers:
Action potentials in presence of sodium channel only
blockers with increasing affinities are plotted here. In
figure (a) the action potentials are plotted without the
presence of any drug. From figure (b) to (f) the action
potentials are plotted for kson=0.001, 0.01, 0.05, 0.1
and 0.5 with A= 200 µm2, Iext = 6.9 µA/cm2 and
reopening flux, F SB= 0.001 ms−1. In all the cases
where Na-blocker is present the action potential train
ultimately dies off, sooner for higher affinity(not shown
for (b) and (c) as they require more than 20,000 ms,
plotting them makes the plots compact and clumsy, the
train dies off similarly as shown in (d)-(f)).
shown the effect of sodium channel blockers on action po-
tentials train with time with increasing drug affinity or
concentration. It is seen that in presence of drug the spik-
ing activity gradually dies off temporally. With increas-
ing affinity the spiking activity dies off in a faster rate.
As seen from figure 3(f) at very high value of kson = 0.5
the system almost fails to regenerate action potentials.
As the affinity increases more channels quickly go to the
drug-bound state and gets trapped in the drug bound
state. Thus the available open sodium channels gradu-
ally decreases which makes it difficult for that particular
patch to generate action potential. Beyond kson = 1.0
we have seen that the patch totally fails to generate even
a single spike after the first spike.
B. Only Potassium Channels Are Blocked
Next we have studied the effect of open state potassium
channel blockers on the action potential generation. Here
the DP state is only considered, not DS state.
In fig-
Effect of Potassium Channel Only Blocker
-1
(I = 6.9 A/cm , A = 200 m , Fp = 0.001 ms )
µ
2
ext
µ
2
B
a
No Drugs
0
1000
2000
3000
4000
5000
6000
7000
8000
9000 10000 11000 12000 13000 14000 15000
b
0
c
1000
2000
3000
4000
5000
6000
7000
8000
kp = 0.0001
on
9000 10000 11000 12000 13000 14000 15000
kp = 0.001
on
0
1000
2000
3000
4000
5000
6000
7000
8000
9000 10000 11000 12000 13000 14000 15000
400
800
1200
1600
2000
2400
2800
3200
3600
4000
kp = 0.01
on
kp = 0.005
on
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
d
e
f
0
0
0
7
larize > hyperpolarize > again comes back to resting po-
tential). But for potassium blocking case we see that
after a few millisecond the spiking activity entirely falls
off with the last spike having incomplete shape. This
happens due to lack of available open potassium chan-
nels. Due to the lack of available open potassium chan-
nels the repolarization process gets hampered, leading to
the incomplete generation of action potential. Once it is
not complete further generation of action potential is not
possible as the refactorization process which was neces-
sary for further generation of action potential could not
complete. Thus no action potential generation can take
place further as seen from figures 4(d-f). But in case of
sodium channels there remains plenty of available potas-
sium channels which can restore the depolarized potential
back to the resting one. Thus in sodium blocking case the
spikes generate with complete shape even if there number
decreases with increasing affinity.
250
500
750
1000
1250
1500
1750
2000
2250
2500
kp = 0.1
on
IV.
EFFECT OF THE CHANNEL BLOCKERS
50
100
150
200
250
300
Time (ms)
FIG. 4: Effect of potassium channel only
blockers: Action potentials in various blocking
affinities of potassium only blockers are plotted here. In
figure (a) the action potentials are plotted without the
presence of any drug. From figure (b) to (f) the action
potentials are plotted for kpon = 0.0001, 0.001, 0.005,
0.01 and 0.1 for A = 200 µm2, Iext = 6.9 µA/cm2 and
reopening flux, F PB = 0.001 ms−1. Although here also
action potential dies off temporally but the nature of
termination is quite different from that of the sodium
channel only blocker.
ure (4) the action potentials for various kpon are plotted.
Here also for all the cases where drug is present ultimately
the spiking activity of the action potential dies off tem-
porally. As usual with increasing affinity the blocking of
action potential occurs in a faster rate. But, surprisingly
here we have observed an interesting trend of action po-
tential termination which is quite different from sodium
blocking case.
If we carefully notice the spikes in figures (3) and (4)
at high affinity regions, we can see that in presence of
sodium channel blockers the number of generated action
potentials although decreases down significantly, but ev-
ery spike has a complete shape(they depolarize > repo-
In this section we have shown how the sodium and
potassium channel blockers and local anesthetics(dual
type blockers) affect the ionic current, spiking frequency,
duration of the action potential and gating dynamics, all
of which plays physiologically very significant roles.
A.
Ionic Current
The sodium and potassium ionic currents across the
membrane are given by the eqution (8). Now here we
would like to point out that although sodium and potas-
sium channels are two different proteins, during an action
potential, they work together and thus sodium and potas-
sium currents have interdeppendent relationship. The
number of spikes of sodium ionic current is equal to the
number of spikes of potassium ionic current. Hampering
sodium channel conductance with sodium channel block-
ers may not directly affect the gating mechanism inside
the potassium channel proteins, but the decreased effi-
ciency of conductance leads to decrease in the number
of sodium ions coming inside and thus less number of
potassium ions are required go out side to balance that
depolarization, leading to a decrease in the magnitude of
potassium current too. On the other hand if we keep on
decreasing the ionic conductance of potassium channels
by applying potassium blockers, the repolarization pro-
cess or refactorization process would gradually get ham-
pered which in turn will also affect the membrane po-
tential which eventually becomes unsuitable for sodium
channels to open as we have seen from figure (4). Thus
both of the currents are affected if anyone of them is
disturbed.
Now in figure (5) we have shown the interdependent
nature of sodium and potassium currents and how the
currents are being affected in presence of ion channel
blockers with increasing drug binding affinity. The cur-
rents are generated in response to a change in the respec-
tive ion channel conductances. The total number of Na+
or K+ ions that permeate inside or outside the membrane
during the action potential is proportional to the areas
under the individual current curves in the figures.
Sodium Only Blocker
Potassium Only Blocker
No Drugs
ks = 0.05
ks = 0.5
on
on
a
I
K
I
Na
1000
800
600
400
200
0
-200
-400
-600
-800
No Drugs
kp = 0.05
kp = 0.5
on
on
1000
800
600
400
200
0
-200
-400
-600
-800
-1000
-80
-60
-20
-40
Action Potential (mV)
20
0
40
-1000
-80
60
-60
-20
40
-40
Action Potential (mV)
20
0
b
60
FIG. 5: Ionic current: In this figure we have plotted
the sodium and potassium channel ionic currents
together with their corresponding action potentials. In
the left panel, figure (a), the effect of sodium channel
only blockers with affinity, kson= 0.05, 0.5 are
compared with the case where no drug is present. The
top trace represents the potassium ionic current and the
bottom trace represents the corresponding sodium ionic
current. In the right panel similar plot has been shown
with potassium only blockers. A= 200 µm2 and
Iext = 6.9 µA/cm2. Both the blockers have different
signatures of blocking.
The red dotted curves in each traces of figure 5(a) rep-
resents the case of drug free situation. After the first
spike the loop area shrinks. Afterwards the area of ionic
currents slightly change due to the presence of the fluc-
tuation in the number of available open channels which
creates the broadening of the band. The blue dot-dashed
curve for affinity kson=0.05 and and black-solid line for
kson= 0.5 shows that both of the ionic currents gradu-
8
ally shrinks over time. If we plot in a 3D diagram with
membrane depolarization in x axis, currents in y axis and
time in z axis we see that with increasing time the area
of the ionic currents shrink gradually with corresponding
decrease in the peak height of the action potentials. This
shows the gradual decrease of sodium ion influx inside the
cell membrane.
Next we have observed an interesting feature of potas-
sium channel blocking on both the ionic currents. Un-
like the case of sodium blockers in 5(a), here we see
that the loop area of both the ionic currents falls down
very rapidly covering the entire space which indicates
the decrease in the potassium channel conductance over
time. Unlike the sodium only blockers, the correspond-
ing action potential magnitude is much higher here when
the currents fall down to zero. This once again clearly
says that the membrane repolarization process is dras-
tically hampered. As the loop areas are proportional
to the number of ions being permitted across the mem-
brane, the potassium only blockers show this amount is
hampered more rapidly than sodium blockers of equal
affinity. The effect of blockers on effective ionic current,
Iint = IN a + IK + IL over the corresponding action po-
tentials can also be found in the supplemental materiel
figure (S1).
B.
Spiking Frequency
Next we have studied the effect of the two types of
blockers on the spiking frequency of the action potentials
with increasing kxon in figure 6(a). Here we have found
an interesting difference of spiking frequency trends be-
tween sodium and potassium blockers. It is seen that in
presence of sodium channel blockers the spiking rate falls
off exponentially with increasing affinity. But in presence
of potassium channel blockers the spiking frequency ini-
tially with increasing kpon increases unlike sodium chan-
nel and then it decreases. This initial increase in spiking
activity in presence of potassium blockers is consistent
with the earlier works done in literature [24, 46]. This
happens because with the increase in affinity number of
potassium channels decrease which causes an increase in
the internal noise and therefore channel number fluctua-
tion starts playing significant role by spontaneously gen-
erating action potentials, as we have already seen from
the figure(2). Actually in presence of potassium blockers
there exists a competition between the overall conduc-
tance of the ions and the channel noise. In the low affin-
)
z
H
(
y
c
n
e
u
q
e
r
F
e
k
i
p
S
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0
Sodium / Potassium Only Blockers
a
Sodium only Blocker
Potassium only Blocker
0
0.01 0.02
0.03
0.04
Affinity
0.05
0.06
0.07
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0
b
Dual Type Blockers
ks > kp : 10 times
on
kp > ks : 10 times
on
ks = kp
on
on
on
on
70
60
50
40
30
20
10
0
0.002
0.004
0.006
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Affinity
FIG. 6: Spiking frequency: In figure (a) the spiking
frequency(Hz) of the action potentials in presence of
increasing sodium and potassium channel only blocking
affinity, kson and kpon have been plotted respectively,
ranging from 0.0001 to 0.07. The spiking frequency is
calculated over a long simulation run upto 20s. The
squared-(black)-solid line and the diamond-(red)-dashed
line shows the frequency of spiking of action potentials
in presence of only sodium and potassium channel
blockers, respectively. In figure (b) the spiking
frequency profile is plotted for the dual type blockers
for the cases: (i) kson > kpon: squared-(black)-solid
line, (ii) kpon > kson: circled-(red)-dashed line and for
(iii) kson = kpon: diamond-(blue)-dot dashed line. For
all the cases A = 200 µm2 and Iext = 6.9 µA/cm2 and
those spikes are only considered which have peak
heights more than -10 mV.
ity region the influence of the channel noise dominates
over the ion conduction and thus the spiking activity in-
creases. After a certain kpon value, the spiking activity
decreases as the decreased overall conductance dominates
here. For sodium channel only blockers, the over all con-
ductance always dominates over channel noise.
C. Action Potential Duration
9
ters with time. First we have chosen three drug affinity
regions such as low, medium and high, corresponding to
the sodium and potassium blocking. For sodium block-
ers we have chosen the affinities, kson= 0.005(low), 0.05
(medium) and 0.1(high). For potassium blockers we have
chosen the affinities, kpon = 0.005(low), 0.01 (medium)
and 0.05(high). The regions are different because with
the same magnitude of affinity potassium blockers affect
the spiking activity more rapidly than sodium blockers as
seen comparing figures (3) and (4). Next for each affinity
we have arbitrarily chosen a spike from few initial spikes
and then compared it to the randomly selected spikes
from intermediate and end positions of the spike trains
and plotted them in a same scale.
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
Effect of Sodium Channel Only Blocker
Initial spike
Intermediate
Ending
onks = 0.01
40
20
0
-20
-40
-60
ks = 0.05
on
b
40
20
0
-20
-40
-60
onks = 0.1
ks = 0.1
c
16
18
20
Time(ms)
-80
14
22
16
18
20
Time(ms)
-80
22
14
16
18
20
Time(ms)
22
Effect of Potassium Channel Only Blocker
kp = 0.005
d
on
40
20
0
-20
-40
-60
kp = 0.01
on
e
kp = 0.1
f
on
40
20
0
-20
-40
-60
40
20
0
-20
-40
-60
-80
14
40
20
0
-20
-40
-60
-80
10
12
16
14
18
Time(ms)
20
22
-80
10
12
14
16
18
20
Time(ms)
-80
10
22
12
14
16
18
20
22
Time(ms)
FIG. 7: Action potential duration: The temporal
change of APD in presence of ion channel blockers have
been shown here. In the top traces, (a-c), the initial,
intermediate and end point spikes are plotted in a same
scale for sodium channel only blockers with affinity kson
= 0.005(low), 0.05(medium) and 0.1(high ). In bottom
traces, (d-f) similar plots have been done for potassium
channel only blockers with affinities, kpon = 0.005(low),
0.01 (medium) and 0.05(high). In both cases, A= 200
µm2 and Iext = 6.9 µA/cm2. The green-solid arrows are
indicating the direction of shortening or broadening of
the incoming action potential.
Next we have seen the effect of channel blockers on
the action potential duration(APD) itself. APD varies
from one neuron to another neuron type. Any alter-
ation of APD of a particular neuron can lead to sig-
nificant complexities in signal transmission process and
can give rise to critical physiological disorders. We have
found that in presence of blockers APD significantly al-
From figure (7) we have again found an interesting dif-
ference in APD for two types of blockers. It is seen that
for sodium channel only blockers with increasing time
the repolarization process gets faster or the shortening
of the APD occurs where as contrastingly, for potas-
sium blockers the repolarization phase is seen to be de-
layed considerably with increasing APD. The shortening
of APD in presence of sodium blocker happens because
the effective number of sodium ions are entering the cell
decreases with time and thus less number of potassium
ions required to go out to bring back the repolarization
which thus gradually takes lesser time than usual. On the
other hand in presence of potassium blockers the repo-
larization phase is seen to be delayed temporally because
the number of available open potassium channels gradu-
ally decrease and thus the repolarization brought by the
decreased number of channels take longer time. Thus
the broadening of APD occurs gradually. This shorten-
ing or the broadening of the APD is actually a special
characteristic feature of sodium and potassium channel
blockers respectively which have great physiological sig-
nificance and clinical use. Patients with longer or shorter
APD due to mutagenic, hereditary or other physiological
conditions are treated with ion channel blockers.
Dual Blockers
Here we want to point out that some ion channel block-
ers such as local anesthetics are well known for their
non-specific blocking nature. Also at an elevated con-
centration some drugs that primarily target Na chan-
nel may also seen to affect K channels. Channel spe-
cific type of blocking as discussed so far is more pro-
nounced for TTX(specifically binds to Na channel only)
and TEA(binds to potassium channel only) types of
blockers only which are mostly used in experimental re-
searches. But for local anesthetics or clinically used drugs
we must consider both type of blocking simultaneously.
For that purpose we have chosen three types of block-
ing schemes which summarizes all possible kinds of non-
specific binding mechanisms. For the first case we have
kept both the sodium(kson) and potassium(kpon) drug
binding affinities equal, i.e, kson = kpon. So these rep-
resents those class of drugs which binds to both type
of channels with equal affinities. For the second case
we have chosen those types of drugs which have ap-
proximately 10 times higher affinity of blocking sodium
channels[12]. The major difference between local anes-
thetic action in K+ currents compared with Na+ cur-
rents is the lower affinity(approximately 10 times lower)
in the former[12]. So here we have kept the sodium block-
ing affinity, kson > kpon ≈ 10 times. Finally we kept
kpon > kson ≈ 10 times. One can change the ratio ac-
cording to the need or according to the knowledge of the
10
binding affinity of a particular drug. The figures of ac-
tion potential trains for these three categories of blocking
can be found in the supplemental figures (S2), (S3) and
(S4).
Spiking Frequency: Now we have plotted the spik-
ing frequency profile for these three cases in figure 6(b).
For the case of kson = kpon(diamond-blue-dot-dashed
line) it is seen that initially frequency decreases up to
around kson=kpon=0.001 and then passing through a
minimum it increases between a small region upto around
0.004 and then a large decrease is observed until 0.01 is
reached. After that again an increase in frequency is ob-
served until 0.03 and then it falls off gradually. These sort
of double minimum and maximum in frequency of spiking
actually indicates a competition between overall sodium
and potassium ionic and channel noise activity. The re-
gions where the maxima exist are basically the regions
where the channel number induced spontaneous spiking
activity dominates on overall ion conductance[24].
In
other words it can be said that the maximums arise due
to the coherence resonance due to channel noise[23]. For
the other two cases such as kson/kpon ≈ 10(squared-
black-solid line) and kpon/kson ≈ 10(circled-red-dashed
line), the previous trends of sodium and potassium only
blockers are observed as seen from figure 6(a). The case
of kson = kpon is different which indeed reduces to the
other two cases when the ratios become 10. However,
for all the three cases it is seen that the action potential
generation gradually ceases off with an the unsuitable
refactorization potential which clearly indicates that to-
wards the end of the action potential train, potassium
channel blocking plays more influential role in the action
potential termination process.
Action Potential Duration: Next we have plotted
the effect of these three types of drug blocking mecha-
nisms on the duration of the action potential in figure
(8) over one individual action potential train. For the
kson > kpon case(the middle trace), we have found that
initially the APD shortening occurs for few ms due to the
dominant(kson 10 times stronger than kpon) blocking of
sodium channels over potassium channels. But shortly
after when sufficient amount of potassium channels are
blocked, the effect of potassium blocking starts playing
significant role by delaying APD. The inward arrow in
the middle trace of figure (8) shows the shortening of
APD followed by an outward arrow indicating the grad-
ual delay of APD as time progresses. We have verified
this for all affinity region. For higher affinity region, such
Effect of Local Anesthetics on Action Potential Duration
ks = kp on
on
ks = 0.01
kp = 0.01
on
on
40
20
0
on
kp > ks : ~10 times
on
kp = 0.01
ks = 0.001
on
on
ks > kp : ~10 times
on
on
ks = 0.01
kp = 0.001
on
on
0
simulation time
t
15 16 17 18 19 20 21 22 23 24 25
15 16 17 18 19 20 21 22 23 24 25 15 16 17 18 19 20 21 22 23 24 25
Time (ms)
Time (ms)
Time (ms)
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
-20
-40
-60
-80
FIG. 8: Action potential duration affected by
dual blockers: The change in the duration of the
action potential in presence of local anesthetic channel
blockers at a higher affinity over their individual course
of action potntial train have been plotted for the cases
kson = kpon(left), kson > kpon(middle) and
kpon > kson(right). The approximate positions of the
five randomly chosen spikes over a train of action
potential simulation run of t ms are shown in the inset
of middle figure. Here also A= 200 µm2 and Iext = 6.9
µA/cm2.
as kson = 0.1, the broadening of APD occurs, preceded
by a quick APD shortening. However, here we want to
point out that the shape of action potentials in this case
have been found to be heavily affected such as the grad-
ual decrease of the peak height of action potentials as
time progresses(see supplemental figure S3). Keeping the
peaks fixed we have found that the initial upward depo-
larization phase of action potentials also changes signif-
icantly, as seen in the left portion of the peak in the
middle trace. So the change of APD here, particularly
in this case is considered as the peak after shortening or
broadening. In the right trace, the case kpon > kson is
shown which shows exactly similar nature of APD broad-
ening as shown in potassium channel only blocking case
in figures 7(d-f). For the kson = kpon case in left trace,
although the frequency response curve came up with dif-
ferent signature but for the duration of the action po-
tential it basically displays the potassium blocking dom-
inance by delaying action potential duration. Following
Table(II) summarizes our results:
D. Gating Dynamics
Here we have studied the percentage of the total pop-
ulation of channels present at a particular time in dif-
11
ferent conformational states including the drug bound
state. In the left traces of figure (9) the population(%)
for sodium only blockers for affinity, kson= 0.1 and the
corresponding population(%) of potassium channel states
has been shown in right traces. Observing both the traces
we find that each of the action potential spike is associ-
ated with either upward or downward spike in all the
population plots, both for sodium and potassium con-
formational states. This indicates that sudden random
action potential generation is associated with sudden dra-
matic changes of the entire population occupancy of all
the conformational states. From the left trace of figure
(9) we see that in 20 seconds( although spikes occurring
ceases long ago) only ∼50% channels are trapped in drug
bound state, DS which slowly increases with time. The
rest of the population is mainly trapped in the inactive
states and closed states. With increasing affinity the drug
bound state populates in a faster rate.
Sodium Channel Only Blocker, ks = 0.1, A= 200, I = 6.9
ext
30
0
-30
-60
80
60
40
20
0
40
20
0
30
20
10
0
80
60
40
20
0
30
20
10
0
15
10
5
0
15
10
5
0
15
10
5
0
Population(%) of Na+ Channel Conformational States
V
m
(mV)
[ m h ]
0
0
[ m h ]
1
0
[ m h ]
0
2
[ m h ]
0
3
[ m h ]
1
0
[ m h ]
1
1
[ m h ]
1
2
[ m h ]
1
3
0
1000
2000
3000
4000
5000
6000
7000
Total Population
[ D ]
S
100
80
60
40
20
0
0
on
Population(%) of K+ Channel Conformational States
V
m
(mV)
[n ]
0
[n ]
1
[n ]
2
[n ]
3
[n ]4
0
1000
2000
3000
4000
5000
6000
7000
Total Population
30
0
-30
-60
15
10
5
0
40
30
20
10
0
40
35
30
25
50
40
30
20
10
30
20
10
0
110
100
5000
10000
Time(ms)
15000
20000
90
0
5000
10000
Time(ms)
15000
20000
FIG. 9: Gating dynamics in presence of sodium
only blocker: The action potential and the
corresponding population of every sodium(left traces)
and potassium(right traces) channel conformational
states are plotted for sodium channel only blocker with
kson= 0.1 at A= 200.0 µm2 and Iext= 6.9 µA/cm2.
Next we have plotted every population(%) of the
sodium and potassium channel states in presence of
potassium only blockers in left and right traces of fig-
ure (10) respectively for kpon = 0.1.
It is very aston-
ishing to see that the trends of population dynamics
has been drastically changed from that of the sodium
only blockers in figure (9). They show different occu-
pancy dynamics. It clearly says that different types of
Blocking mechanism
Sodium channel only blocking
Spiking Frequency
Exponential decrease
Potassium channel only blocking
Initially increase and then decrese
APD
Shortening occurs
Broadening occurs
Equal affinity :kson = kpon
Maxima and minima arises
Broadening occurs(potassium blocking dominates)
Preferential sodium blocking :kson > kpon
Exponential decrease
APD shortens initially, then broadening occurs
Preferential potassium blocking :kson < kpon Initially increase and then decrese
Broadening occurs
TABLE II: Various blocking mechanism and their effect on spiking frequency and action potential duration(APD).
12
blocking agents can considerably change the gating dy-
namics of the system. This inference is consistent with
the earlier literature[13, 47]. It is a very important re-
alization which could be obtained only from a Markov
model oriented studies like this. Another interesting fea-
ture is that the population of the drug bound state, DP
very rapidly grows towards 100% for kon=0.1, unlike the
sodium only blocking. The sodium channel population
is mostly trapped in m3h0 state, which is a closed state.
The altered gating dynamics and hampered membrane
potential originating from disrupted ion conduction in
presence of blockers make them trapped in this state, as
already discussed in Section III. This is a cross verifica-
tion of our previous statement.
Potassium Channel Only Blocker, kp = 0.1, A= 200, I = 6.9
ext
40
0
-40
-80
90
60
30
0
60
40
20
0
40
30
20
10
0
100
75
50
25
0
40
20
0
15
10
5
0
15
10
5
0
20
10
0
110
100
90
Population(%) of Na+ Channel Conformational States
V
m
(mV)
[ m h ]
0
0
[ m h ]
0
1
[ m h ]
0
2
[ m h ]
0
3
[ m h ]
1
0
[ m h ]
1
1
[ m h ]
1
2
[ m h ]
1
3
Total Population
0
50
100
150
200
250
300
Time(ms)
on
Population(%) of K+ Channel Conformational States
40
0
-40
-80
15
10
5
0
30
20
10
0
40
20
0
45
30
15
0
40
20
0
100
80
60
40
20
0
V
m
(mV)
[n ]
0
[n ]
1
[n ]
2
[n ]
3
[n ]4
[ D ]
P
Total Population
0
50
100
150
200
250
300
Time(ms)
FIG. 10: Gating dynamics in presence of
potassium only blocker: The action potential and
the corresponding population(%) of every sodium(left
traces) and potassium(right traces) channel
conformational states are plotted for potassium only
blocker with kon= 0.1 at A= 200.0 µm2 and Iext= 6.9
µA/cm2.
V. OUR KINETIC DRUG BLOCKING MODEL
VERSUS OTHER STUDIES IN THE
LITERATURE
Hodgkin-Huxley squid axon model is a very basic rep-
resentation of a neuronal cell. Since 1952 many Markov
models with more coupled states or more complexities
have come up that mimic the biological responses more
accurately. But due to the diverse nature of living cell
types there exist different types of models for differ-
ent systems.
In this paper our goal is to understand
kinetically how the ion channel blockers affect a neu-
ronal cell, in general. Besides it is simply not possible
to study all kinds of model and compare them. As we
are looking at two criteria mainly: the variety of firing
rate dynamics that can be reproduced and the shape of
action potentials affected by ion channel blockers, the
Hodgkin-Huxley(HH) model is a well accepted model[48]
in this regard compared to threshold models such as
the Leaky Integrate and Fire(LIF)[49] or the Izhike-
vich model(IZH) [50], Morris-Lecar model[51], FitzHuge-
Nugamo(FHN)[52] because the parameters of the HH
model have biophysical significance. Moreover one can
modify the original model parameters to easily include
the cell response of different systems[53]. The stochastic
drug binding approach on HH model is a very good choice
since it is capable of bringing out the essential physiolog-
ical features of ion channel blocking phenomena which
corresponds to many experimental observations till date
tested on different types of systems. In the following we
present the references of such experiments done.
1. Broadening of APD: From the study of potas-
sium only blockers and the local anesthetic block-
ing that have more blocking potency to potassium
channel than sodium blockers(case of kpon > kson),
we have seen that these blockers tend to broaden
or delay the action potential duration. There
exist lots of experimental evidences that support
this finding. TEA, 4-AP(4-Aminopyridine), CTX
(charybdotoxin) etc mainly known for potassium
channel blocking potency show broadening of ac-
tion potential in demyelinated rat sciatic nerve[46],
rat hippocampal pyramidal cell[26], rat superior
cervical sympathetic neurons[27], on supraoptic
neurons[28], on Hippocampal CA3 Neurons[29] and
mammalian central neurons[30]. The effect of Den-
drotoxin, a potassium blocker from mamba snakes,
on cerebellar basket cells, has also shown a delayed
action potential duration[31].
2. Multiple spike discharge: It is also verified that
4-AP can lead to multiple spike discharge, spon-
taneous impulse activity and alteration of refrac-
tory periods[46]. These results are totally con-
sistent with our frequency response curves in fig-
ure (6) and action potential plots in presence of
potassium blockers. From the figure 4(b) and
(c) we can also see that the spike train shows
sudden spontaneous and enhanced spiking density
temporally(spike train seen to be very dense) as
also seen in ERG-K+ channel blockade[32] and
ABP(Ankyrin-binding peptide) blockade on Neu-
ral KCNQ(Kv7) channels[33]. The effect of chan-
nel blockers on effective ionic current loop is also
consistent with literature[29].
3. Shortening of APD: On the other hand the
shortening of action potential duration in presence
of sodium channel blockers is more pronounced
for cardiac cells, as far as our literature survey
is concerned. Shortening of the action potential
by ion channel blockers has been observed in sev-
eral systems like sheep cardiac purkinje fibers[34],
in rabbit cardiac Purkinje fibers[35], in dog ven-
tricular cardiomyocytes[36], in guinea pig ventric-
ular myocytes[37] etc. Antiarrhythmic agents such
as Lidocaine, Phenytoin, Mexiletine, Tocainide etc
sodium channel blockers shortens the action poten-
tial duration[38]. Antiarrhythmic agents such as
bretylium, amiodarone, ibutilide, sotalol etc pre-
dominantly block the potassium channels, thereby
prolonging repolarization[38]. The mechanism of
cardiac cells are quiet different due to the presence
of other important types of ion channels and also
the shape and length of action potential is quite
different than neuronal cell . We do not compare
our simple Hodgkin-Huxley action potential result
to the cardiac action potentials. Hodgkin-Huxley
model shows good agreement with the neuronal
13
cells. We just mention that cardiac cells also show
similar effects as the neuronal cells show in presence
of channel blockers.
4. Altered gating dynamics: The state transi-
tions that underlay the gating process(opening,
closing or inactivation etc) are altered by local
anesthetics[13] and that such altered gating itself
becomes the essential modulator of local anesthetic
block[47]. From the results discussed for gating dy-
namics in figure (9) and (10), we have also shown
that the sodium and potassium blockers have very
distinct population dynamics. Thus both type of
drugs drastically change the gating dynamics.
A. Comparison Between Langevin and Markov
Model Simulation
The transient properties such as spiking interval, co-
efficient of variation of spiking due to channel block-
ers has been studied using Langevin description of
stochasticity[24] in HH model. Here we have provided a
brief comparative study between the original determinis-
tic description of HH model using gating variables(m,h,n
without white noise terms added), Langevin descrip-
tion(equation 3) used in reference [24] and our Gillespie
simulation of HH-Markov chain model used in this pa-
per. Ignoring the channel noise fluctuation we want to
see how close the Langevin[24] and Gillespie simulation
we adapted matches each other with or without the pres-
ence of blockers. In Langevin description the dynamics
of the gating variables are considered to be stochastic as
follows,
z = αz(V )(1 − z) − βz(V )z + ηz(t),
(3)
where z = n, m, h. n represents the potassium and m and
h together represents sodium gating dynamics as consid-
ered by Hodghkin-Huxley originally. ηz(t)(s) are inde-
pendent Gaussian white noise which makes the sodium
and potassium gating dynamics stochastic in nature. The
strength of the individual noises are given as follows[24]:
< ηz(t)ηz(t′) >=
2
Ni
αz(V )βz(V )
[αz(V ) + βz(V )]
δ(t − t′),
(4)
where i = NN a for z=m and h and i = NK for z=n. The
blocking of sodium and potassium channel conductance
are considered as fractional conductance[24] as follows,
GK(V, t) = gmax
K xKn4
and GN a(V, t) = gmax
N a xN am3h,
(5)
where the factors xK and xN a are the fractions of working
or unblocked ion channels to the overall number of potas-
sium and sodium channels, respectively. 0 < xK/N a ≤ 1.
Solving the conductance equation (5) and equation (1)
one obtains the action potential using this Langevin type
of description.
Similarity: From figure 11(a) it is seen that with-
out the presence of drug and stimulus both Langevin
and Markov description matches the deterministic-HH
description when channel number fluctuation is ignored
with high patch size which also validates that our
simulation-code results to be correct. Next in figure (b)
we have shown that the Gillespie and Langevin model
even in presence of stimulus, matches each other when
the channel noise is ignored. However as the external
current is applied the determiistic match requires higher
patch size, such as A= 20,000 µm2. Next, in figure (c) we
have compared the models in presence of high drug bind-
ing affinity(for example, we showed the potassium block-
ing case here only). In high patch size where the channel
noise is ignored it is expected that the Langevin approach
should produce similar result to that of the model we con-
sidered having high drug binding affinity. Now as seen
from figure(c), we find that for obvious reason the origi-
nal deterministic-HH description matches with Langevin
description exactly showing that the action potentials, at
this very high blockade region(xK=0.1), quickly damps
down and as the refactorization is hampered no further
spikes could regenerate. Now this result is compared to
the very high affinity region of our model with kpon=0.5.
We can see that the Langevin simulation shows simi-
lar trend of action potential termination as our Markov
model does. Although they do not match each other ex-
actly, not expected to match each other either(blocking
schemes are very different) but their nature is mostly
same. Thus our model reduces to the Langevin model
to an extent. The difference in the termination poten-
tial arises due to their inherent model differences. The
termination potential in Langevin scheme is around -30
mV which is also unsuitable for further action potential
generation.
Difference: The difference between the Langevin
scheme and our Markov scheme of drug binding is that
the Langevin scheme discards a certain fraction of the
total number of sodium or potassium channel and their
contribution to the total conduction from the patch at
the very beginning. The action potentials are then cal-
culated on the basis of remaining number of available
Comparison Between Deterministic-HH, Langevin & Gillespie-Markov Simulation
a
b
c
14
Deterministic
Langevin, A = 20,000
Gillespie, A = 20,000
Blocker
Potassium Only
Kx = 0.1
kp = 0.5
on
50
0
50
100
150
200
250
300
Time(ms)
0
20
40
Time(ms)
Iext
60
= 6.9
80
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
100
80
60
40
20
0
-20
-40
-60
-80
Deterministic
Langevin, A = 2,000
Gillespie, A = 2,000
I ext = 0.0
No Drug
0
10
20
30
Time(ms)
40
100
80
60
40
20
0
-20
-40
-60
-80
Deterministic
Langevin, A = 20,000
Gillespie, A = 20,000
No Drug
Iext= 6.9
100
80
60
40
20
0
-20
-40
-60
-80
FIG. 11: Comparison between original
Deterministic-HH, Stochastic Langevin and
Markov-Gillespie drug binding schemes: channel
noise ignored: In figure (a) the action potential
generated from Deterministic-HH, Langevin and
Markovian simulations has been compared with out the
presence of drug and external current and channel
noise. In figure (b) similar plot has been done in
presence of an external current. In figure (c) we have
compared three results in presence of very high
potassium channel blocking affinity.
channels, thereby assuming that during the time of ob-
servation, the number of blocked channels do not change,
or as if the drug binding kinetics is in a standby mode.
Actually in real situation during the course of obser-
vation the available number of open channels or num-
ber of blocked channels change due to the drug binding.
With higher affinity the channels are blocked in a faster
rate and vice verse. This is the main reason why the
Langevin approach can not show the gradual change in
shape or APD with time. As the Langevin approach does
not have any specific states of the channel(closed, open
or inactivated state etc) a lot of information regarding
the population dynamics, altered gating dynamics is lost
which makes our Markov model approach a better alter-
native for studying kinetic drug blocking model. Most
importantly, the intricate details of various drug binding
schemes such as closed or inactivated state blocking etc
can also be adapted and implemented using simple con-
sideration of suitable drug-bound states which are not
possible in the Langevin approach. Also the Gillespie
simulation of Markov model gives the opportunity to cal-
culate the population of individual states which opens the
door of carrying out various nonequilibrium thermody-
namical analysis[54, 55] such as entropy production rates,
free energy change etc of different binding schemes also.
In short our Markov model based study has some ad-
vantages over Langevin approach in the context of state
specific ion channel blockade.
VI. CONCLUSION
The definite presentation of specific channel states in
the Markov model helps to describe not only the macro-
scopic current contributing for the action potential, but
also the probability and transitions of each channel state,
which gives a mechanistic and detailed link between the
whole-cell action potential and the structure or function
of ion channels. Thus Markovian simulation is an essen-
tial part of modern day model based elcetrophysiological
and pharmaceutical investigations. Here we have tried
to understand how one can theoretically bring out the
essential features of ion channel blocking. By the suit-
able adaption of Gillespie algorithm along with the direct
numerical simulation of voltage dynamics we have been
able to show how a simple extension of Hodgkin-Huxley
model can prove to be an important tool to understand
the effect of ion channel blockers to predict the impor-
tant drug binding features like altered spiking frequency,
altered duration of action potential of a neuronal cell and
altered gating dynamics etc. This study establishes a link
between the theoretical understanding of drug binding
kinetics and the observed experimental findings[26 -- 38].
Our study gives an opportunity to further investigate dif-
ferent types of drug binding features that can originate
from other types of drugs such as closed state blockers
or inactive-state blockers. The major conclusions of this
study in kinetic drug binding scenario is as follows.
1. In the case of only sodium channel blocking al-
though the spiking activity decreases with blocking
affinity, the train of action potentials generated are
of complete shape because there are plenty of ac-
tive potassium channels available to bring the sys-
tem back to its resting state. However, in case of
potassium blocking at higher value of binding affin-
ity, due to the lack of active potassium channels
the shape of action potential at the end can not
complete. Consequently the re-factorization pro-
cess gets hampered which eventually affects the
sodium channel activation process to regenerate
more spiking activity.
In case of sodium channel
blockers spiking activity slowly decreases mainly
due to absence of adequate number of available
open sodium channels, where as in case of potas-
sium channel blockers the repolarization leading to
re-factorization process plays a vital role in destroy-
ing spike generation process.
15
2. We have also shown two distinct types of spik-
ing trends in presence of two types of blockers.
For sodium blocker we observe that the spiking
frequency falls off exponentially with the increas-
ing binding affinity.
In contrast, with increasing
potassium blocking affinity, initially the spiking fre-
quency increases towards a maximum and then it
gradually falls off exponentially. This initial in-
crease in spiking activity arises due to the domi-
nance of increased internal potassium channel num-
ber fluctuations or channel noise over total ionic
conductances. In case of local anesthetics or dual
blockers where the drugs have equal binding affin-
ity to both the channels, show multiple maxima in
spiking frequency trend. A drug with more binding
affinity to a particular channel shows similar spik-
ing frequency trends as in the case of only sodium
or potassium blockers. However there exists a crit-
ical value of affinity for potassium channels above
which irrespective of mechanisms all spiking activ-
ity is destroyed due to incomplete re-factorization
process.
3. Both the sodium and potassium blocking agents
considerably changes the gating and population dy-
namics of the system which we could show using our
description.
4. Our approach satisfactorily agrees as well as pro-
vides suitable explanation regarding the experi-
mentally observed change in action potential du-
ration in presence of channel blockers. Sodium
channel blockers shorten the action potential dura-
tion and the potassium blockers delays it. However
in case of dual type blockers we have shown that
potassium blockers play influential role by prolong-
ing the repolarization process ultimately.
Thus using our basic mechanisms of ion channel block-
ing one can develop a systematic understanding of the
physiological effects of channel blockers for simple neu-
ronal cell with sufficient details which opens the possi-
bility of exploring many other important drug binding
features with the incorporation of desired level of struc-
tural and functional details.
Appendix
A1: Gillespie Algorithm for Site Selective Ion
Channel Blocking
To consider site selective binding of sodium or potas-
sium blockers in a single neuron the Gillespie's algorithm
we used to study the drug binding kinetics has been
implemented using the following steps.
1. Initially we have fixed the number of sodium and
potassium channels to be simulated. NN a = ρN aA and
NK = ρKA are the numbers of sodium and potassium
ion channels in a particular patch of area A µm2 with
sodium channel density, ρN a= 60 µm−2 and potassium
channel density, ρK= 18 µm−2[23] .
2. Then at a resting membrane potential, i.e. at -
70 mV the steady state values of the gating variables
n,m and h are solved using the following steady state
equations[17],
n =
αn
αn + βn
, m =
αm
αm + βm
and h =
αh
αh + βh
.
(6)
3. Next at t=0 we have assigned the population of the
drug-bound state, DN a/K = 0 for respective sodium and
potassium blocker simulations. All the other 13 states
are binomially distributed[40],
N kj
N a =(cid:18)3
j(cid:19) hk(1 − h)(1−k)mj(1 − m)(3−j)NN a,
N j
K =(cid:18)4
j(cid:19) nj(1 − n)(4−j)NK,
(7)
N a and N j
where N kj
K denotes the population of the states
where k number of h gates, j number of m gates for
sodium channel and j number of n gates for potassium
channels are open. Here N kj
K are integers which
fluctuate around their expected values.
N a and N j
4. Then we put an initial voltage of -60 mV and start
the simulation for a paticular value of Iext. First we
calculate the individual propensities, aK(nj → nj ′ ) for
potassium ion channel and aN a(mjhk → mj ′ hk′ ) for
sodium channels of all the 30 reactions (including the
drug-bound state). Here (mj hk → mh′ kj ′ ) or (nj → nj ′ )
indicates the transition between neighboring states. The
propensities aN a(mj hk → mj ′ hk′ ) or aK(nj → nj ′ ) for
individual transitions are expressed as the transition rate
multiplied by the population of the state from which the
16
transition is taking place. As for example, the transi-
tion from the state m1h0 to m2h0 has the propensity,
aN a(m1h0 → m2h0) = 2αmNm1h0 . Similarly for potas-
sium channel the propensity for the transition between
n3 to n4 is given by, aK(n3 → n4) = 2αnNn3. Only the
backward transitions from DN a or DK, the propensities
has been given a constant value as mentioned earlier.
This way we calculate 30 propensities corresponding to
30 reactions.
5. Next we calculate the total propensity, aT by sum-
ming over all the propensities as calculated in the previ-
ous step. The sum can be expressed simply for sodium
channel blockers as,
aT =" 22
Xν=1
aν
N a# +" 8
Xν=1
aν
K# .
and for potassium channel blockers,
aT =" 20
Xν=1
aν
N a# +" 10
Xν=1
aν
K# .
6. Next we calculate the random time required for the
next transition to occur by calculating τ , given as
τ =
1
aT
ln(1/r1),
where r1 is a pseudo-random number called from an uni-
form distribution[0,1]. As soon as the τ is obtained, the
time is incremented by t = t + τ .
7. In the next step we calculate which one of the 30
reactions has taken place in that τ time. An integer µ
is assigned which designates the transition number. So
µ varies from 1 to 30 in our case including the drug-
binding step. Then another random number, r2 from an
uniform distribution[0,1] of unit interval is called. Then
we calculate the quantity, (r2 × aT ). Then the transition
number µ is calculated using the following relation
aν
N a/K < (r2 × aT ) ≤
µ−1
Xν=1
aν
N a/K.
µ
Xν=1
This is actually adding the successive propensities, such
as aN a(m0h0 → m1h0) + aN a(m1h0 → m0h0) +
aN a(m1h0 → m2h0) + aN a(m2h0 → m1h0) + ..... +
aK(n0 → n1) + aK(n1 → n0) + aK(n1 → n2) + aK(n2 →
n1)+.... under the µ do-loop (µ = 1 to 30) until their sum
is equal or just exceed (r2 × aT ), and the loop number
index or the transition index, µ is then set equal to the
loop index or transition number index of the last aν term
added. This is how the reaction taken place is identified.
8. As soon as the reaction number index, µ is iden-
tified, the population of the states associated with that
reaction is updated by ±1. The population of the state
from which the transition occurred is updated with -1
and the population of the state where to the transition
has occurred is updated with +1.
9. Then the sodium and potassium conductances
are calculated using equation 2 and the corresponding
sodium,IN a and potassium , IK ionic currents are given
by the equation
IK = GK(V, t)(V −EK)
and
IN a = GN a(t)(V −EN a).
(8)
10. Then membrane potential is simply integrated
with the calculated time step τ as follows,
1
Cm
where Iint = IK + IN a + IL.
Vi = Vi−1 +
[(Iext − Iint)τ ] ,
11. Then again the process is repeated from step 4,
with the new value of membrane potential and updated
transition rates and propensities at the obtained value of
membrane potential, V.
12. To calculate the spike frequency of action poten-
tials we run the program for a very long time and then
convert the spike number in Hz unit.
Using the above mentioned algorithm we have written
a Fortran 90 code and complied in GFORTRAN to
study the physiological effect of drug binding on action
potentials.
A2: Time Dependent Propensity vs. Constant
Propensity
Here we want to mention that the method that we
have adapted for the calculation of the propensities
is a very popular and widely used method and of-
ten claimed to be an exact method for simulating ac-
tion potenial using Gillespie algorithm. However con-
trary to this popular belief this method is not an ex-
act method.
In systems where the rate constants or
the transition rates are time dependent through volt-
age change[kz(V (t)), k = α or β, z = m, n, h] or change
with time due to temperature change or volume change,
propensity functions does not remain constant between
reactions, i.e. aν
N a/K(X(t), t), where X(t) =
{[m0h0](t), .., [m3h1](t), [n0](t), .., [n4](t)} is the popula-
tion state vector of different states[56]. When the propen-
sity functions depend only on the state of the system i.e.
N a/K(t) = aν
17
aν
N a/K(t) = aν
N a/K(X(t)) the Gillespie algorithm calcu-
lates the time until the next transition takes place by
considering the first transition time of Rν (total number
of transitions) time homogeneous transitions. How-
ever, as the transition rates become time dependent,
the first firing time has to be calculated from Rν time-
inhomogeneous transitions. The amount of time that
must pass until the next transition takes place, τ , is given
by the distribution function[56]:
1 − exp(cid:18) −
Rν
Xν=1Z t+τ
t
aν
N a/K(X(t), s)ds(cid:19),
(9)
where X(t) is constant in the above integrals as no reac-
tions take place within the time interval [t, t + τ ). Using
the above equation, τ is obtained by first letting r1 be
uniform(0,1) and then solving the following equation:
Rν
Xν=1Z t+τ
t
aν
N a/K(X(t), s)ds = ln(1/r1).
(10)
Then the transition that occurs at that time is chosen
according to the probabilities[56] aν
N a/K(X(t), t + τ )/aT ,
N a/K(X(t), t + τ ). So it is seen
that this time dependent case is very different from
the time homogeneous case which are most frequently
used in literature where τ is exponentially distributed as
where aT = PRν
ν=1 aν
ν=1 aν
P (τ ) = aT (X(t) exp(cid:2) − aT (X(t))τ(cid:3), where aT (X(t)) =
PRν
N a/K(X(t)) and the probability that the next re-
action is ν-th reaction is given as aν
N a/K(X(t))/aT (X(t)).
However, solving equation(10) by both analytically and
numerically is extremely hard and time consuming. Ow-
ing to the difficulty of solving equation (10) one can by-
pass it by using next reaction method or modified next
reaction method[56]. However those methods are also
restricted to very few number of channels. The rigorous
Gillespie algorithm remains a challenge for the stochastic
Hodgkin-Huxley model. For very low patch size where
very few number of channels are considered the Gille-
spie algorithm with constant propensity assumption is
very questionable, indeed it is wrong. However, for many
channels with very large total propensity of a large popu-
lation of channels the typical τ becomes so small that one
can consider propensities to be approximately constant.
The larger is the number of ion channels the better the
approximation works. The reason of this brief discussion
is to just inform the readers that the widely claimed ex-
act one is basically an approximate method, not exact
method. It diverges from the true action potential tra-
jectories as time propagates(as seen using Morris-Lecar
model in reference [57]) for sufficiently small number of
channels. Although it is not at all bad for a patch size
of A= 200µm2. We have used the constant propensity
method. The results with this much of patch area is negli-
gibly affected and moreover our aim is to understand how
the physiological effects of drug blocking can be obtained
in general. Theoretical development on exact simulation
is not our focus here. For better understanding about
the discussion between the piece wise constant propen-
sity and time dependent propensity readers are advised
to read the following three references: [56 -- 58].
18
Acknowledgment
K. Pal wants to thank the anonymous referees for their
valuable suggestions and The Department of Biotechnol-
ogy, Govt. of India for fellowship.
[1] F. M. Ashcroft, From molecule to malady, Nature 440,
440 (2006).
[2] B. Hille,
Ionic channels of excitable membranes (2nd
edition), (Sinauer Associates, Sunderland, Mass., 1992).
[3] J. J. Clare, Functional expression of ion channels in
mammalian systems. J. J. Clare and D.T. Trezise (Eds.)
Expression and Analysis of Recombinant Ion Channels,
(Wiley-VCH, Weinheim, Germany, p 79, 2006).
[4] S. Gilman, Neurobiology of disease, (Academic Press, p
319, 2007).
[5] G. F. Ballester, A. F. Carvajal, J. M. Gonzalez-Ros and
A. F. Montiel, Ionic channels as targets for drug design,
A review on computational methods, Pharmaceutics 3,
932 (2011).
[6] W. A. Catterall, Sodium channels, inherited epilepsy,
and antiepileptic drugs, Annu. Rev. Pharmacol. Toxicol.
54, 317 (2014).
[7] T. Narahashi,
J. W. Moore and W. R. Scott,
Tetrodotoxin blockage of sodium conductance increase
in Lobster giant axons, J. Gen. Physiol. 47, 965 (1964).
[8] G. M. Lipkind and H. A. Fozzard, Molecular modeling
of local anesthetic drug binding by voltage-gated sodium
channels, Mol. Pharmacol. 68, 1611 (2005).
[9] D. B. Tikhonov and B. S. Zhorov, Mechanism of sodium
channel block by local anesthetics, antiarrhythmics, and
anticonvulsants, J. Gen. Physiol. 149, 465 (2017).
[10] S. I. Judge and C. T. Bever, Potassium channel blockers
in multiple sclerosis: neuronal Kv channels and effects of
symptomatic treatment, Pharmacology and Therapeu-
tics 111, 224 (2006).
[11] T. J. Campbell, K. R. Wyse and R. Pallandi, Differen-
tial effects on action potential of class IA, B and C an-
tiarrhythmic drugs: modulation by stimulation rate and
extracellular KC concentration, Clinc. Exp. Pharmacol.
Physiol. 18, 533 (1991).
[12] A. Scholz, Mechanisms of (local) anaesthetics on voltage-
gated sodium and other ion channels. Br. J. Anaesth.
89, 52 (2002); H. A. Fozzard and G. M. Lipkind, The
tetrodotoxin binding site is within the outer vestibule of
the sodium channel, Marine Drugs 8, 219 (2010); K.
Pal and G. Gangopadhyay, Probing kinetic drug binding
mechanism in voltage-gated sodium ion channel, open
state versus inactive state blockers, Channels 9(5), 307
(2015).
[13] G. K. Wang and G. R. Strichartz, State-dependent inhi-
bition of sodium channels by local anesthetics: A 40-year
evolution, Biochem. (Mosc) Suppl. Ser. A Membr. Cell
Biol. 6, 120 (2012).
[14] K. N. Piasta, D. L. Theobald , C. Miller, Potassium-
selective block of barium permeation through single KcsA
channels, J. Gen. Physiol. 138, 421 (2011);
D. J. Posson, J. G. McCoy and C. M. Nimigean, The
voltage-dependent gate in MthK potassium channels is
located at the selectivity filter, Nature Structural &
Molecular Biology 20, 159 (2013);
R. Guo , W. Zeng, H. Cui, L. Chen and S. Ye,
Ionic
interactions of Ba2+ blockades in the MthK K+ channel,
J. Gen. Physiol. 144, 193 (2014).
[15] Y. Rudy and J. R. Silva, Computational biology in the
study of cardiac ion channels and cell electrophysiology.
Q. Rev. Biophys. 39, 57 (2006);
C. E. Clancy , Z. I. Zhu and Y. Rudy, Pharmacogenetics
and anti arrhythmic drug therapy, a theoretical investi-
gation, Am. J. Physiol. Heart. Circ. Physiol. 292, H66
(2007);
M. Fink and D. Noble, Markov models for ion channels:
versatility versus identifiability and speed, Phil. Trans.
R. Soc. A 367, 2161 (2009);
Y. Y. Vladimir, T. W. Allen, C. E. Clancy, Computa-
tional models for predictive cardiac ion channel pharma-
cology, Drug Discovery Today: Disease Models 14, 3
(2014).
[16] A. Nekouzadeh and Y. Rudy, Statistical properties of
ion channel records. Part I. Relationship to the macro-
scopic current, J. Math. Biosci.210, 291 (2007); A. Nek-
ouzadeh and Y. Rudy, Statistical properties of ion chan-
nel records. Part II. Estimation from the macroscopic
current, J. Math. Biosci.210, 315 (2007).
[17] A. L. Hodgkin and A. F. Huxley, A quantitative descrip-
tion of membrane current and its application to conduc-
tion and excitation in nerve, J. Physiol. 117, 500 (1952).
[18] L. J. DeFelice and W. N. Goolsby, Order from random-
ness, spontaneous firing from stochastic properties of ion
channels: Fluctuations and Order, The New Synthesis,
M. Millonas, editor,
(Springer, New York. p 331-342,
1996).
[19] J. Clay and L. DeFelice, Relationship between membrane
excitability and single channel open-close kinetics, Bio-
phys. J. 42, 151-157 (1983).
[20] A. Strassberg and L. J. DeFelice, Limitations of the
Hodgkin-Huxley formalism, effects of single channel ki-
netics on transmembrane voltage dynamics, Neural Com-
putation 5, 843 (1993).
[21] J. T. Rubinstein, Threshold fluctuations in an N sodium
channel model of the node of Ranvier, Biophys. J. 68,
779 (1995).
[22] P. Jung and J. W. Shuai, Optimal sizes of ion channel
clusters, Europhysics Letters 56, 29 (2001).
[23] G. Schmid, I. Goychuk and P. Hanggi, Stochastic reso-
nance as a collective property of ion channel assemblies,
Europhys. Lett. 56, 22 (2001);
P. Hanggi, G. Schmid and I. Goychuk, Excitable mem-
branes: Channel noise, synchronization, and stochastic
resonance, Advances in Solid State Physics 42, 359
(2002);
G. Schmid, I. Goychuk and P. Hanggi, Channel noise
and synchronization in excitable membranes, Physica A
325, 165 (2003).
[24] G. Schmid, I. Goychuk and P. Hanggi, Effect of channel
block on the spiking activity of excitable membranes in
a stochastic Hodgkin-Huxley model, Phys. Biol. 1, 61
(2004).
[25] D. T. Gillespie, Exact stochastic simulation of coupled
chemical reactions, J. Phys. Chem. 81, 2340 (1977).
[26] J. F. Storm, Action potential repolarization and a
fast after-hyperpolarization in rat hippocampal pyrami-
dal cells, J. Physiol. 385, 733 (1987).
[27] S. J. Marsh and D. A. Brown, Potassium currents con-
tributing to action potential repolarization in dissociated
cultured rat superior cervical sympathetic neurons, Neu-
roscience Letters 133, 298 (1991).
[28] M. D. Hlubek and P. Cobbett, Differential effects of K+
channel blockers on frequency-dependent action poten-
tial broadening in supraoptic neurons, Brain Research
Bulletin 53, 203 (2000).
[29] J. Mitterdorfer and B. P. Bean, Potassium currents dur-
ing the action potential of Hippocampal CA3 neurons, J.
Neurosci. 22, 10106 (2002).
[30] B. P. Bean, The action potential in mammalian central
neurons, Nature Reviews, Neuroscience 8, 451 (2007).
19
[32] F. Gullo et al., ERG K + channel blockade enhances
firing and epinephrine secretion in rat chromaffin cells:
the missing link to LQT2-related sudden death?, The
FASEB Journal 17, 330 (2003).
[33] D. A. Brown and G. M. Passmore, Neural KCNQ (Kv7)
channels, British Journal of Pharmacology 156, 1185
(2009).
[34] E. Carmeliet and T. Saikawa, Shortening of the action
potential and reduction of pacemaker activity by Lido-
caine, Quinidine, and Procainamide in Sheep Cardiac
Purkinje Fibers, An effect on Na or K Currents? Circ
Res. 50, 257 (1982).
[35] T. J. Colatsky, Mechanisms of action of lidocaine and
quinidine on action potential duration in rabbit cardiac
Purkinje fibers. An effect on steady state sodium cur-
rents? Circ. Res. 50, 17 (1982).
[36] A. Szabo et al., Effects of Ropivacaine on action potential
configuration and ion currents in isolated canine ventric-
ular cardiomyocytes, Anesthesiology 108, 693 (2008).
[37] Y. Song and L. Belardinelli, Basal late sodium current
is a significant contributor to the duration of action po-
tential of guinea pig ventricular myocytes, Physiol. Rep.
5(10), e13295 (2017).
[38] A. J. Trevor and B. G. Katzung et al., Katzung and
Trevor's Pharmacology Examination and Board Review,
(Chapter 14, 11e: McGraw-Hill Education (2015) and
12e: McGraw-Hill Medical(2012), New York).
M. Perez-Neut, V. Rao, L. Haar , K. W. Jones and S.
Gentile, Current and potential antiarrhythmic drugs tar-
geting voltage-gated cardiac ion channels, Cardiol. Phar-
macol. 4, 139 (2015).
[39] R. F. Fox, Stochastic versions of the Hodgkin-Huxley
equations, Biophys. J. 72, 2069 (1997);
R. F. Fox and Y. Lu, Emergent collective behavior in
large numbers of globally coupled independently stochas-
tic ion channels, Phys. Rev. E 49, 3421 (1994).
[40] E. Skaugen and L. Walløe, Firing behaviour in a stochas-
tic nerve membrane model based upon the Hodgkin-
Huxley equations, L. Acta Physiol. Scand. 107, 343
(1979).
[41] A. Manwani, P. N. Steinmetz and C. Koch, Channel
Noise in Excitable Neuronal Membranes, Advances in
neural information processing systems 12, S. A. Solla, T.
K. Leen and K. R. Muller (eds.), (MIT Press, pp. 143-
149, 2000).
[42] C. Koch, Biophysics of computation, (Oxford University
Press, New York, pp. 202-210, 1999).
[43] C. C. Chow and J. A. White, Spontaneous action poten-
tials due to channel fluctuations, Biophys. J. 71, 3013
(1996).
[31] R. Begum, Y. Bakiri, K. E. Volynski and D. M. Kull-
mann, Action potential broadening in a presynaptic
channelopathy, Nat. Commun. 6(7), 12102 (2016).
[44] H. I. Mino, J. T. Rubinstein and J. A. White, Compari-
son of algorithms for the simulation of action potentials
with stochastic sodium channels, Annals of Biomedical
20
response of a voltage gated sodium ion channel and bio-
physical characterization of dynamic hysteresis, J. of
Theoretical Biology 415, 113 (2017).
[55] Hao Ge and Hong Qian, Physical origins of entropy pro-
duction, free energy dissipation, and their mathematical
representations, Phys. Rev. E 81, 051133 (2010).
[56] D. F. Anderson, A modified next reaction method
for simulating chemical systems with time dependent
propensities and delays, J. of Chemical Physics 127,
214107 (2007).
[57] D. F. Anderson, B. Ermentrout and P. J. Thomas,
Stochastic representations of ion channel kinetics and ex-
act stochastic simulation of neuronal dynamics, J. Com-
put. Neurosci. 38, 67 (2015).
[58] D. F. Anderson and T. G. Kurtz, Design and analy-
sis of biomolecular circuits, chapter 1. Continuous Time
Markov chain models for chemical reaction networks,
(Springer, 2011).
Engineering 30, 578 (2002).
[45] D. E Becker and K. L. Reed, Local anesthetics: Review
of pharmacological considerations, Anesth. Prog. 59, 90
(2012);
O. Afolabi, A. Murphy, B. Chung and D. Lalonde, The
effect of buffering on pain and duration of local anesthetic
in the face: A double-blind, randomized controlled trial,
Can. J. Plast. Surg. 21, 209 (2013);
G. B. Silva and et.al, Duration and efficacy of differ-
ent local anesthetics on the palmar digital nerve block
in horses, Journal of Equine Veterinary Science 35, 749
(2015);
J. A. Patacsil, M. S. McAuliffe, L. S. Feyh and L. L.
Sigmon, Local anesthetic adjuvants providing the longest
duration of analgesia for single- injection peripheral nerve
blocks in orthopedic surgery: A literature review, AANA
Journal 84, 95 (2016).
[46] E. F. Targ and J. D. Kocsis, Action potential character-
istics of demyelinated rat sciatic nerve following applica-
tion of 4-aminopyridine, Brain Research 363, 1 (1986).
[47] G. R. Strichartz, The inhibition of sodium currents in
myelinated nerve by quaternary derivatives of lidocaine,
J. Gen. Physiol. 62, 37 (1973).
[48] B. Maisel and K. Lindenberg, Channel noise effects on
first spike latency of a stochastic Hodgkin-Huxley neuron,
Phys. Rev. E 95, 022414 (2017);
S. Jamasb, Extension of the neuronal membrane model
to account for suppression of the action potential by a
constant magnetic field, Biophysics J. 62, 428 (2017);
N. Stankevich and E. Mosekilde, Coexistence between
silent and bursting states in a biophysical Hodgkin-
Huxley-type of model, Chaos 27, 123101 (2017);
R. Baravalle, O. A. Rosso and F. Montani, A path inte-
gral approach to the Hodgkin-Huxley model, Physica A
486, 986 (2017).
[49] G. Indiveri, Synaptic plasticity and spike-based com-
putation in VLSI networks of integrate-and-fire neurons,
Neural Inf. Process. Lett. Rev. 11, 135 (2007).
[50] E. M. Izhikevich, Simple model of spiking neurons, IEEE
Trans. Neural. Netw. 14, 1569 (2003).
[51] C. Morris and H. Lecar, Voltage oscillations in the bar-
nacle giant muscle fiber, Biophys. J. 35, 193 (1981).
[52] R. FitzHugh, Impulses and physiological states in the-
oretical models of nerve membrane, Biophys J. 1, 445
(1961).
[53] A. R. Willms, D. J. Baro, R. M. Harris-Warrick and
J. Guckenheimer, An improved parameter estimation
method for Hodgkin-Huxley models, J. Comput. Neu-
rosci. 6(2), 145 (1999);
D. Csercsik et al., Hodgkin-Huxley type modeling and
parameter estimation of GnRH neurons, IEEE BioSys-
tems 100, 198 (2010).
[54] K. Pal, B. Das and G. Gangopadhyay, Nonequilibrium
Supplemental
Material
1. Effective ionic current: The effective ionic cur-
rent is given as, Iint = IN a + IK + IL.
1
3. Dual Blockers, case: kson > kpon The action po-
tential for the case kson > kpon shows very noisy action
potential trains. The peak height seems to fall gradually.
Thus the spikes which have peaks above -10 mV has been
considered in the paper works.
100
0
-100
-200
-300
P
A
g
n
i
r
u
D
t
n
e
r
r
u
C
c
i
n
o
I
e
v
i
t
c
e
f
f
E
Sodium Only Blocker
No Drug
0.05
0.5
-400
-80
-60
-40
20
Action Potential(mV)
-20
0
100
0
-100
-200
-300
P
A
g
n
i
r
u
D
t
n
e
r
r
u
C
c
i
n
o
I
e
v
i
t
c
e
f
f
E
40
-400
-80
-60
Potassium Only Blocker
-40
20
Action Potential(mV)
-20
0
40
FIG. S1: Effective ionic current is plotted here in
presence of sodium and potassium only blockers for
kon= 0.05 and 0.5 has been plotted with the case when
no drug is present. Here A= 200.0 µm2 and Iext= 6.9
µA/cm2. The left panel is for sodium only blockers and
the right panel is for potassium only blockers.
)
s
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
a
b
c
d
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
0
0
0
0
Effect of Dual Blocking, Case ks > kp
on
ks = 0.001, kp = 0.0001
on
on
on
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
ks = 0.01, kp = 0.001
on
on
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
ks = 0.1, kp = 0.01
on
on
1000
2000
3000
4000
5000
6000
ks = 0.5, kp = 0.05
on
on
200
400
600
800
1000
1200
1400
1600
1800
2000
Time(ms)
FIG. S3: The trend of action potential termination the
case, kson > kpon is plotted here with A= 200.0 µm2
and Iext= 6.9 µA/cm2.
2. Dual Blockers, case: kson = kpon. The trend of
action potential termination for the case, kson = kpon is
shown below.
4. Dual Blockers, case: kpon > kson is similar to
that of potassium only blockers.
Effect of Dual Blocking, Case: ks = kp
on
on
a
ks = kp = 0.001
on
on
0
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 11000 12000 13000 14000 15000 16000 17000
b
0
200
400
600
800
1000
1200
1400
1600
1800
ks = kp = 0.01
on
on
2000
2200
2400
0.01
ks = kp = 0.1
on
on
60
30
0
-30
-60
60
30
0
-30
-60
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
c
d
60
30
0
-30
-60
0
60
30
0
0
-30
-60
0
0
50
100
150
200
250
300
350
400
ks = kp = 0.5
on
on
20
40
60
80
100
Time(ms)
120
140
160
180
200
FIG. S2: The action potential for the case kson = kpon
is plotted for affinity, kson = kpon= 0.001 to 0.5 with
A= 200.0 µm2 and Iext= 6.9 µA/cm2.
)
V
m
(
l
a
i
t
n
e
t
o
P
n
o
i
t
c
A
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
60
30
0
-30
-60
a
0
b
c
d
0
0
0
Effect of Dual Blocking, Case kp > ks
on
on
kp = 0.001, ks = 0.0001
on
on
3000
6000
9000
12000
15000
kp = 0.01, ks = 0.001
on
on
500
1000
1500
2000
kp = 0.1, ks = 0.01
on
on
2500
50
100
150
200
250
300
350
400
kp = 0.5, ks = 0.05
on
on
20
40
60
80
100
120
140
160
180
200
Time (ms)
FIG. S4: The action potential for the case kpon > kson
is plotted here for various affinities with A= 200.0 µm2
and Iext= 6.9 µA/cm2.
|
1702.02939 | 1 | 1702 | 2017-02-09T18:54:14 | cellGPU: massively parallel simulations of dynamic vertex models | [
"physics.bio-ph",
"cond-mat.soft",
"physics.comp-ph"
] | Vertex models represent confluent tissue by polygonal or polyhedral tilings of space, with the individual cell interacting via force laws that depend on both the geometry of the cells and the topology of the tessellation. This dependence on the connectivity of the cellular network introduces several complications to performing molecular-dynamics-like simulations of vertex models, and in particular makes parallelizing the simulations difficult. cellGPU addresses this difficulty and lays the foundation for massively parallelized, GPU-based simulations of these models. This article discusses its implementation for a pair of two-dimensional models, and compares the typical performance that can be expected between running cellGPU entirely on the CPU versus its performance when running on a range of commercial and server-grade graphics cards. By implementing the calculation of topological changes and forces on cells in a highly parallelizable fashion, cellGPU enables researchers to simulate time- and length-scales previously inaccessible via existing single-threaded CPU implementations. | physics.bio-ph | physics | cellGPU: massively parallel simulations of dynamic vertex models
Daniel M. Sussman1, ∗
1Department of Physics, Syracuse University, Syracuse, New York 13244, USA
Vertex models represent confluent tissue by polygonal or polyhedral tilings of space, with the indi-
vidual cell interacting via force laws that depend on both the geometry of the cells and the topology
of the tessellation. This dependence on the connectivity of the cellular network introduces several
complications to performing molecular-dynamics-like simulations of vertex models, and in particular
makes parallelizing the simulations difficult. cellGPU addresses this difficulty and lays the foun-
dation for massively parallelized, GPU-based simulations of these models. This article discusses
its implementation for a pair of two-dimensional models, and compares the typical performance
that can be expected between running cellGPU entirely on the CPU versus its performance when
running on a range of commercial and server-grade graphics cards. By implementing the calcula-
tion of topological changes and forces on cells in a highly parallelizable fashion, cellGPU enables
researchers to simulate time- and length-scales previously inaccessible via existing single-threaded
CPU implementations.
I.
INTRODUCTION
Simple models that represent confluent tissue by polyg-
onal or polyhedral tilings of space have a rich history of
biophysical application [1–4]. In these "vertex models"
the degrees of freedom have traditionally been the ver-
tices of the polygons in the tessellation, although some
recent models take the degrees of freedom to be the posi-
tions of the cells themselves (with the cell geometry fol-
lowing from, e.g., a Delaunay triangulation of those posi-
tions) [5, 6]. Although restricting cell shapes to Voronoi
volumes does limit the range of cell configurations rela-
tive to more permissive vertex models, this restriction in
the spirit of early work proposing Voronoi tessellations
as a description of the arrangement of cells in epithelial
tissue [7].
Regardless of the choice for the underlying degrees of
freedom, the models are united by similar equations of
motion and force laws that depend on both local geom-
etry and the topology of the cellular network.
In this
sense vertex models share many similarities with mod-
els of soaps and foams [8], but in the context of living
tissues frequently add features like active motion, cell
division, and cell death. Vertex models have greatly con-
tributed to the understanding of collective cell growth
and migration, which in turn inform processes as diverse
as cell sorting [9], collective cell motion in epithelial tissue
[10, 11], wound healing [12], and embryonic development
[13–15].
Although the equations of motion used to simulate
these models are often quite similar to those used in stan-
dard molecular dynamics, a number of computational
challenges have prevented the adoption of modern, heav-
ily parallelized implementations common to other simu-
lations. In particular, as noted above, rather than inter-
acting via (typically) short-ranged forces vertex models
have interactions that depend on both local geometry
∗ [email protected]
and the topology of the interacting units. For example,
models that represent cells as Voronoi volume elements
[5, 6] clearly require that a either a Voronoi or Delaunay
triangulation be maintained at all times. From that tri-
angulation the cell shapes and cell neighbors are defined,
which in turn are the direct inputs to determining the
forces which enter the equations of motion governing the
model.
Two important consequences of this stand out. First,
although the computational cost of both (1) perform-
ing a Delaunay triangulation on a point set of size N
and (2) advancing a standard molecular dynamics algo-
rithm by one time step given N particles interacting via
short ranged forces scale like O (N log N ), the prefactor
for the Delaunay triangulation is typically much worse
than the prefactor for simulating, e.g., Lennard-Jones
particles with a reasonable finite-range cut off [16, 17].
Second, and more seriously in light of modern trends in
scientific computing, the triangulation problem is much
less amenable to parallelized computation. Indeed, a re-
cent review has explicitly highlighted the lack of paral-
lelization in these cell models as the main bottleneck to
further research [18], and even the most sophisticated
open-source implementations of vertex models currently
rely on serial, single-threaded CPU-based implementa-
tions [6, 19]. Because of this, vertex models struggle to
tackle both long-timescale problems and simulations of
large numbers of interacting cells. The latter is particu-
larly problematic for two-dimensional vertex models, in
which finite-size effects may be especially strong.
Written in C++/CUDA [20], cellGPU lays the founda-
tion for moving to highly parallelized, GPU-based simu-
lations of these models. Two classes of vertex models
are described in this article: a fully GPU-accelerated
"active vertex model" (AVM) in which vertices move
according to both forces and the intrinsic activity of
nearby cells, and a hybrid GPU/CPU implementation of
a Delaunay-triangulation-based model of active cells (the
"self-propelled Voronoi" (SPV) model [5]). The remain-
der of the paper is organized as follows. Sec. II begins
with a more in-depth overview of the two vertex models
7
1
0
2
b
e
F
9
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
9
3
9
2
0
.
2
0
7
1
:
v
i
X
r
a
2
values of the area and perimeter for cell i, and KA and
KP are area and perimeter moduli. Many variations of
the above functional have been written; to be concrete we
discuss the implementation of the simple one above. Here
the quadratic dependence on cell area can be interpreted
as a cell monolayer's resistance to height fluctuations due
to adhesions between cells and cell incompressibility, and
the quadratic dependence on cell perimeter as a compe-
tition between active contractility of the actomyosin sub-
cellular cortex and tension due to both cell-cell adhesions
and cortical tension [5]. From the energy, the force on a
degree of freedom (either a vertex or a cell position), j,
is defined by
(cid:126)Fj = ∇jE.
(2)
Note that energy functional in Eq. 1 leads to forces which
are fundamentally many-body (not pairwise) in nature.
In addition to these intercellular forces, active ver-
sions of these models include additional cell motility
terms. For instance, in the spirit of simple models of
self-propelled particles [21], the SPV model assigns a po-
larization vector to every cell, (cid:126)ni = (cos θi, sin θi), and
posits that cells exert a self-propulsion force of constant
magnitude v0/mu, where µ is the mobility (with units
of an inverse frictional drag) and v0 is a self-propulsion
speed [5]. The polarization vector of every cell rotates
randomly according to ∂tθi = ηi(t), where ηi is a white-
noise process with zero mean and variance 2Dr; the ro-
tational diffusion constant Dr in the model determines
a time scale of persistent motion that would arise in the
absence of other interactions. Taken together, then, the
SPV equation of motion for cell i is [5]
d(cid:126)ri
dt
= µ (cid:126)Fi + v0(cid:126)ni.
(3)
We define the active vertex model by implementing the
self-propulsion at the level of cells and taking the equa-
tion of motion for the vertices to be
d(cid:126)hi
dt
= µ (cid:126)Fi +
v0
n
(cid:126)nj,
(4)
(cid:88)
(cid:104)ij(cid:105)
where (cid:104)ij(cid:105) represents the n cells indexed by j that are
neighbors of vertex i. That is, each cell has a constant
self-propulsion as in the SPV, and the activity of each
vertex is a simple average of the activity of the adjoining
cells.
B. Computational overview
We first give an overview of the general computational
process that is common to both models described above,
breaking down the simulation time step into a sequence
of logically distinct calculations. Since the forces in the
models depend on the geometry of the cells at the be-
ginning of the time step, the area and perimeter of every
FIG. 1. (Left) A self-propelled Voronoi model, in which each
cell is a Voronoi area, shown in blue solid lines. Interactions
depend on both the cell shapes and the connectivity of the
dual Delaunay triangulation of the cell positions, which is
shown in dotted red lines. (Right) A T1 transition, in which
the edge connecting vertices (cid:126)v1 and (cid:126)v2 is rotated and an ex-
change of cell neighbors occurs, with cells i and k adjacent
before the transition and cells j and l adjacent after the tran-
sition. The active vertex model has the vertices as the degrees
of freedom and cellular topology evolves via these (and other)
transitions.
mentioned above, and then describes the details of the
parallelized implementation. Section III provides per-
formance benchmarks, comparing the GPU-accelerated
algorithms with existing CPU-based implementations.
Section IV closes with a discussion of future directions
for the code, and has additional information on its avail-
ability.
II. DESIGN AND IMPLEMENTATION
A. Vertex model overview
In two dimensions vertex models describe a confluent
patch of tissue as a polygonal tiling of space, each poly-
gon corresponding to a coarse-grained representation of
a cell. The degrees of freedom are either the collection
of vertex positions, here denoted as (cid:126)hi, or the positions
of the cells themselves, denoted as (cid:126)ri. In the SPV model
each cell is represented not by a generic polygonal region,
but by a Voronoi area element, and so the topology of the
network refers to an underlying Delaunay triangulation of
the cell positions. In contrast, in the AVM the vertices
themselves are the degrees of freedom, and changes to
the topology occur via cell neighbor exchange (T1 tran-
sitions), cell removal or addition (T2 transitions), vertex-
edge merging (T3 transitions) [8]. Schematic illustrations
of these models are shown in Fig. 1.
The forces acting on the degrees of freedom in these
simplified cell models are derived by first defining a sim-
ple energy functional for the entire tissue, for example
E =
KA(Ai − A0,i)2 + KP (Pi − P0,i)2.
(1)
i=1
Here N is the total number of cells, Ai and Pi are the
area and perimeter of cell i, A0,i and P0,i are "preferred"
N(cid:88)
nnkkssccqqlljjmmiibbaarrppaaijkl𝑣⃗#𝑣⃗$𝑎⃗𝑏𝑐⃗𝑑⃗ijkl𝑣⃗#𝑣⃗$𝑎⃗𝑏𝑐⃗𝑑⃗ cell must first be computed. Following that, the forces on
the degrees of freedom must be evaluated from Eq. 2, and
the appendix shows in detail how to calculate these forces
in a computationally efficient manner. Since computing
the many-body forces from Eq. 2 is expensive relative
to, e.g., evaluating pairwise Lennard-Jones interactions,
even parallelizing these geometry and force calculations
can lead to substantial performance gains.
With the forces in hand the degrees of freedom can
be displaced according to either Eq. 3 or 4. The next
challenge is determining whether the cellular topology
needs to be updated;
implementing vertex models in
a highly parallelizable fashion has required abandoning
some of the data structures commonly used to represent
the topology of cellular networks (or, more generally, to
represent collections of vertices, edges, and cells). For ex-
ample, the doubly connected edge list (DCEL) is a par-
ticularly convenient data structure when working on the
CPU [22]. In that representation every edge is composed
of two "half-edges," each of which has pointers to the face
that it bounds, the vertex that serves as its origin, and
the other half-edge making up the edge (the half-edge's
"twin"). This sort of structure makes it easy to update
a triangulation or traverse it in some desired order, but
at the cost of having very non-local memory accesses.
Even on the CPU this can lead to significant performance
costs, where if not carefully implemented the resulting
cache-access patterns can change the effective observed
algorithmic complexity of vertex model simulations. On
the GPU these non-local memory requests can be even
more of an imposition, and the performance of accessing
and operating on data is radically enhanced by working
with much flatter data structures in which the majority
of memory requests are local.
Thus, cellGPU takes the approach of initializing and
then maintaining many redundant data structures rep-
resenting the topology – such as a one-dimensional list
keeping track of vertices around each cell in counter-
clockwise order, a separate list keeping track of the ver-
tices each vertex is connected to, and yet a third list
keeping track of the cells that border each vertex. The
code has been profiled to confirm that the cost of main-
taining these redundant data structures is much less than
the performance gains that come with their use. These
performance considerations will be revisited in Sec. III.
Given its relatively simple rules for when topological
transitions can occur, the AVM branch of cellGPU can
manage the step of checking and updating the topology
entirely on the GPU, avoiding the need for costly mem-
ory transfers between the host (CPU) and device (GPU).
The SPV branch of cellGPU instead adopts a hybrid
approach of testing the topology fully on the GPU but
performing any necessary repairs on the CPU. Eventu-
ally, though, even this step will be moved to the GPU,
and algorithmic choices have been made (discussed be-
low) to facilitate this eventual move. Note that the
current implementation of all CPU routines in cellGPU
is strictly single-threaded, although here too algorith-
3
mic choices have been made so that multithreaded re-
implementations would be straightforward.
As a final detail, note that in these simulations there
is a strong incentive to try to order all of the data arrays
so that particles which are physically close to each other
are also close to each other in memory. This is a conse-
quence of the slow memory access on the GPU, and the
improvements to memory access speeds that can come if
all read/write operations are coalesced [20]. That is, a
thread that has to read data from many different parts
of a given array – e.g., reading off the vertex positions
for cells close to a given cell from an un-ordered array of
vertex data in order to test the emptiness of a circum-
circle – would be much slower than a thread which could
read the same information from a contiguous block of the
array.
As is common in modern particle-based simulations
[17, 23], cellGPU uses a Hilbert-curve-based spatial sort-
ing scheme. This takes an approximation to the Hilbert
curve (a particular space-filling fractal curve chosen for
its optimal data-locality properties [24]), finds the loca-
tion on the curve closest to each cell, and sorts cells based
on the order in which they occur along the approxima-
tion to the Hilbert curve. As demonstrated in Sec. III for
the AVM, this represents a (small) optimization even for
the CPU branch of the code, but radically improves per-
formance on the GPU. Thus, cellGPU provides the op-
tion to periodically resort the degrees of freedom, which
should be done at a frequency related to how much the
cells move and rearrange neighbors.
C. Self-propelled Voronoi model implementation
As noted above, cellGPU adopts a hybrid GPU/CPU
approach to simulating the two-dimensional SPV model
defined by Eq. 3. The hybrid approach is currently re-
quired to maintain the Delaunay triangulation, but fu-
ture development is planned to eventually implement the
model simulation entirely on the GPU. A highly paral-
lelizable approach to testing the validity of the triangu-
lation has already been implemented, and algorithms for
maintaining the triangulation have been chosen with an
eye towards their eventual implementation on the GPU.
The following sub-sections highlight the sequence of GPU
and CPU function calls that form the basic loop of ad-
vancing the simulation by one time step.
1. Computing the geometry of cells
At the beginning of a simulation time step, the geom-
etry of every cell (i.e. the current area and perimeter of
each cell) is calculated on the GPU as a prerequisite to
computing the forces on the cells. The parallelism here
is at the level of independently computing the geometry
of each cell, and a different GPU thread can be assigned
to each cell index. In the course of computing the geom-
etry, the kernel computes the Voronoi vertices that form
the boundary of each cell by evaluating the circumcenter
of every triangle in the Delaunay triangulation (we re-
iterate that in the AVM the vertices themselves are the
degrees of freedom – they do not need to form a Voronoi
configuration, and computing the geometry of the cells
must be done very differently). These arrays of Voronoi
positions are saved for later use (after determining that
the cost of saving and loading these Voronoi vertices is
less than the cost of recomputing them on the fly).
2. Calculating the forces on cells
With the geometry computed and the Voronoi vertex
positions saved, the force on every cell is computed. Note
that additional parallelism can be exposed here: as shown
in the appendix, the net force on a cell can be decom-
posed into the contribution to that force from each of the
cell's Voronoi vertices. Thus, the force calculation step is
broken into two GPU kernel calls. First "force sets" are
computed, so that a different thread computes the force
on cell i due to vertex v for each i and v independently.
Second, a different kernel takes this data, together with
neighbor lists containing information on which vertices
are associated with which cell, and assigns one thread
per cell to add the individual force set contributions to
get the net force on each cell.
3. Displacing particles and testing the topology
A straightforward kernel takes the net force on each
cell, updates the cell positions according to both these net
forces and the self-propulsion of each cell, and rotates the
cell directors via a GPU-based pseudo-random number
generator.
Next, the triangulation must be maintained. Depend-
ing on the parameter values of the SPV model that are
chosen, it is entirely possible that the network topology
– the connection between cells and vertices – remains un-
changed. One approach to solving this problem is based
on constructing the digital Voronoi diagram to test for
changes in the topology of the triangulation [25]. Here,
instead, we directly test the empty-circumcircle property
that defines the Delaunay triangulation. Each cell in the
SPV model corresponds to a Delaunay vertex, and by def-
inition a valid Delaunay triangle has the property that a
circumcircle defined by the vertices of that triangle con-
tains no other vertices. Thus, a kernel is launched in
which each thread checks the circumcircle defined by one
of the triangles in the Delaunay triangulation to see if
any cell has moved in such a way that an update of the
triangulation is needed.
4
4. Repair Delaunay triangulation as needed
The kernel that tests the empty-circumcircle property
copies a flag back to the host CPU, indicating whether
any of the circumcircle tests failed.
If no tests failed
(i.e. the topology of the Delaunay triangulation did not
change as the cells moved during that time step) then
the next time step can be immediately started on the
GPU. If a circumcircle came back non-empty, though, a
list of cells that need an update to their neighbor list is
returned to the CPU. For each of these flagged cells, the
triangulation is locally repaired. Typically this would be
done via an edge-flipping or star-splaying algorithm [26],
and such algorithms form the basis for existing hybrid
CPU/GPU triangulation routines for computing large-
scale Delaunay meshes [27].
As an alternative, and with the aim of eventually im-
plementing the triangulation repair step on the GPU and
eliminating the need for a hybrid algorithm, a different
approach is followed. We first note a locality lemma that
can be used to limit the possible set of Delaunay neigh-
bors of a given cell: given a set of Delaunay vertices
that form a polygon bounding the cell i, the set of De-
launay neighbors of i is a strict subset of the cells that
are contained in the circumcircles defined by i and any
two consecutive vertices of on the bounding polygon [28].
These points form the candidate "Delaunay 1-ring" of
cell i. This set is easily found on the CPU, and the data
is handed off to an efficient CGAL library to reduce the
candidate 1-ring to the actual set of neighbors of the cell.
One advantage of this algorithm is that even on the CPU
it can be readily parallelized to find the candidate 1-ring
for all cells that need updating simultaneously. When
all flagged cells have been repaired, the auxiliary data
structures describing the topology of the triangulation
are updated, and all data is sent back to the GPU.
D. Active vertex model implementation
Implementing the AVM described by Eq. 4 is, in some
ways, more straightforward than the hybrid implementa-
tion of the SPV model. Many of the components of the
simulation time step are carried out exactly as described
above, beginning with the calculation of the geometry of
every cell and moving on to the net force on each ver-
tex. The kernel for displacing vertices is modestly more
complicated, since each vertex needs to know about the
director defining the self-propulsion of the vertex' neigh-
boring cells. This is optimized by having separate kernel
calls to first move the vertices and then update the cell
directors.
The advantage of simulating the AVM model is that
maintaining the topology is much simpler. As mentioned
above, rather than referencing an underlying Delaunay
triangulation, topological changes in the AVM are man-
aged by T1, T2, and T3 transitions. At present cellGPU
only implements T1 transitions, but all three transition
types are readily compatible with being performed only
via local access to the auxiliary data structures used to
define the topology. As such, the topological changes can
be carried out entirely on the GPU, removing the need
for hybrid CPU/GPU operation. The primary difficulty
is the need to avoid race conditions that could be caused
by simultaneously executing multiple topological transi-
tions involving the same vertex or cell. To deal with this,
cellGPU adopts a strategy of only allowing a single topol-
ogy update per cell per kernel call. All of the topological
changes called for in a given time step are guaranteed
by calling the topology-updating kernel repeatedly, i.e.,
until no more changes are called for.
III. PERFORMANCE BENCHMARKS
Benchmarking of cellGPU was carried out on two sys-
tems: (1) a workstation running Ubuntu 14.04 with a 3.5
GHz Xeon E5-1620 V3 processor with 32 GB of RAM
with the option of using either a Tesla K40, Quadro K620,
or GeForce 980Ti graphics card, or (2) a laptop running
Ubuntu 16.04 with a 2.2 GHz Core i5 5200U processor
with 8 GB RAM and a GeForce 950M graphics card.
Figure 2 shows the performance of cellGPU when simu-
lating the SPV model in a favorable regime, i.e., in the
solid-like regime where cells move slowly and the Delau-
nay triangulation only needs to be updated occasionally.
The performance is shown for a range of consumer- and
server-grade graphics cards; as can be seen, the perfor-
mance can be as much as three orders of magnitude faster
than existing CPU-based implementations for large sys-
tem sizes. The super-linear scaling seen in the DCEL-
based implementation results not from true a difference
in algorithmic complexity but from the effect of repeated
cache misses when accessing data scattered throughout
memory.
When topological changes happen more frequently, the
hybrid implementation results in less dramatically accel-
erated performance. As the model is simulated in the
increasingly fluid-like regime cells move much more and
the topological updating scheme needs to be invoked on
the CPU more and more frequently. In addition to the
cost of locally repairing the triangulation, this also causes
expensive data transfers between the host and the device.
This overall degradation of performance can be seen for
a particular choice of parameters in Fig. 3.
In contrast, since the AVM branch of the cellGPU code
is fully GPU accelerated it has almost no dependence on
the parameter regimes in which the simulations are per-
formed. Figure 4 shows the performance of the simple
active vertex model implemented by cellGPU , which can
again be more than an order of magnitude faster than its
own CPU-based implementation, and multiple orders of
magnitude faster than non-C++ implementations still in
common use [3]. The figure also shows the importance
of using the spatial sorting schemes described above to
maintain data locality. While Hilbert-curve sorting has
5
FIG. 2. Computational performance of cellGPU for SPV
model parameters in the solid-like regime (P0 = 3.8, v0 =
0.01) with a simulation time step of ∆t = 0.05. From top to
bottom, the gray curve corresponds to a standard CPU imple-
mentation of the SPV, based on doubly-connected edge lists
(DCEL), and the black curve corresponds to cellGPU running
only on a single CPU thread. The remaining curves compare
the algorithm's performance when running on different graph-
ics cards: in order the green curve corresponds to using on a
Quadro K620, the purple to a GeForce 950M, the blue solid
curve to a GeForce 980Ti, and the red to a Tesla K40. All
of these curves represent calculations in double precision; the
dashed blue curve shows the performance on a GeForce 980Ti
when floating-point precision is used instead.
FIG. 3. Performance of cellGPU for the SPV model at fixed
v0 and but varying P0 from the solid to fluid regimes. The top
two curves are again the CPU-based implementations, which
on this scale have a computational speed that is independent
of simulation parameters. From bottom to top the solid blue
curve corresponds to P0 = 3.8, the dashed blue curve to P0 =
3.9, and the dot-dashed blue curve to P0 = 4.0, all running
on a Tesla K40 graphics card.
only a modest impact on the CPU branch of the code, ne-
glecting it on the GPU branch causes a change in the ef-
fective observed computational complexity of the model,
from roughly linear in the number of vertices simulated
per time step to a distinctly super-linear scaling.
IV. CONCLUSIONS
We have demonstrated the computational gains that
come from implementing molecular dynamics simula-
tions of off-lattice cell models in a highly parallelized
manner, allowing access to simulation time- and length
scales unavailable to existing single-threaded imple-
mentations. Natural extensions include incorporating
100100010410510-40.0010.0100.100110numberofcellscostpertimestep(s)2DSPV(existingstandard)cellGPU:CPU-onlycellGPU:QuadroK620GPUcellGPU:GeForce950MGPUcellGPU:GeForce980TiGPUcellGPU:TeslaK40GPUcellGPU:980TiGPU(float)100100010410510-40.0010.0100.100110numberofcellscostpertimestep(s)2DSPVcellGPU(CPU-only)p0=3.8(TeslaK40)p0=3.9p0=4.0 6
FIG. 4. Performance of cellGPU AVM implementation. The
linear black curves correspond to the strictly CPU imple-
mentation, and the blue curves correspond to the GPU-
accelerated implementation running on a Tesla K40. For each
set of curves the dashed curve shows the performance when
no Hilbert sorting scheme is used.
FIG. 5. Schematic diagram of cell i and some of its neighbors,
along with the associated Voronoi vertices (the circumcenters
of three adjacent cells) labeled for convenient reference.
1. AVM forces
more general sets of boundary conditions and support-
ing additional classes of off-lattice cell models, includ-
ing extensions to three-dimensional models. Addition-
ally, work to move the self-propelled Voronoi branch
away from its current hybrid implementation and to-
wards a fully GPU-accelerated algorithm is currently
planned. Code availability and more details about
the planned future directions of the code and devel-
opments that are currently underway can be found at
Ref.
[29], with additional documentation maintained at
http://dmsussman.gitlab.io/cellGPUdocumentation.
ACKNOWLEDGMENTS
I would like to thank Lisa Manning, Matthias Merkel,
Michael Czajkowski, and David Yllanes for fruitful dis-
cussions and comments on this manuscript. This work
was supported by NSF-POLS-1607416; the Tesla K40
used for this research was donated by the NVIDIA Cor-
poration.
Appendix A: Efficiently computing vertex model
forces
If the topology of the cellular network is known – ei-
ther from an underlying triangulation of space or via a
direct enumeration of vertex-vertex connections – com-
puting the forces on the degrees of freedom of a vertex
model is straightforward. For reference, Fig.
5 pro-
vides a schematic picture of a relevant patch of a two-
dimensional tissue model. To be explicit, each vertex is
labeled by the three cells it is adjacent to.
Recall that the energy functional has the form
N(cid:88)
i
E =
(cid:0)KA(Ai − A0)2 + KP (Pi − P0)2(cid:1) ,
(cid:33)
(cid:32)
∂Ei
∂(cid:126)hijk
+
∂Ej
∂(cid:126)hijk
+
∂Ek
∂(cid:126)hijk
− ∂E
∂(cid:126)hijk
= −
and that the force on vertex (cid:126)hijk is
(A1)
,
(A2)
i.e., the motion of vertex (cid:126)hijk only changes the shape of
cells i, j, and k. The energy derivatives are straightfor-
ward, e.g.,
∂Ei
∂(cid:126)hijk
= 2Ka(Ai−A0)
∂Ai
∂(cid:126)hijk
+2KP (Pi−P0)
∂Pi
∂(cid:126)hijk
, (A3)
and where the area and perimeter derivatives with re-
spect to Voronoi vertex positions can themselves be writ-
ten as follows. Let (cid:126)tij = (cid:126)hijk − (cid:126)hgij, (cid:126)tik = (cid:126)hikl − (cid:126)hijk,
with tij and tik being the unit vectors in those directions.
Similarly, let lij be the length of the edge between cell
i and j, and nij be the unit vector pointing outwardly
normal to that cell edge. Then one can write
∂Ai
∂(cid:126)hijk
∂Pi
∂(cid:126)hijk
(lij nij + lik nik)
1
2
=
= −(cid:0)tij + tik
(cid:1)
(A4)
(A5)
2. SPV forces
Following Bi et al. [5], the force on cell i in cartesian
direction λ can be computed as
Fiλ = − ∂E
∂riλ
= − ∂Ei
∂riλ
− (cid:88)
<ij>
∂Ej
∂riλ
,
(A6)
100100010410510610-510-40.0010.0100.100110numberofverticescostpertimestep(s)CPU-onlyCPU,nosortingGPU(TeslaK40)GPU,nosortingigjklm !hijk !hikl !hgij !hjkm where (cid:104)ij(cid:105) refers to all cell neighbors j of cell i, referring
to the configuration in Fig. 5. The terms in the above
can be expanded via the chain rule, for instance:
(cid:19)
(cid:88)
(cid:18) ∂Ek
∂hijk,ν
ν
∂Ek
∂riλ
=
∂hijk,ν
∂riλ
+
∂Ek
∂hikl,ν
∂hikl,ν
∂riλ
.
(A7)
These expressions are related to how the positions of the
Voronoi vertex was written in Ref.
[5]. Further writing
the derivatives
7
Here these are the only terms needed, since the other
voronoi vertices associated with cell k (the middle of the
three neighboring cells in clockwise order) do not depend
on the position of cell i. The partial derivatives here
depend on the positions of (cid:126)hjkm and (cid:126)hkln, where m is
the cell other than i that has both j and k as neighbors,
and n is the cell other than i that has both k and l as
neighbors, so that the forces in the SPV model depend
on nearest and next-nearest Delaunay neighbors of cell i.
The derivative of the energy with respect to the voronoi
vertices was calculated above.
The derivatives of the Voronoi vertices with respect
(∂(cid:126)hijk)/(∂(cid:126)ri), can be
to the position of the cell, e.g.
calculated efficiently as follows. Let (cid:126)rij denote the vector
from i to j, and define:
c = (cid:126)rij,x(cid:126)rkj,y − (cid:126)rij,y(cid:126)rkj,x
d = 2c2
(cid:126)z = βd(cid:126)rij + γd(cid:126)rik
βd = −(cid:126)rik2 · ((cid:126)rij · (cid:126)rjk)
γd = (cid:126)rij2 · ((cid:126)rik · (cid:126)rjk) .
(A8)
(A9)
(A10)
(A11)
(A12)
∂(βd)
∂(cid:126)ri
∂(γd)
∂(cid:126)ri
1
∂d
d
∂(cid:126)ri
= 2 ((cid:126)rij · (cid:126)rjk) (cid:126)rik + (cid:126)rik2 (cid:126)rjk
= −2 ((cid:126)rik · (cid:126)rjk) (cid:126)rij + (cid:126)rij2 (cid:126)rjk
=
2
c
{−(cid:126)rjk,y, (cid:126)rjk,x} .
(A13)
(A14)
(A15)
Finally, with I2 representing the 2 × 2 identity matrix
and ⊗ the dyadic product, the desired change in Voronoi
vertex position with respect to cell position is
(cid:20)
(cid:126)rij ⊗
(cid:18) ∂(βd)
(cid:19)
∂(cid:126)ri
(cid:18) ∂(γd)
(cid:19)
∂(cid:126)ri
+ (cid:126)rik ⊗
∂(cid:126)hijk
∂(cid:126)ri
= I2 +
1
d
− (βd + γd)I2 − (cid:126)z ⊗
(cid:18) 1
d
∂d
∂(cid:126)ri
(cid:19)(cid:21)
.
(A16)
[1] G. W. Brodland, Applied Mech. Rev. 57, 47 (2004).
[2] T. Nagai and H. Honda, Philosophical Magazine Part B
81, 699 (2001).
[3] M. L. Manning, R. A. Foty, M. S. Steinberg, and E.-M.
Schoetz, Proc. Natl. Acad. Sci. USA 107, 12517 (2010).
[4] D. Bi, J. H. Lopez, J. M. Schwarz, and M. L. Manning,
Soft Matter 10, 1885 (2014).
[5] D. Bi, X. Yang, M. C. Marchetti, and M. L. Manning,
Phys. Rev. X 6, 021011 (2016).
(2006).
[11] N. Sep´ulveda, L. Petitjean, O. Cochet, E. Grassland-
Mongrain, P. Silberzan, and V. Hakim, PLoS Comput.
Biol. 9, e1002944 (2013).
[12] J. H. Kim et al., Nat. Mater. 12, 856 (2013).
[13] L. Hufnagel, A. A. Teleman, H. Rouault, S. M. Cohen,
and B. I. Shraiman, Proc. Natl. Acad. Sci. USA 104,
3835 (2007).
[14] R. Farhadifar, J.-C. Roper, B. Aigouy, S. Eaton, and
[6] D. L. Barton, S. Henkes, C. J. Weijer, and R. Sknepnek,
F. Julicher, Curr. Biol. 17, 2095 (2007).
arXiv.org (2016).
[7] H. Honda, J. Theor. Biol. 72, 523 (1978).
[8] A. G. Fletcher, M. Osterfield, R. E. Baker, and S. Y.
Shvartsman, Biophys. J. 106, 2291 (2014).
[9] J. M. Belmonte, G. L. Thomas, L. G. Brunnet, R. M. C.
de Almeida, and H. Chat´e, Phys. Rev. Lett. 100, 248702
(2008).
[15] D. B. Staple, R. Farhadifar, J. C. Roper, B. Aigouy,
and F. Julicher, Euro. Phys. J. E 33, 117
S. Eaton,
(2010).
[16] J.-D. Boissonnat, O. Devillers, M. Teillaud,
and
M. Yvinec, in Proceedings of the sixteenth annual sympo-
sium on Computational geometry (ACM, 2000) pp. 11–
18.
[10] B. Szab´o, G. J. Szollosi, B. Gonci, Z. Jur´anyi,
D. Selmeczi, and T. Vicsek, Phys. Rev. E 74, 061908
[17] S. Plimpton, J. Comput. Phys. 117, 1 (1995).
[18] P. Van Liedekerke, M. Palm, N. Jagiella, and D. Drasdo,
8
Comput. Part. Mech. 2, 401 (2015).
[19] G. R. Miriams et al., PLoS Comput. Biol. 9, e1002970
[25] M. Bernaschi, M. Lulli,
Phys. Commun. (2016).
and M. Sbragaglia, Comput.
(2013).
[20] J. Nickolls, I. Buck, M. Garland, and K. Skadron, Queue
6, 40 (2008).
[21] S. Henkes, Y. Fily, and M. C. Marchetti, Phys. Rev. E
84, 040301 (2011).
[22] D. E. Muller and F. P. Preparata, Theoretical Computer
Science 7, 217 (1978).
[23] J. A. Anderson, C. D. Lorenz, and A. Travesset, J. Com-
put. Phys. 227, 5342 (2008).
[24] B. Moon, H. V. Jagadish, C. Faloutsos, and J. H. Saltz,
IEEE Transactions on knowledge and data engineering
13, 124 (2001).
[26] R. Shewchuk, in Proceedings of the twenty-first annual
symposium on Computational geometry (ACM, 2005) pp.
237–246.
[27] T.-T. Cao, A. Nanjappa, M. Gao,
and T.-S. Tan,
in Proceedings of the 18th meeting of the ACM SIG-
GRAPH Symposium on Interactive 3D Graphics and
Games (ACM, 2014) pp. 47–54.
[28] R. Chen and C. Gotsman, in Voronoi Diagrams in Sci-
ence and Engineering (ISVD), 2012 Ninth International
Symposium on (IEEE, 2012) pp. 24–31.
[29] https://gitlab.com/dmsussman/cellGPU, (2017).
|
1211.0119 | 1 | 1211 | 2012-11-01T08:37:30 | Application of magnetically induced hyperthermia on the model protozoan Crithidia fasciculata as a potential therapy against parasitic infections | [
"physics.bio-ph",
"cond-mat.mes-hall",
"q-bio.CB",
"q-bio.SC"
] | Magnetic hyperthermia is currently an EU-approved clinical therapy against tumor cells that uses magnetic nanoparticles under a time varying magnetic field (TVMF). The same basic principle seems promising against trypanosomatids causing Chagas disease and sleeping sickness, since therapeutic drugs available display severe side effects and drug-resistant strains. However, no applications of this strategy against protozoan-induced diseases have been reported so far. In the present study, Crithidia fasciculata, a widely used model for therapeutic strategies against pathogenic trypanosomatids, was targeted with Fe_{3}O_{4} magnetic nanoparticles (MNPs) in order to remotely provoke cell death using TVMFs. The MNPs with average sizes of d approx. 30 nm were synthesized using a precipitation of FeSO_{4}4 in basic medium. The MNPs were added to Crithidia fasciculata choanomastigotes in exponential phase and incubated overnight. The amount of uploaded MNPs per cell was determined by magnetic measurements. Cell viability using the MTT colorimetric assay and flow cytometry showed that the MNPs were incorporated by the cells with no noticeable cell-toxicity effects. When a TVMF (f = 249 kHz, H = 13 kA/m) was applied to MNP-bearing cells, massive cell death was induced via a non-apoptotic mechanism. No effects were observed by applying a TVMF on control (without loaded MNPs) cells. No macroscopic rise in temperature was observed in the extracellular medium during the experiments. Scanning Electron Microscopy showed morphological changes after TVMF experiments. These data indicate (as a proof of principle) that intracellular hyperthermia is a suitable technology to induce the specific death of protozoan parasites bearing MNPs. These findings expand the possibilities for new therapeutic strategies that combat parasitic infections. | physics.bio-ph | physics | Application of magnetically induced hyperthermia on the
model protozoan Crithidia fasciculata as a potential therapy
against parasitic infections
V. Grazú1
A.M. Silber2
M. Moros1
L. Asín1
T.E. Torres1,3,4
C. Marquina3,4
M.R. Ibarra1,3
G.F. Goya1,3
1Instituto de Nanociencia de Aragón (INA),Universidad de Zaragoza, Zaragoza, Spain.
2Departamento de Parasitologia, Instituto de Ciências Biomédicas, Universidade de São Paulo, São
Paulo, Brazil.
3Departamento de Física de la Materia Condensada, Facultad de Ciencias, Universidad de Zaragoza,
Spain.
4Instituto de Ciencia de Materiales de Aragón (ICMA), CSIC, Universidad de Zaragoza, Zaragoza, Spain.
Correspondence: Gerardo F. Goya
Instituto de Nanociencia de Aragón (INA).
Universidad de Zaragoza
Ed. I+D - Calle Mariano Esquillor s/n
Campus Rio Ebro
50018-Zaragoza - Spain
Tel: (+34) 876 555 362
Fax: (+34) 976 76 2776
Email : [email protected]
Abstract:
Purpose: Magnetic hyperthermia is currently an EU-approved clinical therapy against tumor cells that
uses magnetic nanoparticles under a time varying magnetic field (TVMF). The same basic principle
seems promising against trypanosomatids causing Chagas´ disease and sleeping sickness, since
therapeutic drugs available display severe side effects and drug-resistant strains. However, no
Magnetic hyperthermia as a potential trypanocidal therapy.
applications of this strategy against protozoan-induced diseases have been reported so far. In the present
study, Crithidia fasciculata, a widely used model for therapeutic strategies against pathogenic
trypanosomatids, was targeted with Fe3O4 magnetic nanoparticles (MNPs) in order to remotely provoke
cell death using TVMFs.
Methods: Iron-oxide MNPs with average sizes of <d> ≈ 30 nm were synthesized using a precipitation of
FeSO4 in basic medium. The MNPs were added to Crithidia fasciculata choanomastigotes in exponential
phase and incubated overnight, removing the excess of MNPs by a DEAE-cellulose resin column. The
amount of uploaded MNPs per cell was determined by magnetic measurements. The MNP-bearing cells
were submitted to TVMFs using a homemade AC-field applicator (f = 249 kHz, H = 13 kA/m), and the
temperature variation during the experiments was measured. Scanning Electron Microscopy was used to
asses morphological changes after TVMF experiments. Cell viability was analyzed using the MTT
colorimetric assay and flow cytometry.
Results: The MNPs were incorporated by the cells, with no noticeable cell-toxicity effects. When a TVMF
was applied to MNP-bearing cells, massive cell death was induced via a non-apoptotic mechanism. No
effects were observed by applying a TVMF on control (without loaded MNPs) cells. No macroscopic rise
in temperature was observed in the extracellular medium during the experiments.
Conclusion: These data indicate (as a proof of principle) that intracellular hyperthermia is a suitable
technology to induce the specific death of protozoan parasites bearing MNPs. These findings expand the
possibilities for new therapeutic strategies that combat parasitic infections.
Keywords: Magnetic Hyperthermia, Magnetic Nanoparticles, trypanosomatids, Crithidia fasciculata
Introduction
Diseases caused by organisms of the Trypanosoma and Leishmania genera affect a global
population of at least 20 million people, with an estimated at risk population of approximately 450 million
people 1,2. These statistics indicate that trypanosomatid-induced diseases are a severe sanitary problem,
with the resulting disease burden affecting much of the population that resides in the tropical and
2
Magnetic hyperthermia as a potential trypanocidal therapy.
subtropical regions of the globe. 3,4 Despite the sanitary relevance of trypanosomatid-induced diseases
for human health, no satisfactory treatments exist to combat these infections. 5-7The two therapeutic
agents that are presently in use for the treatment of Chagas´ disease are Nifutrimox and Benznidazole;
however, these drugs were developed approximately 40 years ago 8. The main disadvantages of these
treatments are a high level of toxicity and a low therapeutic efficiency during the chronic phase of the
disease. The latter disadvantage is a serious problem because Chagas´ disease is mainly diagnosed
during the chronic phase; therefore, the majority of infected people miss the opportunity to be treated
using effective chemotherapy 9. In addition, several cases of drug-resistant or partially resistant strains
have been reported for both of these drugs 10. The initial stages of Trypanosoma brucei infections, when
the central nervous system is not compromised, can be treated using suramin or pentamidine 11. Again,
these drugs are not effective during the late stages of the disease (when the majority of cases are
diagnosed) because these drugs do not traverse the brain blood barrier. For these cases the first-line
treatment is melarsoprol, a drug that can cross the blood-brain barrier. Melarsoprol is a highly toxic drug
that causes a myriad of serious undesired side effects, including reactive encephalopathy, in
approximately 20% of the patients receiving treatment 12. For the treatment of diseases caused by
Leishmania spp., Pentavalent antimonials are most often used 13. Although these drugs are effective for
treating the cutaneous form of the disease, treatments for the parenteral form of the disease are limited.
In addition, these drugs are highly toxic. Amphotericin B and pentamidine are considered to be second-
line drugs because of their serious or irreversible toxic effects. However, these drugs are now being
reconsidered on the basis of new formulations or dosage regimens 13-15. The fact that the majority of the
drugs that are currently used for trypanosomatid-induced diseases were developed approximately 40
years ago reveals the limited success of the strategies to develop novel therapeutic treatments. This lack
of success highlights the necessity for new strategies and tools to address this important public health
issue.
Magnetic hyperthermia is a relatively new medical protocol 16 that uses magnetic nanoparticles to
heat areas of the body using the application of time-varying magnetic fields (TVMFs). The physical
mechanisms underlying energy absorption by MNPs are related to the existence of the magnetic
3
Magnetic hyperthermia as a potential trypanocidal therapy.
relaxation of single domains by Arrhenius-Néel processes 4,17. With the advent of nanotechnology, it
became possible to engineer efficient MNPs that have the ability to absorb large amounts of energy from
TVMFs (up to several kW per gram of material) to induce a local rise in temperature. 18 Because of their
size, these nanoparticles can be incorporated inside target cells, making possible to heat up small foci at
the single cell level.19 The application of hyperthermia using MNPs, alone or in combination with other
therapeutic strategies, was proposed more than a decade ago as a therapeutic technique to treat
cancer.16 In the present study, we used the non-pathogenic trypanosomatid Crithidia fasciculata, a well-
accepted model of other pathogenic trypanosomatid parasites 20, to evaluate the use of magnetic
hyperthermia as a potential trypanocidal treatment. The mechanisms of cell death and the application of
these principles of magnetic hyperthermia to treat parasitic diseases are discussed in later sections of the
present work.
Material and methods
Reagents and Culture media
The chemicals and fetal calf serum (FCS) used for the present study were purchased from Sigma
(Missouri -- USA). The other components of the culture medium were purchased from Difco (Lawrence-
USA). The apoptosis detection kit was purchased from Immunostep (Coimbra -- Portugal). The
diethylaminoethyl-cellulose (DEAE-cellulose, DE52) was purchased from Whatman (Dassel- Germany).
Cells
Crithidia fasciculata choanomastigotes were grown at 28ºC in a Warren culture medium [37 g/l
brain heart infusion, 100 ng/l hemin, 100 mg/l folic acid] supplemented with 10% fetal calf serum. The
cells were seeded in 75 cm3 tissue culture plates at 1 x 106 cells/ml. C. fasciculate cultures were
incubated until they reached the exponential phase (approximately 24 h of incubation). Using daily
subculturing, the cells were maintained in the exponential growth phase for use in the studies. Cell
4
Magnetic hyperthermia as a potential trypanocidal therapy.
counting was performed in a Neubauer chamber. The cells were evaluated for viability using the optical
microscopic observation of flagellar motility and by counting of the number of viable cells after incubation
with 2% trypan blue in PBS.
Magnetic Nanoparticles
The MNPs used in the present study were synthesized using a precipitation of iron (II) salt
(FeSO4) in the presence of a base (NaOH) and a mild oxidant (KNO3) under a Nitrogen atmosphere, as
previously described in the literature 21. Mixing the reactants during a 24 h period resulted in Fe3O4
particles with average sizes of <d> ≈ 30 nm and with a colloidal stability in aqueous medium at pH 7.
Cell uptake of MNPs and separation of the non-incorporated
nanoparticles.
The cells collected during exponential growth phase were centrifuged and resuspended in fresh
culture medium, adjusting concentrations to 2.5 x 108 cells/ml. The MNPs were added to a final
concentration of 0.425 mg/ml and were incubated at 28oC overnight with gentle agitation. To separate the
cells from the non-incorporated MNPs, we took advantage of the fact that MNPs (isoelectric point= 5.0)
adsorb to DEAE-cellulose resin at pH 7.0 (isoelectric point = 2.5), whereas C. fasciculata does not
interact with the resin under these conditions. Briefly, the cells were washed twice with PBS, resuspended
in 12 ml of 2% glucose in PBS (gPBS), and incubated with 6 g DEAE-cellulose ionic exchange resin that
was previously equilibrated with gPBS for 10 min with gentle agitation at room temperature. The cells
were recovered with an efficiency of more than 95%.
Determination of cell-incorporated MNPs by magnetization
measurements
The amounts of cell-associated MNPs were determined by measuring the saturation
magnetization using a Superconducting Quantum Interference Device (SQUID) (MPMS-7T, Quantum
5
Magnetic hyperthermia as a potential trypanocidal therapy.
Design). The magnetization measurements were conducted on dried MNPs or lyophilized MNP-loaded
cells and were performed as a function of the applied magnetic field up to 5 kOe (0.4 MA/m) at different
temperatures between 5 and 300 K.
Experiments of time-varying magnetic field application
Aliquots of MNPs (10 mg/ml) or MNP-bearing cells (1.7x109 cells/ml) were submitted to
alternating magnetic fields using a homemade AC-field applicator. The magnetic field applicator,
consisting of a resonant LC tank working close to the resonant frequency, was used to measure the
specific power absorption (SPA) of the samples. A magnetic field (f = 249 kHz, H = 13 kA/m) was
achieved inside a gap of four (2+2) turns of a copper tube around high-permeability polar pieces. The
SPA values were obtained from adiabatic measurements inside an insulated Dewar. The temperature
data was measured using a fiber-optic temperature probe (ReflexTM, Neoptix) that was immune to the
radiofrequency environment. Prior to each experiment, the temperature evolution was measured between
the 5 and 10 min time points with the RF-source turned off to establish a T-baseline. Next, the power was
turned on, and the temperature rise was monitored for a 30 min interval.
Scanning Electron Microscopy (SEM)
Cells in the exponential growth phase were fixed with 2.5 % glutaraldheyde in 0.1 M sodium
cacodylate 3% sucrose solution for 90 min at 4ºC. The dehydration process was conducted by incubating
the cells in increasing concentrations of methanol at 30, 50, 70, and 100%. Each of these concentrations
were used for 5 minutes in duplicate, and a final step was conducted using anhydride methanol for 10
minutes. A drop of the dehydrated cells in suspension was placed over a coverslip. Next, when the
methanol was evaporated, the coverslip was gold coated. The samples were then observed using
scanning electron microscopy (EDX Hitachi S-3400 N, Instituto Carboquimica CSIC). Secondary electron
images were also performed.
6
Magnetic hyperthermia as a potential trypanocidal therapy.
Viability analysis
MTT assay
Cell viability was analyzed using the MTT colorimetric assay. For the cytotoxicity assay, 5x106 cells
(MNPs-/TVMF-, MNPs-/TVMF+, MNPs+/TVMF- or MNPs+/TVMF+) were resuspended in 100 µl of Warren
culture medium. Next, 40 µL of MTT dye solution (5 mg/mL in PBS) was added to each aliquot. After 4 h
of incubation in eppendorf tubes at 28°C, formazan crystals were dissolved by the addition of 100 µL of
10% SDS. All of the cell debris, which has been shown to interfere with the assay, was removed by
centrifugation (10 min at 12000 r.p.m). Next, the absorbance of each supernatant was read using a
microplate reader (Biotek ELX800) at 570 nm. The spectrophotometer was calibrated to zero absorbance
using a culture medium without cells. The relative cell viability (%) compared with the control cells (the
exponential-phase cells not submitted to any treatment) was calculated by
[Absorbance]test/[Absorbance]control x 100. Each measurement was repeated at least five times to
obtain mean values with standard deviations.
Flow cytometry
The cell viability was also measured using flow cytometry with the commercial apoptosis detection kit
purchased from Immunostep. Briefly, 1x106 cells of each sample were resuspended in Annexin-binding
buffer, were stained with 5 ml of Annexin and 5 ml of propidium iodide, and were incubated for 15 minutes
at room temperature in the dark. The cell analysis was performed using the FACSAria Cytometer
(Becktom Dickinson) and FACSDiva Software.
Results and discussion
The magnetic and physicochemical properties of the MNPs used in the present study were
reported elsewhere.22 The magnetic colloids were composed of cubic Fe3O4 nanoparticles of average
size <d> = 30 ± 8 nm with saturation magnetization at room temperature of 85 emu/g, which is close to
7
Magnetic hyperthermia as a potential trypanocidal therapy.
that of the bulk magnetite magnetization.23 This synthesis route resulted in magnetic colloids with
isoelectric point of 5.0, electrostatically stabilized due to the adsorption of SO4
= groups on the particle
surface. The ability to dissipate heat under a TVMF of any type of MNPs is measured by the specific
power absorption (SPA) given in watts per gram of magnetic material, which for the present MNPs and
experimental conditions (f = 249 kHz, H = 13 kA/m) was found to be 83.6 W/g,22 comparable to other SPA
values reported in the literature for Fe3O4 particles of similar size.24-26 The selection of MNPs was made
based on the known dependence of the SPA values on the average particle size and size distribution.
Indeed, the strong dependence of the Nèel relaxation-based model on particle size yields a maximum
value of SPA for magnetite MNPs within a narrow range of diameters <d> around 15-30 nm,4 as
experimentally confirmed in many colloidal systems yields. 3,19,21The precise value for this maximum will
depend on other magnetic properties of the MNPs such as magnetic anisotropy and saturation
magnetization.
The optimal experimental conditions for the C. fasciculata to incorporate the MNPs were
determined by a series of experiments as a function of the incubation time, incubating C. fasciculata with
a fixed MNP concentration for increasing times from 15 to 240 min. After the incubation, the non-
incorporated MNPs were separated by the column method already detailed in the Materials and Methods
section. The incorporated mass of magnetic material per cell was determined by measuring the saturation
magnetization MS of 100 ml of culture medium containing ≈109 cells and comparing these values with the
MS of the pure colloid.27 From these data and the average particle size, the average number of
incorporated MNPs per cell was calculated (Table 1). The number of incorporated MNPs decreased from
a maximum of approximately 108 NPs/cell after 15 min of incubation to approximately 106 NPs/cell after 1
h of incubation. Shorter incubation times resulted in a low and highly variable cell charging, possibly due
to the fact that the cells require an induction time to activate the biological mechanisms to incorporate the
MNPs. The time course of this decrease of the cell-associated MNPs followed an exponential decay
(R2=0.9998), reaching a near-steady state after approximately 1 h incubation, which remained essentially
constant up to 12 h of incubation (Fig. 1 and Fig. S1 in the supplementary material). From these
8
Magnetic hyperthermia as a potential trypanocidal therapy.
experiments, we established the 15 min incubation as being the optimal condition for charging the cells
with a reproducible and defined number of MNPs/cell.
To evaluate the suitability of the MNPs for hyperthermia applications, it was necessary to assess
the influence of these NPs on cell viability and on the incorporation and separation conditions. It was
observed that, when compared with exponential-phase cells, the NP-treated or mock-treated cultures that
were submitted to separation conditions (incubated with DEAE-cellulose resin in gPBS) remained more
than 95% viable (data not shown). This finding illustrated that the observed cell death in the subsequent
experiments was not due to the previous exposure of the cells to toxic conditions.
Hyperthermia experiments
After the best conditions for maximum uptake of MNPs by the cells were determined (15 min
incubation, as stated above), the hyperthermia experiments were performed. The experiments were
designed in a 2x2-row-column format (Fig. 2), in which the four groups were defined as follows: a. cells
not-bearing MNPs that were not-submitted to TVMFs (MNPs-/TVMF-); b. cells bearing MNPs that were
not-submitted to the magnetic field application (MNPs+/TVMF-); c. cells not-bearing the MNPs that were
submitted to the magnetic field application (MNPs-/TVMF+); and d. cells bearing the MNPs that were
submitted to the magnetic field application (MNPs+/TVMF+). The effects on cell viability of the above
mentioned treatments were evaluated using several criteria, including the direct observation of the cell
motility using optical microscopic observation, the mitochondrial activity using a MTT assay, and the
detailed morphology using scanning electronic microscopy (SEM). The qualitative observation of the
samples that were examined using the previously described treatments revealed that the MNPs+/TVMF+
population caused 100% cell death (Fig. 2, supplementary material). The quantitative analysis of the cells
using a MTT assay revealed that the cell viability of all of the other samples was not affected compared
with the control (MNPs-/TVMF-) (Fig. 3). The analysis of all four samples using SEM revealed severe
structural damage in the cell morphology only for the MNPs+/TVMF+ population, particularly at the level of
the cell surface, indicating severe plasma membrane damage (Fig. 4).
9
Magnetic hyperthermia as a potential trypanocidal therapy.
Cell death
Because the application of the TVMFs was performed in an adiabatic device, the hyperthermia
treatment could have produced a transient macroscopic increase of the sample temperature due to
differences in the rates of heat generation and dissipation. When a positive control experiment was
conducted by applying a TVMF to a suspension that only contained MNPs (Fig. 5), the sample showed a
large increase of temperature, of about 50ºC along the 30 minutes of the experiment. Therefore, we
tested whether a similar increase in average temperature might have contributed to the amount of cell
death by monitoring the temperature of the extracellular medium during the application of the TVMF. The
results on both control and magnetically loaded cells showed only a slight macroscopic increase in
temperature (about 2-4 ºC) after 30 min of TVMF application. This result (i.e., the absence of temperature
increase in the samples composed of magnetically-loaded cells) is the expected based on the much lower
'average concentration' of MNPs in these samples, since the small amounts of uploaded MNPs are
contained within a total volume of about 0.5 ml of liquid cell medium.
These results clearly demonstrate that the heat released from the MNPs was not enough to
increase the average temperature of the cell culture in such a way that would compromise the viability of
the cells. Therefore, the origin of the cell death that was measured after the application of the TVMF
should not be related to thermal stress. This is in agreement with previous works on magnetically-loaded
human dendritic cells19,28, which demonstrated that application of TVMF for 30 min yielded up to 90-95 %
of cell death, without affecting blank (i.e., without MNPs) cells. Similar reports have been reported in
HeLa cell line 29 loaded with MNPs. Some theoretical models on metal nanoparticles have suggested this
possibility.30 Since the temperature was essentially constant during experiments, the observed cell death
suggests an intracellular, MNP-triggered mechanism different from the temperature-induced apoptosis by
hyperthermia. However, as the temperature was measured with a macroscopic sensor, the possibility of
intracellular heating up to apoptotic temperatures cannot be excluded.
To evaluate the plausibility of intracellular heating during our experiments, a simplified heat
transfer model at the single-cell scale was considered, e.g. a single cell magnetically loaded with MNPs,
surrounded by a large matrix of unloaded cells (or just culture medium free of MNPs). We estimated the
10
Magnetic hyperthermia as a potential trypanocidal therapy.
expected intracellular temperature increase in such a case based on a) the measured mass of MNPs at
the intracellular medium; b) the calculated average cell volume from SEM images (230±50 mm3), and c)
the measured SPA values of the pure magnetic colloid. For simplicity, we further approximated the
specific heat capacity of a single cell to the pure water value CP = 4.18 J/(g.K). From the average cell
volume and MNPs upload (1-10 pg/cell), we estimated a temperature increase rate of 0.02-0.87 K/s. The
above calculations were made considering that no heat was dissipated from the intracellular medium to
the cell environment, a clearly unrealistic hypothesis. Since the cell membrane has a non-negligible
thermal conductivity, the calculated heating rates are not enough to rise the intracellular temperature up
to the 41-45 ºC needed for triggering thermally induced apoptotic mechanisms.
Several prior studies suggest that apoptosis-related mechanisms are a main cause of hyperthermia-
associated cell death 31,19. Apoptosis (cellular programmed death) is a precise mechanism in which cells
follow a programmed sequence of events to induce their death with minimal disturbance to the total cell
population. 32 This phenomenon appears to be present in a wide range of organisms, from primitive
single-cell to higher multi-cellular eukaryotes. In the present work, we investigated whether cell death by
magnetic hyperthermia of C. fasciculata was attributable to apoptosis. An early event that is considered
as a marker of apoptosis is the exposition of phosphatidylserine on the external surface of the plasma
membrane. The cells that incorporated nanoparticles, and the controls, that were either submitted or not
submitted to the magnetic field, were incubated with annexin (used to detect the presence of
phosphatidylserine) and propidium iodide (used to detect damage to the plasma membrane). Next, these
four populations of cells were analyzed using flow cytometry. As illustrated in Fig. 6, the results confirm
the 100% viability for the MNPs-/TVMF-, MNPs+/TVMF-, and MNPs-/TVMF+ cell samples and the 0%
viability of the MNPs+/TVMF+ cells. In this last case, it was observed that the cells were reactive to
annexin and permeable to propidium iodide, indicating plasma membrane damage, which was confirmed
using SEM (Fig. 4). These results suggest that the application of a TVMF to cells that have incorporated
magnetic nanoparticles results in cell death via a non-apoptotic mechanism. Recent works on the
application of TVMF on magnetically-loaded cells showed that a large decrease in cell viability can be
achieved without actual temperature increase of the cell medium.29,33 Furthermore, it has been reported
11
Magnetic hyperthermia as a potential trypanocidal therapy.
that in the case of magnetically-loaded dendritic cells the percentage of cell death was proportional to the
amount of uptaken MNPs. As the cell death observed in our work correspond to the maximum amount of
uploaded MNPs (i.e., after 15 min of co-cultivation), it is still to be determined whether a similar effect
could be achieved with smaller amounts of uploaded MNPs.
Taken together, our results lead us to propose that in this case, the whole irreversible cell injury due to
mechanical stress (evidenced by SEM) in MNPs+/TVMF+ cell samples is the main cause of death.
Theoretical calculations on the effect of the power released by MNPs on the cell membrane supports this
hypothesis.34 However, it is worth mentioning that other factors, such as the liberation of toxic proteins
into the cytoplasm due to the disruption of membranes compartmentalizing them inside specific
organelles like lysosomes, cannot be ruled out as simultaneous cause of cell injury and death.
Conclusion
In conclusion, the series of experiments reported here demonstrated, as a proof of principle, that
magnetically induced hyperthermia causes microorganism death. Our results also illustrate that
hyperthermia is a thermal phenomenon of a subcellular scale because no macroscopic increase of
temperature was observed. We were able to show that hyperthermia was specific for cells that
incorporated MNPs and were submitted to the TVMF because neither NPs nor the TVMF alone resulted
in a loss of viability. Lastly, the cells that were submitted to the hyperthermia treatment were dramatically
damaged at the plasma membrane level. It should be stressed that although this methodology is being
extensively investigated and is currently used for mammalian cells, to our knowledge, this method is not
currently being proposed to treat diseases that are caused by microorganisms. The present study
highlights this method (at least as a proof of principle) as a potential and novel alternative to treat
infections caused by microorganisms. A major advantage of this method is that it causes selective
physical damage to the target cells. Therefore, the probability of the emergency of resistance strains is
small. More detailed studies are also being conducted to identify the mechanisms involved in cell death at
a greater level of detail. Developing delivery systems that have the capability of specifically directing
12
Magnetic hyperthermia as a potential trypanocidal therapy.
magnetic nanoparticles to infectious agents remains a challenge. These findings lead us to propose this
method as a novel strategy to develop new therapeutics against pathogenic microorganisms.
Acknowledgments/Disclosures
The author reports no conflicts of interest in this work. This work was supported by the Spanish Ministry
Ministerio de Ciencia e Innovación (project MAT2010-19326 and Consolider NANOBIOMED CS-27 2006)
and IBERCAJA. Partial support from Brazilian grants 08/57596-4 and 11/50631-1 from FAPESP and
INBEQMeDI is also acknowledged. We gratefully recognize Dr. M. Vergés, Dr. M.P. Morales and Dr. A.G.
Roca for their kind donation of MNPs. We also thank Dr. J. Godino (Instituto Aragonés de Ciencias de la
Salud I+CS, Zaragoza) for his help with flow cytometer measurements; I. Echaniz for technical support
and L. Casado for helping with scanning electron microscopy imaging.
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
Rascalou G, Pontier D, Menu F, Gourbiere S. Emergence and prevalence of human vector-borne
diseases in sink vector populations. PloS one. 2012 (Epub 2012 May 2012;7(5):e36858.
Hotez PJ, Molyneux DH, Fenwick A, et al. Control of neglected tropical diseases. N Engl J Med.
Sep 6 2007;357(10):1018-1027.
Goya GF, Fernandez-Pacheco R, Arruebo M, Cassinelli N, Ibarra MR. Brownian rotational
relaxation and power absorption in magnetite nanoparticles. J. Magn. Magn. Mater. Sep
2007;316(2):132-135.
Lacroix LM, Malaki RB, Carrey J, et al. Magnetic hyperthermia in single-domain monodisperse
FeCo nanoparticles: Evidences for Stoner-Wohlfarth behavior and large losses. Journal of
Applied Physics. Jan 2009;105(2).
Santos DM, Carneiro MW, de Moura TR, et al. Towards development of novel immunization
strategies against leishmaniasis using PLGA nanoparticles loaded with kinetoplastid membrane
protein-11. Int. J. Nanomed. 2012 2012;7:2115-2127.
Danesh-Bahreini MA, Shokri J, Samiei A, Kamali-Sarvestani E, Barzegar-Jalali M, Mohammadi-
Samani S. Nanovaccine for leishmaniasis: preparation of chitosan nanoparticles containing
Leishmania superoxide dismutase and evaluation of its immunogenicity in BALB/c mice. Int. J.
Nanomed. 2011 2011;6:835-842.
Vyas SP, Gupta S. Optimizing efficacy of amphotericin B through nanomodification. Int. J.
Nanomed. 2006;1(4):417-432.
Boscardin SB, Torrecilhas AC, Manarin R, et al. Chagas' disease: an update on immune
mechanisms and therapeutic strategies. J Cell Mol Med. Jun 2010;14(6B):1373-1384.
Pinto Dias JC. The treatment of Chagas disease (South American trypanosomiasis). Ann Intern
Med. May 16 2006;144(10):772-774.
13
Magnetic hyperthermia as a potential trypanocidal therapy.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
Filardi LS, Brener Z. Susceptibility and natural resistance of Trypanosoma cruzi strains to drugs
used clinically in Chagas disease. Trans R Soc Trop Med Hyg. 1987;81(5):755-759.
Wilkinson SR, Kelly JM. Trypanocidal drugs: mechanisms, resistance and new targets. Expert
Rev Mol Med. 2009;11:e31.
Pepin J, Milord F, Khonde A, Niyonsenga T, Loko L, Mpia B. Gambiense trypanosomiasis:
frequency of, and risk factors for, failure of melarsoprol therapy. Trans R Soc Trop Med Hyg. Jul-
Aug 1994;88(4):447-452.
Goto H, Lindoso JA. Current diagnosis and treatment of cutaneous and mucocutaneous
leishmaniasis. Expert Rev Anti Infect Ther. Apr 2010;8(4):419-433.
Berman JD. Human leishmaniasis: clinical, diagnostic, and chemotherapeutic developments in
the last 10 years. Clin Infect Dis. Apr 1997;24(4):684-703.
Berman JD. U.S Food and Drug Administration approval of AmBisome (liposomal amphotericin
B) for treatment of visceral leishmaniasis. Clin Infect Dis. Jan 1999;28(1):49-51.
Jordan A, Wust P, Fahling H, John W, Hinz A, Felix R. Inductive heating of ferrimagnetic particles
and magnetic fluids: Physical evaluation of their potential for hyperthermia. International Journal
of Hyperthermia. 2009;25(7):499-511.
Goya GF, Grazu V, Ibarra MR. Magnetic nanoparticles for cancer therapy. Current Nanoscience.
Feb 2008;4(1):1-16.
Lee J-H, Jang J-t, Choi J-s, et al. Exchange-coupled magnetic nanoparticles for efficient heat
induction. Nature Nanotechnology. Jul 2011;6(7):418-422.
Levy M, Wilhelm C, Siaugue JM, Horner O, Bacri JC, Gazeau F. Magnetically induced
hyperthermia: size-dependent heating power of gamma-Fe(2)O(3) nanoparticles. J. Phys.-
Condes. Matter. May 2008;20(20).
Comini M, Menge U, Wissing J, Flohe L. Trypanothione synthesis in Crithidia revisited. J Biol
Chem. Feb 25 2005;280(8):6850-6860.
Verges MA, Costo R, Roca AG, et al. Uniform and water stable magnetite nanoparticles with
diameters around the monodomain-multidomain limit. Journal of Physics D-Applied Physics.
2008;41(13).
Gonzalez-Fernandez MA, Torres TE, Andres-Verges M, et al. Magnetic nanoparticles for power
absorption: Optimizing size, shape and magnetic properties. Journal of Solid State Chemistry.
Oct 2009;182(10):2779-2784.
Goya GF. Handling the particle size and distribution of Fe3O4 nanoparticles through ball milling.
Solid State Communications. 2004;130(12):783-787.
Hergt R, Dutz S, Roder M. Effects of size distribution on hysteresis losses of magnetic
nanoparticles for hyperthermia. Journal of physics. Condensed matter : an Institute of Physics
journal. Sep 24 2008;20(38):385214.
Hilger I, Fruhauf K, Andra W, Hiergeist R, Hergt R, Kaiser WA. Heating potential of iron oxides for
therapeutic purposes in interventional radiology. Academic radiology. Feb 2002;9(2):198-202.
Pineiro-Redondo Y, Banobre-Lopez M, Pardinas-Blanco I, Goya G, Lopez-Quintela MA, Rivas J.
The influence of colloidal parameters on the specific power absorption of PAA-coated magnetite
nanoparticles. Nanoscale research letters. 2011;6(1):383.
Goya GF, Marcos-Campos I, Fernandez-Pacheco R, et al. Dendritic cell uptake of iron-based
magnetic nanoparticles. Cell biology international. 2008;32(8):1001-1005.
Marcos-Campos I, Asin L, Torres TE, et al. Cell death induced by the application of alternating
magnetic fields to nanoparticle-loaded dendritic cells. Nanotechnology. May;22(20):13.
Villanueva A, de la Presa P, Alonso JM, et al. Hyperthermia HeLa Cell Treatment with Silica-
Coated Manganese Oxide Nanoparticles. J. Phys. Chem. C. Feb 2010;114(5):1976-1981.
Richardson HH, Carlson MT, Tandler PJ, Hernandez P, Govorov AO. Experimental and
theoretical studies of light-to-heat conversion and collective heating effects in metal nanoparticle
solutions. Nano letters. Mar 2009;9(3):1139-1146.
Schildkopf P, Frey B, Mantel F, et al. Application of hyperthermia in addition to ionizing irradiation
fosters necrotic cell death and HMGB1 release of colorectal tumor cells. Biochemical and
Biophysical Research Communications. Jan 2010;391(1):1014-1020.
14
Magnetic hyperthermia as a potential trypanocidal therapy.
32.
33.
34.
Peter ME. Programmed cell death: Apoptosis meets necrosis. Nature. Mar 17
2011;471(7338):310-312.
Creixell M, Bohorquez AC, Torres-Lugo M, Rinaldi C. EGFR-targeted magnetic nanoparticle
heaters kill cancer cells without a perceptible temperature rise. ACS Nano. Sep 27;5(9):7124-
7129.
Lunov O, Zablotskii V, Pastor JM, et al. Thermal Destruction on the Nanoscale: Cell Membrane
Hyperthermia with Functionalized Magnetic Nanoparticles. In: Hafeli U, Schutt W, Zborowski M,
eds. 8th Internatioanl Conference on the Scientific and Clinical Applications of Magnetic Carriers.
Vol 1311. Melville: Amer Inst Physics; 2010:288-292.
15
Magnetic hyperthermia as a potential trypanocidal therapy.
Table 1 Average number of MNPs incorporated within a single cell as a function of incubation
time, as calculated from the saturation magnetization MS of the magnetically loaded cells.
Time (minutes)
Mass of Fe3O4
(pg/cell)
15 min 30 min 45 min
1 h
1,5 h
2 h
4 h
12 h
12.3
0.55
0.31
0.14
0.22
0.15
0.23
0.18
Number of MNPs
(x106)/cell
87
3.9
2.2
1.0
1.6
1.1
1.6
1.3
16
Figure 1 Number of MNPs uploaded per cell as a function of incubation time. The observed
decrease of incorporated MNPs followed an exponential decay, and reached a near-steady state
for incubation times larger than 1 h.
Magnetic hyperthermia as a potential trypanocidal therapy.
Figure 2: Schematic view of the experimental '2x2' design for evaluating the effect of MNPs and
TVMF on C. fasciculata: a) cells without MNPs not submitted to magnetic fields; b) cells with
MNPs without magnetic field application; c) application of magnetic fields on unloaded cells and
d) application of magnetic fields on MNP-loaded cells.
18
Magnetic hyperthermia as a potential trypanocidal therapy.
Figure 3: MTT results of the four situations displayed in Figure 2. All samples showed 100% of
cell viability except in the case of TVMF applied on magnetically loaded cells, where caused
95±5% cell death.
19
Magnetic hyperthermia as a potential trypanocidal therapy.
Figure 4: SEM images of MNPs+/TVMF+ sample before (A) and after (B) application of magnetic
fields. In the latter case, the changes in cell morphology can be clearly observed, reflecting the
severe cell damage after TVMF.
20
Magnetic hyperthermia as a potential trypanocidal therapy.
Figure 5: Specific Power absorption (SPA) of magnetic colloid (solid squares) at 1 % wt.
concentration, and the NP-loaded protozoa (solid line) during application of ac magnetic field (H =
160 Oe, f = 250 kHz).
21
Magnetic hyperthermia as a potential trypanocidal therapy.
Figure 6: Flow citometry results of the '2x2' experiments showed in Figure 2(see text for details).
Experiments A); B) and C) showed 89-91% of cell viability, whereas for experiment D) the
application of magnetic fields during 30 minutes on magnetically charged cells resulted in only 9%
of cell survival.
22
Magnetic hyperthermia as a potential trypanocidal therapy.
Supporting information 1:
Magnetic response at T = 10 K from (a) unloaded cells, (b) the response from co-cultured C.
fasciculata with MNPs (sample incubated for t = 15 min), (c) the difference between loaded and
unloaded cells (b-a), and (d) the pure magnetic colloid. Note that for (d), the pure colloid, the
curve was divided by 1.35x104 to fit the same scale than the magnetic signal from loaded cells
(fig. S1.c). To calculate the amount of magnetic material mmag incorporated by the cells, the MS
values from the pure colloids and from the MNP-loaded cells were used as 𝑚𝑚𝑎𝑔�𝑔𝑐𝑒𝑙𝑙�=
𝑀𝑆×𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑒𝑙𝑙𝑠. The number of MNPs per single cell was estimated from the known average
𝑀
particle diameter.
23
Magnetic hyperthermia as a potential trypanocidal therapy.
Supporting information 2:
Movies corresponding to the 2 x 2 design for evaluating the effect of MNPs and TVMF on C.
fasciculata.
a) File "a.mov": cells without MNPs not submitted to magnetic fields;
(http://www.youtube.com/watch?v=rZmw8NhsDt0&feature=youtu.be)
b) File "b.mov": cells with MNPs without magnetic field application;
(http://www.youtube.com/watch?v=1wcwnadAjTY&feature=youtu.be)
c) File "c.mov": application of magnetic fields on unloaded cells
(http://www.youtube.com/watch?v=LfIIpnPJ5bk&feature=youtu.be)
and;
d) File "d.mov": application of magnetic fields on MNP-loaded cells.
(http://www.youtube.com/watch?v=EiTx28AZ1gg&feature=youtu.be)
24
|
1807.08994 | 1 | 1807 | 2018-07-24T09:44:13 | Forced- and Self-Rotation of Magnetic Nanorods Assembly at the Cell Membrane: A Biomagnetic Torsion Pendulum | [
"physics.bio-ph",
"cond-mat.soft"
] | In order to give insights into how anisotropic nano-objects interact with living cell membranes, and possibly self-assemble, we designed magnetic nanorods with average size around 100 nm x 1$\mu$m by assembling iron oxide nanocubes within a polymeric matrix under a magnetic field. We then explored the nano-bio interface at the cell membrane under the influence of a rotating magnetic field. We observed a complex structuration of the nanorods intertwined with the membranes. Unexpectedly, after a magnetic rotating stimulation, the resulting macrorods were able to rotate freely for multiple rotations, revealing the creation of a bio-magnetic torsion pendulum. | physics.bio-ph | physics |
Forced and Self Rotation of Magnetic Nanorods Assembly at the Cell
Membrane: a Bio-Magnetic Torsion Pendulum
François Mazuel, Samuel Mathieu, Riccardo Di Corato, Jean-Claude Bacri, Thierry
Meylheuc, Teresa Pellegrino, Myriam Reffay*, Claire Wilhelm*
Dr. F. Mazuel, S. Mathieu, Dr. R. Di Corato, Prof. J.-C. Bacri, Dr. M. Reffay, Prof. C.
Wilhelm
Laboratoire Matière et Systèmes Complexes (MSC), UMR 7057, CNRS and Université Paris
Diderot, 75205 Paris Cedex 05, France.
E-mails: [email protected], [email protected]
Dr. R. Di Corato
Dipartimento di Matematica e Fisica "Ennio De Giorgi", Università del Salento
Via Arnesano, 73100 Lecce, Italy
Dr. T Meylheuc
INRA, UMR1319 Micalis, F-78352 Jouy-en-Josas, France
Prof. T. Pellegrino
Istituto Italiano di Tecnologia, I-16163 Genoa, Italy
Keywords: magnetic nanorods, assembly, membranes, rotating magnetic field
Abstract
In order to give insights into how anisotropic nano-objects interact with living cell membranes,
and possibly self-assemble, we designed magnetic nanorods with average size around 100 nm
x 1µm by assembling iron oxide nanocubes within a polymeric matrix under a magnetic field.
We then explored the nano-bio interface at the cell membrane under the influence of a
rotating magnetic field. We observed a complex structuration of the nanorods intertwined
with the membranes. Unexpectedly, after a magnetic rotating stimulation, the resulting
macrorods were able to rotate freely for multiple rotations, revealing the creation of a bio-
magnetic torsion pendulum.
1
In addition
to
their recognized potential for advancing diagnostics and
therapies,
nanomagnetic materials were recently described as tools to probe or act on the cell membrane.
In magnetic manipulation of cell membranes, an external rotating magnetic field is used to act
at a distance, with precisely controlled intensity, direction and localization of the applied
magnetic torque. Magnetically driven rotating or oscillating magnetic nanoparticles could thus
be used to probe cells' mechanical properties,[1-3] to deliver drugs,[4-8] or to kill cells by
physical membrane rupture.[9-16] In this last application, anisotropic nanoparticles in disk- or
rod-like shapes, stimulated with low-frequency magnetic fields, could compromise the cell
membrane and thereby trigger apoptotic or necrotic programmed cell death. Another
promising and recent field of research takes advantage of field-mediated magnetic
nanoparticle bioassembly to activate biochemical signaling mechanisms.[17-23] When
localized to the cell membrane,[17-21] the resulting clustering or orientation of targeted
receptors can be seen as a nanomagnetic switch to trigger cell responses.
In parallel, chemical synthesis has yielded magnetic nanomaterials with anisotropic
geometries, often rod-like shapes. [15, 24] Synthesis or self-assembly of magnetic particles
under a magnetic field is then frequently used to form super-organized anisotropic structures.
[25-29] However, how magnetic anisotropic nano-objects interact and assemble with cell
membranes under the action of a remote magnetic field has rarely been investigated. Here, we
addressed this field by producing highly magnetic nanorods, by assembling them at the cell
membrane into larger macrorods through remote spinning by a rotating magnetic field, and by
exploring the macrorod bio-structuration and dynamic response.
Magnetic macrorod formation at the cell membrane
Nanorods were prepared by self-aggregation of iron oxide nanocubes (20-nm edge) embedded
within a polymeric matrix (loaded with FITC dye) in the presence of a 0.05 T magnetic field.
2
The resulting fluorescent magnetic nanorods (Figure 1A1) are (1.1±0.1) µm long and
(110±15) nm wide (Figure 1A1) and exhibit strong magnetization (80 emu/g). The nanorods
spontaneously attach to the cell membrane (Figure 1A2) and are then internalized when
incubated overnight in absence of a magnetic field (Figure 1A3). However, if incubation
takes place in a magnetic field (Figures 1B and Figure 1C), internalization is prevented. The
nanorods are then forced to assemble at the membrane into larger anisotropic structures
aligned in the direction of the magnetic field, which we call macrorods. The formation of such
macrorods is a two-step process. First, dipolar magnetic interactions (created by a 0.2-T static
vertical magnetic field) forces the assembly of multiple nanorods into thin micron-long
elongated clusters (Figure 1B) during their first stage of attachment to the cell membrane. In
the second step, cells are subjected to a spinning 0.28-T magnetic field rotating around a 45°C
cone, for 30 min at 1 Hz. This conical rotating field is created by two permanent magnets that
generate a horizontal 0.2-T magnetic field in-between the magnets and spin around the sample,
itself within the vertical 0.2-T magnetic field (see experimental. The thin nanorod clusters
then merge, forming larger ellipsoid "macrorods" (average length: 22±11μm and diameter:
4±1μm). SEM and confocal images of typical macrorods are shown in Figures 1C1-3.
First step was to evaluate the impact of the magnetic stimulations on cell viability (Figure 1D).
It demonstrated that neither the static magnetic field nor the rotating one result in any cell
damage, as quantified by measuring the cells metabolic activity (relative to untreated control
cells) the day after the magnetic stimulations. The nanorods / cell membrane interactions thus
do not induce any significant alteration of cells viability and proliferation capacity, even in the
rotating setting, contrary to some other studies reporting membrane physical rupture triggered
by rotating magnetic nanoparticles, generally targeted to a specific receptor.[9-16] Here it
shows that cells can also adapt to a rotating stress and avoid massive harm.
3
Free rotation
Right after the rotating stimulation, we removed the cells from the magnetic fields (rotating
and static) and observed them under the microscope. As expected from the cell viability
measurements, cells morphologies were similar to control, with no cell rounding or
detachment. However, and surprisingly, some of the macrorods continued rotating, with no
external energy source (magnetic or other) to explain this motion (Figure 1D, see also
supplementary Movie S1 -- accelerated 5 times). This "spontaneous" rotation reveals in fact
the remarkable adaptability of cell membrane materials. Indeed, the macrorods rotated in the
direction opposite to the magnetic field rotation to which the rods were exposed. This
phenomenon is impressive, as it implies that the cell membranes confer a very efficient elastic
energy storage ability on the macrorods. This energy accumulated during the forced magnetic
excitation is release by free rotations in the opposite direction when the stimulation is stopped
(hundreds of free rotations were sometimes recorded).
Macrorods: intertwined nanorods and membrane filaments
The explanation for this elastic response comes first from the microscopic structure of the
macrorods. Macrorods are composed not only of nanorods but are mingled with membrane
structures all along their axis. This is clearly illustrated by the typical confocal image of a
macrorod in Figure 2A, with the cell membranes labeled with the red Pkh26 membrane
marker, and nanorod fluorescence collected in the green channel (FITC). Note that the
membrane labels are present all along the macrorod. Looking at the same samples with
scanning electron microscopy (SEM, Figure 2B) an intertwining of membrane filaments and
nanorods are observed (Figure 2B2). Membrane fragments are also detected on transmission
electron microscopy (TEM) images (Figure 2C) of macrorods detached from cells.
The use of the rotating magnetic conical field is decisive to structure the anchor point of the
nanorods within the membrane. It is responsible for membrane-rods entanglement and for
4
radial organization of membrane filaments around the macrorod. Indeed, these hybrid
membrane/nanorod structures are absent when rods are assembled without rotation (Figure
2B1).
Biomagnetic macrorods at the cell membrane: a torsion pendulum?
To model and quantify the observed free rotation, the rotating stimulation process was
explored. A programmable electromagnetic rotating field was used to impose the same
rotating cone under the microscope as that generating macrorod formation. Both the magnetic
rotating stimulation and the free relaxation could then be video-monitored (supplementary
Movie S2 -- accelerated 5 times). Because the field intensity was now lower (30 mT) and thus
not sufficient to form the macrorods, they were first formed as previously described with
rotating permanent magnets (130 mT). The electromagnetic field was programmed to spin the
macrorods counterclockwise, for a given number of rotations (generally 20), at 2 Hz. When
the forcing was stopped, and thus in absence of a magnetic field, the macrorods rotated
clockwise, some of them up to 20 times, while others rotated only a few degrees (Figure 3).
We propose a torsion pendulum model to describe this free rotation.
The membrane structure at the anchor point during the forced rotation generates an elastic
torque C and the rod is subjected to a rotational viscous friction when it moves. As inertia
can be neglected (low Reynolds number ≈ 10-2), the rod's angle obeys the equation
, with R the total of relaxation rotation (in degrees), and can thus be
simply written as
with
. The fitting
in Figure 3A
demonstrates the correct matching of this torsion pendulum model. The parameters and R
reflect the peculiar coiling of cellular membranes around and within the macrorod. Figure 3B
shows the plot of as a function of R for different rods. Two groups emerge: the first one
5
RCdtdt=tRexp1C=
corresponds to rods relaxing only few tens of degrees (black symbols), while the second
corresponds to rods relaxing through more than a thousand degrees (> 3 rotations). For this
second group, the rod's anchor point is organized such that it enables extremely efficient
energy storage during forced rotations as the membrane filament tension increases. In contrast,
for the first group, either the organization is inappropriate or the coiling structure may be
damaged during the forced rotation, resulting in an abrupt energy release that would prevent
free rotation.
Interestingly, the typical relaxation curve shown in Figure 3A presents a succession of angular
jumps that follow the exponential trend described above. This could be explained by different
coiling modes of the membrane at the anchor point. Jumps would be associated with twisted
membrane filaments, whereas the overall relaxation would correspond to membrane filaments
coiled around the rod. The supplementary movie S2 illustrate these two modes.
Within the framework of the torsion pendulum model, successive excitations of a rod is
expected to damage the anchor point organization as tension increases and exceeds the
membrane filament resistance. This is illustrated by Figure 3C. In this case, rods were
subjected to successive periods of magnetic field rotation and immobilization. The intensity of
the magnetic field in the immobilized phase is half the intensity of the rotating field, so that a
rod with a proper coiling at the anchor point can relax if the elastic restoring force is stronger
than the magnetic torque. In the particular example shown in Figure 3C, this is the case for the
two first excitations: the rod is able to relax for one rotation despite the magnetic field
(indicated with black stars). The increase in tension against the magnetic forcing at the anchor
point is materialized in Figure 3C with events (indicated with red crosses) when the rod slows
down and stops before being caught up by the magnetic field after an additional rotation. The
occurrence of such events increases with successive excitations until the structure at the
anchor is damaged (red arrow). After that the rod follows perfectly the magnetic field: the
damaged structure cannot store anymore elastic energy efficiently.
6
Taken together, all those observations confirm the simple torsion pendulum model based on a
specific membrane coiling that enable energy storage and restitution.
In summary, by remote assembly of magnetic nanorods during their first stage of attachment
to cell membranes through the application of a rotating magnetic field, we managed to create
a super-assembly (macrorod) trapping within and wrapping around membrane filaments. As a
result, this biomagnetic structure exhibited an elastic behavior which provided an impressive
numbers of free rotations, when the rotating field is released and no other stimulation applied.
This movement, evidences a remarkable pool of membranes available and the possibility to
arrange them in a biomagnetic torsion pendulum.
Experimental section
Magnetic nanorods preparation
The preparation method is based on iron oxide nanocubes assembly into a polymeric matrix,
in presence of a static magnetic field, as adapted by a procedure previously described for
spherical magnetic nanoparticles [29]. Briefly, the polymeric anisotropic construct is obtained
by mixing nanocubes with the poly(maleic anhydride alt-1 octadecene) polymer, in
chloroform. The nanoparticles dispersion was placed in an ultrasound bath under the influence
of two opposite permanent magnets placed against each other on the vial wall, generating a
0.05 T magnetic field. The slow and controlled addition of acetonitrile induces a change in the
solubility of both the nanocubes and polymer and promotes their aggregation. The application
of a permanent magnetic field promoted the formation of elongated superstructure. The
magnetic nanorods were finally magnetically sorted and resuspended in water.
The iron oxide nanocubes, about 20 nm in cube-edge, were synthesized by thermal
decomposition technique detailed as detailed in [30]. Briefly, they were prepared by mixing 1
mmol of iron(III) acetylacetonate and 4 mmol of decanoic acid in 25 mL of dibenzyl ether,
7
heating the solution to 200°C (5°C/min) for 2.5 h. The temperature is then increased further to
reflux temperature (300°C at a rate of 10°C/min) for 1 h. At the end of the process, nanocubes
are dispersed in chloroform.
Magnetic field stimulations
Different magnetic devices were designed and fabricated. The simplest one dedicated to
macrorods formation (Figure 4A) consists in a permanent magnet (neodymium, 50x40x20
mm, Supermagnet) placed below the cell sample. It creates a magnetic field of 130 mT in the
sample region. The second one (Figure 4B) is adapted to this first one by adding a set of two
permanent magnets facing one to the other and fixed to a motor controlling a gear wheel
(frequency up to 2 Hz). This set of magnets creates a horizontal field of 130 mT, so that the
resulting field (about 180 mT) makes an angle of 45° with the vertical axis. As a result, when
the motorized magnets are rotated, the magnetic field spins, describing a 45° cone. Both
devices contain a cylindrical 20 mm cradle to welcome the cell sample, and are thermostated
by water circulation. The third device (Figure 4C) was designed to be adapted to a
microscope and to allow switching on and off the magnetic field at will. It is composed of
four coils which cores are made of soft iron, connected by pairs and supplied by an alternative
current. The space between each core is about 1cm and it creates in between a 30 mT
magnetic field. To generate a rotating 45°C rotating cone, this magnetic device is placed 2 cm
beneath the cell sample. The magnetic field on the cells is then reduced to 10 mT. To generate
the rotating field in the plane of the magnetic coils, the two pairs of coils are supplied with
sinusoidal currents (2A amplitude) displaying the same frequency (up to 5 Hz) but 90° out of
phase. Finally, the whole set-up was mounted on the vertical arm of a Leica DMIRB with
control of the z position, and the microscope was thermostated at 37°C by cube&box (Life
Imaging Services).
Nanorods incubation with cells and macrorods formation
8
PC3 cancer cells were cultured in DMEM supplemented with 10% fetal bovine serum, L-
glutamine and penicillin/streptomycin. Prior to experiments, cells were transferred into 20
mm cylindrical home-made cuves with glass bottom which fit within the magnetic fields
devices. When cells reached 80-90 % of confluency, nanorods were incubated at an iron
concentration of 1 mM, corresponding roughly to 106 nanorods per ml, or equivalently to 103
nanorods per cell. This concentration was adjusted by testing a range of concentration,
observing the cells with confocal microscopy, and selecting concentration where nanorods
were numerous on the cells membrane, but still quasi-individual.
After the 2-hour incubation period, the cells were immediately transferred into the
thermostated magnetic field devices, first the permanent static field for 30 min, then the
spinning magnetic field for 30 min.
Cell viability was assessed by Alamar Blue metabolic assay (Thermo Fisher Scientific).
Fluorescence (appearing post-metabolization of the active ingredient, Resaruzin), was
quantified with a microplate reader (excitation 550 nm, detection 590 nm). In brief, 100 000
cells were first seeded in the cuves (3 per conditions). 24 hours after, nanorods incubation was
performed (2-hours at 1mMFe, except for the control conditions) and the cuves were submitted
either to the static magnetic field only (30 min, condition B in Figure 1), or to both the static
and rotating magnetic field (30 min + 30 min, condition C in Figure 1). 24 hours after, all
cuves were incubated (800 µl total) with 10 % Alamar Blue in DMEM for 2 hours, and the
reagent was transferred to 96-well plate for analysis (200 µl per well). All values are
expressed relative to control (normalized at 100% viability).
Imaging
Nanorods and macrorods were imaged by scanning (SEM) and transmission (TEM) electron
microscopy, by confocal microscopy, and by conventional transmission microscopy. TEM
was used to observe nanorods in aqueous suspension, or macrorods extracted from the cells.
For this second case, the cells were first fixed with paraformaldehyde (2%) for 30 min right
9
after the whole stimulation protocol. Culture medium was then washed and replaced with
ultra-pure water to avoid salts contaminations, and cells were scratched in order to detach
cells and macrorods. For both nanorods and macrorods TEM observation, a small drop of
suspension (5µl) was pipetted onto a copper grid, and observed after total evaporation of the
liquid with a Phillips Tecnai 12.
SEM was used to observe whole cells. Cells were fixed with glutaraldehyde (2% in 0.1 M
cacodylate buffer), and dehydrated by soaking in a graded series of ethanol before critical-
point drying under CO2. Samples were mounted on aluminum stubs with conductive silver
paint and sputter coated with gold palladium for 200 s at 10 mA. Samples were then imaged
with a Hitachi S4500 instrument.
For confocal microscopy, cells were labeled either by Pkh26 label (20 min incubation,
according to manufacturer's instruction, Sigma) which binds to the plasma membrane, or with
phalloidin to see actin filaments, or with DAPI to image the nuclei. Cells were observed by
means of an Olympus JX81/ BX61 Device/Yokogawa CSU Device spinning disk microscope
(Andor Technology plc, Belfast, Northern Ireland), equipped with a 60x Plan-ApoN oil
objective lens.
Conventional transmission microscopy was carried out with a DMIRB Leica microscope,
equipped with a Cube & Box device to maintain the cells at 37°C during experiment.
Macrorods imposed and free rotations were captured with an ultra-fast camera. Time
sampling was used in between 100 to 10 images per second. Image J home-made plugins
(succession of thresholding, selecting and measuring) were used to analyze the macrorods
rotational movements.
10
Acknowledgements
This work was supported by the European Union (ERC-2014-CoG project MaTissE 648779).
We acknowledge the ImagoSeine core facility of the Institut Jacques Monod, member of
IBiSA and France-BioImaging (ANR-10-INBS-04) infrastructures.
Received: ((will be filled in by the editorial staff))
Revised: ((will be filled in by the editorial staff))
Published online: ((will be filled in by the editorial staff))
11
References
1.
2.
Wang, N.; Butler, J. P.; Ingber, D. E., Science 1993, 260 (5111), 1124-1127.
Isabey, D.; Pelle, G.; Dias, S. A.; Bottier, M.; Nguyen, N.-M.; Filoche, M.; Louis, B.,
Biomechanics and modeling in mechanobiology 2015, 1-17.
3.
Robert, D.; Aubertin, K.; Bacri, J.-C.; Wilhelm, C., Physical Review E 2012, 85 (1),
011905.
4.
5.
Dobson, J., Drug development research 2006, 67 (1), 55-60.
McBain, S.; Griesenbach, U.; Xenariou, S.; Keramane, A.; Batich, C.; Alton, E.;
Dobson, J., Nanotechnology 2008, 19 (40), 405102.
6.
Katagiri, K.; Imai, Y.; Koumoto, K.; Kaiden, T.; Kono, K.; Aoshima, S., Small 2011, 7
(12), 1683-1689.
7.
8.
Mura, S.; Nicolas, J.; Couvreur, P., Nature materials 2013, 12 (11), 991-1003.
Griffete, N.; Fresnais, J.; Espinosa, A.; Wilhelm, C.; Bée, A.; Ménager, C., Nanoscale
2015, 7 (45), 18891-18896.
9.
Kim, D.-H.; Rozhkova, E. A.; Ulasov, I. V.; Bader, S. D.; Rajh, T.; Lesniak, M. S.;
Novosad, V., Nature materials 2010, 9 (2), 165-171.
10.
11.
Liu, D.; Wang, L.; Wang, Z.; Cuschieri, A., Nano letters 2012, 12 (10), 5117-5121.
Bouchlaka, M. N.; Sckisel, G. D.; Wilkins, D.; Maverakis, E.; Monjazeb, A. M.; Fung,
M.; Welniak, L.; Redelman, D.; Fuchs, A.; Evrensel, C. A., PloS one 2012, 7 (10), e48049.
12. Wang, B.; Bienvenu, C.; Mendez-Garza, J.; Lançon, P.; Madeira, A.; Vierling, P.; Di
Giorgio, C.; Bossis, G., Journal of Magnetism and Magnetic Materials 2013, 344, 193-201.
13.
Leulmi, S.; Chauchet, X.; Morcrette, M.; Ortiz, G.; Joisten, H.; Sabon, P.; Livache, T.;
Hou, Y.; Carrière, M.; Lequien, S., Nanoscale 2015, 7 (38), 15904-15914.
14.
Zhang, E.; Kircher, M. F.; Koch, M.; Eliasson, L.; Goldberg, S. N.; Renstrom, E., ACS
nano 2014, 8 (4), 3192-3201.
15. Martínez-Banderas, A. I.; Aires, A.; Teran, F. J.; Perez, J. E.; Cadenas, J. F.; Alsharif,
N.; Ravasi, T.; Cortajarena, A. L.; Kosel, J., Scientific reports 2016, 6.
16.
Kilinc, D.; Dennis, C. L.; Lee, G. U., Advanced Materials 2016.
17. Mannix, R. J.; Kumar, S.; Cassiola, F.; Montoya-Zavala, M.; Feinstein, E.; Prentiss,
M.; Ingber, D. E., Nature nanotechnology 2008, 3 (1), 36-40.
18.
Perica, K.; Tu, A.; Richter, A.; Bieler, J. G.; Edidin, M.; Schneck, J. P., ACS nano
2014, 8 (3), 2252-2260.
19.
Kilinc, D.; Lesniak, A.; Rashdan, S. A.; Gandhi, D.; Blasiak, A.; Fannin, P. C.; von
Kriegsheim, A.; Kolch, W.; Lee, G. U., Advanced healthcare materials 2015, 4 (3), 395-404.
12
20.
21.
Lee, K.; Yi, Y.; Yu, Y., Angewandte Chemie 2016, 128, 7510-7513.
Seo, D.; Southard, K. M.; Kim, J.-w.; Lee, H. J.; Farlow, J.; Lee, J.-u.; Litt, D. B.;
Haas, T.; Alivisatos, A. P.; Cheon, J., Cell 2016, 165 (6), 1507-1518.
22.
Etoc, F.; Lisse, D.; Bellaiche, Y.; Piehler, J.; Coppey, M.; Dahan, M., Nature
nanotechnology 2013, 8 (3), 193-198.
23.
Etoc, F.; Vicario, C.; Lisse, D.; Siaugue, J.-M.; Piehler, J.; Coppey, M.; Dahan, M.,
Nano letters 2015, 15 (5), 3487-3494.
24. Milosevic, I.; Jouni, H.; David, C.; Warmont, F.; Bonnin, D.; Motte, L., The Journal
of Physical Chemistry C 2011, 115 (39), 18999-19004.
25. Wang, Y.; Zhang, L.; Gao, X.; Mao, L.; Hu, Y.; Lou, X. W. D., Small 2014, 10 (14),
2815-2819.
26.
27.
Luo, D.; Yan, C.; Wang, T., Small 2015, 11 (45), 5984-6008.
Singh, G.; Chan, H.; Baskin, A.; Gelman, E.; Repnin, N.; Král, P.; Klajn, R., Science
2014, 345 (6201), 1149-1153.
28.
Sahoo, Y.; Cheon, M.; Wang, S.; Luo, H.; Furlani, E.; Prasad, P., The Journal of
Physical Chemistry B 2004, 108 (11), 3380-3383.
29.
Allione, M.; Torre, B.; Casu, A.; Falqui, A.; Piacenza, P.; Di Corato, R.; Pellegrino,
T.; Diaspro, A., Journal of Applied Physics 2011, 110 (6), 064907.
30. Guardia, P.; Di Corato, R.; Lartigue, L.; Wilhelm, C.; Espinosa, A.; Garcia-Hernandez,
M.; Gazeau, F.; Manna, L.; Pellegrino, T., ACS nano 2012, 6 (4), 3080-3091.
13
Figure 1 (2 columns / 16 cm): Assembly process of magnetic nanorods into a macrorod at
the cell membrane. A. Nanorods dispersed in the cells culture medium interact individually
with the cell membrane without magnetic field. (A1) shows transmission electron microscopy
(TEM) image of the magnetic nanorods before cellular interaction (in aqueous dispersion) and
identifies iron oxide nanocubes embedded in polymer (insert). (A2) shows confocal image
after 1 hour incubation, where nanorods (FITC, green channel) are detected on the cell
14
membranes (Pkh26, red channel). Nuclei are labeled with DAPI (blue channel). (A3) shows
confocal image of the nanorods after a 24h incubation and illustrates their complete
internalization in cell cytoplasm. B. Formation of small clusters still attached to the membrane
under the application of a static vertical magnetic field. (B1) shows a scanning electron
microscopy (SEM) picture of these nanorods clusters on cell membrane; (B2-3) shows
confocal microscopy of nanorods clusters spread on all membranes (nanorods in green,
membranes in red, nuclei in blue): 4-µm width stacked image (B2) or Z views (B3:
reconstruction from Z stacks acquired with a 0.5 μm interslice with Image J Volume Viewer
plugin, with 45° x tilt angle). C. Rotating stimulation process: the magnetic field spins,
describing a cone around its initial vertical direction. This rotating motion forces the nanorods
clusters to interact over wider range, and form a larger magnetic macrorod. (C1) and (C2)
show SEM images of macrorods attached to cell membranes. (C3) shows confocal
microscopy (3D stack reconstruction) of a macrorod (green) fixed by one end on the cell
surface (here F-actin in red -- phalloidin staining). C4 illustrates the free rotations of the
macrorods: large view (right) of the cells at the end of the magnetic field rotation process, and
superimposition of 60 images (separated by 1s) for the 4 zones delimited by a square (left). It
clearly shows that these macrorods rotate, freely, in absence of magnetic field stimulation.
The corresponding movie can be seen as supplementary movie S1. D. Cell viability measured
by quantifying the metabolic activity (Alamar blue) of cells incubated with the magnetic
nanorods in presence of the static magnetic field (condition "static", similar to part B), and
under the influence of the rotating magnetic field (condition "rotating", similar to part C), and
compared to control cells (seeded at the same exact number of cells, see Experimental
Section).
15
Figure 2 (1.5 column / 12 cm): A. Confocal microscopy image of a macrorod in the red
channel (membrane, left), green channel (nanorods, middle), and superposition of both (plus
nucleus in the blue channel, right). One can clearly see that membranes are trapped all
through the macrorod. B. SEM pictures of two macrorods either formed with magnetic field
rotation (B2), or without (B1). On the right image, membrane filaments are present all along
the macrorod (arrows). By contrast, on the left image, it corresponds to "clean" nanorods
clusters without membrane entanglements. C. TEM pictures of a magnetic macrorod after
removal from cell surface. Membrane fragments are detected all around the rod, stuck
between the nanorods. This macrorod was probably fixed to the cell by its left end.
16
Figure 3 (1.5 column / 12 cm): A. Typical angular evolution of two macrorods after the same
forced excitation (20 rotations at 2Hz). One relaxes on 20 rotations (grey) whereas the other
relaxes only a few degrees (black). The inset shows a magnification around the time when the
magnetic
stimulation
stops.
Both
relaxation
curves
are
fitted
with
. The fitting curve is superimposed (dotted line). B. Fitting
parameters as a function of R for all analyzed macrorods. Two populations of macrorods
emerge. C. Successive coiling of the anchor point and its damage. The spinning excitation
follows the periodic pattern: 8 counterclockwise rotations at 3.2Hz followed by 5 seconds
where the magnetic field position is fixed. Four stimulation cycles are applied. The
cumulative angle of a typical macrorod (grey) as well as the magnetic field position modulo
17
0exp1tt=tR
360° (black) are plotted as a function of time. Of note, the intensity of the spinning magnetic
field is twice the intensity of the fixed magnetic field. Stars () indicate the macrorod
relaxations. Instants when the macrorod does not follow the magnetic field are marked with
crosses (x). The arrow indicates the time when the anchor point is probably damaged. After
that the macrorod follows perfectly the rotating magnetic field.
Figure 4 (one column / 8 cm): Magnetic set-ups. A. For macrorods formation: a home-made
culture dish is placed on a permanent magnet generating a 30 mT vertical magnetic field for
30 min. B. The dish is then placed in between two magnets fixed on a motorized axis
generating the spinning conical magnetic field (130 mT). C. Electromagnetic set-up designed
to manipulate the macrorods once they are formed: two pairs of coils are arranged
perpendicularly to create a rotating magnetic field in an upper plane over the dish, resulting in
a spinning field in the plane of the dish (left = side view / right = top view).
18
Table of contents entry
Assembling nanorods at the cell membrane by a remote spinning magnetic field leads to a
complex bio-magnetic macrorod constituted of both nanorods and intertwined cell membrane.
This new object behaves as a torsion pendulum able to store elastic energy: it is able to freely
rotate over multiple turns in response to the spinning stimulation.
Keywords: magnetic nanorods, assembling, membranes, rotating magnetic field
19
|
1701.09095 | 1 | 1701 | 2017-01-31T15:35:34 | Mechanics of active surfaces | [
"physics.bio-ph",
"cond-mat.soft"
] | We derive a fully covariant theory of the mechanics of active surfaces. This theory provides a framework for the study of active biological or chemical processes at surfaces, such as the cell cortex, the mechanics of epithelial tissues, or reconstituted active systems on surfaces. We introduce forces and torques acting on a surface, and derive the associated force balance conditions. We show that surfaces with in-plane rotational symmetry can have broken up-down, chiral or planar-chiral symmetry. We discuss the rate of entropy production in the surface and write linear constitutive relations that satisfy the Onsager relations. We show that the bending modulus, the spontaneous curvature and the surface tension of a passive surface are renormalised by active terms. Finally, we identify novel active terms which are not found in a passive theory and discuss examples of shape instabilities that are related to active processes in the surface. | physics.bio-ph | physics | Mechanics of active surfaces
Guillaume Salbreux1, 2 and Frank Julicher1
1Max Planck Institute for the Physics of Complex Systems, Nothnitzer Str. 38, 01187 Dresden, Germany
2The Francis Crick Institute, 1 Midland Road, NW1 1AT, United Kingdom
(Dated: February 1, 2017)
We derive a fully covariant theory of the mechanics of active surfaces. This theory provides a
framework for the study of active biological or chemical processes at surfaces, such as the cell cortex,
the mechanics of epithelial tissues, or reconstituted active systems on surfaces. We introduce forces
and torques acting on a surface, and derive the associated force balance conditions. We show
that surfaces with in-plane rotational symmetry can have broken up-down, chiral or planar-chiral
symmetry. We discuss the rate of entropy production in the surface and write linear constitutive
relations that satisfy the Onsager relations. We show that the bending modulus, the spontaneous
curvature and the surface tension of a passive surface are renormalised by active terms. Finally, we
identify novel active terms which are not found in a passive theory and discuss examples of shape
instabilities that are related to active processes in the surface.
Biological systems exhibit a stunning variety of complex morphologies and shapes. Organisms form from a fertilized
egg in a dynamic process called morphogenesis. Such shape forming processes in biology involve active mechanical
events during which surfaces undergo shape changes that are driven by active stresses and torques generated in the
material.
Important examples are two-dimensional tissues, so called epithelia. They represent surfaces that can
deform their shape as a result of active cellular processes [1]. Cells also exhibit a variety of different shapes and can
undergo active shape changes. For example during cell division, cells round up to a spherical shape due to an increase
of active surface tension [2]. Cell shapes are governed by the cell cortex, a thin layer of an active contractile material
at the surface of the cell [3]. Epithelial tissues and the cell surface are examples of active surfaces. In addition, recent
experiments have reconstituted thin shells of active material in-vitro [4]. These are thin sheets of active matter that
can deform due to the generation of internal forces and torques that are balanced by external forces (Figure 1A).
The theory of active gels describe the large-scale properties of viscoelastic matter driven out-of-equilibrium due to
a source of chemical free energy in the system [5]. A number of processes in living systems have been successfully
described using this theoretical framework [6]. Living or artificial active systems often assemble into nearly two-
dimensional surfaces. To understand the physics of such active surfaces, requires a systematic analysis of force and
torque balances in curved two-dimensional geometries, taking into account active stresses and material properties.
The shapes of passive fluid membranes has been described with considerable success by the Helfrich free energy, a
coarse-grained description of membranes with an expansion of the free energy in powers of the curvature tensor [7].
Expressions for the stress and torque tensors within a Helfrich membrane have been obtained. The associated force
and torque balance equations are equivalent to shape equations for miminal energy shapes [8, 9]. Active membranes
theories have expanded the description of passive membranes to include external forces induced by pumps contained
in a membrane [10–12].
The morphogenesis of epithelial tissues is a highly complex problem involving forces generated actively within the
cells. Distribution of forces acting along the cross-section of a sheet-like tissue give rise to in-plane tensions, but also
to internal torques resulting from differential stresses acting along the cross-section of the tissue (Fig. 1B). These
differential stresses are crucial to generate tissue shape changes [13]. However, no framework currently allows to
describe the mechanics of active thin surfaces with internal stresses and torque densities.
In this work, we present such a general framework for the mechanics of active surfaces, driven internally out of
equilibrium by molecular processes such as a chemical reaction. We start by considering forces and torques generated
in a surface of arbitrary shape. The corresponding expression for the virtual work shows that components of the
tension and torque tensors are coupled to the variation of the metric, of the curvature tensors and of the Christoffel
symbols defined for the surface (Eq. 15). Using these expressions, we then derive the entropy production for a fluid
surface undergoing chemical reactions. We analyse the symmetries of surfaces with rotational symmetry in the plane,
and show that they can have up-down, chiral or planar-chiral broken symmetry. We write down the corresponding
constitutive equations for the components of the tension and torque tensors and for the fluxes of the chemical species.
Interestingly, the generic constitutive equations involve couplings of the curvature tensor with the chemical potential
of the surface chemical species. We then discuss the stability of a flat active fluid with broken up-down symmetry.
Finally, we show that generic equations for an active elastic thin shell can be obtained using the same framework.
7
1
0
2
n
a
J
1
3
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
9
0
9
0
.
1
0
7
1
:
v
i
X
r
a
2
A
B
Apical
Basal
Tissue
Extracellular matrix
C
OUT
D
IN
FIG. 1: A. Filaments and motors near a surface and epithelial tissues are examples of active surfaces. B. The distribution of
stresses within a thin layer give rise to stresses and torques when integrated across the thickness of the layer. C. Local basis
of tangent vectors e1, e2 and normal vector n associated to the surface X(s1, s2). D. Internal and external forces and torques
acting on a surface element with surface area dS.
I. FORCE AND TORQUE BALANCE ON A CURVED SURFACE
We consider a curved surface X(s1, s2) parametrised by two generalised coordinates s1, s2 (Figure 1C). We use
latin indices to refer to surface coordinates and greek indices to refer to 3D euclidean coordinates. We introduce the
metric tensor gij = ei · ej where ei = ∂iX with ∂i = ∂/∂si. The curvature tensor is defined as Cij = −∂i∂j X · n,
where n = e1 × e2/e1× e2 is the unit normal vector, which we usually consider to point outward for a closed surface.
We denote dl with dl2 = gijdsidsj a line element on the surface, and dS = √gds1ds2 a surface element, where
g = det gij is the determinant of the metric tensor (Appendix A).
The force f and torque Γ across a line of length dl with unit vector ν = νiei, tangential to the surface and normal
to the line can be expressed as
f = dl νiti = dl νiti,
Γ = dl νimi = dl νimi
,
(1)
(2)
where we have introduced the tension ti and moment mi per unit length (Figure 1B,D). Decomposing ti and mi in
tangential and normal components as
ti = tij ej + ti
nn,
mi = mij ej + mi
nn ,
defines the tension and moment per unit length tensors tij, ti
n, mij and mi
n.
By expressing the total force acting on a region of surface S with contour C and using Newton's law, one finds
ZS
dSρa =IC
dlνiti +ZS
dSf ext,
(3)
(4)
(5)
where ρ is the surface mass density, a is the local center-of-mass acceleration, f ext is an external force surface density.
When the surface is embedded in a medium, the external force surface density is related to stresses exerted by the
medium on the surface, f ext
α = σβαnβ with σαβ the 3-dimensional stress tensor in the medium. The total torque obeys
ZS
dS[X × ρa] =IC
dlνi(cid:2)mi + X × ti(cid:3) +ZS
dS(cid:2)Γext + X × f ext(cid:3) .
(6)
where Γext is the external torque surface density, and where the left-hand side is the torque stemming from inertial
forces. Here, we ignore the moment of inertia tensor for simplicity. This results in the force balance expression
(Appendix C):
3
These equations can be expressed in terms of the components of the tension and torque tensors:
∇iti = −f ext + ρa,
∇imi = ti × ei − Γext.
∇itij + Ci
∇iti
n = −f ext,j + ρaj,
jti
n − Cij tij = −f ext
n + ρan,
n − Γext,j,
jti
n − Cij mij = −ǫijtij − Γext
n ,
∇imij + Ci
∇imi
n = ǫi
jmi
(7)
(8)
(9)
(10)
(11)
(12)
where the tangential and normal component of a vector v on the surface are written vi = v · ei and vn = v · n.
II. VIRTUAL WORK
We introduce the virtual work δW , which is the mechanical work acting on a region of surface S enclosed by a
contour C, upon a small deformation δX of the surface, with X′(s1, s2) = X(s1, s2) + δX(s1, s2). Here δX(s1, s2)
represents a displacement of a material point on the surface specified by (s1, s2). The virtual work can be defined as
δW =IC
dlνi(ti · δX +
1
2
mi · (∇ × δX)) +ZS
dS((f ext − ρa) · δX +
1
2
Γext · (∇ × δX)),
(13)
where S is the surface region enclosed by C, and we have introduced the rotational operator in euclidian space (Eq.
A29):
(∇ × δX)α = ǫαβγei
β(∂iδXγ) + ǫαβγnβ(∂nδXγ).
(14)
In Eq.14, we have introduced the normal derivative of the surface deformation, ∂nδX. We consider here ∂nδX =
−(∂iδX · n)ei (Appendix B).
The terms in the expression of the virtual work 13 describe the work due to forces and torques acting at the
boundary C as well as external forces and torques acting on the surface S. Using force balance and the divergence
theorem, the virtual work can be re-expressed as (see Appendix D)
δW =ZS
dS(cid:18)t
ij δgij
2
+ mi
jδCi
j + mi
n
ǫj
kδΓk
ij
2 (cid:19).
(15)
j and variation of Christoffel symbols
ij as a function of the surface variation δX are given in Appendix B. We have introduced the in-plane tension and
Here, the explicit expression of the metric variation δgij, curvature variation δCi
δΓk
bending moment tensors:
1
s +
tij = tij
mij = −mikǫk
2(cid:0) ¯mkiCk
j.
j + ¯mkj Ck
i(cid:1) ,
(16)
(17)
where the s subscript denotes the symmetric part of the tensor (Eq. A14). In Eq. 15, we have used a reference frame
that deforms with the material.
The virtual work given by equation 15 can be interpreted physically as the mechanical work due to different types of
deformations. The in-plane surface stress tensor ¯tij is conjugate to the variation of the metric tensor δgij, describing
internal shear and area compression. The in-plane tension tensor tij introduced in Eq. 16 differs from the tension
tensor tij introduced in Eq. 3: this is because in a thin shell, a deformation leading to a change of metric of the
surface mid-plane corresponds to a three-dimensional shear within the shell. As a result, the work to deform the
surface mid-plane depend on the in-plane bending moment tensor, which reflects the distribution of stresses across
the thickness of the shell. The in-plane tensor ¯mij of bending moments is conjugate to the variation of the curvature
kδΓk
tensor δCi
ij.
j due to bending of the surface. The normal torque mi
n is conjugate to gradients of local rotations ǫi
4
The expression of the virtual work 15 does not include shear perpendicular to the surface: this would require the
introduction of an additional variable.
The virtual work given in Eq. 15 is very general.
In order to evaluate the virtual work for a given surface
deformation, the values of the internal stresses characterised by the in-plane stress tensor tij , the in-plane bending
moment tensor mij, and the normal torque mn, have to be known. In general, they are provided by constitutive
relations describing the properties of the material associated with the surface.
We now discuss constitutive relations for active fluid and elastic curved surfaces. The case of a passive membrane
is discussed in Appendix H.
III. CURVED ACTIVE FILM
We now use concepts for irreversible thermodynamics to derive constitutive equations for a curved isotropic fluid.
We consider a fluid consisting of several species α = 1..N with concentrations cα. The local mass density is given
by ρ = Pα mαcα with mα the molecular mass of species α. The free energy density in the rest frame is denoted
j is the curvature tensor of the film in mixed coordinates, and T the temperature. The
j, T ) where Ci
f0(cα, Ci
differential of f0 is
where µα is the chemical potential of component α, K i
The total free energy density is
j is the passive bending moment and s the entropy density.
df0 = µαdcα + K i
jdCi
j − sdT,
(18)
f =
1
2
ρv2 + f0,
(19)
where the kinetic energy is given by 1
chemical potential of the chemical species α.
2 ρv2 = 1
2 ρ(cid:2)vivi + (vn)2(cid:3). We denote µα
tot = df /dcα = µα + mαv2/2 the total
A. Conservation equations
FIG. 2: Free energy balance on a surface element in the isothermal case. Free energy density is exchanged between the surface
element and the surrounding surface with a flux Jf , with the surrounding bulk with flux J f
n , and is produced with rate −T π.
We start by deriving conservation equations for the surface mass, concentration of chemical species, energy, entropy
and free energy. Using an Eulerian representation (Appendix E), mass balance reads
∂tρ + ∇i(ρvi) + vnCi
iρ = J ρ
n,
(20)
where J ρ
velocity.
n is a source term due to mass exchange with the environment and v = viei + vnn is the center-of-mass
The concentrations cα obey the balance equation
where J αi = cαvi + jαi is the tangential flux in the surface of molecule α, jα,i is the flux relative to the center of
mass, J α
n describes exchanges between the surface and its surrounding environment, and rα denote source and sink
∂tcα + ∇iJ αi + vnCi
icα = J α
n + rα,
(21)
terms corresponding to chemical reactions in the surface. Mass conservation implies the following relation between
fluxes of molecules and chemical rates
5
mαJ α
n = J ρ
n,
mαjα,i = 0,
mαrα = 0.
Xα
Xα
Xα
(22)
(23)
(24)
In the remaining of this work, summation over α is implicit. The conservation of energy and the balance of entropy
and free energy density have the form (Figure 2)
∂te + ∇iJ e,i + vnCi
∂ts + ∇iJ s,i + vnCi
∂tf + ∇iJ f,i + vnCi
ie = J e
n,
is = J s
n + π,
if = J f
n − T π − J i
s∇iT − ∂tT s,
(25)
(26)
(27)
where e and s are the energy and entropy density respectively, J e
n are energy and entropy fluxes entering
the surface from the adjacent bulk, J e,i and J s,i are tangential energy and entropy fluxes within the surface, and
J f
n = J e
n and J f,i = J e,i − T J s,i are the normal and tangential fluxes of free energy. The entropy production
rate within the surface is denoted π. Eq. 27 is obtained from the relation f = e − T s and Eqs. 25 and 26. In the
following, we consider for simplicity the isothermal case.
n − T J s
n and J s
B. Translation and rotation invariance
We now discuss relations between equilibrium tensions and torques implied by invariance of the surface properties
under a rigid translation or rotation.
1. Gibbs-Duhem relation
FIG. 3: Two Gibbs-Duhem relations for a fluid surface are obtained by considering a rigid translation of the surface by a
uniform infinitesimal vector δa or a rigid rotation with infinitesimal vector δθ. Coordinates on the new surface are obtained
by following the normal n of the original surface.
Using translation invariance of the free energy, we can derive a Gibbs-Duhem relation. We consider a infinitesimal
translation of the surface by a constant vector δa. The condition ∂iδa = 0 implies using Eq. A20
During translation, we reparametrize the new surface such that each point (s1, s2) moves normal to the original surface
on the new translated surface (Figure 3). Translation invariance then implies the relation (see Appendix F)
∇iδaj + Ci
jδan = 0,
∂iδan − Cij δaj = 0.
(28)
(29)
Eq. 30 is a covariant generalisation for surfaces of the Gibbs-Duhem relation for a three-dimensional multi-component
fluid [14, 15], with an additional term arising from the passive bending moment tensor.
∇j(cid:2)(f0 − µαcα)gi
j − K jkCik(cid:3) + Cik∇jK jk = −cα∂iµα.
(30)
2. Rotation invariance
6
We can derive a generalised Gibbs-Duhem relation describing torque balances using infinitesimal rotation described
by the pseudo vector δθ, such that the surface is deformed as:
The deformation defines a new surface X′ = X + δX, which is reparametrized such that (s1, s2) is constant along the
normal to the original surface. Rotation invariance then implies (see Appendix F):
δXα = ǫαβγδθβXγ.
(31)
K ijǫjkCi
k = 0.
(32)
implying that the tensor K ijCi
k is symmetric.
3. Equilibrium tensions and torques
The equilibrium tension and bending moment tensors can be obtained by calculating the change of free energy
under a surface deformation and using the expression of the virtual work Eq. 15 (Appendix H). The equilibrium
tension and bending moments are given by
¯tij
e = (f0 − µαcα)gij,
¯mij
e = K ij,
mi
n,e = 0 .
(33)
(34)
(35)
e,s = (f0−µαcα)gij−(Kk
with γ = f0−µαcα the bare membrane surface tension. Using Eqs 16 and 17, one also obtains the symmetric part of the
j.
e = K ikǫk
equilibrium tension tensor tij
iΓext,j. Using
Using the tangential torque balance equation 11 then yields the equilibrium tension ti
Eq. 32 and 35, the normal torque balance equation 12 yields the equilibrium antisymmetric part of the stress,
ǫijtij = −Γext
n .
symmetry relation 32, leads to the equilibrium condition relating chemical equilibrium gradients to external forces:
Combining the Gibbs-Duhem relation 30 and the tangential force balance given by Eq. 9, taking into account the
jC ki)/2 and the bending moment tensor mij
e,n = ∇j K ji − ǫj
iC kj +Kk
cα∂jµα = f ext
j −
1
2
ǫi
j(∂iΓext
n ) − Cij ǫk
iΓext,k
In the second line, the external force and torque surface densities derive from a potential U α(s1, s2, n) acting on
component α (Eqs. H10 and H11). Eq. 36 shows that one can then introduce the effective chemical potential
eff (s1, s2) = µα(s1, s2) + U α(s1, s2, n(s1, s2)), for which cα∂iµα
µα
The remaining normal force balance equation 10 then provides a shape equation for the equilibrium surface shape.
eff = 0.
= −cα(cid:2)∂jU α + Cij(∂U α/∂n) · ei(cid:3) .
(36)
C. Entropy production rate
We can now calculate the entropy production rate using the variation of the free energy and the Gibbs-Duhem
relation derived above. We consider a region of surface S enclosed by a fixed contour C, which can deform in 3
dimensions. The rate of change of the free energy F can be written as (see Appendix I):
dF
dt
= ZS
dS(cid:20) −(cid:16)tij − tij
+(∂iµα)jα,i + µαrα + µα
e (cid:17) vij − ( ¯mij − K ij)
totJ α
n(cid:0)∂iωn − Cij ωj(cid:1)
DCij
Dt − mi
n + f ext · v + Γext · ω(cid:21)
dlνi(cid:20) − f vi − µαjα,i + ti · v + mi · ω(cid:21),
+ZC
(37)
Flux
Force
In-plane shear tensor vij
Bending rate tensor DCij
Vorticity gradient (∂iω) · n
Dt
In-plane tension tensor ¯tij
d
In-plane bending moment tensor ¯mij
d
Normal moment mi
n
Diffusion flux ji,α
Chemical potential gradient −∂iµα
Chemical reaction rate rα
Chemical potential µα
7
TABLE I: List of pairs of conjugate thermodynamics fluxes and forces in a thin active surface.
where we have introduced the symmetric in-plane shear tensor vij , the rotational of the flow ω = ωiei + ωnn, and
the corotational derivative of the curvature tensor:
vij =
1
2
(∇ivj + ∇jvi) + Cij vn,
1
ǫij (∇ivj)n,
2
j + vk∇kCij + ωn(ǫi
ω = ǫij (∂jvn − Cjkvk)ei +
= −∇i(∂jvn) − vnCikC k
DCij
Dt
kCkj + ǫj
kCki)
(38)
(39)
(40)
Note that the in-plane shear tensor vij is the sum of a contribution from in-plane flows, equal to the symmetric part
of the covariant gradient of flow ∇ivj, and a contribution arising from normal flows vn, corresponding to in-plane
shear induced by the deformation of the surface in three-dimensions. The vorticity ω of the flow has a normal part
arising from the two-dimensional vorticity of the flow ǫij∇ivj /2, and a tangential part specific to curved surfaces. The
bending rate tensor DCij /Dt has the form of a corotational derivative, with the third term in Eq. 40 corresponding
to advection of the curvature, and the last two terms to a corotational term.
In Eq. 37, we have not included
contributions from the antisymmetric part of ¯mij . Note that the bending moment tensor can always be chosen to be
symmetric in the force balance equations, see Appendix I.
We can read off the entropy production rate in the surface per unit area from Eq. 37:
T π = tij
d vij + ¯mij
d
DCij
Dt
+ mi
n(cid:0)∂iωn − Cij ωj(cid:1) − (∂iµα)jα,i − µαrα,
(41)
ij
d = ¯tij − γgij and ¯mij
d = ¯mij − K ij are the dissipative part of the in-plane stress and bending moment tensor.
where t
The mechanical contribution to dissipation can be also understood starting from Eq. 15 using T πm = δWd/δt, where
δWd is the work done by dissipative forces, together with Eqs. E6 and E8. Note that the entropy production rate is a
sum of products of conjugate thermodynamic fluxes and forces, which all vanish at thermodynamic equilibrium. The
pairs of conjugate fluxes and forces are listed in Table I.
We now briefly discuss the conjugate fluxes and forces. The dissipative in-plane tension tensor tij
d is conjugate to
the in-plane shear rate vij , corresponding to the dissipative cost of introducing in-plane deformations in the surface.
The coupling between the in-plane dissipative bending moment mij
d and the bending rate tensor DCij /Dt arises only
for curved surfaces and is associated to the dissipative cost of changing the surface shape in three dimensions. The
coupling between the normal moment mi
n and the vorticity gradient of flow (∂iω)· n = ∂iωn− Cij ωj is a generalisation
to curved surface of a coupling which also arises for planar surfaces, and is associated to the dissipative cost of gradients
of rotations within the surface [16]. Finally, the two last terms in Eq. 41 correspond to couplings of the chemical
potential and its gradient to the rates of reactions and the flux of diffusion of species in the surface [15].
The flux of free energy entering the surface from the adjacent bulk reads
n = f ext · v + Γext · ω + µα
J f
totJ α
n ,
(42)
which corresponds to the sum of the mechanical power acting on the surface and of the influx of chemical energy in
the surface. The flux of free energy tangential to the surface reads:
where f vi is the advection of free energy, µαjα,i is the flux of chemical free energy, and the remaining terms describe
the mechanical power tangential to the surface at its boundaries.
J f,i = f vi − ti · v − mi · ω + µαjα,i
(43)
A
Up-down mirror symmetry
Normal plane mirror symmetry
8
Up-down rotation symmetry
Symmetry with respect to
rotation of pi around the normal
B
(i) Up-down symmetric,
(O)
non chiral surfaces
(ii) Surfaces with broken
up-down symmetry
(UD)
(iii) Chiral surfaces preserving up-down rotation symmetry
(C)
(iv) Planar-chiral surfaces with broken planar mirror symmetry ,
preserving up-down mirror symmetry
(PC)
(v) Surfaces with broken chiral and up-down symmetry
FIG. 4: Classification of surfaces with in-plane rotation symmetry. A. The surface state can change under up-down mirror
symmetry Mn, mirror symmetry Mt, up-down rotation symmetry Rt, and rotation by π around the normal Rn. The symmetry
Rn is not broken for a surface with in-plane rotation symmetry. B. Surfaces with in-plane rotation symmetry can be categorised
in 5 classes according to their symmetries. Schematics give examples of actual surfaces belonging to each category. Red and
green letters indicate respectively broken and preserved symmetries.
D. Mirror and rotation symmetries of surfaces
Constitutive relations describing the active surface must respect the symmetries satisfied by the surface [17]. We
therefore classify surfaces by asking whether the state of an element of surface is preserved under application of
symmetries (Fig. 4).
We restrict ourselves to surfaces with rotation symmetry in the plane. We then find that that 3 sets of discrete
symmetries can be associated to thin shells: up-down mirror symmetry Mn, mirror symmetry with respect to a plane
9
normal to the surface Mt, and up-down rotation symmetry Rt (Fig. 4A). Mt corresponds to a mirror symmetry by
a normal plane going along an arbitrary tangent vector t, Rt to a rotation of π around an arbitrary tangent vector
t. The corresponding transformations rules are given in Appendix G. Because inversion of space can be written
as the combination of Mn and the rotation of π around the normal Rn, inversion of space and Mn are broken or
preserved simultaneously for a surface with in-plane rotation symmetry. Furthermore, combination of two of the
symmetries Mn, Mt and Rt yield the third one, such that at least two of these symmetries must be broken. As a
result, surfaces can be classified into 5 different classes: (i) up-down symmetric, non-chiral surfaces (type 0) preserve
all three symmetries, (ii) non-chiral surfaces with broken up-down symmetry (type UD) preserve Mt but break Mn
and Rt, (iii) chiral surfaces with up-down rotation symmetry break all mirror symmetries Mt and Mn but preserve
Rt (type C) (iv) planar-chiral surfaces preserve up-down mirror symmetry Mn but break Mt and Rt (type PC) (v)
up-down asymmetric and chiral surfaces break Mn, Mt and Rt (Fig. 4). Note that we choose to denote surfaces
breaking Mt and not Mn planar-chiral surfaces because they break mirror-symmetry in the plane, but these surfaces
are not necessarily made of chiral molecules (Fig. 4B).
E. Constitutive and hydrodynamic equations
Using the conjugate thermodynamic forces and fluxes obtained from Eq. 41 and listed in Table I, we write a generic
linear response theory taking into account the symmetries of an active fluid surface. For simplicity, we consider that
a single chemical reaction occurs in the surface converting a fuel species F into a product species P . The fuel and
product species have the same mass. We denote ∆µ = µF − µP the difference of chemical potential between the field
and product species, r = −rF = rP the rate of fuel consumption and j = jF = −jP its flux. We also assume that no
chemical exchange exists between the membrane and its surrounding, such that the normal fluxes J α
n vanish.
In the linear response theory, we expand the tensors ¯tij
d , ¯mij
n, diffusion flux ji,α and chemical reaction rate rα
j/Dt, (∂iω) · n, chemical potential ∆µ and chemical potential
to linear order in the rates of deformation vij , DCi
gradient ∂i∆µ.
n and J ρ
d , mi
The stress and moment tensor can then be decomposed as
tij
d = tij
d = mij
mij
mi
n = mi
0 + tij
0 + mij
n0 + mi
UD + tij
C + tij
PC,
UD + mij
nUD + mi
C + mij
PC,
nC + mi
nPC,
(44)
0 is the part of the stress tensor that exists for any surface, tij
where tij
breaks up-down symmetry, tij
decomposition of the bending moment tensor and normal moment tensor.
C exist for chiral surfaces, and tij
UD correspond to terms present when the surface
PC for planar-chiral surfaces. Similar rules apply for the
To express constitutive equations for each of the components, we then write all possible terms of the expansion of
the generalised forces in the fluxes at linear order, and ask whether the corresponding terms break the symmetry Mn,
Mt, Rt according to the signatures given in Appendix G. The contributions to the stress tensor then read
gij + 2ζ C ij ∆µ + ζ ′Ck
kgij∆µ
Dt (cid:19) + ζC(cid:0)ǫi
kC kj + ǫj
kC ki(cid:1) ∆µ
(45)
kgij for the traceless part of a tensor A. The moment tensor
kvkj + ǫj
where we have introduced the notation Aij = Aij − 1
reads
tij
0 = 2ηvij + ηbvk
ij
UD = 2¯η
t
D C ij
Dt
+ ¯ηb
t
ij
C = ηC(cid:18)ǫik DCk
PC = ηPC(cid:0)ǫi
Dt
tij
j
k
kgij + ζgij ∆µ
DCk
Dt
+ ǫjk DCk
kvki(cid:1) ,
i
2 Ak
D C ij
Dt
mij
+ ηcb
0 = 2ηc
mij
UD = 2¯ηvij + ¯ηbvk
mij
kvkj + ǫj
j
C = −ηC(cid:0)ǫi
PC = ηcPC(cid:18)ǫik DCk
Dt
mij
kgij + ζcgij∆µ
kvki(cid:1)
+ ǫjk DCk
k
DCk
Dt
gij + 2 ζc C ij ∆µ + ζ ′
cCk
kgij∆µ
i
Dt (cid:19) + ζPC(cid:0)ǫi
kC kj + ǫj
kC ki(cid:1) ∆µ.
(46)
In Eq. 46, we have only introduced symmetric contributions to the bending moment tensor. The normal moment
reads
10
mi
mi
n0 = λ(∂iωn − C ij ωj) + χǫij∂j∆µ
nUD = 0
mi
nC = χCC ij ∂j∆µ
nPC = λPCǫij(∂jωn − Cjkωk) + χPC∂i∆µ.
mi
The rate of fuel consumption then reads
r = −(ζ + ζ ′Ck
−(ζc + ζ ′
cCk
k)vk
k)
k − 2ζ C ij vij − 2ζCǫi
Dt − 2 ζc C ij D Cij
DCk
k
kC kj vij
Dt − 2ζPCǫi
kC kj DCij
Dt
+ Λ∆µ.
and the fuel flux relative to the centre of mass is given by
(47)
(48)
(49)
ji = −L∂i∆µ +(cid:0)χǫij + χCC ij + χPCgij(cid:1) (∂jωn − Cjkωk).
η, ηb, ¯η, ¯ηb, ηc, ηcb, ηC, λ, Λ and L are dissipative couplings, ζ, ζ ′, ζ, ζc, ζc, ζ ′
c, χ, χC and χPC are reactive
couplings. The viscosities depend in general on the curvature tensor Cij ; here we have not taken this dependency
into account for simplicity. We have introduced terms proportionals to ηPC, ηcPC and λPC corresponding to odd or
Hall viscosities which do not contribute to dissipation. These are reactive coefficients, and the time signatures of the
constitutive equations imply that they change sign under time reversal, which could exist for example in the presence
of a magnetic field [18]. Active tensions and bending moments proportional to the difference of chemical potential
∆µ depend on the curvature tensor. In the constitutive equations 45-49, we have expanded these terms to first order
in the curvature tensor Cij . Although we have not written explicitly this dependency here, the phenomenological
coefficients also depend in general in the concentration fields cα. Positivity of entropy productions implies that the
viscosities η, ηb, ηc, ηcb, and λ are positive, however the up-down asymmetric viscosities ¯η, ¯ηb and chiral viscosity ηC
can be positive or negative.
In the equations above, the contribution to the two-dimensional stress ¯tij
0 is the generalisation for curved surfaces
of the generic hydrodynamic equations of a three-dimensional active gel [5]: η and ηb are respectively the planar shear
and bulk viscosity of the surface, and ζ∆µ is an active tension arising in the surface from active processes. Additional
viscous tensions proportional to ¯η, ¯ηb, and ηC arise for a curved surface due to the dissipative cost of changing the
surface curvature. We also find new active terms for the tension tensor of a curved surface proportional to ζ, ζ ′,
ζC , that depend on the curvature tensor Cij . In particular, anisotropic active stresses can arise in a curved surface
isotropic in the plane, due to the anisotropy of the curvature.
Active terms for the moment tensor introduced in Eq. 46 are specific to thin films and correspond to actively
induced torques in the film. The active torque ζc, arising in a surface with broken up-down symmetry, can induce
active bending of a flat surface.
Combining the constitutive equations 45-49, the force and torque balance equations 7 and 8, and the concentration
balance equations 21 yield dynamic equations for the surface shape, the velocity field on the surface v and the
concentration fields on the surface ck. While the constitutive equations obtained here are linear, the dynamics
equations for the surface shape are non-linear due to geometric couplings.
F.
Instabilities of a homogeneous active Helfrich membrane
In this section, we restrict ourselves to non-chiral surfaces with broken up-down symmetry and discuss low Reynolds
numbers where inertial terms can be neglected. Starting from a description of a passive surface with the Helfrich free
energy, we consider effects introduced by additional active terms.
1. General equations
A passive fluid membrane described by the Helfrich energy with membrane tension γ, bending modulus κ, gaussian
bending modulus κg and spontaneous curvature C0 has the equilibrium tension and bending moment tensor (Appendix
H)
¯mi
¯tij = (cid:0)γ + (κ + κg)(Ck
j = (cid:0)2(κ + κg)Ck
k)2 − 4κCk
k − 4κC0(cid:1) gi
kC0 − κgCl
j − 2κgCi
j.
kCk
l(cid:1)gij ,
(50)
(51)
Starting from such a passive fluid membrane, the constitutive relation for the tension and bending moment tensor
of an active surface reads, neglecting viscous terms for this discussion and only keeping terms to first order in the
curvature:
11
tij ≃ (cid:0)γ + ζ∆µ + (−4κC0 + ζ ′∆µ)Ck
c − ζc(cid:17) ∆µ(cid:17) Ck
¯mij ≃ (cid:16)(cid:16)2κ + 2κg +(cid:16)ζ ′
k(cid:1) gij + 2ζ∆µ C ij ,
k + (ζc∆µ − 4κC0)(cid:17) gij − 2(cid:16)κg − ζc∆µ(cid:17) C ij.
(52)
(53)
(54)
(55)
Introducing a surface tension renormalised by activity ¯γ = γ + ζ∆µ, and similarly the renormalized bending moduli
¯κg = κg − ζc∆µ, ¯κ = κ + ( ζc + ζ ′
c)∆µ/2 and spontaneous curvature ¯C0 = κC0/¯κ − ζc∆µ/4¯κ, one obtains
tij = (cid:0)¯γ + (−4¯κ ¯C0 + (ζ ′ − ζc)∆µ)Ck
¯mij = (cid:0)(2¯κ + 2¯κg) Ck
k − 4¯κ ¯C0(cid:1) gij − 2¯κgC ij .
k(cid:1) gij + 2ζ∆µ C ij ,
Two active terms proportional to ∆µ remain in the constitutive equation 54. Active terms therefore do not simply
renormalise the physical parameters of the Helfrich membrane, but introduce other physical effects. To clarify the
role of these terms, we discuss below simple surface geometries of active Helfrich membranes and show that they can
result in instabilities of a flat surface.
2.
Instabilities of a flat surface
Tension-curvature coupling
instability
Active
buckling
instability
FIG. 5: Phase diagram for the stability of an flat active Helfrich membrane with up-down asymmetry, as a function of the
active tension ζ∆µ and the active tension-curvature coupling term (ζ ′ + ζ − ζc)∆µ. For simplicity, we consider here the case
ζc = ζ ′
c = 0.
We consider here a flat, homogeneous and compressible membrane. We ignore here the surrounding medium and
the membrane is therefore free from external forces and torques. Perturbations of the flat shape are described in the
Monge gauge by the height h(x, y), such that the surface position is given by X(x, y) = xux + yuy + h(x, y)uz. We
take here for simplicity the bulk viscosities ηb = ηbc = 0 and we obtain the shape equation (Appendix J)
q4(cid:20)(ηc −
¯η2
η
)∂t + 2κ + ( ζc + ζ ′
c)∆µ +
¯η(ζc − ζ ′ − ζ)∆µ
η
(cid:21)h + q2(γ + ζ∆µ)h = 0,
(56)
where we have introduced the Fourier transform of the height, h(qx, qy) = 1
x + q2
y. The second law of thermodynamics imposes that ηcη > ¯η2. We find that the active flat surface undergoes
2π R dxdyh(x, y)e−i(qxx+qy y), and q =
qq2
shape instabilities for (Figure 5):
ζ∆µ < −γ,
¯η(ζ ′ + ζ − ζc)∆µ
η
> 2κ + (ζc + ζ ′
c)∆µ.
12
(57)
(58)
The first condition corresponds to the classical buckling instability occurring when active stresses are compressive and
establish a negative surface tension ¯γ = γ + ζ∆µ < 0 in the membrane.
In the second condition, the instability is favoured by negative values of ζc + ζ ′
c or positive values of ¯η(ζ ′ + ζ − ζc).
Negative values of ζc + ζ ′
c lower the effective bending modulus ¯κ. Positive values of ¯η(ζ ′ + ζ − ζc) induce an instability
coupling the membrane shape to tangential flows. This instability can be understood from the dependency of the
tension on curvature (Eqs. 16 and 45). Because of this dependency, a perturbation of the surface shape results
in regions of low and high surface tension, depending on the sign of the local mean curvature and the sign of the
coefficient ζ ′ + ζ − ζc which couples the tension tensor to the curvature tensor. Differences of surface tensions result
in flows towards region of higher surface tension. These flows generate further in-plane torques when the surface has
a non-zero up-down asymmetric viscosity ¯η or ¯ηb. A shape instability occurs when the sign of this additional torque
leads to further deformation of the surface.
IV. ACTIVE ELASTIC THIN SHELL
In addition to fluid surfaces, the formalism presented here can also be used for elastic surfaces. We discuss here
isotropic active elastic thin shells.
A. Hookean elasticity
We first write generic constitutive equations for a Hookean elastic shell. Rather than inferring tensions and moment
tensors from three-dimensional stresses, we directly obtain generic two-dimensional constitutive equations [19, 20].
We consider a surface with reference shape X and a deformation field u, such that the deformed surface has position
X′ = X + u. Using the differential virtual work expression Eq. 15, the change of virtual work induced by the
deformation field u reads to first order in the deformation field:
∆W ≃ZS
dS(cid:2)¯tij uij + ¯mij cij + mi
nΩi(cid:3)
(59)
where we have introduce the deformation tensors uij = ∆gij /2, cij = (gik∆Cj
ij /2,
and we have assumed as for the fluid case that ¯mij can be taken to be symmetric. The deformation tensors read to
first order in the deformation field:
k)/2 and Ωi = ǫj
k + gjk∆Ci
k∆Γk
uij =
1
2
(∇iuj + ∇jui) + Cij un,
cij = −∇i(∂jun) − unCikC k
Ωi =
j + (∇kCij )uk +
k∇juk) − Cij ǫjk(∂kun − Cklul).
1
2∇i(ǫj
1
2
ǫkl∇kul(ǫi
kCki + ǫj
kCki),
(60)
(61)
(62)
where we have used Eqs. B13, B17 and D10. We can then identify that the in-plane stress ¯tij , in-plane bending
moments ¯mij , and normal moments mn are conjugate to the deformation tensor uij, cij and Ωi. We can therefore
express Hookean elasticity by the following constitutive relations:
¯te
ij = Eijklukl + Gijklckl,
ij = Kijklukl + Fijklckl,
¯me
n = H ijΩj,
mi
(63)
(64)
(65)
where we have introduced the Hookean elastic moduli tensors E, F , G, K and H. For a shell in thermodynamic equi-
librium with free energy F , Gijkl = ∂2F/∂ckl∂uij = ∂2F/∂uij∂ckl = Kklij and Hij = ∂2F/∂Ωi∂Ωj = ∂2F/∂Ωj∂Ωi =
Hji. On an homogeneous elastic shell, the metric, curvature and Levi-Civita tensors can be used to define these
elasticity tensors. We therefore simplify the general relations above in the form
13
t
ij
kgij
kgij
tij
0 = E1uij + E2uk
UD = G1cij + G2ck
tij
tij
C = GC(cid:2)ǫi
PC = EPC(cid:2)ǫi
kckj + ǫj
kukj + ǫj
kcki(cid:3)
kuki(cid:3) ,
kgij
mij
0 = F1cij + F2ck
mij
UD = K1uij + K2uk
mij
mij
C = −KC(cid:2)ǫi
PC = FPC(cid:2)ǫi
kgij
kukj + ǫj
kckj + ǫj
kuki(cid:3)
kcki(cid:3) ,
mi
n0 = H1Ωi
mi
nUD = 0
nC = HCǫikCkj Ωj + ¯HCǫjkC kiΩj
mi
nPC = HPCǫijΩj.
mi
(66)
(67)
(68)
where we have decomposed the tension and bending moment tensors according to the symmetry class of the shell,
as in the fluid case (Eqs. 44). The coefficient Ek, Fk, Gk, Kk and Hk are elastic moduli, and we have written all
terms allowed by symmetry for an homogeneous isotropic elastic material, at lowest order in the curvature tensor
Cij . For an elastic shell at thermodynamic equilibrium, G1 = K1, G2 = K2, GC = KC , HC = ¯HC , as a result of
the tensor symmetries Gijkl = Kklij and Hij = Hji (see after Eq. 65). A linear shell theory for a homogeneous
elastic shell yields E1 = Eh/(1 + ν), E2 = Ehν/(1 − ν2), F1 = Eh3/(12(1 + ν)), F2 = Eh3ν/(12(1 − ν2)) and other
coefficients equal to zero, with E, ν the 3D elastic modulus and Poisson ratio of the shell material and h the thickness
of the shell [20, 21]. The elastic moduli EPC, FPC and HPC do not contribute to the work Eq. 15 and only exist for
non-equilibrium systems: they vanish for an elastic shell at equilibrium as they do not derive from a free energy.
B. Constitutive relations for an active elastic shell
For an active elastic shell, an active contribution to the tension and bending moment tensors can be added to the
elastic contribution in Eqs. 66-68:
¯ta
¯ma
ij = hζgij + ζ ′Ck
ij = hζcgij + ζ ′
cCk
kgij + 2ζ C iji ∆µ,
kgij + 2 ζc C iji ∆µ,
(69)
(70)
where we restrict ourselves here for simplicity to the case of an non-chiral surface with broken up-down symmetry.
In the expansions above, the terms proportional to ζ and ζc, which are to lowest order in the curvature, introduce
respectively an active tension and an active torque within the elastic shell.
We can perform a stability analysis of an elastic flat active surface, similar to the calculation of section III F 2 for
the fluid case (Appendix J). We find
ξ∂th = −ζ∆µq2h −(cid:20)F + ζ ′
c + ζc −
K
E (cid:16)G + (ζ ′ + ζ − ζc)∆µ(cid:17)(cid:21) q4h.
(71)
where we have introduced an effective external friction force normal to the surface, with friction coefficient ξ, the
Fourier transform of the height h, and the coefficients F = F1 + F2, K = K1 + K2, E = E1 + E2 and G = G1 + G2.
The elastic surface is then unstable for
ζ∆µ < 0
K(ζ ′ + ζ − ζc)∆µ
E
> F −
KG
E
+ (ζ ′
c + ζc)∆µ.
(72)
(73)
14
As for the fluid case (Eqs. 57-58), an instability can arise from active compressive stresses in the surface, or from
active couplings between tension and curvature.
The deformation induced by a gradient of active stress and torques in a cylindrical elastic shell have been discussed
in Ref. [22]. In this work, it was shown that deformation profiles depend on two characteristic lengths which depend
on the shell bending modulus, elastic modulus, cylinder radius and on the active tension acting within the shell.
V. DISCUSSION
We have developed a general, covariant theory for the dynamics of active surfaces. Starting from balances of forces
and torques, we have derived an expression for the virtual work. We have identified the entropy production on a
curved surface which generalises the entropy production of bulk fluids known from irreversible thermodynamics to
surfaces of arbitrary shapes [23]. Using this entropy production, we have identified conjugate fluxes and forces for
an active fluid membrane. Our approach can also be directly applied to the study of active elastic surfaces. Our
constitutive relations for active surfaces include the derivation of a fully generalized Hooke's law for elastic thin shells.
We have classified active surfaces in 5 different symmetry classes: (i) up-down symmetric, non-chiral surfaces, (ii),
non-chiral surfaces with broken up-down symmetry, (iii) chiral surfaces, (iv) planar chiral surfaces, and (v) chiral
surfaces with broken up-down symmetry. Classes (i) and (ii) have been characterised before. Chiral surfaces (iii)
must consist of chiral constituents and are up-down asymmetric, while planar-chiral surfaces (iv) do not have to be
built from chiral subunits and only appear chiral when viewed from one side (Fig. 4). The constitutive equations for
the surface have to obey these symmetries and coupling terms in the constitutive equation can be associated with
these symmetry classes.
We have here neglected some degrees of freedom of the surface, such as the local rotation rate of molecules Ωij which
relaxes rapidly to the vorticity flow ωij [16]. We have also identified the normal derivative of the surface deformation
with the rotation of the normal vector to the surface (Eqs. B21 and 14). This corresponds to neglecting a component
of the shear normal to the surface. This additional contribution could be taken into account by adding an additional
polar field tangential to the surface. We have considered the physics of an isolated surface, not taking into account the
environment and external forces. Furthermore, we have restricted ourselves to isotropic surfaces. It will be interesting
to expand the theory presented here to the case of active nematic or polar surfaces.
When the surface is embedded in a viscous fluid, external forces and torques acting on the surface arise from stresses
acting within the bulk fluid. The hydrodynamics of the 3D fluid and of the membrane are then coupled to each other.
It would be interesting to expand the theory obtained here to include these couplings between the surface and the
bulk fluid.
Our work is related to previous works on active membrane and membrane dynamics [10–12, 14, 24–28] as well as
on works on thin active films [29, 30]. We propose here a generic framework for active surfaces that captures many
aspects of the physics discussed in earlier works. In addition, we identify new active terms associated with internal
tensions and bending moments. In particular we show the existence of active torque terms that can induce curvature
changes.
Our general approach provides a framework for the study of complex morphological changes of active surfaces in
biology, for example during morphogenesis of an organism or the formation of complex cell shapes. We have introduced
a limited number of phenomenological parameters which capture the generic effects of a large variety of molecular
processes in cells and tissues. We expect in particular that biological processes such as tissue folding, invagination
and twisting can be captured by our theory [31, 32]. Fold formation could occur through apical constriction [13],
which corresponds to the establishment of a difference in apical and basal surface tension in an epithelium, resulting
in a gradient of active bending moment. Our theoretical framework provides a formalism to study how such gradients
can result in tissue folding. By quantifying forces and deformations in tissues, the phenomenological parameters we
introduce could be experimentally measured. Active tensions and bending moments could be related to the spatial
and temporal distribution of force-generating elements such as myosin molecular motors in a tissue, as has been done
to estimate active stresses distribution in the cell cortex [33–35].
In general, biological systems have both elastic and viscous properties that could be captured by a viscoelastic
generalisation of our theory. However, in many cases either elastic or viscous properties dominate: plant morphogenesis
is often described as an active elastic medium, while long-time behaviour of tissue flows during animal morphogenesis
can be captured by a viscous limit, on time scales where cells can rearrange their neighbours [36, 37]. It will be a
future challenge to find analytic and numerical solutions for the complex shape changes predicted by our theory.
Acknowledgments
15
We thank Jacques Prost, Andrew Callan-Jones and Marino Arroyo for critical reading of the manuscript, and H´el`ene
Berthoumieux, Karsten Kruse, Stephan Grill and Vijaykumar Krishnamurthy for useful discussions. G.S acknowledges
support by the Francis Crick Institute which receives its core funding from Cancer Research UK (FC001317), the UK
Medical Research Council (FC001317), and the Wellcome Trust (FC001317).
Appendix A: Differential geometry
We give here definitions of differential geometry quantities used in the text. We consider a two-dimensional surface
parametrized by two coordinates X(s1, s2). Two tangent vectors and a normal vector are associated to every point
on the surface, according to
e1 =
∂X
∂s1 , e2 =
∂X
∂s2 , n =
e1 × e2
e1 × e2
.
(A1)
Lower indices correspond to covariant coordinates and upper indices to contravariant coordinates. The metric gij and
curvature tensor Ci
j associated to X are defined by
j = Cikgkj . The inverse of the metric tensor gij = gij
gij = ei.ej , Cij = −(∂i∂j X).n,
−1 verifies
where Ci
The contravariant basis is defined by
gijgjk = δi
k.
ei.ej = δi
j,
(A2)
(A3)
(A4)
with ei = gij ej. Indices can be raised and lowered by contraction with the metric tensor according to ai = gijaj and
ai = gijaj for a tangent vector a = aiei = aiei.
The derivatives of the basis and normal vectors are given by the Gauss-Weingarten equations
where the Christoffel symbols Γk
ij ek,
j ej,
∂in = Ci
∂iej = −Cij n + Γk
ij are obtained from the metric by
The surface area element is denoted dS = √gds1ds2 where g = det(gij) is the determinant of the metric.
The Levi-Civita tensor on the curved surface can be defined as:
Γk
ij =
1
2
gkm [∂jgim + ∂igjm − ∂mgij] .
It is antisymmetric when expressed in a purely contravariant or covariant basis:
ǫij = n · (ei × ej).
Furthermore, it satisfies the identity
ǫij = √g 0 1
−1 0 ! , ǫij =
1
√g 0 1
−1 0! .
ǫijǫjk = −δk
i .
The Levi-Civita tensor can be used to express vectorial products of the basis vectors:
j ej,
n × ei = ǫi
ei × ej = ǫij n.
(A5)
(A6)
(A7)
(A8)
(A9)
(A10)
(A11)
(A12)
The second relation implies
a × b = ǫijaibj,
for two tangent vectors a and b.
A tensor with two indices can generally be decomposed into a symmetric and antisymmetric part:
Aij = As
ij
ij + Aa
1
2
ij +
= As
Aklǫklǫij.
16
(A13)
(A14)
(A15)
We denote ∂i = ∂/∂si and ∇i the covariant derivative, which has the property for a tangent vector a = aiei and
tensor t = tij ei ⊗ ej:
The definitions above then correspond to the following expressions:
∇iaj = (∂ia) · ej,
∇itjk = ej · (∂it) · ek,
For a general vector f = f iei + fnn, we have
∇ivj = ∂ivj + Γj
∇itjk = ∂itjk + Γj
ikvk,
iltlk + Γk
iltjl.
The curvature tensor satisfies the Mainardi-Codazzi equation [38]:
∂if =(cid:2)∇if j + Ci
jfn(cid:3) ej +(cid:2)∂ifn − Cijf j(cid:3) n.
∇iCjk = ∇jCik.
The curvature tensor also satisfies the identity
and the Gauss equation [38]
j − Ck
kδi
∇i(cid:2)C i
j(cid:3) = 0,
CikC k
j = Ck
kCij − gij det(Ck
l).
The covariant derivatives of the metric and of the Levi-Civita antisymmetric tensor vanish
∇igjk = 0,
∇iǫjk = 0.
The coordinates of the tangent vectors in the 3D space with cartesian euclidian basis uα are written
ei = ei,αuα
ei = ei
αuα.
The gradient of a vector field v(xα) in the 3D space can be evaluated on the surface X through
∂vβ
∂xα
= (∂ivβ)ei
α + (∂nvβ )nα,
(A16)
(A17)
(A18)
(A19)
(A20)
(A21)
(A22)
(A23)
(A24)
(A25)
(A26)
(A27)
(A28)
where ∂nv is the derivative normal to the surface. In particular, the curl of a vector field on the surface is given by
The divergence theorem on a curved surface can be expressed using the covariant derivative [8]:
(∇ × v)α = ǫαβγei
β∂ivγ + ǫαβγnβ∂nvγ.
ZS
dS∇if i =ZC
dlνif i,
(A29)
(A30)
where S is the surface enclosed by C, ν is a unit vector tangent to S, outward-pointing and normal to the contour C,
and dl is an infinitesimal line element going along the contour C. Eq. A30 results from the identity [38]
∂i√g = √gΓk
ki.
17
(A31)
Indeed, denoting s a coordinate going along the closed contour C in a trigonometric orientation around the normals
to the surface S, one obtains
ds1ds2∂i(√gf i)
ds√g(cid:20) ∂s2
f 1 −
∂s
∂s1
∂s
f 2(cid:21)
ZS
dS∇if i = ZS
= ZC
= −ZC
= ZC
ds
∂si
∂s
dlνif i,
ǫij f j
(A32)
where the second line results from the usual divergence theorem, the third line from Eq. A9, and the fourth line from
the relations
dl = dses,
es × n
ν =
es × n
∂si
∂s
= −
ǫi
j ej/es,
(A33)
(A34)
with es = ∂sX = (∂si/∂s)ei the vector tangent to the contour C.
Appendix B: Variation of surface quantities
We consider here that the surface X is modified to a new surface X′:
We derive here expressions for the perturbations of the associated differential geometry quantities. The tangent vector
variation reads
X′(s1, s2) = X(s1, s2) + δX(s1, s2).
(B1)
Using gij = ∂iX · ∂j X, one finds
Using n · ei = 0 and n · n = 0,
δei = ∂iδX.
δgij = (∂iδX) · ej + (∂jδX) · ei.
δn = −((∂iδX) · n)ei.
Using ei · ej = δi
j, resulting in ei · δej + δei · ej = 0,
δei = −((∂j δX) · ei)ej + ((∂iδX) · n)n.
Using Cij = −(∂i∂j X) · n and Ci
j = Cikgkj ,
δ Cij = −(∇i∂jδX) · n,
δCi
j = δ Cikgkj + Cikδgkj.
(B2)
(B3)
(B4)
(B5)
(B6)
(B7)
Note that we distinguish δ Cij = C ′
Using Γk
ij = (∂i∂j X) · ek,
ij − Cij and δCi
j = C ′
i
j − Ci
j, which are two different tensors, related by Eq. B7.
δΓk
ij = (∇i∂jδX) · ek − Cij (∂kδX) · n
(B8)
Separating δX into a tangent and normal part:
δX = δX iei + δXnn,
we obtain the expressions in terms of components of the shape perturbation:
jδX j)n,
jδXn)ej + (∂iδXn − Cij δX j)n,
δei = (∇iδX j + Ci
δei = −(∇jδX i + C ij δXn)ej + (∂iδXn − C i
δn = (−∂iδXn + Cij δX j)ei,
δgij = ∇iδXj + ∇j δXi + 2CijδXn,
δgij = −∇iδX j − ∇jδX i − 2C ijδXn,
δ√g =
δ Cij = −∇i(∂jδXn) + (∇jδX k)Cik + (∇iδX k)Ckj + (∇iCjk)δX k + δXnCikC k
j,
k)δX k − δXnCikC kj ,
δCi
δΓk
k(∂jδXn) − Cij (∂kδXn)
j = −∇i(∂jδXn) + (∇iδX k)Ck
ij = ∇i∇jδX k + (Cij C k
l − CjlCi
j − (∇kδX j)Cik + (∇iC j
k)δX l + Cj
k(∂iδXn) + Ci
√ggijδgij,
1
2
+∇iCj
kδXn.
18
(B9)
(B10)
(B11)
(B12)
(B13)
(B14)
(B15)
(B16)
(B17)
(B18)
In order to define the the normal derivative of an infinitesimal surface deformation, ∂nδX, we introduce material
coordinates for the points in the volume around the surface:
X(s1, s2, z) = X(s1, s2) + zn,
(B19)
with z a coordinate going along the normal to the surface. When the surface is deformed with infinitesimal vector
deformation δX, we assume that the volume around the surface is deformed by
This choice implies that only in-plane shear occurs. We then obtain
δX(s1, s2, z) = δX(s1, s2) + zδn.
∂zδX = δn = −(∂iδX · n)ei.
(B20)
(B21)
We identify ∂nδX with ∂zδX in Eq. 14. This choice is equivalent to assume that points along the normal to the initial
surface before deformation are along the normal to the new surface after deformation.
Appendix C: Force balance derivation
We discuss here the force and torque balance for an element of surface. We consider a force balance equation taking
into account the contribution of mass accretion or ejection from the surface. For simplicity, we assume here that mass
accretion or ejection occurs only on one side of the surface. Applying the law of Newton on a surface region S of
contour C yields
∂t(cid:18)ZS
dSρv(cid:19) = ZS
dSJ ρ
n(v + u) +IC
dlνiti +ZS
dSf ext
0
,
(C1)
where the second term arises from the change of momentum due to mass being absorbed by the surface with velocity
u relative to the surface, and the third term arises from the force acting on the surface S from the surface outside of
S. f ext
is the external stress acting on the surface in addition to the momentum of incoming molecules. The flux of
mass towards the surface J ρ
n is introduced in Eq. 20. The surface momentum rate of change can be rewritten using
Eqs. E15, 20, and the divergence theorem A30:
0
∂t(cid:18)ZS
dSρv(cid:19) = ZS
= ZS
= ZS
dνiviρv
kρv(cid:3) +IC
dS(cid:2)∂t(ρv) + vnCk
dS(cid:20)ρ∂tv − ∇i(ρvi)v + J ρ
nv + ∇i(ρviv)(cid:21)
nv(cid:21),
dS(cid:20)ρ(∂tv + vi∇iv) + J ρ
(C2)
such that the force balance equation C1 can be rewritten
ZS
dSρa =IC
dlνiti +ZS
dSf ext,
where we have introduced the total external force f ext = f ext
0 + J ρ
nu , and the acceleration a is defined by
with d/dt = ∂t + vi∇i the convected derivative. Using the divergence theorem A30, Eq. C3 can be rewritten
a =
dv
dt
19
(C3)
(C4)
(C5)
Because this equation has to be valid for any surface element, this results in Eq. 7, which can also be written in the
form of a local conservation of momentum:
ZS
dS(cid:2)ρa − ∇iti − f ext(cid:3) = 0.
∂t(ρv) = ∇i(ti − ρviv) + f ext
0 − ρCk
kvnv + J ρ
n(v + u).
(C6)
Here the three last terms arise from exchange of momentum normal to the surface.
Ignoring the moment of inertia tensor for simplicity, the total torque acting on on a surface region S of contour C
vanishes:
IC
dlνimi +IC
dlX × νiti +ZS
dSX × (f ext − ρa) +ZS
dSΓext = 0,
(C7)
where Γext is the external torque density acting on the surface. Using the divergence theorem and the force balance
equation 7, the torque balance equation can be rewritten
which results in the torque balance expression Eq. 8.
ZS
dS(cid:2)∇imi + ei × ti + Γext(cid:3) = 0
We note that the force balance equations 9-12 are invariant under the variable transformation
kC kj ,
n + ǫij(∂j m),
tij → tij + mǫi
ti
n → ti
mij → mij + mgij.
(C8)
(C9)
(C10)
(C11)
where m is an arbitrary function on the surface, and we have used equations A21 and A23.
Appendix D: Differential work
The virtual work defined in Eq. 13 can be rewritten using the divergence theorem on a curved surface A30 and the
force and torque balance equations 7 and 8:
δW = ZS
dS(cid:20)ti · ∂iδX +
1
2
mi · ∂i(∇ × δX) +
1
2
(ti × ei) · (∇ × δX)(cid:21).
Projecting ti and mi along the tangent and normal directions and using Eqs. A11-A12, one finds
δW = ZS
−
1
2
dS(cid:20)tij ej · ∂iδX + ti
tijǫij n · (∇ × δX) +
1
2
ti
nǫij ej · (∇ × δX)(cid:21).
nn · ∂iδX +
1
2
mij ej · ∂i(∇ × δX) +
1
2
mi
nn · ∂i(∇ × δX)
(D1)
(D2)
Using the definition of the curl operator Eq. 14, the relations A11 and A12, the expression of the normal derivative
of the displacement B21, the variation of the curvature tensor B6 and of the Christoffel symbols B8, the following
identities can be obtained:
20
n · (∇ × δX) = ǫi
j∂iδX · ej
ei · (∇ × δX) = 2ǫij∂jδX · n
ej · ∂i(∇ × δX) = (cid:2)2Ci
n · ∂i(∇ × δX) = ǫj
lǫj
kδΓk
ij.
k + Cij ǫkl(cid:3) ∂kδX · el − 2ǫj
kδ Cik
Using these relations, the virtual work can be rewritten
δW = ZS
dS(cid:20)(cid:18)tij −
1
2
tklǫklǫij + mklCk
jǫl
i +
1
2
mklCklǫij(cid:19)∂iδX · ej − mijǫj
kδ Cik +
1
2
mi
nǫj
kδΓk
ij(cid:21).
(D3)
(D4)
(D5)
(D6)
(D7)
Using the in-plane torque tensor introduced in Eq. 17 ¯mij = −mikǫk
the expression for the variation of the metric B3, one finds
j (with inverse relation mij = ¯mikǫk
j) and using
δW = ZS
dS(cid:20) 1
2
(tij
s −
1
2
( ¯mkiCk
j + ¯mkjCk
i))δgij + ¯mij δ Cij +
Using Eq. B7 leads to the alternative expression of the virtual work
δW = ZS
dS(cid:20) 1
2
(tij
s +
1
2
( ¯mkiCk
j + ¯mkj Ck
i))δgij + ¯mi
jδCi
j +
which leads to Eq. 15 with the definition 16.
1
2
1
2
mi
nǫj
kδΓk
ij(cid:21).
mi
nǫj
kδΓk
ij(cid:21),
(D8)
(D9)
The deformation term in factor of mi
n is a generalisation to curved surface of the gradient of rotations [16]. This
can be seen from its explicit expression in term of the deformation coordinates:
where we have used Eq. B18.
ǫj
kδΓk
ij = ∇i(ǫj
k∇jδX k) − 2Cij ǫjk(∂kδXn − CklδX l),
(D10)
Appendix E: Eulerian and Lagrangian representation of surface flows
1. Lagrangian representation
In a Lagrangian representation, the parameters s1 and s2 label the center of mass of a specific volume element. The
surface is characterised by the time-dependent parametrisation X(s1, s2, t). The center-of-mass velocity is given by
v = ∂tX(s1, s2, t).
(E1)
The mass density conservation equation without exchange between the surface and its environment reads in La-
grangian coordinates
∂tρ + ρ(∇ivi + Ci
ivn) = 0.
This can be seen from the conservation of mass of a region of surface S:
dSρ(s1, s2)(cid:19) = 0
ds1ds2 (ρ∂t(√g) + √g∂tρ) = 0
ZS
i(cid:1) ρ + ∂tρ(cid:1) = 0.
dS(cid:0)(cid:0)∇ivi + vnCi
dt(cid:18)ZS
ZS
d
which leads to Eq. E2. In this derivation, we have obtained d√g/dt by setting δX = vdt in Eq. B15.
(E2)
(E3)
(E4)
(E5)
The rate of change of metric in Lagrangian coordinates
∂tgij = ∇ivj + ∇jvi + 2vnCij
21
(E6)
obtained by setting δX = vdt in Eq. B13, relates to the gradient of flow defined in Eq. 38 through ∂tgij = 2vij.
Similarly, the rate of change of the curvature tensor in Lagrangian coordinates, obtained by setting δX = vdt in
Eq. B17, defines a convected Lagrangian derivative ¯D/ ¯Dt of the curvature tensor:
j
¯DCi
¯Dt
= −∇i(∂jvn) − vnCikC kj + (∇iC j
k)vk + (∇ivk)Ck
j − (∇kvj)Ci
k
(E7)
and its symmetric part is introduced in Eq. 40. The rate of change of the Christoffel symbols is related to the gradient
of rotations introduced in Eqs. 37 and 39 through the identity
where we have used Eq. B18.
1
2
ǫj
k
¯DΓk
ij
¯Dt
= ∇iωn − Cijωj = (∂iω) · n,
2. Eulerian coordinates
In Eulerian coordinates, the center-of-mass velocity field is given by
v = viei + vnn,
where viei is the tangential velocity field, and the normal velocity field is given by
In addition, one requires the condition
vn = (∂tX(s1, s2, t)) · n.
(∂tX(s1, s2, t)) · ei = 0.
such that coordinates do not change when the surface is not deforming. Here, s1 and s2 do not describe a specific
material element.
In the Eulerian perspective, the time derivative of the tangent vectors, normal, metric, surface element area and
curvature are given by
(E8)
(E9)
(E10)
(E11)
(E12)
(E13)
(E14)
(E15)
(E16)
(E17)
(E18)
(E19)
(E20)
j ej + (∂ivn)n
∂tei = vnCi
∂tn = −(∂ivn)ei
∂tgij = 2vnCij
∂t√g = √gvnCi
∂tCi
i
where we have used Eqs. B10, B12, B13, B15 and B17 with δX i = 0 and δXn = vndt.
Mass conservation without exchange between the surface and its environment has the form
j = −∇i(∂jvn) − vnCikC kj
which follows from the mass conservation of an element of surface S with fixed contour C:
ivn = 0,
d
∂tρ + ∇i(ρvi) + ρCi
dSρ(cid:19) = −IC
dt(cid:18)ZS
dlν.vρ = −ZS
ds1ds2 ((∂t√g)ρ + √g∂tρ) = −ZS
ZS
dS∇i(viρ)
dS∇i(viρ)
i(cid:1) ρ + ∂tρ + ∇i(ρvi)(cid:1) = 0.
ZS
dS(cid:0)(cid:0)vnCi
which leads to Eq. E17.
The acceleration reads (Eq. C4),
a = ∂tv + vi∂iv
where we have used Eqs. E12 and E13.
= (cid:2)∂tvi + vj∇jvi + 2vnvjCj
i − vn∂ivn(cid:3) ei +(cid:2)∂tvn + 2vi∂ivn − vivj Cij(cid:3) n,
22
Appendix F: Translation and rotation invariance
We derive here relations for the stress and torque tensor of a fluid surface, obtained from the invariance of the free
energy under rigid translation and rotations of the surface. We consider for this derivation a surface in the absence
of external forces. The fluid surface contains N species α = 1...N with concentration cα and its free energy density is
given by Eq. 19. The deformation by an infinitesimal rigid translation or rotation defines a new surface X′ = X + δX.
The new surface is then reparameterized by new coordinates, such that a point (s1, s2) on the initial surface finds its
new position on the new surface by going along the normal to the initial surface (Figure 3):
X′′(s1, s2) = X(s1, s2) + δX · n .
1.
Invariance by translation
(F1)
We consider here a rigid translation of the surface by an infinitesimal uniform vector δa, implying ∂iδa = 0 and
the relations 28-29. With the choice of coordinates F1, the concentration, density and velocity fields on the surface
are modified only by the tangential contributions of displacement:
δcα = −∂icαδai,
1
2
δ(ρv2) = −
1
2
∂i(ρv2)δai
(F2)
(F3)
The geometric quantities on the new surface can be obtained by using Eqs. B15 and B17, with the normal displacement
F1, and using Eqs. 28 and 29:
The variation of surface free energy after the rigid translation must vanish, and is given by
δ√g = √gCi
k = −∇iCj
δCj
iδan
kδai
dνi(f δai)
δF = ZS
= ZS
= δaiZS
= δaiZS
k + δf(cid:3) +IC
k − µα∂icαδai −
dS(cid:2)f δanCk
dS(cid:20)f δanCk
dS(cid:2)−(µα∂icα + K jk∇iCjk) + ∂if0(cid:3)
dS(cid:20)∂iµαcα + (∇jK jk)Cik + ∇j((f0 − µαcα)gi
= 0,
1
2
∂i(ρv2)δai − K jk∇iCjkδai + ∇i(f δai)(cid:21)
j − K jkCik)(cid:21)
(F4)
(F5)
(F6)
where we have used the expression of the differential of the free energy density 19 at constant temperature, Eq. 28,
and the Mainardi-Coddazi equation A21.
Because Eq. F6 is valid for any surface element and any infinitesimal vector δa, we obtain Eq.30.
2.
Invariance by rotation
We now consider a uniform rotation of the surface with vector δθ, such that the surface is deformed by δX = δθ×X.
One can verify that the following identity holds for such a deformation:
As for a rigid translation, the concentration, density and velocity fields on the surface are modified only by tangential
contributions of displacements. One finds then
∇iδX i = −Ci
iδXn.
(F7)
δc = −(∂icα)δX i
1
2
δ(ρv2) = −
1
2(cid:0)∂i(ρv2)(cid:1) δX i
(F8)
(F9)
As for translations, changes in geometric quantities can be obtained from Eqs. B15 and B17, with the normal
displacement F1:
23
j = −(∇kCi
δCi
δ(√g) = √gCi
iδXn
j)δX k − δωjkCki − δωikC kj
with δωij = ǫij( 1
(F10)
(F11)
(F12)
(F13)
dνi(f δX i)
1
k + δf(cid:3) +IC
k − µα(∂icα)δX i −
2∇kδXlǫkl) = ǫijδθ · n. The associated variation of free energy reads
δF = ZS
dS(cid:2)f δXnCk
dS(cid:20)f δXnCk
= ZS
−K ijδωjkCi
dS(cid:20) − (µα∂icα + K jk∇iCjk)δX i + (∂if0)δX i − δθ · n(cid:2)K ijǫjkCi
= ZS
= −δθ · nZS
= −2δθ · nZS
2(cid:0)∂i(ρv2)(cid:1) δX i − K ij(∇kCij )δX k
j + ∇i(f δX i)(cid:21)
dS(cid:2)K ijǫjkCi
dSǫjkK ijCi
k − K ijδωikC k
= 0,
k + K ijǫikC k
j(cid:3)
k
k + K ijǫikC k
j(cid:3)(cid:21)
where we have used Eqs. 30 and F7.
Because Eq. F13 is valid for any surface element and any infinitesimal rotation vector δθ, this results in Eq. 32.
Appendix G: Up-down asymmetry, chirality and planar-chirality of surfaces
We discuss here the symmetries of a surface with rotational symmetry in the plane. The symmetries that can be
broken for to these surfaces are the up-down mirror symmetry (Mn), the mirror symmetries in the plane (Mt, a mirror
symmetry by a plane going along an arbitrary tangent vector t), and the up-down rotation symmetries (Rt, a rotation
of π around an arbitrary tangent vector t). Rotations around the normal with angle π, Rn preserve the state of a
surface with rotational symmetry in the plane. Composition of these symmetries are indicated in the multiplication
table II.
Mn Mt
Rt
Mn 1
Rt Mt
Rn
I
Mt Rt
1 Mn MtRn
Rt Mt Mn
1 RtRn
Rn
I MtRn RtRn
1
TABLE II: Multiplication table of discrete symmetries. 1 denotes the identity operation, I denotes inversion of space.
Because inversion of space can be written as a composition of the up-down mirror symmetry and the rotation Rn,
I = RnMn, inversion of space and up-down mirror symmetry are preserved and broken simultaneously for a surface
with rotation symmetry in the plane.
Under these symmetries, the stress, torque, curvature, Levi-Civita tensor, as well as vectors and pseudo vectors are
modified. We list in Table III the signatures of transformations of components of these tensors under the symmetries
introduced above. Additional transformations arise from the combinations MnRn, MtRn, and RtRn, which are not
relevant to discuss symmetry properties of our equations.
The force and torque balance equations 9-12 are invariant under these transformations. One can further verify that
transformations under Rn preserve all the constitutive equations 45-49.
24
1
tij
ti
n
mij
mi
n
Cij
ǫij
vi
vn
Symmetry Mn Mt Rt Rn
1
1
1 -1 -1
-1 1
1
-1 -1 -1
1 -1 1
-1 -1 1
1
1 -1
1 -1 1
1
1 -1
1
1
1
-1 1 -1
-1 -1 1
1
-1
-1
1
-1
1
1
-1
1
1
-1
1
∇i
vij
ωi
ωn
TABLE III: Signature of symmetries on vector and tensor fields of the surface.
Appendix H: Equilibrium tension and moment tensors, external force and torque surface densities for a fluid
membrane
Using the expression of the virtual work given in Eq. 15, we obtain in this appendix the equilibrium tension tensor
and moment tensor for a fluid membrane, first for the generic case, and then for the specific case of an Helfrich
membrane. We then obtain the external force and torque surface densities induced by an external potential acting
on the surface.
1. Tension and moment tensors for a generic equilibrium fluid membrane
We start here from a fluid membrane with a free energy density given by Eq. 18, such that the free energy for
a region of surface S is given by F = RS dSf0 with df0 = µαdcα + K i
j − sdT . We now calculate the change
of free energy following a change of shape of the surface. A change of the surface metric results in a dilution of
the concentrations, according to δcα/cα = −δ(√g)/√g. As a result and using Eq. B15, the free energy differential
jdCi
following a shape change reads
δF =ZS
dS(cid:20)(f0 − µαcα)gij δgij
2
+ K i
jδCi
j(cid:21)
(H1)
At equilibrium, inertial forces vanish and we have δW = δF for infinitesimal deformations. Using Eq. 15, one can
then identify the equilibrium tensors ¯tij
e , ¯mij
e , mi
e,n given in Eqs. 33-35.
2. Tension and moment tensors for a Helfrich membrane
The Helfrich free energy functional for a region of surface S of a fluid membrane reads:
F = ZS
dS(cid:20)γ + κ(cid:16)(cid:0)Ci
i(cid:1)2
− 4C0Ci
i(cid:17) + 2κg det(Ci
j)(cid:21),
where γ is the surface tension, κ is the bending rigidity, C0 is the spontaneous curvature and κg the gaussian bending
modulus. Using relation Eq. 15 and the relation δW = δF for infinitesimal deformations, one finds for the in-plane
stress tensor and bending moment tensor:
(H2)
(H5)
(H6)
where we have used det(Ci
stress tensor tij and bending moment tensor mij are then given by
k)2 − CikC ki(cid:3) by taking the trace of Eq. A23, and Eq. B15. Note that the
k)2 − 4C0Ck
k(cid:1) + 2κg det(Ck
kgij − 4κC0gij − 2κgC ij ,
l)(cid:1)gij ,
(H3)
(H4)
¯tij
¯mij
j) = 1
e = 2(κ + κg)Ck
e = (cid:0)γ + κ(cid:0)(Ck
2(cid:2)(Ck
e = γgij − 2κ(Ck
tij
mij
e = 2(κ + κg)Ck
k − 2C0)C ij + κCk
kǫij − 4κC0ǫij − 2κgC ikǫkj,
k(Ck
k − 4C0)gij,
δU = ZS
= ZS
dScα(cid:2)(∂iU α)δX i + (∂U α/∂n) · δn(cid:3) ,
dScα(cid:2)(∂iU α)δX i +
1
2
((∂U α/∂n) · ei)ǫikek · (∇ × δX)(cid:3).
(H8)
25
where we have used the identity A23. The stress tensor tij therefore does not depend on the gaussian bending modulus
κg. Furthermore, from the force balance equation 11 and because of Eq A22, the normal shear stress tn is given in
the absence of external torque by
tj
n = ∇i ¯mij
e = 2κ∇jCk
k,
(H7)
which also does not depend on the Gaussian bending modulus. Eqs. H5 and H7 are in accordance with Ref. [8], with
an opposite sign convention for the force density t.
Note that while mij
e depend on κg, terms proportional to κg cancel when using force balance equations 7 and
8. Therefore, in accordance with the Gauss-Bonnet theorem, the Gaussian bending modulus only enters boundary
conditions when solving the force balance equations to find the the surface shape.
3. External force and torque density for an equilibrium fluid membrane
If molecules in the surface are subjected to an external potential U = cαU α(s1, s2, n), where U α acts on component
α, the variation of this external potential induced by a deformation of the surface element S reads
where we have used the identities B4 and D4. The contribution of external forces and torques to the virtual work
given in Eq. 13 on the other hand reads, ignoring here inertial terms,
δWext =ZS
dSf ext · δX +
1
2
Γext · (∇ × δX)).
Using δWext = −δU , one then obtains the external surface force density and external surface torque density:
f ext = −cα∂iU αei,
Γext = −cα((∂U α/∂n) · ej)ǫjiei.
(H9)
(H10)
(H11)
Appendix I: Entropy production rate for a fluid surface
We derive here the entropy production rate for a region of a fluid surface S enclosed by a fixed contour C. The time
derivative of the free energy of the surface F is
dF
dt
=ZS
dS(cid:20) 1
2
∂t(ρvivi + ρ(vn)2) + K i
j∂tCi
j + µα∂tcα(cid:21) +ZS
dSvnCi
i(cid:20) 1
2
ρv2 + f0(cid:21) ,
(I1)
Using the following relation obtained from Eq. E14 ∂tvi = ∂t(gij vj) = gij∂tvj + 2vnCij vj, as well as the mass balance
equation 20, one obtains
1
2
∂t(ρvivi + ρ(vn)2) +
1
2
ρvnCi
iv2 = ρaivi + ρanvn −
1
2∇i(ρv2vi) +
1
2
nv2,
J ρ
(I2)
where we have used the expression of the acceleration a obtained in Eq. E20. Using then the force balance equation
7 and the concentration balance equation 21, we find
dF
dt
=ZS
dS(cid:20) −
+ K i
j∂tCi
1
2
1
2∇i(ρviv2) +
j + (f0 − µαcα)Ci
nCj
nv2 + ∇jtjivi + tj
J ρ
ivn − µα∇i(cαvi + jα,i) + µα(J α
ivi + f ext,ivi + ∇iti
n + rα)(cid:21).
nvn − tijCij vn + f ext
n vn
Using the divergence theorem A30, this can be rewritten
dF
dt
=ZS
dS(cid:20) − tji∇j vi + tj
nCj
ivi − ti
+ (∂iµα)(cαvi + jα,i) + µα (J α
n + rα) +
n∂ivn − tij Cijvn + f ext · v + (f 0 − µαcα)Ci
ρv2vi + tij vj + ti
J ρ
nv2(cid:21) +IC
dνi(cid:20)−
1
2
1
2
ivn + K i
j∂tCi
j
nvn − µα(cαvi + jα,i)(cid:21) .
(I3)
(I4)
Using the Gibbs-Duhem equality 30 and the balance of fluxes 22 and 24,
dF
dt
n(Ci
dS(cid:20) − tij∇ivj − tijCij vn + ti
=ZS
+ viKjk∇iC jk − (∂i(f0 − µαcα))vi + (∂iµα)jα,i + (µα +
+IC
ρv2vi + ti · v − µα(cαvi + jα,i)(cid:21) .
dsi(cid:20)−
jvj − ∂ivn) + f ext · v + (f0 − µαcα)Ci
n + µαrα(cid:21)
mαv2)J α
1
2
1
2
ivn + K i
j∂tCi
j
Reorganizing, performing an integration by part, introducing the convected derivative of the curvature tensor,
and using the total chemical potential µα
tot = µα + 1
j
dCi
dt
j,
= ∂tCi
j + vk∇kCi
2 mαv2, one finds:
dF
dt
dS(cid:20) −(cid:2)tij − (f0 − µαcα)gij(cid:3) ∇ivj − tij Cij vn + ti
j
n(Ci
ivn + K i
j
+ (∂iµα)jα,i + µα
dCi
dt
=ZS
+ f ext · v + (f0 − µαcα)Ci
+IC
dsi(cid:20)−
1
2
ρv2vi + ti · v − µαjα,i − f0vi(cid:21) .
jvj − ∂ivn)
n + µαrα(cid:21)
totJ α
26
(I5)
(I6)
(I7)
Using the torque balance equation 8, splitting the tension tensor tij into a symmetric and an antisymmetric part, and
introducing the equilibrium tension tensor ¯tij
dF
dt
=ZS
dS(cid:20) − (tij
s − ¯tij
e )vij −
+ K i
j
j
dCi
dt
+ f ext · v + (∂iµα)jα,i + µα
1
2
e = (f0 − µαcα)gij , we obtain
tklǫklǫij∇ivj + (∇k ¯mki − Cl
kml
n + µαrα(cid:21) +IC
totJ α
jvj − ∂ivn)
i)(Ci
nǫki − Γext,kǫk
dlνi(cid:2)−f vi + ti · v − µαjα,i(cid:3) ,
where we have introduced the symmetric velocity gradient vij defined in Eq. 38 and used the symmetry of the
curvature tensor, ǫij C ij = 0, and the symmetry of the tensor KC implied by rotational invariance (Eq. 32). Using
the torque balance equation 12, performing an integration by part, and using the definition of the vorticity of the flow
(Eq. 39),
dF
dt
=ZS
dS(cid:20) − (tij
+ mj
nCj
iωi + K i
s − ¯tij
e )vij + (∇imi
dCi
dt
j
j
n − Cij ¯mikǫk
j + Γext
n )ωn − ¯mki∇k(Ci
jvj − ∂ivn)
+ f ext · v + Γext,iωi + (∂iµα)jα,i + µα
totJ α
n + µαrα(cid:21)
+IC
dlνi(cid:2)−f vi + ti · v + ¯mik(Ck
jvj − ∂kvn) − µαjα,i(cid:3) .
Rearranging and performing an integration by part,
j − (∇kvj)Cik(cid:3)
(I8)
(I9)
=ZS
− mi
+IC
dF
dt
dS(cid:20) − (tij
s − ¯tij
e + Ck
j ¯mki)vij − ¯mi
n∂iωn + mi
nCi
jωj + K i
j
j
dCi
dt
dlνi(cid:2)−f vi + ti · v + mijωj + mi
In Eq.
¯DCi
j/ ¯Dt, defined in Eq. E7:
j(cid:2)−∇i(∂jvn) − CikC jkvn + vk∇iC jk + (∇ivk)C k
+ f ext · v + Γext · ω + (∂iµα)jα,i + µα
nωn − µαjα,i(cid:3) .
totJ α
n + µαrα(cid:21)
I9, the term in factor of ¯mi
j corresponds to the Lagrangian convected derivative of the curvature tensor,
j
¯DCi
¯Dt
=
j
dCi
dt
+ (∇ivk)Ck
j − (∇kvj)Cik,
(I10)
where we have used the Mainardi-Coddazzi equation A21. We define the bending rate tensor as the symmetric part
of this tensor:
27
whose explicit expression is given in Eq. 40. In addition, one can verify that K i
j ¯DCi
j/ ¯Dt = K i
jdCi
j/dt; indeed
DCij
Dt
=
1
2(cid:18)gjk
k
¯DCi
¯Dt
+ gik
¯DCj
k
¯Dt (cid:19) ,
(I11)
K i
j(cid:2)(∇ivk)C kj − (∇kvj)C ki(cid:3) = ∇ivj(cid:2)K ikC j
k − K kjC i
k(cid:3) = 0,
as a result of the invariance by rotation, Eq. 32, and the symmetry of K ij. Using these relations and the symmetry
of the tensor K ij, we then find the expression for the rate of change of free energy:
dF
dt
dS(cid:20) − (tij − tij
= ZS
+f ext · v + Γext · ω + (∂iµα)jαi + µα
ǫln ¯mln(ǫikCk
e +
1
2
j + ǫjkCk
i))vij − ( ¯mij − K ij)
totJ α
n + µαrα(cid:21) +IC
DCij
Dt − mi
jωj(cid:3)
dlνi(cid:2)−f vi + ti · v + mi · ω − µαjαi(cid:3) .
n(cid:2)∂iωn − Ci
(I12)
(I13)
In Eq. 37, we have not included the contribution ǫln ¯mln of the antisymmetric part of ¯mij. The antisymmetric part
of ¯mij is related to the trace of mij through the relation mk
k = − ¯mijǫij . The transformation invariance in Eqs. C9-
C11 implies that the contribution of the antisymmetric part of the bending moment tensor ¯mij to the force balance
equation can be absorbed in a redefinition of the stress tensor.
Appendix J: Stability of an homogeneous flat active surface
We discuss here the stability of a homogeneous flat active surface, in the absence of external forces and torques.
Perturbations of the flat shape are described in the Monge gauge by the height h(x, y) such that the surface position
is given by X(x, y) = xux + yuy + h(x, y)uz. Calculations are performed for a weakly bent surface, ∂ih ≪ 1, at
linear order in the height h and velocity v. In this limit, covariant and contravariant indices can be used indifferently,
and
(∂ivj + ∂jvi)
gij ≃ δij
Cij ≃ −∂i∂jh
vij ≃
ǫij∂ivj
ωn ≃
ωi ≃ ǫij ∂jvn.
1
2
1
2
For a fluid surface, the tensions and torque tensors are given by
a. Fluid surface
tij = η(∂ivj + ∂jvi) + (ηb − η)∂kvkδij − 2¯η∂i∂j∂th − (¯ηb − ¯η)(∂t∆h)δij
¯mij = −2ηc∂i∂j ∂th − (ηcb − ηc)(∂t∆h)δij + ¯η(∂ivj + ∂jvi) + (¯ηb − ¯η)∂kvkδij +
+(cid:16)γ + ζ∆µ −(cid:16)−4κC0 + (ζ ′ − ζ)∆µ(cid:17) ∆h(cid:17) δij − 2ζ∆µ∂i∂jh
+(cid:16)ζc∆µ − 4κC0 −(cid:16)2κ + 2κg + (ζ ′
c − ζc)∆µ(cid:17) ∆h(cid:17) δij − 2h−κg + ζc∆µi ∂i∂jh
mn,i =
∂iǫkl∂kvl.
λ
2
(J1)
(J2)
(J3)
(J4)
(J5)
(J6)
(J7)
(J8)
where we have used the Laplacian operator ∆ = ∂2
inertial terms at low Reynolds number
x +∂2
y . The force and torque balance equations then yield, neglecting
28
η∆vj + ηb∂j∂kvk − (¯ηb + ¯η)∂j ∂t∆h − (ζ ′ + ζ − ζc)∆µ∂j ∆h = −
1
2
tn,j = −(ηcb + ηc)∂j ∂t∆h + ¯η∆vj + ¯ηb∂j∂ivi −h2κ + (ζ ′
ǫijtij = −
λ
2
ǫij ∂i(ǫkltkl)
∂itn,i = −∆h [γ + ζ∆µ]
c + ζc)i ∂j∆h
∆(ǫkl∂kvl).
We then obtain the shape equation
(cid:20)ηcb + ηc −
(¯η + ¯ηb)2
η + ηb (cid:21) ∂t∆∆h = −(cid:18)2κ +(cid:18)ζ ′
c + ζc +
¯η + ¯ηb
η + ηb
(ζc − ζ ′ − ζ)(cid:19) ∆µ(cid:19) ∆∆h
+ (γ + ζ∆µ) ∆h.
(J9)
(J10)
(J11)
(J12)
(J13)
b. Elastic surface
For an elastic surface, the tensions and torque tensors are given in the limit of small displacements by
tij = E1
1
2
(∂iuj + ∂jui) + E2∂kukδij − G1∂i∂jh − G2∆hδij + ∆µ(ζδij − (ζ ′ − ζ)∆hδij − 2ζ∂i∂jh)
(J14)
mij = −F1∂i∂jh − F2∆hδij + K1
mn,i = H1
ǫjk∂i∂juk
.
1
2
1
2
(∂iuj + ∂jui) + K2∂kukδij + ∆µ(ζcδij − (ζ ′
c − ζc)∆hδij − 2ζc∂i∂jh) (J15)
(J16)
A calculation similar to the fluid case then yields the equation for the surface height
ξ
dh
dt
= ζ∆µ∆h −(cid:20)F1 + F2 + ζ ′
c + ζc −
K1 + K2
E1 + E2 (cid:16)G1 + G2 + (ζ ′ + ζ − ζc)∆µ(cid:17)(cid:21) ∆∆h ,
(J17)
where we have introduced an effective external friction force f ext = −ξv · n.
[1] Thomas Lecuit and Pierre-Francois Lenne. Cell surface mechanics and the control of cell shape, tissue patterns and
morphogenesis. Nature Reviews Molecular Cell Biology, 8(8):633–644, 2007.
[2] Patricia Kunda and Buzz Baum. The actin cytoskeleton in spindle assembly and positioning. Trends in cell biology,
19(4):174–179, 2009.
[3] Guillaume Salbreux, Guillaume Charras, and Ewa Paluch. Actin cortex mechanics and cellular morphogenesis. Trends in
cell biology, 22(10):536–545, 2012.
[4] Felix C Keber, Etienne Loiseau, Tim Sanchez, Stephen J DeCamp, Luca Giomi, Mark J Bowick, M Cristina Marchetti,
Zvonimir Dogic, and Andreas R Bausch. Topology and dynamics of active nematic vesicles. Science, 345(6201):1135–1139,
2014.
[5] Karsten Kruse, Jean-Francois Joanny, F Julicher, Jacques Prost, and Ken Sekimoto. Generic theory of active polar gels:
a paradigm for cytoskeletal dynamics. The European Physical Journal E, 16(1):5–16, 2005.
[6] J Prost, F Julicher, and JF Joanny. Active gel physics. Nature Physics, 11(2):111–117, 2015.
[7] Wolfgang Helfrich. Elastic properties of lipid bilayers: theory and possible experiments. Zeitschrift fur Naturforschung C,
28(11-12):693–703, 1973.
[8] Riccardo Capovilla and Jemal Guven. Stresses in lipid membranes. Journal of Physics A: Mathematical and General,
35(30):6233, 2002.
[9] Jean-Baptiste Fournier. On the stress and torque tensors in fluid membranes. Soft Matter, 3(7):883–888, 2007.
[10] Sriram Ramaswamy, John Toner, and Jacques Prost. Nonequilibrium fluctuations, traveling waves, and instabilities in
active membranes. Physical review letters, 84(15):3494, 2000.
[11] Hsuan-Yi Chen. Internal states of active inclusions and the dynamics of an active membrane. Physical review letters,
92(16):168101, 2004.
[12] N Gov. Membrane undulations driven by force fluctuations of active proteins. Physical review letters, 93(26):268104, 2004.
29
[13] Jacob M Sawyer, Jessica R Harrell, Gidi Shemer, Jessica Sullivan-Brown, Minna Roh-Johnson, and Bob Goldstein. Apical
constriction: a cell shape change that can drive morphogenesis. Developmental biology, 341(1):5–19, 2010.
[14] Michael Andersen Lomholt, Per Lyngs Hansen, and Ling Miao. A general theory of non-equilibrium dynamics of lipid-
protein fluid membranes. The European Physical Journal E, 16(4):439–461, 2005.
[15] JF Joanny, F Julicher, K Kruse, and J Prost. Hydrodynamic theory for multi-component active polar gels. New Journal
of Physics, 9(11):422, 2007.
[16] S Furthauer, M Strempel, SW Grill, and F Julicher. Active chiral fluids. The European physical journal. E, Soft matter,
35:89, 2012.
[17] Pierre Curie. On symmetry in physical phenomena, symmetry of an electric field and of a magnetic field. Journal de
Physique, 3:401, 1894.
[18] JE Avron. Odd viscosity. Journal of statistical physics, 92(3-4):543–557, 1998.
[19] Warner Tjardus Koiter and James G Simmonds. Foundations of shell theory. Springer, 1973.
[20] WT Koiter. On the mathematical foundation of shell theory. In Proceedings of the international congress on mathematics
Nice 1970, volume 3, pages 123–130, 1970.
[21] Junuthula Narasimha Reddy. Theory and analysis of elastic plates and shells. CRC press, 2006.
[22] H´el`ene Berthoumieux, Jean-L´eon Maıtre, Carl-Philipp Heisenberg, Ewa K Paluch, Frank Julicher, and Guillaume Salbreux.
Active elastic thin shell theory for cellular deformations. New Journal of Physics, 16(6):065005, 2014.
[23] Sybren Ruurds De Groot and Peter Mazur. Non-equilibrium thermodynamics. Courier Corporation, 2013.
[24] Akira Onuki. Dynamic equations of surfactants and surfaces. Journal of the Physical Society of Japan, 62(2):385–389,
1993.
[25] PA Kralchevsky, JC Eriksson, and S Ljunggren. Theory of curved interfaces and membranes: mechanical and thermody-
namical approaches. Advances in colloid and interface science, 48:19–59, 1994.
[26] Michael A Lomholt and Ling Miao. Descriptions of membrane mechanics from microscopic and effective two-dimensional
perspectives. Journal of Physics A: Mathematical and General, 39(33):10323, 2006.
[27] Marino Arroyo and Antonio DeSimone. Relaxation dynamics of fluid membranes. Physical Review E, 79(3):031915, 2009.
[28] Mohammad Rahimi and Marino Arroyo. Shape dynamics, lipid hydrodynamics, and the complex viscoelasticity of bilayer
membranes. Physical Review E, 86(1):011932, 2012.
[29] G Salbreux, J Prost, and JF Joanny. Hydrodynamics of cellular cortical flows and the formation of contractile rings.
Physical review letters, 103(5):058102, 2009.
[30] S Furthauer, M Strempel, SW Grill, and F Julicher. Active chiral processes in thin films. Physical review letters,
110(4):048103, 2013.
[31] Adam C Martin, Michael Gelbart, Rodrigo Fernandez-Gonzalez, Matthias Kaschube, and Eric F Wieschaus. Integration
of contractile forces during tissue invagination. The Journal of cell biology, 188(5):735–749, 2010.
[32] Kiichiro Taniguchi, Reo Maeda, Tadashi Ando, Takashi Okumura, Naotaka Nakazawa, Ryo Hatori, Mitsutoshi Nakamura,
Shunya Hozumi, Hiroo Fujiwara, and Kenji Matsuno. Chirality in planar cell shape contributes to left-right asymmetric
epithelial morphogenesis. Science, 333(6040):339–341, 2011.
[33] Mirjam Mayer, Martin Depken, Justin S Bois, Frank Julicher, and Stephan W Grill. Anisotropies in cortical tension reveal
the physical basis of polarizing cortical flows. Nature, 467(7315):617–621, 2010.
[34] Jakub Sedzinski, Mat´e Biro, Annelie Oswald, Jean-Yves Tinevez, Guillaume Salbreux, and Ewa Paluch. Polar actomyosin
contractility destabilizes the position of the cytokinetic furrow. Nature, 476(7361):462–466, 2011.
[35] Martin Behrndt, Guillaume Salbreux, Pedro Campinho, Robert Hauschild, Felix Oswald, Julia Roensch, Stephan W Grill,
and Carl-Philipp Heisenberg. Forces driving epithelial spreading in zebrafish gastrulation. Science, 338(6104):257–260,
2012.
[36] Jonas Ranft, Markus Basan, Jens Elgeti, Jean-Fran¸cois Joanny, Jacques Prost, and Frank Julicher. Fluidization of tissues
by cell division and apoptosis. Proceedings of the National Academy of Sciences, 107(49):20863–20868, 2010.
[37] Raphael Etournay, Marko Popovi´c, Matthias Merkel, Amitabha Nandi, Corinna Blasse, Benoıt Aigouy, Holger Brandl, Gene
Myers, Guillaume Salbreux, Frank Julicher, et al. Interplay of cell dynamics and epithelial tension during morphogenesis
of the drosophila pupal wing. Elife, 4:e07090, 2015.
[38] Erwin Kreyszig. Introduction to differential geometry and Riemannian geometry, volume 16. University of Toronto Press,
1968.
|
1506.06339 | 1 | 1506 | 2015-06-21T09:51:35 | Nanodiamond Landmarks for Subcellular Multimodal Optical and Electron Imaging | [
"physics.bio-ph",
"physics.med-ph"
] | There is a growing need for biolabels that can be used in both optical and electron microscopies, are non-cytotoxic, and do not photobleach. Such biolabels could enable targeted nanoscale imaging of sub-cellular structures, and help to establish correlations between conjugation-delivered biomolecules and function. Here we demonstrate a subcellular multi-modal imaging methodology that enables localization of inert particulate probes, consisting of nanodiamonds having fluorescent nitrogen-vacancy centers. These are functionalized to target specific structures, and are observable by both optical and electron microscopies. Nanodiamonds targeted to the nuclear pore complex are rapidly localized in electron-microscopy diffraction mode to enable "zooming-in" to regions of interest for detailed structural investigations. Optical microscopies reveal nanodiamonds for in-vitro tracking or uptake-confirmation. The approach is general, works down to the single nanodiamond level, and can leverage the unique capabilities of nanodiamonds, such as biocompatibility, sensitive magnetometry, and gene and drug delivery. | physics.bio-ph | physics | SUBJECT AREAS
Electron microscopy, fluorescence microscopy, nitrogen-vacancy center,
nanodiamonds, subcellular microscopy, correlated imaging.
Correspondence and requests for materials should be addressed to L.S.B.
([email protected])
Nanodiamond Landmarks for Subcellular Multimodal Optical and Electron
Imaging
Mark A. Zurbuchen1,2,3, Michael P. Lake4, Sirus A. Kohan5, Belinda Leung4, Louis-S.
Bouchard2,4,6,7
1Department of Materials Science and Engineering, 2California NanoSystems Institute,
3Western Institute of Nanoelectronics, Department of Electrical Engineering. 4Department of
Chemistry and Biochemistry, 5The Electron Microscopy Services Center of the Brain
Research Institute, 6Department of Bioengineering; University of California, Los Angeles,
CA 90095, USA. 7Jonsson Comprehensive Cancer Center (UCLA).
There is a growing need for biolabels that can be used in both optical and electron
microscopies, are non-cytotoxic, and do not photobleach. Such biolabels could enable
targeted nanoscale imaging of sub-cellular structures, and help to establish correlations
between conjugation-delivered biomolecules and function. Here we demonstrate a sub-
cellular multi-modal imaging methodology that enables localization of inert particulate
probes, consisting of nanodiamonds having fluorescent nitrogen-vacancy centers. These
are functionalized to target specific structures, and are observable by both optical and
electron microscopies. Nanodiamonds targeted to the nuclear pore complex are rapidly
localized in electron-microscopy diffraction mode to enable “zooming-in” to regions of
interest for detailed structural investigations. Optical microscopies reveal
nanodiamonds for in-vitro tracking or uptake-confirmation. The approach is general,
works down to the single nanodiamond level, and can leverage the unique capabilities of
nanodiamonds, such as biocompatibility, sensitive magnetometry, and gene and drug
delivery.
Nanoparticles have emerged in recent years as a promising approach to particulate
labeling probes for multimodal imaging, and also for targeted drug or gene delivery. They can
also be used for tracking at the single-molecule level.1 In general, little is known about the
immediate environment of targeted nanoparticles due to a lack of suitable visualization
protocols. Knowledge of the local environment would enable clear delineation of
relationships between the activity of a labeled agent and the local biological environment.
One would like to know, for example, the nature and proximity to adjacent macromolecular
assemblies. Despite considerable investigation, the identification and optimization of such a
cross-platform biomarker has remained elusive.
Optical microscopy offers versatility, specificity, and sensitivity in fixed and live cell
settings.2 The development of super-resolution techniques has improved spatial resolution to
tens of nanometers, but these techniques are restricted to a subset of cellular processes.3-5
Transmission electron microscopy (TEM) offers higher spatial resolution.6 But, it lacks
reliable multimodal markers to align the field-of-view with a desired subcellular region. This
is particularly true if multiple iterations between imaging techniques is desired. Routine
multimodal correlated imaging, which could enable the study of live cells with concurrent
visualization of ultra-structural details,7 would require the development of inert biomarkers
and protocols for correlating optical and electron microscopies.
Reporter systems suitable for both optical and TEM imaging are lacking. For example,
fluorescent dyes and fluorophores, required for fluorescence microscopy, cannot be resolved
by TEM, can be cytotoxic, and photobleach under optical excitation. Several candidate
markers for multimodal imaging have been pursued, but none has yet been found to be
practical and universally applicable. Gold nanoparticles and nano-gold cluster compounds are
readily bio-conjugated and resolvable by TEM due to their strong electron scattering. Gold
can be cytotoxic by itself, and frequently becomes isolated in lysosomes in live cells.
Detergent can be used after fixation to allow penetration into cells, but penetration and
immunolabeling efficiency can be low, particularly with colloidal gold, due to steric
hindrance.8 Particulate markers such as gold particles are frequently applied after cryo-slicing
and thawing, after cryo-fracture, or post-embedding after slicing. That is, after cell death.
These approaches have higher immunolabeling success rates, which is dependent upon
particle size and whether or not a slice is embedded.8 While the cryo-slicing approach has
produced some impressive correlative imaging results,9 it is an exceptionally difficult
approach, and is also not amenable to multiple iterations between imaging techniques. Last,
immunogold without fluorophores can be used as a particulate marker, but for typical particle
sizes, requires silver-enhancement to enlarge the particles to render them visible in optical
and/or electron microscopies of slices.10
Colloidal quantum dot (QD) biolabels are resolvable by TEM, do not photobleach, and
can be functionalized with antigen-specific antibodies for targeting. They are, however,
cytotoxic, so their surfaces must be decorated with organic conjugants to provide a barrier. 11
They (QDs) also suffer from "blinking" problems12 and are incompatible with the osmium
tetroxide stain frequently used in sample preparation for TEM.11,13
Fluorescent nanodiamonds (NDs) have emerged as promising biomarkers. They have
been used as optical labels,14 magnetic sensors,15 magnetic resonance imaging (MRI) contrast
agents,16 cell division monitors,17 drug and gene delivery vectors,18,19,20 and for single-
molecule tracking in live cells,21 and for nanometer-scale thermometry in living cells.22
Fluorescence can be enhanced by implanting strongly fluorescing nitrogen vacancy (NV-)
color centers23 to enable optical imaging. Such diamond labels are amenable to transfection
into the cytoplasm,33 have surfaces that can be conjugated directly to proteins for targeting
sub-cellular structures, are non-toxic to cells24,25 and microorganisms,26 do not photobleach or
blink, and are compatible with traditional TEM stains and labels.
In this Letter, we describe the use of bioconjugated NDs6,27 as markers for locating
target cell structures. Locating ND landmarks by regular TEM is problematic due to the
strong background signal originating from the amorphous fixation medium. Nanodiamonds
tend to agglomerate,28,29 reducing the bio-availability of conjugants, which tend to remain
outside of cells, adhere to the outer cell membrane,30 and when taken up generally remain
trapped in endosomes,30-32 all of which render them biologically inert. While the successful
transfection of conjugated NDs has been reported using fluorescence microscopy, the results
have not been confirmed independently.33 To this end, lattice-resolution TEM or electron
diffraction is required.
Our NDs are implanted with NV- centers and targeted using a nuclear membrane-
specific localizer protocol where single NDs are transfected into the cytoplasm of live HeLa
cells. After embedding and slicing, the cells are imaged by fluorescence lifetime imaging
(FLIM), confocal microscopy and by TEM –– bright-field, lattice-fringe imaging, and
electron diffraction –– to unambiguously differentiate NDs from similar-looking features. The
bright fluorescence of NDs yields excellent cellular labels. Confocal fluorescence
microscopy confirms that control (untransfected) cells incubated with an excess of NDs
primarily form aggregates of NDs outside cells, with occasional NDs internalized and
retained near cell surfaces (Section A, Supplementary Information), consistent with reported
observations.28-32
Results
2
For lower ND-concentration imaging, we turned to FLIM. DAPI serves as a positive
control for lifetime imaging (Fig. 1a), where the nuclei are clearly resolved from the
cytoplasm and from areas outside the cell by differential fluorescence lifetimes. The
secondary peak in fluorescence is consistent with DAPI having a stronger signal and a clear
peak at the expected lifetime of 2.2 ns. This peak can also be resolved into components
below 1 ns and above 3 ns, representing bound and unbound DAPI, consistent with the
expected emission of DAPI from two and three-photon absorption processes which both
occur at 910 nm.34 This signal dominates. Cellular auto-fluorescence, which had the strongest
measured lifetimes between 1 and 3 ns, is consistent with literature values (Fig. 1b). Together
these signals serve as strong positive controls for the accurate measurement of fluorescence
lifetimes. Nanodiamonds exhibit an extremely short optical emission lifetime, on the order of
200 ps (Fig. 1c). The histograms of lifetimes weighted by pixel intensity show two peaks, one
that corresponds to the short emission lifetime of the NDs (~250 ps), and a second that
corresponds to the cellular auto-fluorescence and/or DAPI signals. The NDs in cells show the
expected distribution, appearing as punctate spots with exclusion from the nucleus (Fig. 1c).
Conjugation and transfection were used to deliver NDs into living cells (in vitro), to
help them escape from endosomes, and to be released into the cytoplasm. Fluorescence
lifetime imaging of untransfected NDs (Figs. 2a,b) confirms the fluorescence imaging results
above. Lifetime imaging of cells transfected with NDs conjugated to anti-actin antibodies
using polypropylene imide (PPI) dendrimers (Fig. 2c) confirms successful transfection by the
broad distribution of NDs throughout the cytoplasm, which is distinct from distributions of
nanodiamonds when confined within endosomes or targeted to specific membrane-bound
structures. The exclusion of signal from the nucleus is consistent with the published literature
on NDs, and provides additional evidence that the NDs are responsible for the short lifetime
component. Conjugation alone does not produce a broad cellular distribution, but PPI
dendrimers significantly enhance the release of NDs into the cytoplasm (Fig. 2d), promoting
conjugation to membrane antibodies.
Imaging of individual NDs in a biological TEM specimen presents several challenges.
Because our ultimate goals are to enable multiple iterations of imaging, and ultimately to
perform correlated imaging, cell cultures were first embedded in epoxy resin before slicing by
ultramicrotome, with slices ranging from 70 to 250 nm in thickness. The mounting resin
creates a significant amorphous background that confounds TEM image interpretation,
particularly for objects under 100 nm in size. A further complication is that NDs and the
mounting resin have essentially identical electron densities. Together, these prevent the use of
the most commonly used methods for achieving image contrast in a nano-inclusion. That is,
underfocusing the image to achieve Fresnel contrast at edges is not feasible, as contrast
delocalization in the mounting medium causes a strongly modulated background that can
obscure any such detail, even at very limited defocus. Second, there is no appreciable mass-
thickness contrast between NDs (carbon) and the mounting medium (carbon, oxygen,
nitrogen). There is also the complicating factor that a stained sample (osmium tetroxide,
uranyl acetate and lead citrate) presents significant cell-structure contrast which must be
discerned from the ND landmarks.
The TEM technique was initially validated on simulated biological slices, comprising
NDs dispersed in agarose gel, fixed, resin-mounted, and sliced for TEM (Section B,
Supplementary Information). Our TEM observations of untransfected cells dispersed with
NDs reveal the morphology and distribution reported in the literature, and are consistent with
our observations made by optical techniques, above. That is, the NDs are loosely
agglomerated, and are almost exclusively present in the extra-cellular matrix (Fig. 3a). We
found during this study that some dark features having the morphology of a ND particle were
in fact not NDs. That is, it is necessary to confirm that any ND-like feature in fact is a ND,
rather than, for example, a nano-precipitated lead citrate agglomeration. The second approach
to confirm that the dark features are indeed NDs is electron diffraction (Fig. 3b) by the ND
lattice. Diffraction spots corresponding to diamond are evident over the diffuse scattering of
the amorphous mounting medium. The shadow is due to a pointer used to block the primary
3
beam from the camera, to enable acquisition of diffraction patterns with adequate dynamic
range. Without this, camera bloom and oversaturation would obscure the fine details.
Discussion
Figure 4 presents the central result of this work. Targeting of the nuclear membrane
(Fig. 4a) was accomplished by a covalently conjugated antibody specific to Nup98 (Fig. 4a),
which is a nucleoporin (NUP) protein normally found at the nuclear pore, embedded in the
nuclear membrane. The particular ND shown in Fig. 4b is composed of three sub-grains.
High-resolution transmission electron microscopy (HRTEM) (Fig. 4c) reveals lattice fringes
of the crystalline diamond structure, and an electron diffraction pattern of the same ND
(Fig. 4d) provides further confirmation that the landmark is a ND. This procedure yielded
transfected NDs that were found to be non-agglomerated and attached to the nuclear
membrane. The results of Fig. 4 enable us to conclude that TEM can characterize the
successful transfection, targeting, and localization (as a landmark or alignment fiducial) of
single NDs at a region of interest.
It is anticipated that this work will have a significant transformative impact in the
biological sciences, enabling the determination of mechanistic descriptions of biological
structures and dynamics. The approach was demonstrated for live-cell labeling, but is
expected to be equally powerful for post-slicing labeling, whether by cryo or post-embedding.
The intrinsic fluorescence of NDs having NV- centers means that no fluorophore is needed for
optical imaging of the nanoparticulate markers, meaning that cytotoxic fluorophores and
detail-obscuring silver enhancement are unneeded. Transfected and targeted NDs could also
serve as image-alignment fiducials for tomographic TEM reconstruction. Last, and perhaps
most importantly, this NV- ND approach is amenable to correlated optical and electron
microscopy, perhaps even in live cell environments over multiple iterations.
Methods section
Nanodiamonds, Transfection Reagents and Conjugates
The nanodiamonds used in these experiments were prepared by ball milling of larger
~100 µm diamonds containing NV- centers and are on average <100 nm in diameter.
Conjugation of nanodiamonds to antibodies was performed by oxygen termination of the
surface using strong acid treatment followed by EDC conjugation to the antibodies.
Polypropylene imide dendrimers were purchased from SymoChem (Holland) and were
conjugated with maltotriose to prevent cytotoxicity as described by Mkandawire et al.33
Cells and Transfections
All cells used were HeLa cells grown in DMEM with 10% FBS, 1% P/S.
Transfections were performed by mixing transfection reagents with nanodiamond conjugates
in HEPES buffer, or nanodiamonds alone in HEPES. Reagents were allowed to precipitate for
20 min at room temperature, followed by drop-wise addition of the solution to cells in serum-
free DMEM, 1% P/S for 6 hours. Afterward, the media was replaced with growth media and
cells were left to grow for 18-24 hours on plastic coverslips, after which cells were fixed with
2% PFA, and some stained with DAPI. Coverslips were mounted onto slides in
10% PBS / 90% Glycerol and sealed.
Fluorescence Microscopy
Confocal microscopy was performed on a Leica SP5-STED microscope. Excitation
was done using two laser lines at 514 nm and 548 nm. Samples were bleached by repetitive
scanning in order to reduce background. On this system, fluorescence can be detected using a
photomultiplier tube (PMT) or avalanche photodiode (APD). Lifetime imaging was
performed on a Leica SP2-FLIM microscope with Becker and Hickl SP-830 imaging
4
hardware. Lifetime excitation was performed with infrared light from 900-910 nm with a
tunable TI sapphire laser.
TEM Sample Preparation
The cell monolayers grown on coverslips were immersed in a solution of 0.1 M PBS,
pH 7.4 containing 2% glutaraldehyde and 2% paraformaldehyde at room temperature for 2
hours, and then at 4 °C overnight. The cells were subsequently washed in 0.1 M PBS buffer
and post-fixed in a solution of 1% OsO4 in PBS, pH 7.2–7.4. Samples were then buffered in
Na acetate, pH 5.5, and stained in 0.5% uranyl acetate in 0.1 M Na acetate buffer, pH 5.5, at
4 °C for 12 hours.
The samples were sequentially dehydrated in graded ethanol (50%, 75%, 95%, 100%)
and infiltrated in mixtures of Epon 812 and ethanol (1:1 ratio) and 2:1 for two hours each
time. The cells were then incubated in pure Epon 812 overnight and subsequently embedded
and cured at 60 °C for 48 hr. Sections of 70-90 nm thickness (gray interference color) were
cut on an ultramicrotome (RMC MTX) using a diamond knife. The sections were deposited
on carbon-film copper grids and double-stained in aqueous solutions of 8% uranyl acetate at
60 °C for 25 min, and lead citrate at room temperature for 3 min prior to TEM.
TEM Sample Examination
TEM examination was performed using an FEI Titan operated at 300 keV. Locations
of regions of interest were recorded in relation to a fiducial on the copper grids, enabling
location of the same area in subsequent optical microscopy experiments. Low electron beam
currents, typically < 0.6 nA, along with standard low-dose imaging techniques, were
employed in order to preserve specimen integrity. For very low-magnification imaging, the
FEI Titan uses the projector lenses for magnification, the optical path of which blocks many
diffracted beams, rendering some NDs visible as dark spots (among real-world artifacts that
may also appear dark). At moderate to high magnifications, a restrictive objective aperture is
used to block all diffracted beams of the nanodiamond, rendering it dark against a relatively
lighter background of the embedded cell. Tilting of an eucentric-height positioned sample
was occasionally employed to achieve "twinkling" of ND particles as they moved through
strongly diffracting (near-zone axis) orientations. HRTEM images were acquired as close to
zero defocus as possible in order to minimize the contribution of the amorphous background,
which would otherwise obscure any visible lattice fringes.
Our solution to locating NDs by TEM is (1) to use a very restrictive objective
aperture to block all diffracted beams from NDs, making them appear darker than
surrounding material, and (2) to exploit their crystalline nature to discern them from sample-
preparation artifacts. HRTEM imaging of the lattice fringes of NDs embedded in the
amorphous mounting medium is feasible, but only near zero defocus of the objective lens. In
the ideal case, lattice fringes in HRTEM images appear strongest at the Scherzer defocus, but
due to the limited useful defocus range for cell/ND specimens, this is not possible.
1
Alivisatos, P. The use of nanocrystals in biological detection. Nat. Biotechnol. 22, 47-
52 (2004).
Ishikawa-Ankerhold, H. C., Ankerhold, R. & Drummen, G. P. Advanced
fluorescence microscopy techniques –– FRAP, FLIP, FLAP, FRET and FLIM.
Molecules 17, 4047-4132 (2012).
Abbe, E. Contributions to the theory of the microscope and microscopic detection
(translated from the German). Arch. Mikroskop. Anat. 9, 413 (1873).
Hell, S. W. Far-field optical nanoscopy Single Molecule Spectroscopy in Chemistry,
Physics and Biology. Gräslund, A., Rigler, R. & Widengren, J. (eds.), 365-398
(Springer, Berlin, 2010).
2
3
4
5
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Deerinck, T. J. The application of fluorescent quantum dots to confocal, multiphoton,
and electron microscopic imaging. Toxicologic Pathology 36, 112-116 (2008).
Muller, S. A., Aebi, U. & Engel, A. What transmission electron microscopes can
visualize now and in the future. J. Struct. Biol. 163, 235-245 (2008).
Su, Y., et al. Multi-dimensional correlative imaging of subcellular events: combining
the strengths of light and electron microscopy. Biophys. Rev. 2, 121-135. (2010).
Robinson, J. M., Takizawa, T. & Vandré, D. D. Applications of gold cluster
compounds in immunocytochemistry and correlative microscopy: comparison with
colloidal gold. J. Microscopy 199, 163-179 (2000).
Pombo, A., Hollinshead, M. & Cook, P. Bridging the resolution gap: imaging the
same transcription factories in cryosections by light and electron microscopy. J.
Histochem. Cytochem. 47, 471-480 (1999).
De Waele, M. Silver-enhanced colloidal gold for the detection of leucocyte cell
surface antigens in dark-field and epipolarization microscopy Colloidal Gold:
Principles, Methods, and Applications. Hayat, M. A. (ed.), 443-467 (Academic Press,
New York, 1989).
Hardman, R. A. Toxicologic review of quantum dots: toxicity depends on
physicochemical and environmental factors. Environ. Health Perspect. 114, 165-172
(2006).
Nirmal, M., et al. Fluorescence intermittency in single cadmium selenide
nanocrystals. Nature (London) 383, 802-804 (1996).
Deerinck, T. J. The Application of fluorescent quantum dots to confocal,
multiphoton, and electron microscopic imaging. Toxicologic Pathology 36, 112-116
(2008).
Fu, C. C., et al. Characterization and application of single fluorescent nanodiamonds
as cellular biomarkers. Proc. Natl. Acad. Sci. USA 104, 727-732 (2007).
McGuinness, L. P., et al. Quantum measurement and orientation tracking of
fluorescent nanodiamonds inside living cells. Nat. Nanotechnol. 6, 358-363 (2011).
Manus, L. M., et al. Gd(III)-nanodiamond conjugates for MRI contrast enhancement.
Nano Lett. 10, 484-489 (2010).
Lien, Z. Y., et al. Cancer cell labeling and tracking using fluorescent and magnetic
nanodiamond. Biomaterials 33, 6172-6185 (2012).
Chow, E. K., et al. Nanodiamond therapeutic delivery agents mediate enhanced
chemoresistant tumor treatment. Sci. Transl. Med. 3, 73ra21 (2011).
Alhaddad, A., et al. Nanodiamond as a vector for siRNA delivery to Ewing sarcoma
cells. Small 7, 3087-3095 (2011).
Ho, D. Beyond the sparkle: the impact of nanodiamonds as biolabeling and
therapeutic agents. ACS Nano 3, 3825-3829 (2009).
Chang, Y. R., et al. Mass production and dynamic imaging of fluorescent
nanodiamonds. Nature Nanotechnol. 3, 284-288 (2008).
Kucsko, G., et al. Nanometre-scale thermometry in a living cell. Nature (London)
500, 54-58 (2013).
Jelezko, F. & Wrachtrup, J. Single defect centres in diamond: a review. Phys. Status
Solidi 203, 3207-3225 (2006).
Horie M., et al. Evaluation of cellular influences induced by stable nanodiamond
dispersion; the cellular influences of nanodiamond are small. Diamond and Related
Mater. 24, 15-24 (2012).
Zhang, X., Hu, W., Li, J., Tao, L. & Wei, Y. A comparative study of cellular uptake
and cytotoxicity of multi-walled carbon nanotubes, graphene oxide, and
nanodiamond. Toxicol. Res. 1, 62-68 (2012).
Mohan, N., Chen, C-S., Hsieh, H-H., Wu, Y-C. & Chang, H-C. In vivo imaging and
toxicity assessments of fluorescent nanodiamonds in caenorhabditis elegans. Nano
Lett. 10, 3692-3699 (2010).
6
27
28
29
30
31
32
33
34
Studer, D., Humbel. B. M. & Chiquet, M. Electron microscopy of high pressure
frozen samples: bridging the gap between cellular ultrastructure and atomic
resolution. Histochem. Cell Biol. 130, 877-889 (2008).
Tzeng, Y-K., et al. Superresolution imaging of albumin-conjugated fluorescent
nanodiamonds in cells by stimulated emission depletion. Angew. Chem. Int. Ed. 50,
2262-2265 (2011).
Schrand, A. M., Schlager, J. J., Dai, L. & Hussain, S. M. Preparation of cells for
assessing ultrastructural localization of nanoparticles with transmission electron
microscopy. Nature Protoc. 5, 744-757 (2010).
Liu, K-K., Wang, C-C., Cheng, C-L. & Chao, J-I. Endocytic carboxylated
nanodiamond for the labeling and tracking of cell division and differentiation in
cancer and stem cells. Biomaterials 30, 4249-4259 (2009).
Schrand, A. M., Lin, J. B., Hens, S. C. & Hussain, S. M. Temporal and mechanistic
tracking of cellular uptake dynamics with novel surface fluorophore-bound
nanodiamonds. Nanoscale 3, 435-445 (2011).
Faklaris, O., et al. Photoluminescent diamond nanoparticles for cell labeling: study of
the uptake mechanism in mammalian cells. ACS Nano 3, 3955-3962 ( 2009).
Mkandawire M, et al. Selective targeting of green fluorescent nanodiamond
conjugates to mitochondria in HeLa cells. J. Biophotonics 2, 596-606 (2009).
Lakowicz, J. R., et al. Time-resolved fluorescence spectroscopy and imaging of DNA
labeled with DAPI and hoechst 33342 using three-photon excitation. Biophysical J.
72, 567-578 (1997).
7
Acknowledgments
This study was supported by AFOSR and DARPA QuASAR. We acknowledge the use
of instruments at the Electron Imaging Center for NanoMachines (EICN) supported by NIH
(1S10RR23057 to ZHZ) at the California NanoSystems Institute (CNSI), UCLA. Confocal
laser scanning microscopy was performed at the CNSI Advanced Light
Microscopy/Spectroscopy Shared Resource Facility at UCLA. We thank Arek Melkonian for
help with preparing solutions. The authors thank Joerg Wrachtrup and Gopi Subramanian for
providing the fluorescent nanodiamonds, Hong Zhou, Laurent Bentolila, and Sergey
Ryazantsev for helpful discussions, and Charles M. Knobler for critical comments on the
manuscript.
Author contributions
M.A.Z. developed the TEM methodology and carried out the TEM experiments. M.P.L
prepared cultures and performed optical microscopies. B.L. helped with nanodiamond
conjugation. S.A.K. devised the agarose control experiment, and prepared TEM samples.
M.A.Z, M.P.L and L.S.B. designed experiments. M.A.Z. and M.P.L performed data analysis.
M.A.Z., M.P.L., and L.S.B. wrote the paper. M.A.Z. and M.P.L. contributed equally to this
work.
Additional Information
Supplementary information: Supplementary information accompanies this paper at
http://www.nature.com/scientificreports
Competing financial interests: The authors declare no competing financial interests.
License: This work is licensed under a Creative Commons Attribution-NonCommercial-
ShareAlike 3.0 Unported License. To view a copy of this license, visit
http://creativecommons.org/licenses/by-nc-sa/3.0/
How to cite this article:
8
Figure captions
Figure 1. Fluorescence lifetime imaging of cells with NDs, with corresponding fluorescence
lifetime spectra. (a) HeLa cells stained with DAPI. (b) Unstained HeLa cells. (c) NDs
incubated with HeLa cells.
Figure 2. Comparison of distributions of nanodiamonds in HeLa cells by fluorescence
lifetime imaging. (a) Cellular distribution observed in untransfected NDs (control) in excess
concentration. (b) Magnified image of an untransfected cell showing nucleus and cytoplasm,
with aggregated nanodiamonds primarily outside of cells. (c) Uniform distributions of NDs
throughout the cytoplasm (no DAPI) in NDs transfected into cells with PPI dendrimers.
(d) Less uniform distribution of NDs added to cells without transfection agent (brightness
enhanced, no DAPI).
Figure 3. TEM characterization of a control sample reveals the typical appearance of NDs in
non-transfected HeLa cells. (a) Image of several loose agglomerations of NDs in the
extracellular matrix. Cytoplasm (Cy), nanodiamonds (ND). Scale bar, 500 nm. (b) Electron
diffraction pattern of the same area, confirming the agglomerations to comprise NDs. The
shadow is from a pointer, used to block the central beam for image acquisition.
Figure 4. TEM verification of ND transfection and adherence to nuclear membrane of HeLa
cells. (a) Image of a portion of a cell with a ND adhered to the nuclear membrane. White
arrows indicate the nuclear membrane. Stripes are knife-marks from the ultramicrotome, and
the wide, horizontal dark band is due to a slight wrinkle of the slice. Scale bar, 500 nm.
(b) Magnified image of single ND (consisting of three domains). Scale Bar, 50 nm.
(c) HRTEM image of the ND, showing lattice fringes (indicated by pairs of lines) which
verify its crystalline character. Scale bar, 1 nm. (d) Electron diffraction pattern of the same
area, further verifying that the particle is a (crystalline) ND. The shadow is from a pointer,
used to block the central beam for image acquisition. Cytoplasm (Cy), Nucleus (Nu),
nanodiamond (ND).
9
|
1302.1344 | 1 | 1302 | 2013-02-06T12:48:00 | Period Variability of Coupled Noisy Oscillators | [
"physics.bio-ph",
"nlin.CD"
] | Period variability, quantified by the standard deviation (SD) of the cycle-to-cycle period, is investigated for noisy phase oscillators. We define the checkpoint phase as the beginning/end point of one oscillation cycle and derive an expression for the SD as a function of this phase. We find that the SD is dependent on the checkpoint phase only when oscillators are coupled. The applicability of our theory is verified using a realistic model. Our work clarifies the relationship between period variability and synchronization from which valuable information regarding coupling can be inferred. | physics.bio-ph | physics | Period Variability of Coupled Noisy Oscillators
Fumito Mori1, ∗ and Hiroshi Kori1, 2
1Department of Information Sciences,
Ochanomizu University, Tokyo, Japan
2PRESTO, Japan Science and Technology Agency, Kawaguchi, Japan
(Dated: July 2, 2018)
Abstract
Period variability, quantified by the standard deviation (SD) of the cycle-to-cycle period, is
investigated for noisy phase oscillators. We define the checkpoint phase as the beginning/end
point of one oscillation cycle and derive an expression for the SD as a function of this phase. We
find that the SD is dependent on the checkpoint phase only when oscillators are coupled. The
applicability of our theory is verified using a realistic model. Our work clarifies the relationship
between period variability and synchronization from which valuable information regarding coupling
can be inferred.
PACS numbers: 05.40.Ca,05.45.Xt,87.18.Yt
3
1
0
2
b
e
F
6
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
4
4
3
1
.
2
0
3
1
:
v
i
X
r
a
∗ [email protected]
1
Oscillators functioning as clocks, such as crystal oscillators [1], spin torque oscillators
[2 -- 5], and circadian and heart pacemakers [6 -- 8], play an important role in various systems.
Although these clocks are subjected to various types of noise, including thermal, quantum,
and molecular noise, they are required to perform temporally precise oscillations; i.e., os-
cillations with only a small variability in the period (known as "period jitter" in electronic
engineering [9]).
In many cases, it is sufficient for the clock to strike precisely at a specific time in each
oscillation cycle, and thus a perfectly regular oscillation waveform is not needed. For cardiac
pacemakers only the moment of stimulation is relevant. Experimental data regarding circa-
dian activity in mice [10] indicate that the variability in the period between each activity
onset is smaller than that between each offset. Similar results have also been obtained in
explant circadian pacemaker tissue (the suprachiasmatic nucleus, SCN) [10]. These obser-
vations suggest that the onset is more important than the offset in a circadian clock, which
may be designed in such a way that the crucial moment is expressed with high precision.
Remember that the definition of an oscillation period requires a fixed beginning/end point
for each oscillation cycle; hereafter referred to as the checkpoint (Fig. 1). Although the
average period does not depend on the particular choice of checkpoint, the period variability
may be sensitive to the checkpoint. In order to clarify whether the checkpoint dependence
in circadian activity is an artifact due to a technical problem in determining the onset and
offset times or an essential property of the circadian clock, we need to investigate under
what conditions the period variability is dependent on the checkpoint; this has received
scant attention to date.
Another important aspect of the period variability is its relationship to synchronization.
A clock is commonly synchronized to its master clock such as in the case of the SCN in
response to the daily variation of sunlight, and in peripheral clocks in response to the SCN.
In addition, most biological clocks, including the SCN, cardiac pacemakers, and pacemakers
in weakly electrical fish, are composed of a population of synchronized oscillators [6, 7, 11].
It is known, both experimentally and theoretically, that period variability is reduced when
the oscillators are coupled and synchronized [6, 12 -- 16]. The question, therefore, arises as to
whether the checkpoint dependence of the period variability is attributable to the interaction
between oscillators.
In this Letter, we discuss this checkpoint dependence for the case of coupled noisy phase
2
oscillators. The period variability can be quantified using the standard deviation (SD) of the
cycle-to-cycle period, and we show that although the SD is not dependent on the checkpoint
in a single phase oscillator, it is dependent in a system of coupled phase oscillators; i.e., the
checkpoint dependence results from the coupling effect. The SD is derived as a function of
the checkpoint phase, which clarifies the relationship between the SD and synchronization.
In particular, we find that in the case of diffusive coupling between oscillators, the checkpoint
dependence of the SD has the same tendency as that of the synchronization: the SD is small
when the oscillators are well synchronized. In other cases, however, the relationship is more
complex. We also apply our theory to a realistic model of the electrical activity in a cell to
demonstrate its validity. We believe that this is the first theoretical study to elucidate the
existence of precise timing and its relationship with synchronization.
To begin, we prove that the period variability is independent of the checkpoint in a
single phase oscillator system. When a limit cycle oscillator is subjected to weak noise, its
dynamics are well described by the following phase oscillator model [17, 18];
= ω + Z(θ)√Dξ(t),
dθ
dt
(1)
where θ and ω are the phase and natural frequency, respectively. The 2π-periodic function
Z(θ) is a phase sensitivity function, which quantifies the phase response of the oscillator to
noise, and ξ(t) denotes independent and identically distributed (i.i.d.) noise; each random
variable ξ(t) for all t obeys the same probability distribution and all are mutually indepen-
dent. The positive constant D denotes the noise strength. Note that our proof below holds
even if we permit ω and the probability distribution of ξ to be 2π-periodic functions of θ:
ω(θ) and ξ(t, θ).
The kth oscillation time of an oscillator, tθcp
k , is defined as the time at which θ passes
through 2πk + θcp (0 ≤ θcp < 2π) for the first time [Fig. 1(b)]. We define θcp as the
checkpoint phase. The kth oscillation period ∆tθcp
k−1, and the
k
is defined as ∆tθcp
k = tθcp
k − tθcp
SD is defined as
SD(θcp) =qE[(∆tθcp
k − τ )2],
(2)
where E[· · · ] represents the statistical average over k, and τ is the average period given by
τ = E[∆tθcp
k ]. Note that E[· · · ] denotes both the statistical average taken over k and the
ensemble average in the present paper, which are identical in the steady state. The system
given by Eq. (1) is always in the steady state.
3
To prove that the SD is independent of θcp, we introduce two checkpoint phases denoted
by α and β [Fig. 1(b)]. Since the processes α → β and β → α for any k are independent,
we arrive at SD(α) = SD(β) for any arbitrary checkpoint phases α and β. A detailed proof
is given in Appendix A.
FIG. 1. (color online). (a) An example of the time series of an oscillation. Periods are observed at
two checkpoints, α and β. (b) The corresponding checkpoint phases in the phase description.
Next, we consider a pair of coupled phase oscillators subjected to noise. When limit cycle
oscillators are weakly coupled to each other and subjected to weak noise, the dynamics can
be described by [17, 18]
θ1 = ω + κJ(θ1, θ2) + Z(θ1)√Dξ1(t),
θ2 = ω + κJ(θ2, θ1) + Z(θ2)√Dξ2(t),
(3)
where θi and κ ≥ 0 are the phase of the oscillator i and the coupling strength, respectively.
The i.i.d. noise ξi(t) satisfies E[ξi(t)] = 0 and E[ξi(t)ξj(t′)] = δijδ(t − t′). The 2π-periodic
function J(x, y) describes the interaction between oscillators, which leads to synchronization.
We assume that, in the absence of noise (D = 0), the oscillators are synchronized in phase,
i.e., θ1,2(t) → φ(t) (t → ∞), where φ(t) is a solution of
φ(t) = ω + κJ(φ, φ).
(4)
The necessary condition for the stability of in-phase synchrony for D = 0 is provided below
[see Eq. (11)]. We also assume that ω + κJ(φ, φ) > 0 for any φ for the coupled system to
be oscillatory.
Our particular interest is in the relationship between the SD [Eq. (2)] and the synchro-
nization of two oscillators. We thus introduce the following order parameter that measures
the phase distance from the in-phase state:
d(θcp) =qE [kθ1 − θ2k2]θ1=θcp
,
4
(5)
where E[x(t)]θ1=θcp represents the average of xk over k (where xk is the value of x(t) taken
when θ1 passes through 2πk + θcp for the first time), and kθ1 − θ2k is the phase difference
defined on the ring [−π, π). The phase distance d(θcp) is zero when the oscillators are
completely synchronized in phase, and increases with the phase difference.
As we demonstrate below, the relationship between SD(θcp) and d(θcp) is qualitatively
different for the two cases where J(φ, φ) is (A) independent of φ and (B) dependent on φ.
Cases (A) and (B) imply that φ given in Eq. (4) is independent of φ and dependent on
φ, respectively. Phase reduction theory indicates that it is appropriate to assume the form
J(x, y) = z(x)G(x, y), where z(x) is the phase sensitivity function for the interaction G(x, y)
[17, 18]. It is known that diffusive coupling between chemical oscillators and gap-junction
coupling between cells yields J(x, y) = z(x)(h(x) − h(y)), where h represents a chemical
concentration [19, 20] or membrane potential, which corresponds to case (A). Case (A) also
allows the form J(x, y) = j(x − y), which has been employed in many models such as the
Kuramoto model [18]; however, we do not employ this form in the demonstration, since the
term j(x−y) is derived as a result of averaging the interaction z(x)G(x, y) over one oscillation
period [18], and, by this approximation, the information about the θcp dependence is lost.
Many other types of coupling, such as J(x, y) = z(x)h(y) employed below, correspond to
case (B) [21].
As an example of case (A), we consider z(θ) = sin θ for 0 ≤ θ < π, z(θ) = 0 for
π ≤ θ < 2π, and h(θ) = cos θ, and the following as an example of case (B): z(θ) = − sin θ
and h(θ) = 1 + cos θ [21]. We set Z(θ) = 1, ω = 2π, √D = 0.03× 2π, and θ1(0) = θ2(0) = 0,
and assume ξ1,2(t) to be white Gaussian noise. We integrate Eq. (3) using the Euler scheme
with a time step of 5 × 10−4 for t = 0 -- 10100 and discard the t = 0 -- 100 data as transient.
Using these examples, numerically obtained SD values for θ1 are plotted as a function
of θcp in Fig. 2(a) and (b). The results indicate clearly the existence of θcp dependence
in both cases, which was absent in the single phase oscillator system. This dependence
In contrast, for κ ≪ ω, the dependence vanishes
becomes stronger for larger κ values.
because J(x, y) is well approximated by j(x − y) [18], and thus, the system effectively has
rotational symmetry. The θcp value at which SD(θcp) assumes its minimum represents the
most precise timing.
The θcp dependence of d(θcp) for the two cases is shown in Fig. 2 (c) and (d). A
comparison with SD(θcp) shows that the checkpoint phase maxima and minima of each
5
κ value coincide in the case of (A). Thus, the most precise timing is obtained when the
oscillators are synchronized. By contrast, the θcp dependence is considerably different in
the case of (B). Therefore, we expect that nontrivial factors, apart from synchronization,
influence the SD. We also examined several other functions, z(θ), h(θ), and Z(θ), and found
a similar relationship between SD(θcp) and d(θcp) (data not shown).
)
%
(
3.0
/
)
p
c
(
D
S
2.5
(a)
2.0
0
0.3
)
d
a
r
(
0.2
)
p
c
(
d
0.1
0
(c)
0
4.0
3.5
3.0
= 0.1
= 0.2
= 1.0
(b)
2.5
0
0.4
2
= 0.4
= 1.0
= 12.0
0.2
= 0.4
= 1.0
= 12.0
(d)
0
0
2
cp
= 0.1
= 0.2
= 1.0
cp
2
2
FIG. 2. (color online). The SD(θcp)/τ for case (A) and (B) is shown in (a) and (b), respectively,
where the vertical scale is expressed as a percentage. The distance from in-phase synchronization,
d(θcp), for case (A) and (B) is shown in (c) and (d), respectively. The points and lines are
the numerical results of the simulation and analytical predictions given by Eqs. (16) and (13),
respectively.
We now derive an expression for the SD. The derivation consists of two steps: (i) calcu-
lation of the phase diffusion σ(θcp) [defined by Eq. (7)] with a linear approximation, and (ii)
transformation from σ(θcp) to SD(θcp). Here, we employ the solution φ(t) of Eq. (4) with
φ(0) = 0 and the time tcp is defined by φ(tcp) = θcp. The oscillation period for D = 0 is
denoted by τ ; i.e., φ(tcp + τ ) = θcp + 2π. After a transient time, our system approaches the
steady state, which is defined by the following equation for all Ψ:
P (kθ1 − θ2k; θ1 = Ψ) = P (kθ1 − θ2k; θ1 = Ψ + 2π),
(6)
where P (kθ1 − θ2k; θ1 = Ψ) is the probability density function of the distance kθ1 − θ2k at
θ1 = Ψ. We assume that the system is in the steady state at t = 0. The ensemble we consider
here is defined by the initial condition at t = tcp, θ1(tcp) = θcp, and θ2(tcp) is distributed in
6
[θ1(tcp)− π, θ1(tcp) + π) according to Eq. (6). From this point, E[· · · ] represents the average
taken over this ensemble. The phase diffusion σ(θcp) is defined by
σ(θcp)2 = E[(θ1(tcp + τ ) − θ1(tcp) − 2π)2].
(7)
We also assume that the noise intensity D is sufficiently small and that the other parameters
and functions are of O(1), so that the phase difference kθ1 − θ2k is small in most cases in
the steady state.
To calculate the phase diffusion, we decompose θ1,2 as θ1,2(t) = φ(t) + ∆1,2(t). We then
consider the time duration 0 ≤ t ≤ O(τ ), in which ∆1,2(t) ≪ 1 is expected in most cases
because D ≪ 1. Therefore, we can linearize Eq. (3). We define the two modes, X = ∆1 + ∆2
and Y = ∆1 − ∆2, which obey
( X, Y ) = κfX,Y (φ(t))(X, Y ) + ξX,Y (t, φ(t)),
(8)
where fX(φ) ≡ ∂J
ξX,Y (t, φ(t)) ≡ √DZ(φ(t))(ξ1(t) ± ξ2(t)). Note that fX (φ) = 0 for all φ in case (A).
∂x(cid:12)(cid:12)x=y=φ − ∂J
∂x(cid:12)(cid:12)x=y=φ + ∂J
, fY (φ) ≡ ∂J
, and
= dJ(φ,φ)
dφ
∂y(cid:12)(cid:12)(cid:12)x=y=φ
The solutions of Eq. (8) can be described as
∂y(cid:12)(cid:12)(cid:12)x=y=φ
, we obtain
(9)
(10)
where FX,Y (φ(t)) ≡R t
(X, Y )(t) = exp [+κFX,Y (φ(t))]
×(cid:26)(X, Y )(0) +Z t
0
exp [−κFX,Y (φ(t′))]ξX,Y (t′, φ(t′))dt′(cid:27) ,
0 fX,Y (φ(t′))dt′. Furthermore, because fX (φ) = dJ(φ,φ)
dφ
FX (φ(t)) =
1
κ
φ(0)! .
ln φ(t)
For ξY = 0, we obtain FY (2π) = (1/κ) ln(Y (τ )/Y (0)). Therefore, in the absence of noise,
in-phase synchronization is stable if
FY (2π) ≡ c < 0.
(11)
The correlations, E[X(t)2], E[Y (t)2], and E[X(t)Y (t)] are given in Appendix B. Since
Eq. (6) can be rewritten as P (∆1−∆2; t) ∼= P (∆1−∆2; t+τ ), then E[Y (t)2] = E[Y (t+τ )2]
holds approximately, leading to
E[Y (0)2] = 2D
exp[2κc]
1 − exp[2κc]Z τ
0
7
Z(φ(t′))2 exp[−2κFY (φ(t′))]dt′.
(12)
In addition, because d(θcp)2 = E[(∆1(tcp) − ∆2(tcp))2] = E[Y (tcp)2], we obtain
d(θcp)2 = exp[2κFY (θcp)]
×(cid:18)E[Y (0)2] + 2DZ θcp
0
Z(φ)2 exp[−2κFY (φ)]
ds
dφ(s)
dφ(cid:19) ,
(13)
which is generally θcp-dependent even if Z(φ) is constant.
Using these correlations and Eq. (10), we obtain the following expression for the phase
diffusion (See Appendix B)
σ(θcp)2 = E[(∆1(tcp + τ ) − ∆1(tcp))2]
= C1
φ(θcp)
2
+ C2d(θcp)2,
(14)
Z(θ)2
where the C1,2 are independent of θcp and are given by C1 = D
φ(θ)3 dθ and C2 =
(1 − exp[κc])/2. The C1 term is an effective diffusion constant for the center of the two
oscillators, which is half that of an uncoupled oscillator, and the C2 term is associated with
2 R 2π
0
the stability of the synchronization.
To transform σ(θcp) to SD(θcp), we note that when the noise intensity is low, most of the
trajectories of θ1(t) are very close to the unperturbed trajectory φ(t) (see Appendix C). In
such a case, the following relation approximately holds true:
σ(θcp)
SD(θcp)
= φ(θcp).
(15)
The same approximation (but for constant φ) was employed in Ref.
[16] and verified nu-
merically.
From Eqs. (B11) and (15), we finally arrive at
SD(θcp) =vuutC1 + C2
d(θcp)2
2 .
φ(θcp)
(16)
The analytical results given by Eqs. (16) and (13) are in excellent agreement with the nu-
merical results (Fig. 2). Although we have only discussed paired identical phase oscillators,
our theory can easily be extended to other cases, e.g., N globally coupled (all-to-all) identical
oscillators or a periodically driven noisy oscillator.
Equation (16) shows that the periodicity of SD(θcp) is based on the synchronization
d(θcp) and phase velocity φ(θcp). For case (A), since φ(θcp) is constant, there is one-to-one
correspondence between SD(θcp) and d(θcp); i.e., the most precise timing (θmin
cp ) is the timing
8
at which the best synchronization is achieved. This was observed in Fig. 2 (a) and (c), where
θmin
cp = π/2 + O(κ−1) can be obtained from dd(θcp)/dθcp = 0. For case (B), however, the SD
also depends on φ(θcp); this is in contrast to that observed for the single phase oscillator
system in which the phase velocity ω(θ) does not contribute to the checkpoint dependence
of the SD. Figures 2(b) and (d) showed that SD(θcp) and d(θcp) are considerably different,
φ in this particular example. Indeed, SD(θcp) assumes
which indicates the strong effect of
its minimum around a maximum φ(θcp) (θmin
cp ≈ 5π/3).
To investigate whether Eq. (16) holds for a more realistic model, we employ the FitzHugh-
Nagumo model given by
V1 = V1(V1 − a)(1 − V1) − W1 + ξ1(t) + KV (V2 − V1),
W1 = ǫ(V1 − bW1) + KW (W2 − W1),
(17)
in which the second oscillator is described in a similar way. We fixed a = −0.1, b = 0.5,
and ǫ = 0.01. This system shows limit-cycle oscillations with a period of τ ≃ 126.5 when
noise and coupling are absent. The white Gaussian noise ξi(t) has an intensity of 0.01. The
interaction is diffusive, i.e., case (A), and we consider the following two types: V -coupling
(KV = 0.01, KW = 0) and W -coupling (KV = 0, KW = 0.01). The phase θ was defined
properly (see Appendix D), and SD(θcp) and d(θcp) were obtained numerically. Figure 3
shows that the θcp-dependence of the SD is different in the two cases, suggesting a significant
effect from the coupling. We estimated the C1 and C2 values using Eq. (16) and the least-
squares method under the condition that both cases have the same C1 value, resulting in
C1 = 5.4, C (V )
2 = 0.48. In Fig. 3, we can see that the SD is described well
2 = 0.20, and C (W )
by Eq. (16) using the fitted C1 and C2 values. This demonstrates that the theory is valid
for this biological model.
In many cases, only the SD measured at a functionally relevant checkpoint characterizes
the performance of a clock. When designing a precise clock, we only have to reduce SD(θcp)
for a specific θcp. Equation (16) implies that SD(θcp) at a given θcp decreases with decreasing
d(θcp) and increasing φ(θcp). Therefore, attractive coupling between oscillators should be
activated around the functionally relevant timing point. In addition, in case (B), the phase
velocity should be increased through coupling.
Our theory enables us to infer the coupling timing or form by measuring SD(θcp) at
several checkpoints. Although this is, in principle, possible with d(θcp), using SD(θcp) has
the added advantages that the SD can be measured from a single time series and that d(θcp)
9
3.6
3.2
)
p
c
(
D
S
2.8
2.4
0
V W
no
SD
Eq.(16)
π
cp
2π
FIG. 3. (color online). Validation of Eq. (16) in the FitzHugh-Nagumo model. Open symbols are
the numerically obtained SD values. Filled symbols are the SD values evaluated from Eq. (16)
with the numerically obtained d values and fitting parameters C1 and C2. The triangles and circles
represent the V - and W -coupling cases. The plus symbols are the numerically obtained SD values
for an uncoupled oscillator.
is sensitive to the definition of phase. From the observations of circadian periods in mice
described in the introduction [10], it is possible that the SCN sends signals to the peripheral
clocks around the onset of a subjective day. An experimental observation of the checkpoint
dependence in other biological clocks would be a new source of coupling information.
We thank Hiroshi Ito for valuable discussions. This work was supported by JSPS KAK-
ENHI Grant Number 23·11148.
Appendix A: Proof that the SD is independent of the checkpoint phase in a single
phase oscillator
We introduce two checkpoint phases denoted by α and β. By defining the intervals
∆tβ→α
k
k
k
k and ∆tα→β
k
= tβ
= tα
k − tβ
k−1, the oscillation periods observed at α and β can be
decomposed as ∆tα
k−1 , respectively. Because
ξ(t) is independent, the processes α → β and β → α for any k are independent. We thus
have
k − tα
+ ∆tα→β
k = ∆tβ→α
k = ∆tα→β
k
+ ∆tβ→α
and ∆tβ
E[∆tβ→α
k
] = E[∆tβ→α
k−1 ],
E[(∆tβ→α
k
)2] = E[(∆tβ→α
k−1 )2],
10
(A1)
(A2)
and
E[∆tα→β
k ∆tβ→α
k−1 ] = E[∆tα→β
k
]E[∆tβ→α
k−1 ] = E[∆tα→β
k ∆tβ→α
k
].
(A3)
The average and the mean square period are independent of the checkpoint phase labels;
i.e.,
and
E[∆tα
k ] = E[∆tβ→α
k
] + E[∆tα→β
k
] = E[∆tβ→α
k−1 ] + E[∆tα→β
k
] = E[∆tβ
k ] = τ
(A4)
E[(∆tα
k )2] = E[(∆tβ→α
k
)2] + E[(∆tα→β
k
)2] + 2E[∆tβ→α
k
][∆tα→β
k
] = E[(∆tβ
k )2].
(A5)
Thus, we arrive at
SD(α) =qE[(∆tα
k − τ )2] =qE[(∆tβ
k − τ )2] = SD(β)
(A6)
for any arbitrary checkpoint phases α and β.
Appendix B: Calculation of the correlations
The correlations of the noise terms, ξX,Y (t, φ(t)) = √DZ(φ(t))(ξ1(t) ± ξ2(t)), are given
as
E[ξX(s, φ(s))ξX(s′, φ(s′))] = 2DZ(φ(s))Z(φ(s′))δ(s − s′),
E[ξY (s, φ(s))ξY (s′, φ(s′))] = 2DZ(φ(s))Z(φ(s′))δ(s − s′),
E[ξX(s, φ(s))ξY (s′, φ(s′))] = 0.
(B1)
(B2)
(B3)
Using Eqs. (9), (B1), (B2), (B3), and E[ξX,Y (t, φ(t))] = 0, we obtain
E[X(t)2] = exp [2κFX(φ(t))](cid:20)E[X(0)2]
exp[−κ{FX (φ(s)) + FX(φ(s′))}]E[ξX(s, φ(s))ξX(s′, φ(s′))]dsds′(cid:21)
0
+Z t
0 Z t
= exp [2κFX(φ(t))](cid:20)E[X(0)2] + 2DZ t
0
Z(φ(s))2 exp[−2κFX (φ(s))]ds(cid:21),
(B4)
E[Y (t)2] = exp [2κFY (φ(t))](cid:20)E[Y (0)2]
exp[−κ{FY (φ(s)) + FY (φ(s′))}]E[ξY (s, φ(s))ξY (s′, φ(s′))]dsds′(cid:21)
+Z t
0 Z t
= exp [2κFY (φ(t))](cid:20)E[Y (0)2] + 2DZ t
0
0
Z(φ(s))2 exp[−2κFY (φ(s))]ds(cid:21),
(B5)
11
and
E[X(t)Y (t)] = exp [κ(FX (φ(t)) + FY (φ(t)))]E[X(0)Y (0)].
(B6)
Substituting t = tcp + τ in Eqs. (B4) and (B6), we obtain
E[X(tcp + τ )2] = exp [2κFX(θcp + 2π)](cid:20)E[X(0)2] + 2DZ tcp+τ
+Z tcp+τ
Z(φ(s))2 exp[−2κFX (φ(s))]ds,
= exp [2κFX(θcp)](cid:20)E[X(0)2] + 2DZ tcp
= E[X(tcp)2] + 2D exp[2κFX(θcp)]Z τ
Z(φ(s))2 exp[−2κFX (φ(s))]ds(cid:21)
Z(φ(s))2 exp[−2κFX (φ(s))]ds(cid:21)
(B7)
tcp
0
0
0
and
E[X(tcp + τ )Y (tcp + τ )] = exp [κ(FX(θcp + 2π) + FY (θcp + 2π))]E[X(0)Y (0)]
= exp [κc]E[X(tcp)Y (tcp)],
(B8)
where we use φ(tcp + τ ) = θcp + 2π, FX(θ + 2π) = FX (θ), FY (θ + 2π) = FY (θ) + c, and
Z(θ + 2π) = Z(θ).
θ1(t) − φ(t), we obtain
Inserting θ1(tcp) = θcp and φ(tcp) = θcp into the definition ∆1(t) =
We then obtain
∆1(tcp) = 0.
E[X(tcp)2] = −E[X(tcp)Y (tcp)] = E[Y (tcp)2] = d(θcp)2.
(B9)
(B10)
Using Eqs. (B7) -- (B10), the relation E[Y (tcp + τ )2] = E[Y (tcp)2], and Eq.(10), we obtain
the following expression for the phase diffusion
σ(θcp)2 = E[(∆1(tcp + τ ) − ∆1(tcp))2]
1
=
=
1
4(cid:8)E[X(tcp + τ )2] + E[Y (tcp + τ )2] + 2E[X(tcp + τ )Y (tcp + τ )](cid:9)
4(cid:26)2D exp[2κFX(θcp)]Z τ
Z(φ(t′))2 exp[−2κFX (φ(t′))]dt′ + 2(1 − exp [κc])d(θcp)2(cid:27)
0
= C1
φ(θcp)
2
+ C2d(θcp)2.
(B11)
12
Appendix C: Transformation from phase diffusion to period variability
Here, we illustrate the relationship between σ(θcp) and SD(θcp). Figure 4(a) presents a
schematic view of the trajectories of φ(t) and θ1(t). An enlarged view of the region around
(tcp + τ, θcp + 2π) is displayed in Fig. 4(b), in which the vertical width between the dotted
lines represents the standard deviation of the phase distribution of θ1(t). In particular, the
vertical arrow represents the standard deviation of θ1(tcp + τ ), which is denoted by σ(θcp).
Because we assume a low noise intensity, the actual trajectories of θ1(t) (thin lines) are very
close to that of φ(t). We can thus expect that the trajectories are approximately straight
and parallel to φ(t) in this enlarged region. Therefore, the horizontal width between the
dotted lines at θ1 = θcp + 2π is approximately equal to SD(θcp) (horizontal arrow), and the
relation σ(θcp)/SD(θcp) = φ(θcp) holds approximately.
FIG. 4. Illustration of the relationship between σ(θcp) and SD(θcp).
Appendix D: Definition of phase in the FitzHugh-Nagumo model
We define the phase θ as a function of (V, W ), which are the state variables of the
FitzHugh-Nagumo model given in Eq. (17) in the main text, as follows (Fig. 5). We first
assign φ values to all points on the limit cycle trajectory generated by Eq. (17) without noise
such that φ identically satisfies φ = 2π/τ , where τ is the period. The limit cycle trajectory
is independent of the coupling strengths. We set φ = 0 at V = 0.6 with V > 0. We then
consider radial lines extending from an arbitrary point inside the limit cycle, which we chose
as (0.6, 0.05) (filled square) in this case. When a radial line intersects the limit cycle at a
13
point that has a value of φ, the phase θ of all points on the radial line is defined by θ = φ.
These radial lines are different from isochrones that give a standard definition of the phase
[18], but the isochrones are usually unknown. As shown in Fig. 3, our theory is valid even
for this practical definition.
FIG. 5. Illustration of the definition of the phase in the FitzHugh-Nagumo model. The limit cycle
trajectory is generated by a coupled FitzHugh-Nagumo model without noise, whose parameters
are given in the main text. The circles are placed at equally spaced intervals of φ. All points on a
straight line radiating from the origin (filled square) have the same phase.
[1] H. Zhou, C. Nicholls, T. Kunz, and H. Schwartz, Frequency accuracy & stability dependencies
of crystal oscillators, Tech. Rep. (Technical Report SCE-08-12, Carleton University, Systems
and Computer Engineering, 2008)
[2] W. Rippard, M. Pufall, S. Kaka, T. Silva, S. Russek, and J. Katine, Phys. Rev. Lett. 95,
67203 (2005)
[3] S. Kaka, M. Pufall, W. Rippard, T. Silva, S. Russek, and J. Katine, Nature 437, 389 (2005)
[4] F. Mancoff, N. Rizzo, B. Engel, and S. Tehrani, Nature 437, 393 (2005)
[5] M. Keller, A. Kos, T. Silva, W. Rippard, and M. Pufall, Appl. Phys. Lett. 94, 193105 (2009)
[6] A. T. Winfree, The Geometry of Biological Time, 2nd ed. (Springer, New York, 2001)
[7] S. M. Reppert and D. R. Weaver, Nature 418, 935 (2002)
14
[8] L. Glass, Nature 410, 277 (2001)
[9] T. Yamaguchi, M. Soma, D. Halter, R. Raina, J. Nissen,
and M. Ishida, in VLSI Test
Symposium, 19th IEEE Proceedings on. VTS 2001 (IEEE, 2001) pp. 102 -- 110
[10] E. D. Herzog, S. J. Aton, R. Numano, Y. Sakaki, and H. Tei, J. Biol. Rhythms 19, 35 (2004)
[11] K. T. Moortgat, T. H. Bullock, and T. J. Sejnowski, J. Neurophysiol. 83, 971 (2000)
[12] J. R. Clay and R. L. DeHaan, Biophys. J. 28, 377 (1979)
[13] J. T. Enright, Science 209, 1542 (1980)
[14] D. J. Needleman, P. H. E. Tiesinga, and T. J. Sejnowski, Physica D 155, 324 (2001)
[15] K. Kojima, T. Kaneko, and K. Yasuda, Biochem. Biophys. Res. Commun. 351, 209 (2006)
[16] H. Kori, Y. Kawamura, and N. Masuda, J. Theor. Biol. 297, 61 (2012)
[17] A. T. Winfree, J. Theor. Biol. 16, 15 (1967)
[18] Y. Kuramoto, Chemical Oscillations, Waves, and Turbulence (Springer, New York, 1984)
[19] H. Kori and Y. Kuramoto, Phys. Rev. E 63, 046214 (2001)
[20] I. Z. Kiss, Y. M. Zhai, and J. L. Hudson, Phys. Rev. Lett. 94, 248301 (2005)
[21] J. Ariaratnam and S. Strogatz, Phys. Rev. Lett. 86, 4278 (2001)
15
|
1809.04900 | 1 | 1809 | 2018-09-13T11:59:48 | Theoretical model of transcription based on torsional mechanics of DNA template | [
"physics.bio-ph",
"q-bio.SC"
] | Transcription is the first step of gene expression, in which a particular segment of DNA is copied to RNA by the enzyme RNA polymerase (RNAP). Despite many details of the complex interactions between DNA and RNA synthesis disclosed experimentally, much of physical behavior of transcription remains largely unknown. Understanding torsional mechanics of DNA and RNAP together with its nascent RNA and RNA-bound proteins in transcription maybe the first step towards deciphering the mechanism of gene expression. In this study, based on the balance between viscous drag on RNA synthesis and torque resulted from untranscribed supercoiled DNA template, a simple model is presented to describe mechanical properties of transcription. With this model, the rotation and supercoiling density of the untranscribed DNA template are discussed in detail. Two particular cases of transcription are considered, transcription with constant velocity and transcription with torque dependent velocity. Our results show that, during the initial stage of transcription, rotation originated from the transcribed part of DNA template is mainly released by the rotation of RNAP synthesis. During the intermediate stage, the rotation is usually released by both the supercoiling of the untranscribed part of DNA template and the rotation of RNAP synthesis, with proportion depending on the friction coefficient in environment and the length of nascent RNA. However, with the approaching to the upper limit of twisting of the untranscribed DNA template, the rotation resulted from transcription will then be mainly released by the rotation of RNAP synthesis. | physics.bio-ph | physics |
Theoretical model of transcription based on torsional mechanics of DNA
template
Xining Xu and Yunxin Zhang∗
Laboratory of Mathematics for Nonlinear Science,
Shanghai Key Laboratory for Contemporary Applied Mathematics,
Centre for Computational Systems Biology, School of Mathematical Sciences,
Fudan University, Shanghai 200433, China.
(Dated: September 14, 2018)
Transcription is the first step of gene expression, in which a particular segment of DNA
is copied to RNA by the enzyme RNA polymerase (RNAP). Despite many details of the
complex interactions between DNA and RNA synthesis disclosed experimentally, much of
physical behavior of transcription remains largely unknown. Understanding torsional me-
chanics of DNA and RNAP together with its nascent RNA and RNA-bound proteins in
transcription maybe the first step towards deciphering the mechanism of gene expression. In
this study, based on the balance between viscous drag on RNA synthesis and torque resulted
from untranscribed supercoiled DNA template, a simple model is presented to describe me-
chanical properties of transcription. With this model, the rotation and supercoiling density
of the untranscribed DNA template are discussed in detail. Two particular cases of tran-
scription are considered, transcription with constant velocity and transcription with torque
dependent velocity. Our results show that, during the initial stage of transcription, rotation
originated from the transcribed part of DNA template is mainly released by the rotation of
RNAP synthesis. During the intermediate stage, the rotation is usually released by both the
supercoiling of the untranscribed part of DNA template and the rotation of RNAP synthe-
sis, with proportion depending on the friction coefficient in environment and the length of
nascent RNA. However, with the approaching to the upper limit of twisting of the untran-
scribed DNA template, the rotation resulted from transcription will then be mainly released
by the rotation of RNAP synthesis.
Keywords: transcription, torsional mechanics, torque dependent velocity
I.
INTRODUCTION
Essential for all cell functions, transcription is the first step of gene expression [1]. It begins
with the binding of RNA polymerase (RNAP) to a specific DNA sequence, which is usually
named promoter [2, 3]. As transcription proceeds, RNAP translocates along the template
strand and uses base pairing complementarity with the DNA template to create an RNA copy
[4].
Due to the nature of DNA's right-handed double-helical structure [5], transcription requires
a relative rotation of RNAP and its nascent RNA around DNA template. With elongation of
RNA, however, it might become much easier to rotate a segment of the DNA being actively
transcribed around its axis than to rotate the RNAC (short for RNA complex including RNA
polymerase, its nascent RNA and any other proteins needed in transcription) [6, 7]. As de-
scribed in the well-known twin-supercoiled-domain model [6], the advancing RNAC generates
∗Email: [email protected]
2
overwound (or positively supercoiling) for DNA downstream in transcription whereas DNA up-
stream becomes underwound (or negatively supercoiling), which has been validated in both in
vivo and in vitro studies as well [8 -- 13].
Recently, many efforts have been made to establish a full physical characterization of the
transcription process. With combination of the simple free energy model of plectonemic su-
percoil [14] and the free energy of extended twisted DNA [15], an analytical theory for the
coexistence of extended and supercoiled DNA came to light, providing closed-form expressions
for torque as a function of force and linking number [16]. Afterwards, [17] highlighted the signif-
icance of DNA supercoling as a major impact on gene expression. Then a framework describing
the coupled RNA elongation, RNAP rotation and DNA supercoiling was introduced in [18].
In this study, inspired by the framework given in [18], a general model to describe the
mechanical properties of transcription will be presented. With this model, the supercoiling
density of DNA, rotation of RNAC, as well as the transcription velocity along DNA template
can be obtained. Roughly speaking, for a long DNA template, transcription velocity and torque
resulted from the supercoiling of DNA will reach their limit values as transcription going on.
Meanwhile, with the motion of RNAP along DNA template, the total rotation shared by DNA
will first increase rapidly, then keep almost constant, and finally decrease rapidly to zero.
The organization of this study is as follows. The model and corresponding theoretical analysis
will be given in section II, and then results obtained by numerical calculations will be presented
in section III. Finally, conclusions and remarks will be given in section IV.
II. THEORETICAL MODEL BASED ON TORSIONAL MECHANICS
As stated in [18], the relative rotation of RNAC θ(x) and DNA sequence φ(x) are tied as
ω0x = φ(x) + θ(x).
(1)
Here, x is the spatial distance of RNAP along gene template away from the transcription start
site (TSS), with x > 0 for RNAP downstream from the TSS. The linking number of DNA
ω0 = 1.85 nm−1 converts distance into angle of rotation, since the two strands of DNA in
relaxed state wind around each other once approximately every 10.5 base pairs(bp), about 3.6
nm, forming a right-handed double helix [16]. The amount of φ(x) and θ(x) are related by the
following balance
τ (φ) = Γ(x, θ),
(2)
where τ (φ) is the torque resulted from the supercoiling of DNA template, and Γ(x, θ) is the
friction drag due to the viscous environment. Generally, Γ(x, θ) can be obtained by Γ(x, θ) =
γ(x) θ, with θ the rotation speed and γ(x) the rotational friction coefficient.
It is well known that the rotational friction coefficient γ depends on the shape and size of
RNAC, see [19]. With the forward motion of RNAP along DNA template, the length of nascent
RNA will increase linearly with distance x. To describe this dependence, this study assumes
that the friction coefficient γ of RNAC is a function of x, and denoted as γ(x). According to
the usual linear approximation between friction drag Γ and rotation speed, we get Γ = γ(x) θ.
Particularly, in Ref. [18], γ(x) = cxα is used.
3
In the following, we approximate the unknown function γ(x) as η(1 + kx), with two pa-
rameters η and k. Actually, η(1 + kx) can be regarded as the first two terms of the Taylor
expansion of γ(x). So, the friction drag Γ can be approximated as Γ = η(1 + kx) θ, in which
the length-independent term η θ can be regarded as the drag resulted from RNAP and other
bounded proteins needed in transcription, and the linear term ηkx θ can be regarded as the drag
resulted from the nascent RNA with length proportional to distance x.
By the chain rule of derivative,
θ = ∂tθ(x(t)) = θ′ x = vθ′. Where v = x is the elongation
speed of RNA. With identities in Eq. (1) and Eq. (2), we have
τ (φ) = η(1 + kx) θ = ηv(1 + kx)(ω0 −
dφ
dx
) .
(3)
In [18], it is assumed that the twisting strain of DNA sequence at the point of transcription,
which is caused by the forward motion of RNAP, spreads immediately throughout the specified
DNA length. That is to say, the twisting is assumed to be shared uniformly by the untranscribed
DNA sequence with length L − x, where L denotes the total length of the coding sequence. This
assumption is reasonable for slow transcription process, or transcription along a DNA template
with large stiffness. To describe more general cases, this study takes the spread process of
twisting strain into account and assumes that the DNA rotation (overwinding) φ is shared by the
untranscribed DNA sequence with length L−x, but through an exponential distribution function
with a parameter λ. Mathematically, the supercoiling density at the downstream position s from
the present transcription position x can be expressed as σ(s; x) = σ(0; x) exp(−λs). Considering
the total rotation φ(x) of DNA sequence, σ(s; x) should satisfy
Z L−x
0
σ(s; x)ds =Z L−x
0
σ(0; x)e−λsds =
φ(x)
ω0
,
(4)
which then gives the supercoiling density of DNA at the present transcription position x as
follows,
σ(0; x) =
λ
ω0
φ(x)(1 − e−λ(L−x))−1 .
(5)
Generally, parameter λ may depend on the dwell time of RNAP at each base pair of DNA
template, or equivalently depend on the transcription velocity v(x). For convenience of theoret-
ical analysis, this study only discusses the simple cases in which parameter λ is constant, this
is reasonable if transcription is almost in steady state.
For limit case λ → 0, σ(s; x) reduces to an uniform distribution, σ(s; x) ≡ σ(0; x) =
φ(x)/ω0(L − x), which is just the one discussed in [18]. For simplicity, we denote σ(0; x)
as σ(x) in the following. Eq. (5) can be reformulated as
φ(x) = ω0σ(x)(1 − e−λ(L−x))/λ .
Substitution of Eq. (6) into Eq. (3) gives
λτ (σ) = ω0ηv(1 + kx)(cid:0)λ − [σ(x)p(x)]′(cid:1) ,
with p(x) := 1 − e−λ(L−x).
(6)
(7)
According to Ref. [16], the supercoiling density σ(x) dependent torque τ resulted from the
overwinding of DNA template, as appeared in Eq. (7), can be approximated as follows,
4
τ (σ) =
Sσ, σ < σs ,
τ0, σs < σ < σp ,
P σ, σp < σ .
(8)
Where parameters S, τ0, P and supercoiling transition values σs, σp are determined by me-
chanical properties of DNA. Phenomenically, σs and σp are supercoiling densities when DNA
exceeds stretched state and reaches plectonemic, respectively.
The relation τ (σ) shown in Eq. (8) formulates Eq. (7) as two types of equations. For constant
torque τ ≡ τ0, i.e., when supercoiling density σ lies between σs and σp, Eq. (7) reduces to
λτ0 = ω0ηv(1 + kx)(λ − [σ(x)p(x)]′).
(9)
While for the other two linear relation cases, i.e., when σ < σs or σ > σp, Eq. (7) reduces to
λW σ = ω0ηv(1 + kx)(λ − σ′(x)p(x) − σ(x)p′(x)) ,
(10)
with W = S or W = P respectively, see Eq. (8).
For cases with constant transcription velocity, i.e., v ≡ v0, the first type equation (Eq. (9))
can be solved by direct integration, while for the other type of equation (Eq. (10)) it can be
solved by multiplying an integrating factor µW (x) = exp(cid:16)λfWR [(1 + kx)p(x)]−1dx(cid:17), withfW :=
W/(ω0ηv0). These lead to the full solution for supercoiling density as follows,
σ(x) =
µS(y) dy(cid:19)(cid:30) p(x)µS(x) ,
ln(1 + kx) + C1(cid:19)(cid:30) p(x),
λeτ0
µP (y) dy + C2(cid:19)(cid:30) p(x)µP (x) ,
k
0
0
(cid:18)λZ x
(cid:18)λx −
(cid:18)λZ x
σ < σs,
σs < σ < σp
,
(11)
σ > σp ,
where eτ0 := τ0/(ω0ηv0), and constants C1, C2 can be determined by matching boundary condi-
tions to keep σ(x) continuous at σ = σs and σ = σp. Finally, the rotation of DNA φ(x) and
torque τ (x) can be obtained by Eqs. (6, 8) and Eq. (11), and by Eq. (1) the rotation of RNAC
θ(x) = ω0x − φ(x) can also be obtained.
Experiments show that positive supercoiling built up by transcription is a impediment that
may slow down the motion of RNAC [20]. So, generally a stalling torque τc of RNAP, under
which the transcription velocity v is vanished, should be introduced. To describe these more
general cases, similar as the discuss in motor proteins [19, 21, 22], the following velocity-torque
relation of RNAP is assumed,
v = v0 [1 − (τ (σ)/τc)n ] ,
(12)
with v0 the torque-free velocity of RNAP. Take Eq. (12) into Eq. (9) and Eq. (10), we obtain
λeτ0 [1 − (τ0/τc)n ]−1
λfW σ(x) [1 − (W σ/τc)n ]−1
= (1 + kx)(λ − [σ(x)p(x)]′),
σs < σ < σp,
= (1 + kx)(λ − σ′(x)p(x) − σ(x)p′(x)),
σ < σs or σ > σp ,
(13)
(14)
5
with eτ0,fW defined the same as in Eq. (11).
Eq. (13) is different from Eq. (9) only in the left-hand side through an additional constant
[1 − (τ0/τc)n]−1, so its solution can be obtained explicitly. But for Eq. (14), it is much difficult
to obtain its solution for general exponent n. We will solve them numerically with the similar
boundary conditions as for Eqs. (9, 10) in the following section.
III. ROTATION AND TORQUE OF DNA TEMPLATE DURING TRANSCRIPTION
7200
4800
2400
0
0
12000
8000
4000
0
0
1600
800
0
500
0
1000
1000
2000
3000
4000
5000
800
400
0
500
0
1000
1000
2000
3000
4000
5000
18
12
6
0
0
3000
2000
1000
0
0
0.2
0.1
0
1000
0
500
1000
2000
3000
4000
5000
40
20
0
1000
0
500
1000
2000
3000
4000
5000
FIG. 1: Rotation φ, torque τ , supercoiling density σ of DNA template, and rotation θ of RNAP, as
a function of length x of the nascent RNA. Parameter values used in calculations are L = 5000 nm,
ηv0 = 1, k = 0.055, λ = 10−5, and S = 582.0, τ0 = 9.5, P = 189.6 pN nm [18]. Supercoiling density
σ(x) is calculated from Eq. (11), rotation φ(x) is calculated from Eq. (6), rotation θ(x) is calculated
by θ(x) = ω0x − φ(x) (see Eq. (1)), and torque τ is calculated from Eq. (8). The vertical dotted line
corresponds to the position at which rotation φ(x) reaches its maximum. The dashed lines in inserted
zoomed figures are calculated from the model employed in [18] for comparison, where friction drag on
RNAC is calculated by Γ = ηx1/2 θ, and rotation φ(x) of DNA template is always assumed to be shared
equally by the untranscribed part with length L − x.
In this section, based on theoretical models given in Sec. II, rotation of RNAC θ and DNA
template φ, as well as the torque τ and supercoiling density σ of DNA template will be discussed
in detail. Firstly, for the simple cases with constant transcription velocity v ≡ v0, the results
are displayed in Fig. 1, which are calculated from Eq. (9) and Eq. (10), or Eq. (11) equivalently.
Results show that, with the forward motion of RNAP, the rotation of DNA φ(x) will increase first
and then decrease to zero rapidly, though the supercoiling density σ(x) and torque τ (x) always
increase. For short nascent RNA (x . 750 nm), the slope of θ(x) decreases while that of φ(x)
increases with x, which means that the proportion of total rotation ω0x shared by DNA/RNAP
increases/decreases with x. One can image that, initially most of the rotation will be shared
6
by RNAP since there is no nascent RNA and therefore the friction drag Γ on RNAP is very
small. With the elongation of RNA, the friction drag Γ becomes larger compared with the strain
in DNA sequence, then more and more rotation will be shared by DNA sequence. However,
contrary to previous cases, for long nascent RNA (x & 750 nm), the slope of θ increases while
that of φ(x) decreases with x, which means that the proportion of total rotation ω0x shared
by DNA/RNAP decreases/increases with x. The reason is that, with increase of x, the friction
drag Γ on RNAC increases only linearly while the strain on DNA sequence increases nonlinearly
(with order larger than 1). Therefore, more and more rotation have to be shared by the RNAC.
Which has been observed previously in [6]. After a critical value x∗ (corresponding to the dotted
vertical line in Fig. 1), the slope of φ(x) will change from positive to negative and with the slope
of θ(x) increasing rapidly. This implies that, for x > x∗, all of the newly produced rotation
ω0(x − x∗) will be shared by RNAC. Meanwhile, the rotation previously stored in DNA's double
helix will also be partly shared by RNAC, since the downstream DNA's double helix almost
reaches it twisting limit. But the supercoiling density σ(x) and torque τ will never decrease.
Finally, all rotation of DNA template ω0L will be released by the rotation of RNAC, and with
φ(L) = 0. Due to their linear relationship (see Eq. (8)), torque τ (x) and supercoiling density σ
have similar behavior, except that τ (x) is a piecewise function with three pieces. Finally, dashed
lines plotted in the inserted zoomed subfigures of Fig. 1 are calculated by the model used in [18]
with friction drag given by Γ = ηx1/2 θ, with the assumption that rotation in the untranscribed
DNA template is distributed uniformly, or equivalently the untranscribed template with length
L − x is always in steady state. It shows that the model used in [18] can be regarded as the
limited case of the one presented in this study, with parameter λ tending to zero.
The influences of parameters k and λ (see Eq. (3) and Eq. (4)) on the mechanical properties
of DNA template are illustrated with Fig. 2. Obviously, the drag on RNAC Γ(x, θ) = ηv(1 +
kx) θ increases with k. For large k, more rotation will be shared by the untranscribed DNA
sequence, so rotation φ(x) and supercoiling σ(x) increases with parameter k. Moreover, our
results surprisingly show that φ ≈ αkβ, with α decreasing with λ while β almost independent of
λ, see Fig. 2(b). For large parameter λ, the twisting strain of DNA helix will concentrate more
at exactly the current transcriptional position x. Therefore, it will become more difficult to
rotate the DNA helix, and consequently, more rotation will be released by RNAC, which means
rotation φ(x) of DNA decreases with parameter λ, see Fig. 2(a). However, the supercoiling
density σ(x) at current transcription position increases with parameter λ, since with large λ
more rotation will be concentrated at the transcription point (see Fig. 2(c)), though the total
rotation φ(x) shared by all the untranscribed DNA helix decreases with λ. As mentioned before,
for λ → 0 the model presented in this study reduces to the simple case as employed in [18].
Meanwhile, plots in Fig. 2(a, c) show that both φ(x) and σ(x) become insensitive to parameter
λ for large or small values of λ, i.e., there are limit values for λ → 0 or λ → ∞. The limit
value of φ(x) for large λ is zero, which means that, when λ is large enough, almost all rotation
resulted from transcription will be released by RNAC. However, for λ → ∞, the limit value of
supercoiling density σ is nonzero, which increases with parameter k. This is because that, with
the increase of k, the drag Γ on RNAC will also increase. Therefore, the twist strain needed
to balance the drag should increase, see Eq. (2). Similarly, the limit values of φ(x) and σ(x)
for λ → 0 also increase with parameter k. Finally, the plots in Fig. 2(d) show that, for large
parameter λ, supercoiling density σ(x) increases almost linearly with k. The reason is that,
7
4200
(a)
2800
1400
0
10-5
(c)
3
2.5
2
1.5
1
0.5
0
-0.5
10-5
(b)
104
103
102
101
10-3
10-1
10-2
10-1
(d)
3
2
1
0
10-3
10-1
0
0.05
0.10
FIG. 2: DNA rotation φ and supercoiling density σ when RNAP reaches a particular position at x = 2500
nm away from the transcription start site, i.e., the midpoint of a DNA sequence of length L = 5000
nm. Here, S, τ0 and P are kept the same as those used in Fig. 1. (a) and (c) show how φ and σ change
with parameter λ for k = 0.1(dash-dot lines), 0.05(dashed lines) or 0.005(solid lines). While (b) and (d)
illustrate the behavior of φ and σ as functions of parameter k for four typical values of λ (dash-dot lines
for λ = 10−5, dashed lines for λ = 10−4, solid lines for λ = 10−2, and dotted lines for λ = 10−1 ).
for large λ almost all the twisting strain needed to balance the drag Γ on RNAC concentrates
on the current transcription position x, while for given value of x the drag Γ increases linearly
with parameter k. It is noted that, due to the linear relationship between τ and σ (see Eq. (8)),
the dependence of torque τ on parameters λ and k is similar as that of σ, and therefore is not
shown in Fig. 2. Finally, for more detailed calculations with regard to the constant transcription
velocity cases, see Figs. S1-S4, and Tab. SI in supplementary materials.
As mentioned in [23], the assumption that RNAP always moves along DNA template with
a constant velocity is not generally true, especially for real transcription in vivo. For example,
supercoiling density σ, and consequently the torque τ , will become pretty large as RNAP
approaches the end of template, especially for RNAP working in a viscous environment. With a
particular form of velocity-torque relation given in Eq. (12), results of rotation φ(x), supercoiling
density σ(x), and torque τ (x) of DNA template are plotted in Fig. 3. Similar as the constant
velocity cases, rotation φ(x) increases first and then decreases to zero with the completion of
transcription, and both σ(x) and τ (x) increase monotonically with the length x of nascent
RNA. But an obvious difference is that rotation φ(x) in Fig. 3 increases more rapidly than
that in Fig. 1, while decreases more slowly than that in Fig. 3. With the increase of length
x of nascent RNA, twist strain induced by the overwinding of DNA helix will increase. This
implies that the transcription velocity v decreases with x, see the left-bottom figure in Fig. 3.
Meanwhile, Fig. 3 shows that, during the initial period of transcript, the torque τ (x) increase
also very rapidly. Thus, the drag on RNAC Γ = η(1 + kx)vθ′ increases very rapidly with x
8
120
80
40
0
0
1.0
0.5
0
0
1000
2000
3000
4000
5000
1 0 0 0 0
5 0 0 0
0
0
2 5 0 0
5 0 0 0
1000
2000
3000
4000
5000
0.072
0.048
0.024
0
0
16
8
0
0
1000
2000
3000
1000
2000
3000
FIG. 3: Rotation φ(x), supercoiling density σ(x), torque τ (x) of DNA template, and the transcription
velocity v for general cases in which velocity v is torque dependent, see Eqs. (12, 13, 14). Parameter
values used in calculations are L = 5000 nm, k = 0.055, λ = 10−3, n = 2, and ηv0, S, τ0, P are kept the
same as those in Fig. 1. The same as [23], stalling torque of the RNAP is assumed to be τc = 12 pN nm.
Similar to Fig. 1, dotted lines show the position x at which DNA rotation φ reaches its maximum.
during the initial period.
In some sense, Fig. 3 shows that, with torque dependent velocity,
transcription (especially the physical state of DNA template) will reach its steady state more
rapidly.
In the initial period of transcription, most of the rotation will be shared by DNA
helix until it approximates its upper limit of capacity, see the plots in Fig. 3 for torque τ and
supercoiling density σ.
Similar to Fig. 2, dependent properties of parameters λ and k obtained by the general model
with torque dependent velocity of RNAP are plotted in Fig. 4. The dependence of DNA rotation
φ on parameter λ is similar to that in Fig. 2, the constant transcription velocity cases, which
changes from one limit constant to zero as the increase of λ. But the supercoiling density σ
is almost independent of λ, especially for large k cases. Consequently, the velocity v is almost
independent of λ, see Eq. (12). These imply that, with the increase of λ, the supercoiling
density σ, or equivalently the torque τ needed to balance the drag Γ on RNAC, is almost
constant at current transcription site, though the total rotation φ of the untranscribed DNA
helix decreases. Or in other words, although the total rotation φ of DNA decreases with λ, the
supercoiling density σ at the current transcription position is almost independent of λ, since
large parameter λ means more proportion of rotation φ will be concentrated at the current
transcription position. In short, two factors that might influence the value of σ balance each
other. On the other hand, the three right subfigures in Fig. 4 show that rotation φ is almost
independent of parameter k, while supercoiling density σ increases slightly with k. With large
value of k, the drag Γ = η(1 + kx) θ on RNAC will be large, therefore the torque τ induced
by the overwinding of DNA helix, which is needed to balance Γ, should be large. One possible
9
200
100
0
10-5
100
0
10-5
100
0
10-5
10-5
100
0
100
0
10-5
100
0
10-5
10-3
10-1
10-3
10-1
10-3
10-1
10-3
10-1
10-3
10-1
FIG. 4: Rotation φ, supercoiling density σ of DNA helix, and velocity v of RNAP, at a typical position
which is x = 3500 nm away from the transcription start site. The parameter values used in calculations
are L = 5000 nm, τc = 12 pN nm, ηv0 = 1, S = 582.0 pN nm, τ0 = 9.5 pN nm and P = 189.6 pN nm.
The λ dependent results are shown on the left with k = 0.1, 0.05, 0.005, respectively. The k dependent
results are shown on the right with λ = 10−5, 10−3, 10−1 respectively.
reason that φ is almost independent of k and σ is almost independent of λ is that, at position
x = 3500 nm, the rotation φ and supercoiling density σ have almost reached their upper limits,
and therefore are not sensitive to the change of drag Γ and parameter λ, respectively. For
calculation results with x = 100 nm, see Fig. S5 in supplementary materials.
IV. CONCLUSIONS AND REMARKS
In this study, theoretical model to describe the mechanical properties of transcription process
is presented. Roughly speaking, our model is mainly based on the balance between the friction
drag on RNAC, which is caused by the viscous environment in cells, and the torque induced by
the overwinding of DNA helix caused by transcription. As usual, the drag is calculated by the
product of friction coefficient and rotation velocity, but with the friction coefficient depending
on the length of nascent RNA, as discussed in [10]. Meanwhile, the torque is calculated by the
supercoiling density of DNA helix. The basic idea of our model is similar to the one used in
[18], while the methods to calculate the friction drag on RANC and the supercoiling density
of DNA helix are generalized. Two types of transcription are discussed in this study, one with
constant transcription velocity and the other with torque dependent velocity. Our results show
that, during the initial period of transcription, the rotation coming from the transcribed part of
DNA template is mainly shared by the untranscribed part of DNA template. With the increase
of supercoiling density and twisting strain induced by overwinding of the untranscribed part
of DNA template, the rotation resulted from transcription will then be mainly released by the
10
rotation of RNAC. But anyway, all the rotation of DNA template will finally be totally released
by the rotation of RNAC. Although the model presented in this study still looks too simple
to get detailed properties of transcription quantitively, it is reasonable enough to understand
the basic principle of transcription, especially the mechanism of how to deal with the natural
rotation stored in DNA template during the transcription process. It may need to note that the
model presented in this study can also be generalized to describe the transcription process with
multiple (cooperated) RNAPs aligning along one single DNA template, which may be helpful
to the understanding of transcriptional bursting in gene expression [20, 24].
[1] G. M. Cooper. The Cell: A Molecular Approach, 2nd Edn. Sinauer Associates, Inc., Sunderland,
Mass., 2000.
[2] P. L. Dehaseth, M. L. Zupancic, and M. T. Record Jr. RNA polymerase-promoter interactions: the
comings and goings of RNA polymerase. Journal of Bacteriology, 180:3019 -- 3025, 1998.
[3] G. M. T. Cheetham, D. Jeruzalmi, and T. A. Steitz. Structural basis for initiation of transcription
from an RNA polymerase promoter complex. Nature, 399:80 -- 83, May 1999.
[4] J. D. Watson, T. A. Baker, S. P. Bell, A. A. Gann, M. Levine, and R. M. Losick. Molecular Biology
of the Gene, 7th Edn. Pearson, 2013.
[5] M. H. Wilkins, A. R. Stokes, and H. R. Wilson. Molecular structure of deoxypentose nucleic acids.
Nature, 171(4356):738 -- 740, 1953.
[6] L. F. Liu and J. C. Wang. Supercoiling of the DNA template during transcription. Proceedings of
the National Academy of Sciences, 84(20):7024 -- 7027, 1987.
[7] G. N. Giaever and J. C. Wang. Supercoiling of intracellular DNA can occur in eukaryotic cells.
Cell, 55(5):849 -- 856, 1988.
[8] Y. P. Tsao, H. Y. Wu, and L. F. Liu. Transcription-driven supercoiling of DNA: direct biochemical
evidence from in vitro studies. Cell, 56(1):111 -- 118, 1989.
[9] A. S. Krasilnikov, A. Podtelezhnikov, A. Vologodskii, and S. M. Mirkin. Large-scale effects of
transcriptional DNA supercoiling in Vivo. Journal of Molecular Biology, 292(5):1149 -- 1160, 1999.
[10] Y. Harada, O. Ohara, A. Takatsuki, H. Itoh, N. Shimamoto, and K. Kinosita Jr. Direct observation
of DNA rotation during transcription by Escherichia coli RNA polymerase. Nature, 409(6816):113 --
115, 2001.
[11] F. Kouzine, J. Liu, S. Sanford, H. J. Chung, and D. Levens. The dynamic response of upstream DNA
to transcription-generated torsional stress. Nature Structural and Molecular Biology, 11(11):1092 --
1100, 2004.
[12] S. Forth, M. Y. Sheinin, J. Inman, and M. D. Wang. Torque measurement at the single-molecule
level. Annual Review of Biophysics, 42(42):583 -- 604, 2013.
[13] J. Ma and M. D. Wang. DNA supercoiling during transcription. Biophysical Reviews, 8(1):75 -- 87,
2016.
[14] V. V. Rybenkov, A. V. Vologodskii, and N. R. Cozzarelli. The effect of ionic conditions on dna helical
repeat, effective diameter and free energy of supercoiling. Nucleic Acids Research, 25(7):1412 -- 1418,
1997.
[15] J. D. Moroz and P. Nelson. Torsional directed walks, entropic elasticity, and DNA twist stiffness.
Proceedings of the National Academy of Sciences, 94(26):14418 -- 14422, 1997.
[16] F. J. Marko. Torque and dynamics of linking number relaxation in stretched supercoiled DNA.
Physical Review E, 76(2):021926, 2007.
[17] C. Lavelle. Pack, unpack, bend, twist, pull, push: the physical side of gene expression. Current
Opinion in Genetics & Development, 25:74 -- 84, 2014.
[18] S. A. Sevier and H. Levine. Mechanical properties of transcription. Physical Review Letters,
11
118(26):268101, 2017.
[19] J. Howard. Mechanics of Motor Proteins and the Cytoskeleton. Sinauer Associates and Sunderland,
MA, 2001.
[20] S. Chong, C. Chen, H. Ge, and X. S. Xie. Mechanism of transcriptional bursting in bacteria. Cell,
158(2):314 -- 326, 2014.
[21] M. J. I. Muller, S. Klumpp, and R. Lipowsky. Tug-of-war as a cooperative mechanism for bidi-
rectional cargo transport by molecular motors. Proceedings of the National Academy of Sciences,
105:4609 -- 4614, 2008.
[22] A. Kunwar and A. Mogilner. Robust transport by multiple motors with nonlinear forcecvelocity
relations and stochastic load sharing. Physical Biology, 7(1):16012, 2010.
[23] J. Jie, L. Bai, and M. D. Wang. Transcription under torsion. Science, 340(6140):1580 -- 1583, 2013.
[24] D. Nicolas, N. E. Phillips, and F. Naef. What shapes eukaryotic transcriptional bursting? Molecular
Biosystems, 13(2), 2017.
|
1806.01902 | 1 | 1806 | 2018-06-05T19:24:16 | Swimming of peritrichous bacteria is enabled by an elastohydrodynamic instability | [
"physics.bio-ph",
"cond-mat.soft",
"physics.flu-dyn"
] | Peritrichously-flagellated bacteria, such as Escherichia coli, self-propel in fluids by using specialised motors to rotate multiple helical filaments. The rotation of each motor is transmitted to a short flexible segment called the hook which in turn transmits it to a flagellar filament, enabling swimming of the whole cell. Since multiple motors are spatially distributed on the body of the organism, one would expect the propulsive forces from the filaments to push against each other leading to negligible swimming. We use a combination of computations and theory to show that the swimming of multi-flagellated bacteria is enabled by an elastohydrodynamic bending instability occurring for hooks more flexible than a critical threshold. Using past measurements of hook bending stiffness, we demonstrate how the design of real bacteria allows them to be safely on the side of this instability that promotes systematic swimming. | physics.bio-ph | physics | Swimming of peritrichous bacteria is enabled by an elastohydrodynamic instability
Emily E. Riley,∗ Debasish Das,† and Eric Lauga‡
Department of Applied Mathematics and Theoretical Physics, University of Cambridge, UK.
(Dated: June 7, 2018)
Peritrichously-flagellated bacteria, such as Escherichia coli, self-propel in fluids by using spe-
cialised motors to rotate multiple helical filaments. The rotation of each motor is transmitted to
a short flexible segment called the hook which in turn transmits it to a flagellar filament, enabling
swimming of the whole cell. Since multiple motors are spatially distributed on the body of the
organism, one would expect the propulsive forces from the filaments to push against each other
leading to negligible swimming. We use a combination of computations and theory to show that
the swimming of multi-flagellated bacteria is enabled by an elastohydrodynamic bending instability
occurring for hooks more flexible than a critical threshold. Using past measurements of hook bend-
ing stiffness, we demonstrate how the design of real bacteria allows them to be safely on the side of
this instability that promotes systematic swimming.
Although out of sight, bacteria dominate chemical pro-
cesses on our planet. They are the most abundant organ-
isms on earth and, equipped with the ability to live in
extreme and hostile conditions, they play crucial roles in
both the environment and human health [1]. Many bac-
teria self-propel in response to physical and chemical cues
by actuating specialised, rotary motors in bulk fluid en-
vironments [2]. Each motor imposes a moment normal to
the surface of the cell body transmitted to a helical flag-
ellar filament via a short elastic segment called the hook
that acts as a universal joint [3, 4]. Due to the helical na-
ture of flagellar filaments, the rotation imposed by each
motor is not time-reversible and as a result bacteria are
able to swim [5]. While a flagellar filament can take one
of eleven polymorphic forms, the normal form used for
swimming is left-handed and rotates counter-clockwise
(CCW, looking from the flagellum to the cell) propelling
the bacterium cell-first [6, 7], a type of swimmer known
as a pusher [8].
If the same left-handed helix were to
rotate in the opposite direction then the cell would swim
flagella first and be a puller [9].
Peritrichous bacteria possess multiple flagella that can
grow from essentially any point on the cell body sur-
face [10, 11]. Well-studied examples include Escherichia
coli (E. coli, Fig. 1A), Bacillus subtilis and Salmonella
enterica. During the swimming of these pusher cells, all
flagellar filaments gather and bundle at one end of the
body propelling the cell forward (Fig. 1B). The main ad-
vantage of possessing multiple flagella is not increased
propulsion [12] but rather the ability to change direction
via tumbling. This occurs when at least one of the rotary
motors slows down [13] or reverses its direction [7] caus-
ing the bundle to break-up (unbundling). At the end of
a tumble, the motors return to their swimming state, the
bundle reforms and the bacterium swims in a new direc-
tion. Through a modulation of the tumbling frequency,
∗ Present address: Centre for Ocean Life, Technical University of
Denmark; These authors equally contributed to this work.
† These authors equally contributed to this work.
‡ [email protected]
bacteria can move towards favourable environments [14].
Crucial for the formation of the filament bundle, and suc-
cessful swimming, is the flexible hook. When the hook
is stiffened, bacteria are stuck in a tumble mode and can
barely swim [15]. Hook flexibility is also crucial for singly
flagellated bacteria, enabling changes in swimming direc-
tions via buckling [16] but causing unstable locomotion
if it is too flexible [17].
Much theoretical work has been devoted to predict-
ing the propulsion mechanisms of self-propelled bacteria
[18]. Most studies assume a fixed relative position be-
tween helical filaments (or bundles) and the cell body,
and modelling tools have been developed to address both
swimming [8, 19–21] and the bundling/unbundling pro-
cess [22, 23].
If peritrichous bacteria have similar filaments dis-
tributed spatially around their cell body, why are the
flagella not all pushing against each other leading to neg-
ligible swimming? In this paper we use a combination of
computations and theory to show that swimming is en-
abled by an elastohydrodynamic instability of the hook.
If hooks are too rigid, flagellar filaments always point
normal to the cell body surface and never bundle.
In
contrast, when the bending rigidity of the hook is be-
FIG. 1. Swimming E. coli bacteria. A: Peritrichous bac-
terium with multiple flagella spatially distributed around the
cell body; B: Flagellar filaments are located behind the swim-
ming cell in a helical bundle whose rotation push the cell
forward. Reproduced from Turner, Ryu & Berg (2000) J.
Bacteriol., 182, 2793–2801 [6]. Copyright 2000 American So-
ciety for Microbiology.
8
1
0
2
n
u
J
5
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
2
0
9
1
0
.
6
0
8
1
:
v
i
X
r
a
2
about the axis of the flagellar filament; (ii) a short flexi-
ble hook treated as a torsion spring about the motor axis
whose hydrodynamics can be neglected [17]; (iii) a heli-
cal flagellar filament of the normal left-handed polymeric
form whose hydrodynamics is captured with slender-
body theory [24]. Motor and filament parameters are
chosen to match those of E. coli bacteria [7] (Table S1 in
supplementary information). Each helical filament has
a tapered end such that the helix radius is zero at its
attachment point to the motor [25]. Flagellar filaments
can rotate but not translate relative to their attachment
point on the cell body and while the rotation about the
helix axis is imposed by the motor, any further rotations
relative to the body are solved for. We neglect hydro-
dynamic interactions between the cell body and flagellar
filaments but include steric interactions to prevent fila-
ments from entering the body. For each hook, we use θ
to denote the tilt angle between the normal to the cell
body at the motor location and the axis of the flagel-
lar filament (i.e. when θ = 0 the filament is normal to
the cell body). The restoring elastic moment imposed by
each motor on its flagellar filament is modelled as torsion
spring of spring constant K = EI/(cid:96)h, where EI and (cid:96)h
are the bending rigidity and length of the hook respec-
tively [16]. The magnitude of the restoring moment is
thus given by Kθ and the elasticity of the hook acts
to align the helix axis with the normal to the cell body.
The computational model solves for the instantaneous
positions of the flagellar filaments and for the swimming
velocity, Ub, and angular velocity, Ωb, of the cell body
as a function of the hook stiffness.
Pusher bacteria with flexible hooks undergo a
swimming instability
Examining the results of our computational model un-
covers a remarkable elastohydrodynamic instability, illus-
trated in Fig. 3 in the case of Nf = 4 flagella, the average
number of flagella on an E. coli cell [6]. The motors are
positioned symmetrically around the surface of the cell
body. We start the computations with each flagellar fila-
ment tilted at some small angle away from the normal to
the surface and march the system forward in time while
tracking the position of the cell in the laboratory frame
and of the flagellar filaments relative to the cell body.
Associated movies are available in supplementary infor-
mation.
In Fig. 3A-C we illustrate the trajectory of a pusher
bacterium (i.e. a cell with flagellar filaments undergo-
ing normal CCW rotation) with two different hook stiff-
nesses over a time scale t = 200 (time nondimension-
alised by the rotation rate of the flagella). While both
start at the same location (A), the cell with the flexible
hook (K = 0.1) ends up with their flagellar filaments all
wrapped in the back and is able to swim five times as
fast (B) as the stiff-hooked cell (K = 100) whose flagel-
lar filaments have remained in the same splayed config-
FIG. 2. A: Computational model of a peritrichous bacterium
actuating Nf helical filaments (radius Rh; angle β) by rotat-
ing them about their axis k with prescribed angular velocity.
The flexible hook acts elastically to align the helix axis with
the normal N to the cell body. B: Simplified model to cap-
ture the elastohydrodynamic instability. Two straight active
filaments of length (cid:96) attached on either side of a spherical
body of radius a are tilted at an angle ±θ away from the cell
body surface normal, N, and act on the cell with tangential
force f (cid:96)t resulting in swimming of the model bacterium with
velocity U y.
low a critical threshold, the feedback between the flow
induced by the flagella and hook bending leads to a con-
formational instability resulting in all flagellar filaments
gathered at the back of the cell, and net locomotion. This
sharp transition from negligible to successful swimming is
observed numerically with decreasing hook stiffness and
we show that this instability can be rationalised using
a simple model of a cell propelled by two straight active
filaments. By examining past measurements of hook flex-
ibility, we demonstrate that bacteria are safely designed
to be on the swimming side of the instability.
RESULTS
Modelling of multi-flagellated bacteria
We start by building a computational model of the
locomotion of a peritrichous bacterium, as outlined in
the Methods section with mathematical details in sup-
plementary information 1. We consider a bacterium pro-
pelled by Nf flagella (Fig. 2A) with a cell body in the
shape of a prolate ellipsoid. Each flagellum consists of:
(i) a rotary motor that generates a fixed rotation rate
1 Supplementary information available upon request by writing to:
[email protected].
3
FIG. 3. Swimming of a bacterium with Nf = 4 flagella with a flexible vs. stiff hook. A: Initial position and conformation of
each cell; B: Pusher cell with flexible hook at t = 200 (times scaled by rotation rate of flagella); C: Pusher cell with stiff hook
at t = 200; D: Distance travelled by each swimmer (nondimensionalised by the pitch of the helical filaments) as a function of
time for four different swimmers: stiff (diamonds) vs. flexible hook (squares) and pusher (filled symbols) vs. puller (empty).
uration (C). This is quantified in Fig. 3D where we plot
the net distance travelled as a function of time (scaled
by the pitch of the helix). The cell with a flexible hook
(filled square) swims consistently faster than the stiff one
(filled diamond).
If alternatively we reverse the direc-
tion of rotation of the flagella to rotate in the clockwise
(CW) direction, the cell becomes a puller and does not
transition to fast swimming for neither a flexible hook
(empty squares) nor a stiff one (empty diamonds). Note
that the two stiff cases (pushers and pullers; diamonds)
have identical swimming magnitude, a consequence of the
kinematics reversibility of Stokes flows [5]. Importantly,
the transition to fast swimming for flexible pusher bac-
teria does not occur smoothly with changes in the hook
stiffness but instead it takes place at a critical dimen-
sionless value of Kc ≈ 1 (nondimensionalised using the
viscosity of the fluid, the pitch of the helical filament and
the frequency of rotation). Above Kc, all flagella remain
normal to the cell (θ ≈ 0) leading to negligible swim-
ming while below Kc, all flagella wrap behind the cell
(θ ≈ π/2) leading to a net locomotion.
This sharp transition does not originate from a buck-
ling instability of the hook which is only modelled here
at the level of a torsional spring [16]. Instead, the insta-
bility arises from the two-way coupling between the con-
formation of the flagella and cell locomotion. To unravel
the physics of this instability, we consider in more detail
the case of a spherical cell body and two flagella, which
is the minimum configuration able to show the instabil-
ity while capturing the same physics as geometrically-
complex cases. The steady-state computational results
in this case are shown in the main part of Fig. 4 (sym-
bols and thin lines) for the angle θ between the axis of the
flagellar filaments and the cell body (A) and for the net
lab-frame swimming speed U of the cell (B). While the
flagella conformation of puller bacteria is independent of
the hook stiffness and leads to zero swimming (light red
circles), pusher cells clearly display a sudden jump to a
wrapped conformation and a net locomotion for a hook
stiffness below Kc ≈ 0.79 (dark blue circles).
Analytical model of the elastohydrodynamic
instability
The observed dynamics can be captured by an analyt-
ical model demonstrating that swimming occurs as the
result of a linear elastohydrodynamic instability. Con-
sider the simple geometrical model illustrated in Fig. 2.
Two straight active filaments of length (cid:96) are symmetri-
cally attached on either side of a spherical cell body of
radius a and are tilted at an angle ±θ away from the body
surface normal, N. Each filament, elastically attached to
the cell body via a hook modelled as a torsion spring
of stiffness K, pushes on the cell along their tangential
direction with propulsive force density ft which results
in the swimming of the bacterium with velocity U y (see
Fig. 2 for all notation). For CCW motion, the propulsion
forces point towards the cell body (f < 0) and the cell
is a pusher. In contrast, for CW motion, the propulsive
forces point away from the cell (f > 0) and the swimmer
is a puller.
The swimming speed (U ) and the rate of change of the
conformation of the filaments ( θ) may be obtained by
enforcing force and moment balance. Using c(cid:107) and c⊥ to
denote the drag coefficients for a slender filament moving
parallel and perpendicular to its tangent respectively, the
balance of forces on the whole cell in the direction of
swimming, y, is written as
−6πµaU − 2(cid:96)U(cid:0)c(cid:107) sin2 θ + c⊥ cos2 θ(cid:1)
(1)
+ θ(cid:96)2c⊥ cos θ = 2f (cid:96) sin θ,
4
FIG. 4. Steady-state flagella tilt angles (θ, A) and lab-frame swimming speeds (U , B) for the full computational model
of Fig. 2A with two flagella (symbols and thin lines) and for the simple active filament model of Fig. 2B (thick lines) as a
function of the dimensionless hook spring constant, K. Light green line and light red symbols: puller bacterium for which the
non-swimming state is always stable; Dark red line and dark blue symbols: pusher bacterium which undergoes a transition to
swimming for K < Kc. The dashed line shows the critical spring constant predicted theoretically, Kc ≈ 0.53.
where the first two terms (the terms on the first line)
are due to the drag on the cell body and on the active
filaments due to swimming, the third term is the drag
on the filaments due to rotation and the last term is the
total propulsive force acting on the cell.
The second equation comes from the balance of mo-
ment on each active filament, written in the z = x × y
direction at the attachment point on the cell surface as
− (cid:96)3
3
c⊥ θ +
(cid:96)2
2
U c⊥ cos θ − Kθ = 0,
(2)
where the first term is the hydrodynamic moment due to
rotation of the filament, the second is the hydrodynamic
moment due to the swimming drag and the last term
is the elastic restoring moment from the hook acting to
return the filament to its straight configuration. Com-
bining Eqs. (1) and (2) leads to the evolution equation
for θ(cid:32)
c⊥ −
(cid:96)3
3
c2⊥ cos2 θ(cid:96)4
12πµa + 4(cid:96)(c(cid:107) sin2 θ + c⊥ cos2 θ)
−f (cid:96)3 sin θ cos θc⊥
6πµa + 2(cid:96)(c(cid:107) sin2 θ + c⊥ cos2 θ)
(cid:33)
θ = (3)
− Kθ.
When the elastic moment dominates, the straight con-
figuration θ = 0 is the only steady state, associated with
no swimming. If instead the elastic moment is negligi-
ble, the swimming states with θ = ±π/2 become possible
equilibria.
To examine how a variation of the hook stiffness allows
transition from one state to the next, we solve Eq. (3)
numerically with the appropriate flagellar filament val-
ues for a wild type swimming E. coli cell and using the
magnitude of f leading to agreement with the full com-
putations at zero hook stiffness. We start with small
perturbations around θ = 0 and compute the long-time
steady state of Eq. (3), with results illustrated in Fig. 4
for both pusher (dark red line) and puller (light green
line). Puller cells never swim for any value of the hook
stiffness, and the straight configuration θ = 0 is always
stable. In contrast, pushers cannot swim for hooks stiffer
than a critical value but undergo a sudden transition to
direct swimming for softer hooks, in excellent agreement
with the computations of the full two-flagella case (sym-
bols in Fig. 4).
The sudden transition to swimming for a critical
hook stiffness can be predicted analytically by linearising
Eq. (3) near the equilibrium at θ = 0, leading to
(cid:19)
(cid:18) 4πµac⊥(cid:96)3 + 1
12πµa + 4c⊥(cid:96)
(cid:19)
(cid:18)
3 c2⊥(cid:96)4
θ ≈ −
K +
f c⊥(cid:96)3
6πµa + 2c⊥(cid:96)
θ.
(4)
If f is positive (puller) then the configuration with θ = 0,
which is associated with no swimming U = 0, is always
linearly stable to small perturbations for any value of K.
In contrast, pushers with f < 0 are linearly unstable for
K < Kc such that the right-hand side of Eq. (4) becomes
positive, i.e. Kc = −f c⊥(cid:96)3/(2c⊥(cid:96) + 6πµa). A linear elas-
tohydrodynamic instability enables therefore pusher bac-
teria with sufficiently-flexible hooks to dynamically tran-
sition to an asymmetric conformation (θ (cid:54)= 0) with net
swimming (U (cid:54)= 0). Note that the simple theoretical
model (linear stability and numerical solution of Eq. 3)
predicts a critical dimensionless stiffness of Kc ≈ 0.53, in
agreement with the computations for the full bacterium
model, Kc ≈ 0.79.
DISCUSSION
METHODS
5
How does this swimming instability affect real bacte-
ria? We first note that for the instability to be relevant,
the rotary motors need to be spatially distributed around
the organism and therefore the instability would not oc-
cur if the rotary motors were all located around the same
position on the cell body. Lophotrichous bacteria whose
multiple flagella are positioned at the pole of the cell (for
example, Helicobacter pylori) would therefore not be sub-
ject to this instability, but peritrichous bacteria such as
E. coli and Salmonella enterica would.
By examining past measurements on the bending stiff-
ness of peritrichous bacterial hooks, we next discover
that swimming bacteria are safely on the unstable side,
explaining their ability to swim despite the presence of
spatially distributed motors. The strength of the torsion
spring in our model is given by K = EI/(cid:96)h, where EI
is the bending rigidity of the hook and (cid:96)h its length. A
recent study measured the hook flexibility for different
species of peritrichous bacteria by extracting and stain-
ing the flagellar hooks and using electron microscopy to
observe their deformations due to thermal fluctuations
[26]. In this study they found that E. coli and Salmonella
enterica had similar hook bending stiffness in the range
EI ≈ 1.6 − 4.8 × 10−28 Nm2 while singly flagellated bac-
teria can have stiffer hooks [16]. Re-dimensionalising our
computational results above using the viscosity of water
(1 mPas) and the pitch (2.22 µm) and frequency (110 Hz)
of E. coli flagella [7], we obtain Kc ≈ 9.6 × 10−19 Nm.
Using a hook length (cid:96)h ≈ 55 nm [2], we therefore pre-
dict a critical hook stiffness of EI ≈ 5.5 × 10−26 Nm2.
The hook stiffness of peritrichous bacteria is thus two or-
ders of magnitude smaller than the critical value for the
instability.
In summary, we showed theoretically and computa-
tionally that pusher peritrichous bacteria can swim by ex-
ploiting an elastohydrodynamic instability while pullers
never can. This instability is due to the bending rigidity
of the hooks and is different from the buckling instabil-
ity displayed by polar bacteria [16]. The physics of this
instability lies in the feedback between the conformation
of the flagella and the swimming of the cell. Flagellar
filaments create propulsive forces which propel the cell
forward. In the frame of the moving cell, the filaments ex-
perience hydrodynamic moments aligning them with the
direction of swimming. The rigidity of the hook balances
these hydrodynamic moments and below a critical rigid-
ity, an elastohydrodynamic instability transitions the fil-
aments from a splayed state to a conformation where they
are gathered behind the cell. Our results rationalise the
ability of real peritrichous bacteria to swim by showing
that they are designed to undergo a successful transition
to swimming after each tumble.
We give here a brief outline of the computational
model, with all details found in supplementary informa-
tion. Our computational model solves for the instan-
taneous positions of the flagellar filaments and for the
swimming velocity, Ub, and angular velocity, Ωb, of the
cell body by enforcing mechanical equilibrium at all time
(inertia is irrelevant at the scale of bacteria). At low
Reynolds numbers, the balance of hydrodynamic forces
and moments for the whole cell at its centre leads to a lin-
ear relationship between the swimming kinematics of the
cell and the angular velocities of each filament, denoted
by ωi for ith filament, of the form
=
Υi · ωi,
(5)
(cid:18)Ub
(cid:19)
Ωb
Nf(cid:88)
i=1
where the tensors Υi depend on the hydrodynamic re-
sistance of each individual component of the cell and on
their geometrical arrangements.
The rotation rate of each filament has a prescribed
value along its helical axis and we need two additional
equations to solve for the other two components. This
is obtained by examining the local balance of moments.
In the frame of a filament, the swimming velocity and
rotations are experienced as background flows which act
to tilt the flagellum away from the normal to the motor
while the hook applies an elastic restoring moment. The
hydrodynamic moment acting on filament i may be writ-
ten as Li = Γi · Ub + Λi · Ωb + ∆i · ωi, where the tensors
Γi, Λi and ∆i dependent on the geometry and relative
configuration of the flagellar filament and cell body, and
are proportional to the fluid viscosity. If we use the unit
vector Ni to denote the direction normal to the cell body
surface at the location of the motor and if ki is the unit
vector along the axis of the filament (see Fig. 2, top) then
the restoring elastic moment acts along the unit vector
Hi = ki× Ni and the balance between the hydrodynamic
moment and restoring elastic moment from the hook is
written for all times as
Li · H + Kθi = 0.
(6)
Finally, we assume that there is no elastic resistance for
the filament to move in the direction Ji = ki × Hi per-
pendicular to both ki and Hi and thus the final moment
equation is Li · J = 0.
Acknowledgements
We thank Lyndon Koens for useful discussions. This
project has received funding from the European Research
Council (ERC) under the European Union's Horizon
2020 research and innovation programme (grant agree-
ment 682754 to EL) and from the EPSRC (EER).
6
[1] P. H. Raven and G. B. Johnson. Part VIII: Viruses and
Simple Organisms. In Biology. McGraw-Hill, 6th edition
edition, 2002.
[2] H. C. Berg. The rotary motor of bacteria flagella. Annu.
Rev. Biochem., 72:19–54, 2003.
[3] S. M. Block, D. F. Blair, and H. C. Berg. Compliance
of bacterial optical polyhooks measured with tweezers.
Cytometry, 12:492–496, 1991.
[4] F. A. Samatey, H. Matsunami, K. Imada , S. Nagashima,
T. R. Shaikh, D. R. Thomas, J. Z. Chen, D. J. DeRosier,
A. Kitao, and K. Namba. Structure of the bacterial flag-
ellar hook and implication for the molecular universal
joint mechanism. Nature, 431:1062–1068, 2004.
[5] E. M. Purcell. Life at low Reynolds number. Am. J.
Phys., 45:3–11, 1977.
[6] L. Turner, W. S. Ryu, and H. C. Berg. Real-time imaging
of fluorescent flagellar filaments. J. Bacteriol., 182:2793–
2801, 2000.
[7] N. C. Darnton, L. Turner, S. Rojevsky, and H. C. Berg.
On torque and tumbling in swimming Escherichia coli.
J. Bacteriol., 189:1756–1764, 2007.
[8] E. Lauga and T. R. Powers. The hydrodynamics of
swimming microorganisms. Rep. Prog. Phys., 72:096601–
096637, 2009.
[9] D. Murat, M. H´erisse, L. Espinosa, A. Bossa, F. Al-
berto, and L.-F. Wu. Opposite and coordinated rota-
tion of amphitrichous flagella governs oriented swimming
and reversals in a magnetotactic Spirillum. J. Bacteriol.,
197:3275–3282, 2015.
[10] S. B. Guttenplan, S. Shaw, and D. B. Kearns. The cell
biology of peritrichous flagella in Bacillus subtilis. Mol.
Microbiol., 87:211–229, 2012.
[11] L. Ping. The asymmetric flagellar distribution and motil-
ity of Escherichia coli. J. Mol. Biol, 397:906–916, 2010.
[12] P. J. Mears, S. Koirala, C. V. Rao, I. Golding, and Y.
R. Chemla. Escherichia coli swimming is robust against
variations in flagellar number. eLife, 3:e01916, 2013.
and pH-induced polymorphic transitions. J. Bacteriol.,
184:5979–5986, 2002.
[14] H. C. Berg. E. coli in Motion. Biological and Medical
Physics, Biomedical Engineering. Springer-Verlag New
York, 2004.
[15] M. T. Brown, B. C. Steel, C. Silvestrin, D. A. Wilkinson,
N. J. Delalez, C. N. Lumb, B. Obara, J. P. Armitage,
and R. M. Berry. Flagellar hook flexibility is essential for
bundle formation in swimming Escherichia coli cells. J.
Bacteriol., 194:3495–3501, 2012.
[16] K. Son, J. S. Guasto, and R. Stocker. Bacteria can exploit
a flagellar buckling instability to change direction. Nature
Phys., 9:494–498, 2013.
[17] H. Shum and E. A. Gaffney. The effects of flagellar hook
compliance on motility of monotrichous bacteria: A mod-
eling study. Phys. Fluids, 24:061901, 2012.
[18] E. Lauga. Bacterial hydrodynamics. Annu. Rev. Fluid
Mech., 48:105–130, 2016.
[19] J. B. Keller and S. I. Rubinow. Swimming of flagellated
microorganisms. Biophys. J., 16:151–170, 1976.
[20] Y. Hyon, T. R. Powers, R. Stocker, and H. C. Fu. The
wiggling trajectories of bacteria. J. Fluid Mech., 705:58–
76, 2012.
[21] B. Rodenborn, C.-H. Chen, H. L. Swinney, B. Lui, and
H. P. Zhang. Propulsion of microorganisms by a helical
flagellum. Proc. Natl. Acad. Sci. USA, 110:E338–E347,
2013.
[22] S. Y. Reigh, R. G. Winkler, and G. Gompper. Synchro-
nization, slippage, and unbundling of driven helical flag-
ella. PLoS ONE, 8:e70868, 2013.
[23] P. Kanehl and T. Ishikawa. Fluid mechanics of swimming
bacteria with multiple flagella. Phys. Rev. E, 89:042704,
2014.
[24] R. E. Johnson. An improved slender-body theory for
Stokes flow. J. Fluid Mech., 99:411–431, 1980.
[25] J. J. L. Higdon. The hydrodynamics of flagellar propul-
sion: helical waves. J. Fluid Mech., 94:331–351, 1979.
[13] B. Scharf. Real-time imaging of fluorescent flagellar fil-
aments of Rhizobium lupini H13-3: Flagellar rotation
[26] A. Sen, R. K. Nandy, and A. N. Ghosh. Elasticity of
flagellar hooks. J. Electron Microsc., 53:305–309, 2004.
|
1509.00765 | 1 | 1509 | 2015-09-02T16:06:05 | Toroidal membrane vesicles in spherical confinement | [
"physics.bio-ph",
"cond-mat.soft"
] | We investigate the morphology of a toroidal fluid membrane vesicle confined inside a spherical container. The equilibrium shapes are assembled in a geometrical phase diagram as a function of scaled area and reduced volume of the membrane. For small area the vesicle can adopt its free form. When increasing the area, the membrane cannot avoid contact and touches the confining sphere along a circular contact line, which extends to a zone of contact for higher area. The elastic energies of the equilibrium shapes are compared to those of their confined counterparts of spherical topology to predict under which conditions a topology change is favored energetically. | physics.bio-ph | physics | Toroidal membrane vesicles in spherical confinement
Lila Bouzar1, Ferhat Menas2,3, and Martin Michael Muller4,5
1 D´epartement de Physique Th´eorique, Facult´e de Physique,
USTHB; BP 32 El-Alia Bab-Ezzouar, 16111 Alger, Algeria
2 Laboratoire de Physique et Chimie Quantique, Universit´e Mouloud Mammeri; BP 17, 15000 Tizi-Ouzou, Algeria
3 Ecole Nationale Pr´eparatoire aux Etudes d'Ing´eniorat, D´ep. de Phys.; BP 05 Rouıba, Alger, Algeria
4 Equipe BioPhysStat, LCP-A2MC, Universit´e de Lorraine; 1 boulevard Arago, 57070 Metz, France
5 Institut Charles Sadron, CNRS-UdS; 23 rue du Loess, BP 84047, 67034 Strasbourg cedex 2, France
(Dated: November 7, 2018)
We investigate the morphology of a toroidal fluid membrane vesicle confined inside a spherical
container. The equilibrium shapes are assembled in a geometrical phase diagram as a function of
scaled area and reduced volume of the membrane. For small area the vesicle can adopt its free form.
When increasing the area, the membrane cannot avoid contact and touches the confining sphere
along a circular contact line, which extends to a zone of contact for higher area. The elastic energies
of the equilibrium shapes are compared to those of their confined counterparts of spherical topology
to predict under which conditions a topology change is favored energetically.
PACS numbers: 87.16.D-, 87.10.Pq
I.
INTRODUCTION
Topological transitions abound in physical and biological systems.
In fluid dynamics they can be induced by
capillary instabilities like the Plateau -- Rayleigh instability which causes a fluid thread to break up into smaller
droplets [1]. Another classic example concerns area-minimizing surfaces: a catenoidal soap film spanning two circular
wires undergoes a transition to two soap films covering both wires separately when the distance between the wires
is increased [2]. A topological transition can also be induced by deforming the wire such as in the case of a Mobius
strip which can change its topology with a twist singularity [3].
In the biological world, cells are controlling the topology of their ingredients at all length scales. At the molecular
level the function of DNA [4] and proteins [5] depends crucially on their topology. On a larger scale membrane-bound
organelles such as the endoplasmic reticulum or mitochondria display intricate geometries whose topological features
are carefully tuned for proper function [6 -- 9]. A healthy cell can change the shape and topology of its membranes
with the help of other constituents such as the cytoskeleton [10, 11]. However, topology changes can also occur
due to pathological conditions: one example is X-linked centronuclear myopathy -- a congenital muscle weakness -- in
which the membrane stacks of the sarcoplasmic reticulum of skeletal muscle cells are remodeled into cubic membrane
structures of higher genus [12]. Similar structures have been observed in the inner membrane of mitochondria upon
starvation [13].
A theoretical ansatz to model the morphology of the mitochondrium consists of studying a fluid membrane vesicle
inside a confining spherical container [14 -- 19]. Although this model is too simple to capture all the details of the
complex folding patterns, one obtains the basic morphological building blocks. Experimental observations in vivo
[20] and in artificial systems [16] confirm the theoretical predictions. However, the spherical topology of the confined
membrane vesicle has been fixed in all existing theoretical studies.
To handle topology changes as well, we will consider confined fluid membrane vesicles of toroidal topology. The
equilibrium shapes of free toroidal vesicles without confinement have been studied since the 1990s [21 -- 28]. Only
recently new stable non-axisymmetric shapes have been found [29]. As in these articles we use the classical curvature
model to study our system (see Sec. II). To obtain a geometrical phase diagram with all equilibrium shapes, we start
by analyzing the free solutions and determine the parameter values at which the vesicle starts to touch the confining
sphere (Sec. III). In the next step we determine the axisymmetric (Sec. IV) and non-axisymmetric (Sec. V) equilibrium
shapes in contact with the confinement. In Sec. VI we finally compare the vesicles of spherical and toroidal topology
to predict under which conditions a topological transition can take place.
We search for the equilibrium shapes of a toroidal membrane vesicle of fixed area ¯A and volume ¯V in spherical
confinement. In the following area and volume are scaled by the respective quantities A0 and V0 of the confining
sphere: a = ¯A/A0 and v = ¯V /V0. The elasticity of the membrane is modeled within the scope of the classical
II. MODEL
5
1
0
2
p
e
S
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
6
7
0
0
.
9
0
5
1
:
v
i
X
r
a
2
FIG. 1: (Color online) Transition from a spherical to a toroidal topology (see text): (bottom left) axisymmetric equilibrium
solution with spherical topology (a, v) = (1.2, 0.8), (top left) ellipsoid-like equilibrium solution with spherical topology (a, v) =
(1.5, 0.8), (top right) confined toroidal membrane (a, v) = (1.2, 0.8) (simulation snapshot), and (bottom right) axisymmetric
equilibrium solution with toroidal topology (a, v) = (1.2, 0.8). The maximum values of the scale bar are emax = 10, 23, 11, and
8 (clockwise starting from bottom left).
curvature model [30 -- 32] with the bending energy
Eb := Eκ + E¯κ =
(cid:90)
κ
2
(cid:90)
dA (2H − C0)2 + ¯κ
dA K ,
(1)
where H is the mean curvature and K the Gaussian curvature. The stiffness of the membrane is reflected in the
material parameters κ and ¯κ which are the bending and the Gaussian rigidity, respectively. We assume in particular
that the membrane prefers to be intrinsically flat which implies that the spontaneous curvature C0 is zero in our
study. The second term of the energy involving the Gaussian curvature is a topological invariant for a closed surface.
According to the Gauss-Bonnet theorem it is given by [33]
(cid:90)
E¯κ = ¯κ
dA K = 4π¯κ(1 − g)
(2)
for a surface of genus g, where g corresponds to the number of "handles" of the vesicle. For a spherical surface such as
those studied in [14, 15] g is zero; for the toroidal surfaces considered here g is equal to one and E¯κ vanishes. Depending
on the sign of ¯κ an increase of g will thus correspond to an abrupt increase or decrease in energy, respectively. For
fluid lipid mono- and bilayers one typically finds ¯κ ≈ −κ < 0 [34], which implies that a change from a spherical to a
toroidal topology will increase the term of the energy involving K. However, the first term of the energy will change
as well and it is not clear a priori whether the total energy will increase or decrease.
Why should the topology of the membrane change at all? The equilibrium shapes of a spherical membrane vesicle
in confinement have been determined recently by some of the authors [14, 15]. For moderate values of area (a > 1)
we have found an axisymmetric solution with a "light bulb"-like invagination which starts to contact itself when the
area of the vesicle is increased (see left part of Fig. 1). If the membrane was able to fuse, for instance, with the
help of embedded specialized molecules such as SNARE proteins [35], the two parts of the surface in contact could
merge. The topology of the resulting surface is toroidal (see right part of Fig. 1): two necks instead of one connect
axisymmetricequilibriumwithsphericaltopologyaxisymmetricequilibriumwithtoroidaltopologynon-axisymmetricequilibriumwithsphericaltopologysimulationsnapshotwithtoroidaltopologyself-contactareaincreasemembranefusionrelaxationtoequilibriumchangeoftopology02H2emax 3
FIG. 2: (Color online) Partial phase diagram (a, vr) showing the free axisymmetric equilibrium shapes together with the
conformal transformations of the Clifford torus. The colored regions indicate the global minima of this set of shapes, which are
either sickle-shaped tori (lime green region), circular tori (orange region), or conformal transformations of the Clifford torus
(hatched orange region). The discoid tori (purple) are local minima of higher energy. The curves 1 - 4 delimit the regions
in which the free shapes are equilibrium solutions of the confinement problem. To their right no such solution exists, which
explains why parts of the phase diagram are white. Inset: Scaled bending energy eb := Eb/(8πκ) as a function of vr for the
different solution types (see Ref. [23, 27]): sickle-shaped (lime green curve 1 ), circular (red curve 2 ), discoid (purple curve
4 ), and stomatocyte (dark green curve 5 ) tori, and the conformal transformations of the Clifford torus (blue curve 3 ). The
long-dashed parts of these curves correspond to shapes of maximum bending energy. The short-dashed line L corresponds to
the limit torus with eb = 2, which consists of two identical spheres connected via two infinitesimal holes.
the invagation to the outer part of the membrane in contact with the spherical container. The toroidal ground state
for the parameters of Fig. 1 is axisymmetric. We will see in the following that the symmetry of the solution depends
on the respective values of area and volume of the vesicle.
III. FREE SOLUTIONS
Because of the scale invariance of the bending energy the only relevant parameters of the free vesicle are the
reduced volume vr = v
a3/2 and the spontaneous curvature C0. For vanishing spontaneous curvature the following
types of axisymmetric local extrema have been found (see Refs. [22, 23, 27] and Fig. 2): (i) sickle-shaped tori, (ii)
(almost) circular tori, (iii) discoid tori, and (iv) toroidal stomatocytes. All but the stomatocytes have an additional
reflection symmetry with respect to the plane perpendicular to the axis of symmetry.
The lower limits of discoid tori and toroidal stomatocytes are shapes with a vanishing hole diameter at the vertical
axis (purple and dark green squares in Fig. 2). The lower limit of the stable sickle-shaped tori, vr → 0, corresponds to
two connected identical spheres with an infinitesimal hole in each of the poles. In the same limit the (almost) circular
tori approach a shape of exact circular cross section while both the diameter of the hole and the cross section vanish
25/4π1/2 ≈ 0.71 the cross section of the corresponding profile of the same branch is exactly circular
[23]. At vr = vcl
as well (blue circles in Fig. 2). This is the Clifford torus whose bending energy is a lower bound to that of any toroidal
shape [36]. Its two different ellipsoidal deformations have been identified as discoid vr < vcl
r and sickle-shaped vr > vcl
r
tori [27]. In this article we will use a slightly different definition by calling only those tori sickle-shaped, which have a
sickle-shaped cross section, i.e., a cross-sectional meridional curvature which changes sign twice. With this definition
the branch of circular tori becomes sickle-shaped for vr ≥ 0.79 (red circles in Fig. 2). All other solutions of this branch
(and the corresponding solutions in partial or full contact with the container, see below) will be called circular tori in
the following.
r =
3
In the inset of Fig. 2 the scaled bending energy eb := Eb/(8πκ) is shown as a function of vr [22]. The branch 5
of the toroidal stomatocytes contains no minima and will thus be omitted in the following. Without the confinement
constraint the global minima can be directly read off: for vr < 0.57 the tori are sickle-shaped. Above this value the
circular tori have a lower bending energy. Increasing vr leads to a discontinuous shape transformation (red squares
in Fig. 2) since the branch of locally stable sickle-shaped tori runs up to vr = 0.65 (lime green squares). For vr > vcl
r
the ground states are non-axisymmetric shapes. They can be constructed with the help of conformal transformations
[22].
Special conformal transformations of the form
4
X(cid:48)
X
X 2 + Λ
( X
X 2 + Λ)2
.
=
(3)
conserve the toroidal topology of the surface X(ξ1, ξ2) ∈ R3 and leave the bending energy invariant. They consist
of an inversion about the origin, followed by a translation along a constant vector Λ, and a second inversion about
the origin. The resulting surface X(cid:48)
(ξ1, ξ2) has a reduced volume which is larger or equal to the reduced volume
of the original surface. X(cid:48)
r the global minima are
thus non-axisymmetric conformal transformations of the Clifford torus since their bending energy eb = π/2 (blue
line 3 in the inset of Fig. 2) is always lower than that of the axisymmetric solution of same reduced volume (red
curve 2 in the inset of Fig. 2) [22, 23, 27]. The most general special conformal transformation (3) involves the three
parameters Λ = (Λx, Λy, Λz). For an axisymmetric surface X only two parameters remain: without loss of generality
we can orient Λ to lie in the plane defined by the axis of symmetry z and the perpendicular axis x. Since conformal
transformations of the Clifford torus along the axis of symmetry rescale the shape, there is only a one-parameter
family of special conformal transformations leading to the global non-axisymmetric minima discussed above [27].
can always be rescaled to stay inside the container. Above vcl
As long as the area of the vesicle is small enough, the confined vesicle does not touch the container at all and
adopts the shape of the free system. To predict the behavior for larger area, one can rescale the free axisymmetric
solutions up to the point at which they come into contact with the spherical container. These solutions have been
obtained numerically with a shooting method (see appendix A for the details of the method). They correspond to
the curves 1 - 4 delimiting the parts in the phase diagram (a, vr) in which the free vesicles are the local minima
(see Fig. 2). To understand how the global energy minima (colored regions of Fig. 2) are found, consider, for example
vr = 0.6. Without the confinement the free circular torus is the global minimum whereas the free sickle-shaped
torus is a local minimum. At a = 0.6 both solutions can be put inside the container without any contact between
membrane and confinement. However, when we increase the parameter a keeping vr constant such that we cross the
red line 2 in Fig. 2, we do not find any circular tori which are not in contact with the confinement. The only free
solution is the corresponding sickle-shaped torus which is thus the global and only minimum of all free shapes. In
general, one observes that the sickle-shaped tori start contacting the confinement for an area which is much larger
than the corresponding value for the circular tori. Consequently, the sickle-shaped tori are the only free minima for
0 < vr < 0.65 provided the area is large enough. In contrast to that we can neglect the locally stable discoid solutions
for the discussion in the following since they start contacting the container for even smaller area than the circular
tori.
To find the delimiting curve of the region of the free non-axisymmetric solutions, i.e., the blue curve 3 of the phase
diagram in Fig. 2, we consider the Clifford torus in contact in the horizontal (xy) plane. Its defining radii are given
by r = 1√
2r. In polar coordinates (θ, φ) its shape can be written as:
√
2+1 and R =
X(θ, φ) =
1√
2 + 1
((
√
√
2 + sin θ) cos φ, (
2 + sin θ) sin φ, cos θ) ,
(4)
on which we perform the conformal transformation, Eq. (3). With an appropriate rescaling and a final translation
along Λ
¯X = X(cid:48)
(1 − Λ2) + Λ ,
(5)
every point of the inner membrane which lies on the confining sphere will stay on the sphere. Choosing Λ = Λxx one
finds a surface with the same circular contact line as the original Clifford torus. In the limit Λx → 1, (a, vr) → (1, 1)
and the conformally transformed torus becomes a sphere with an infinitesimal handle. The resulting shapes are Dupin
cyclides whose centers of curvature are located on two confocal conics: an ellipse and a hyperbola [26, 37]. For the
Clifford torus the ellipse corresponds to a circle of radius R and the hyperbola becomes a line which coincides with
the axis of symmetry. Using the results of Refs. [26, 37] one obtains for the area and volume of the transformed torus:
a = 4(1 − λ)λ[E(e2) +
v = 8λ3[
(1 − λ)2
λ2
E(e2) +
e2 − 1
K(e2)] ,
2
e2 − 1
2
(1 +
and
e2 − 1
2
3
4
)K(e2)] ,
(6a)
(6b)
5
FIG. 3: (Color online) Equilibrium profiles for fixed reduced volume vr = 0.7 and increasing area (from left to right) a = 0.5,
0.8, 0.9, 1.1, and 1.15. The background colors and labels are in accordance with the final phase diagram (Fig. 6). The color
code of the profiles is the same as in Fig. 1 with emax = 10, 5.5, 6, 10, and 10.5.
(cid:113) λ2−4λ+2
λ2
where K(e2) and E(e2) are the complete elliptic integrals of the first and second kind, respectively [38]. The parameter
∈ {0.5, R} is the semi-major axis of the
e =
generating ellipse.
∈ {0, 1} is the eccentricity and λ = 2+
3+2
√
√
√
2+(−2+
2)Λ2
√
x
2+(−3+2
2)Λ2
x
Fig. 2 does not show the whole story yet. We have not yet discussed the conformal transformations of the sickle-
shaped tori. The resulting surfaces have to be compared energetically to the solutions which are in contact with the
confinement. We will do this in two steps. We first identify all axisymmetric solutions of the confinement problem.
In a second step we determine the non-axisymmetric confined shapes and assemble everything in one phase diagram.
IV. CONFINED AXISYMMETRIC SOLUTIONS
When you press a cylindrical sheet of paper gently against a wall, the contact between the two will be a line.
The curvature of the sheet perpendicular to the line of contact will generally not be zero. Increasing the force will
decrease this curvature until it is equal to zero, i.e., equal to the curvature of the flat wall. Above this threshold a
contact area will start to form. The same phenomenon is observed in our system (see the first four profiles in Fig. 3
for an example): increasing the area for a fixed reduced volume one reaches a point at which the free vesicle starts
to touch the spherical confinement. The corresponding curves in the phase diagram (a, vr) have been discussed in
the previous section. Interestingly, all these solutions are in contact with the container in one circle which lies in
the horizontal symmetry plane. Their curvature perpendicular to the contact line, K⊥, is always larger than that of
the sphere, K⊥ = 1 . Increasing the area further thus leads to solutions which are still confined by the circle (called
contact-circle solutions in the following). Above a certain threshold area, K⊥ = K⊥, and the circular contact line
becomes a spherical segment leading to contact-area solutions.
The corresponding confined axisymmetric solutions have been determined numerically with the same shooting
method as in the previous section (see appendix A). The appropriate boundary conditions depend on the case under
consideration and are summarized in Tab. II of the appendix. Fig. 4 shows the resulting phase diagram together
with a comparison of bending energies. The dark green line C1 in the diagram separates the region of contact-circle
solutions from the region of contact-area solutions. For vr < 0.65 these solutions coexist with the free sickle-shaped
tori. Comparing the respective bending energies one finds a discontinous shape transition (lime green line with long
dashes D1 in the phase diagram), below which the free sickle-shaped tori are the global minima. A first estimate of
the upper boundary of the region of confined axisymmetric tori (short-dashed lime green curve Dp
2 ) can be found by
comparing the bending energies of these solutions with those of the conformal transformations of the free tori. We
would like to stress that these free shapes are not equilibrium solutions in general but serve as a means to determine the
regions of the phase diagram in which non-axisymmetric shapes with a lower bending energy than the axisymmetric
solutions exist. We will see below that the curve including all solutions with and without axisymmetry (black curve
D2 in Fig. 6, see next section) lies below the boundary discussed here.
Consider first the free sickle-shaped torus with a reduced volume of vr = 0.65 (lime green square).
It has the
smallest bending energy of the whole branch 1 of sickle-shaped tori (see Fig. 4(b)). A conformal transformation of
this shape along the x axis will not change the energy but increase the reduced volume (black dashed-dotted line T ).
With a rescaling of such a shape or a subsequent conformal transformation along z (which both do not change the
energy as well) every point of the phase diagram left of the curve T can be reached. The resulting surface has to
be compared to the other equilibrium shapes of same (a, vr). Along the black dashed-dotted line one finds that the
energy of the conformally transformed free sickle-shaped torus is lower than the energy of the axisymmetric contact-
area solution for vr ≥ 0.76 (black square). Repeating this comparison for smaller area, one obtains the short-dashed
2C1C2D2 6
(a)
(b)
FIG. 4: (Color online) (a) Partial phase diagram (a, vr) showing the axisymmetric extrema together with the conformal
transformations of the free vesicles. The colored regions indicate the global minima of this set of shapes. Regions of sickle-
shaped tori are shown in green, regions of circular tori in orange/red. Non-axisymmetric regions are hatched. The colors vary
from light to dark with increasing confinement. The curve C1 separates the region of contact-circle solutions from the region of
contact-area solutions, whereas C2 is the boundary between the confined circular and sickle-shaped solutions (for a description
of the curves D1 , Dp
2 , and T see text). (b) Scaled bending energy eb as a function of vr for the different axisymmetric
solutions. In addition to the free solutions (solid lime green 1 , red 2 , and blue curves 3 ) lines of constant area a are shown
as well (dashed yellow I (a = 0.7), purple II (a = 0.9), and pink III (a = 1.1) curves). The short-dashed line L corresponds
to the limit torus with eb = 2, which consists of two identical spheres connected via two infinitesimal holes.
lime green curve Dp
2 along which the energy of the conformally transformed free sickle-shaped torus equals that of
the corresponding axisymmetric contact-area/contact-line solution. The curve ends in the point at which it intersects
with the branch 2 of the free solutions (lime green circle). The part of the curve Dp
2 to the right of the black square
can be found by comparing the conformal transformations of all other free tori of the sickle-shaped branch 1 with
the axisymmetric contact-area solutions. Both curves, Dp
2 and 1 , end in the point (a, vr) = (2, 0), which corresponds
to a surface consisting of two connected identical spheres with an infinitesimal hole in each of the poles (see Sec. III).
The short-dashed black curve L in Fig. 4(a) indicates the conformal transformations of the limit shape along x and
delimits the regions of shapes with and without self-contact (see next section). The analytical expression for this
curve is given by:
vr =
1 − (a − 1)
3
2
a 3
2
.
(7)
To understand the region of confined tori in the center of the phase diagram, consider curves of constant area a (curves
I - III in Fig. 4). For a small reduced volume vr the solutions are free sickle-shaped tori which can be mapped on
one another via a simple rescaling. Their bending energy does not depend on a and corresponds to the curve 1
in Fig. 4(b). When we cross the curve D1 in the phase diagram (yellow, purple, and pink squares, respectively), a
discontinous shape transition to a confined state is energetically favorable. Depending on the value of a the membrane
can either adopt a contact-circle (curve I ) or a contact-area (curves II and III ) solution. In the former case (a = 0.7),
the solution adopts a free shape again above vr = 0.63 (yellow circle). The corresponding bending energy is given by
the curves 2 (between yellow and blue circle) and 3 (to the right of the blue circle) in Fig. 4(b). In the latter case
(a = 0.9 and a = 1.1), we can observe in more detail how the (preliminary) upper limit Dp
2 of the region of confined
axisymmetric solutions has been obtained. The respective bending energies of the confined solutions (curves II and
7
(a)
(b)
(c)
FIG. 5: (Color online) Comparison between (a) the numerical Runge-Kutta result (black dotted) and the result of a finite
element simulation (red solid) for (a, vr) = (1.1, 0.7), and (b) the conformal transformation of minimal bending energy (black
dotted) and the result of a finite element simulation (red solid) for (a, vr) = (1.15, 0.7) (see Fig. 3 for a three-dimensional
representation of the shapes). The red dotted curve outside of the container indicates position and relative value of the
container force f C
(c) Scaled bending energy eb and container force density f0 at the
contact line as a function of a for fixed reduced volume vr = 0.7. The green dotted line corresponds to non-axisymmetric
contact-area solutions and was obtained with the help of the finite element simulations. All other curves were determined
with the Runge-Kutta method. Colors and labels are in accordance with the final phase diagram (Fig. 6). The black triangles
correspond to the shapes shown in Fig. 3.
a in the finite element simulations.
III ) decrease first before increasing up to the point where the energy is equal to one of a conformally transformed
free solution (purple and pink circles on Dp
2 , respectively). One would thus expect a transition to a non-axisymmetric
free torus for higher vr. In reality, however, the membrane will not become a free torus again but prefers to adopt a
non-axisymmetric confined shape as we will see when including all solutions in our discussion.
V. CONFINED NON-AXISYMMETRIC SOLUTIONS
As a starting point we consider conformal transformations given by the combination of the equations (3) and (5)
and apply them on the confined axisymmetric solutions X of the previous section:
¯X =
XX2 + Λ
XX2 + Λ2
(1 − Λ2) + Λ , with Λ = cst ∈ R3 .
(8)
Every point of the membrane in contact with the spherical confinement is mapped on some point of the sphere again
via Eq. (8). The resulting contact surface/contact line is different from the original spherical segment/circle. However,
one can show that the local curvature condition at the boundary/boundaries of the free part of the membrane is still
fulfilled (see appendix B). To be sure that the obtained solutions are in equilibrium, one also has to check whether
the total force on the membrane exerted by the container is zero. This is the case for the axisymmetric solutions X,
where the container exerts a constant normal delta force in each point of the contact line(s) to equilibrate the system.
The corresponding container force density f0 follows from the boundary conditions of the Runge-Kutta scheme and is
given by f0 = ψ0 (see appendix A, Eq. (B11), and -- for an example -- the lower part of Fig. 5(c)). A closer look at the
conformal transformations ¯X of the axisymmetric shapes reveals, however, that the total force on the transformed
membrane exerted by the container is not necessarily zero (see appendix B).
In analogy to Ref. [27] one might, nevertheless, hope that the resulting surfaces are close to the true equilibrium
non-axisymmetric solutions. Assuming this for the moment we can compare bending energies in the same manner as
8
(a)
(b)
FIG. 6: (Color online) (a) Phase diagram (a, vr) showing all global minima. Regions of sickle-shaped tori are shown in green,
regions of circular tori in orange/red. Non-axisymmetric regions are hatched. The black triangles correspond to the shapes
shown in Fig. 3. The curve D2 separates the region of confined axisymmetric and confined non-axisymmetric solutions. For
a description of the other curves, see Fig. 4. (b) Scaled bending energy eb as a function of vr. This figure is equivalent to
Fig. 4(b) except that we have included the confined non-axisymmetric solutions as well.
in the previous section. For every confined axisymmetric solution one can identify the region in the phase diagram
in which its conformal transformations can be found. To identify the shape of minimal bending energy for a fixed
(a, vr) in this set of solutions, one thus has to identify all axisymmetric shapes (of different (a(cid:48), v(cid:48)
r)) which can be
transformed in such a way that the new surface has the area and reduced volume (a, vr). Comparing the energies
of all these shapes with the axisymmetric solution of same (a, vr) allows to determine an approximate boundary D2
which separates the region of axisymmetric and non-axisymmetric global minimum solutions.
To check how well the obtained solutions approximate the real ones, we have searched for equilibrium shapes
with the help of numerical simulations. The respective finite element method has already been used for confined
membranes of spherical topology [14, 15]. In the simulations the membrane is discretized and evolved according to
Newton's equations of motion until a balance of forces is reached (see appendix C). To find an equilibrium surface
for a fixed (a, vr), we have used a discretization of the obtained shape of minimal energy as initial mesh. It turns
out that the axisymmetric solutions and the corresponding results of the finite element simulations coincide within
the numerical error which is about 10−3 in scaled units for all parameters (see Fig. 5(a)). The delta forces exerted
by the container are smeared out slightly but situated at the contact lines as expected. The necks diffuse away from
the axisymmetric ground state very easily for non-vanishing noise during the simulations, similar to what has been
observed for unconfined toroidal vesicles before [29]. The non-axisymmetric solutions, and consequently the phase
boundary, are approximated remarkably well by the respective conformal transformations (with an error of about
10−2 in eb for shapes close to the phase boundary). In addition to the smeared-out delta forces at the contact lines
one now observes additional forces at the contact area (see Fig. 5(b)).
In summary one obtains Fig. 6 which shows all global minima assembled in one figure. We can now observe what has
been anticipated in the previous section: a membrane of large constant area (curves II and III ) breaks axisymmetry
but stays in contact with the container when we cross the curve D2 towards higher vr. Depending on the value of
a the membrane either adopts the shape of a free conformal transformation of the Clifford torus when crossing the
curve 3 (purple triangle) or tends towards the limit torus, which consists of two identical spheres connected via two
infinitesimal holes (pink triangle on curve L ).
It is instructive to consider the case of constant reduced volume as well. The upper part of Fig. 5(c) shows the
scaled bending energy as a function of membrane area for a reduced volume of vr = 0.7 (compare Fig. 3 in Sec. IV).
For small area the membrane is not in contact with the confining sphere and the bending energy does not depend on
esph
κ
(a, v)
etor
κ
(1.1, 0.9) 1.91 1.88
(1.2, 0.9) 1.97 1.97
(1.1, 0.8) 1.95 1.87
(1.2, 0.8) 1.98 1.94
(1.3, 0.8) 1.99 1.97
(1.1, 0.7) 1.99 1.94
(1.2, 0.7) 1.99 1.97
(1.3, 0.7) 2.00 1.98
(1.4, 0.7) 2.00 1.98
9
toroidal membrane shape
non-axi. confined
non-axi. confined
axi. confined
axi. confined
non-axi. confined
not confined
not confined
not confined
axi. confined
TABLE I: Comparison of energies eκ of a vesicle of spherical (from Ref. [14]) and toroidal topology as a function of scaled area
a and volume v.
a. When increasing a contact emerges and the energy starts increasing monotonically. We first find contact-circle,
then contact-area solutions. At a = 1.12 the membrane can break axisymmetry to lower its energy (dotted green line
in Fig. 5(c)) and adopts a confined non-axisymmetric shape.
For larger values of fixed reduced volume vr a similar evolution is observed: for small area a the solution corresponds
to a conformal transformation of the Clifford torus (hatched orange region). By increasing a we cross the blue curve
defined by Eqs. (6). The non-axisymmetric solutions in this region are in contact with the container in one horizontal
circle (hatched light red region). When crossing the left black solid line the contact line extends to a zone of contact.
The corresponding non-axisymmetric tori are first circular (hatched dark red region) and, upon increasing a even
further, sickle-shaped (hatched dark green region). All the shapes lying to the right of the curve (7) exhibit self-
contacts and are not studied in the present paper.
VI. COMPARISON TO SPHERICAL TOPOLOGY
Knowing the global minima of the confined toroidal membrane vesicles we are now in a position to compare its
bending energy (1) to that of a confined spherical membrane of same area and volume. For a better comparison we
again scale the energy with the constant 8πκ, i.e., the energy Eκ of a perfect spherical vesicle: eb = Eb/(8πκ) = eκ+e¯κ.
For the free membrane vesicles one finds that the toroidal ground state has a lower energy eκ than the corresponding
spherical one as long as 0 < vr < 0.72 (see Ref. [23]). For larger values of vr the opposite is true. The second term
in the energy, e¯κ, equals 1
¯κ
κ for spherical vesicles, whereas it is zero for vesicles of toroidal topology (see Sec. II).
2
Absolute value and sign of ¯κ
κ will decide which topology is preferred energetically. Taking a typical value for fluid
κ ≈ −1 [34] the systems increases its energy by ∆e¯κ ≈ 1
lipid mono- and bilayers ¯κ
2 when passing from a spherical to
a toroidal topology. Since this energy is not compensated by the change ∆eκ (compare inset of Fig. 2 with Fig. 8 of
Ref. [39]) we find that the spherical topology is always preferred for the free system.
A similar comparison can be drawn for the confined vesicles. We focus on the case a ≥ 1 for which we have found
axisymmetric solutions of spherical topology [14, 15] such as the one shown in Fig. 1 (bottom left). Table I depicts
the values of eκ for both topologies. For a better comparison with the results of Refs. [14, 15] we have fixed v instead
of vr. One observes that the system decreases its energy eκ when changing its topology from spherical to toroidal.
However, as is the case of the free system, this change does not compensate the increase of ≈ 1
2 in the Gaussian
term of the energy. Note that the situation changes dramatically if we allow for other values of ¯κ. For positive ¯κ
the toroidal topology is favored energetically implying the possibility of a topological transition without any loss of
energy.
VII. CONCLUSION
In this article we have studied the morphology of toroidal fluid membrane vesicles in spherical confinement. A
combination of analytical theory, numerical calculations and finite element simulations allowed us to determine the
ground states as a function of area and volume of the vesicle. In contrast to confined membrane vesicles of spherical
topology we have found non-axisymmetric shapes which do not exhibit any self-contacts. A comparison of energies
revealed, that the spherical topology is preferred for typical values of the material parameters.
10
FIG. 7: Parametrization of an axisymmetric toroidal membrane vesicle in spherical confinement. The membrane consists of a
spherical segment in contact with the container and a free part which is given by the solution of the Hamilton equations (A3)
together with the appropriate boundary conditions (Eqs. (A4)/Tab. II).
However, this study was performed assuming zero temperature. To be closer to the real biological world, one would
have to include the effects of thermal fluctuations. In extensive bilayer fluid membrane phases they can significantly
affect the competition between spherical morphologies and surfaces of higher genus [40 -- 42]. In these cases an attractive
interaction appears in addition to Helfrich's steric repulsion [43]. Guided by experiments on real mitochondria [20],
we expect a similar competition in a biologically significant setting, whenever the membrane is in self-contact. This
gives a hint to what can happen in the region of self-contacts in the final phase diagram (Fig. 6): the toroidal vesicle
touches itself, inducing eventually a topology change towards a higher genus ("a sphere with two handles").
A study of these shapes and the effects of the temperature goes beyond the scope of this paper but remains an
interesting open problem for the future. The present model can as well be extended to incorporate a constant adhesion
energy density due to the contact between membrane and confinement.
The authors thank Osman Kahraman and Norbert Stoop for helpful discussions and Etienne Gallant for his support.
The PMMS (Pole Messin de Mod´elisation et de Simulation) is acknowledged for providing the computer time.
Acknowledgments
Appendix A: Numerical method for the axisymmetric case
The details of the numerical method we have used to find axisymmetric equilibrium solutions can be found in
appendix A of Ref. [15] and the references cited therein. The confined membrane consists of two parts, one which is
in contact with the spherical container of unit size and one which bulges inside to minimize its elastic energy. We
parametrize the shape of the membrane in terms of the angle-arc length parametrization ψ(s) (see Fig. 7), where ψ
is the angle between the horizontal axis and the tangent of the membrane. The equation describing the evolution of
ψ(s) along the cross section of the free part is an ordinary differential equation of fourth order which can be rewritten
with the help of Hamilton's formalism.
The scaled energy of the free membrane is given by:
(cid:90) ¯s
(cid:90) ¯s
(cid:34)
(cid:18)
(cid:19)2
E = E/(πκ) =
ds L =
ds
ρ
s
s
ψ +
sin ψ
ρ
+ 2σρ + P ρ2 sin ψ + λρ( ρ − cos ψ) + λz( z − sin ψ)
,
(A1)
where σ = σ/κ and P = P/κ are the scaled surface tension and pressure difference, respectively. For the Clifford
torus one finds σ = P = 0. The Lagrange multiplier functions λρ and λz couple the angle-arc length parametrization
to the Euclidean coordinate system (ρ, z). The conjugate momenta pi = ∂ L
ρ ),
∂ qi
of the system are: pψ = 2ρ( ψ + sin ψ
(cid:35)
zρ Section
Free solutions (xy sym.)
Free toroidal stomatocytes
Circular contact line
Transition contact line → area
Contact area
III
III
IV
IV
IV
z0
0
0
0
0
ρ0
1
1
1
1
sin α − cos α bisection
α
π
2
π
2
π
2
π
2
ψ0
bisection
bisection
bisection
ψ0
− ψ2
0
scan − ψ2
scan
1
1
bisection
see text
0 + 1 + 2σ + P
0 + 1 + 2σ + P
pz
"
"
11
pψ
2( ψ0 + 1)
2( ψ0 + 1)
"
"
pρ
0
2 ψ0
"
"
scan
via Eq. (A4a)
4 sin α
TABLE II: Boundary conditions for the different cases discussed in this article.
pρ = λρ, and pz = λz. In contrast to the case considered in Ref. [15] the momentum pz does not vanish [44]. The
Hamiltonian is thus:
(cid:88)
i
H =
qipi − L =
p2
ψ
4ρ
− pψ sin ψ
ρ
+ pρ cos ψ + pz sin ψ − 2σρ − P ρ2 sin ψ ,
(A2)
(A3a)
(A3b)
(A3c)
(A3d)
(A3e)
(A3f)
(A4a)
(A4b)
(A4c)
yielding the following shape equations:
ψ =
pψ
2ρ
− sin ψ
ρ
,
ρ = cos ψ ,
z = sin ψ ,
(cid:18) pψ
(cid:18) pψ
pψ =
ρ
pψ
ρ
pz = 0 ,
pρ =
+ P ρ2 − pz
− sin ψ
ρ
4ρ
(cid:19)
(cid:19)
cos ψ + pρ sin ψ ,
+ 2σ + 2 P ρ sin ψ ,
which are solved using a standard shooting method [45] subject to the boundary conditions:
H = 0 ,
ρ(s) = ρ0 ,
ψ(s) = α .
z(s) = z0 ,
The values ρ0, z0, and α depend on the case under consideration (see Tab. II). It is sufficient to scan σ and P to
obtain the free solutions with xy symmetry as well as the transition line of shapes where the contact line extends to
a zone of contact. For the other cases a third parameter has to be scanned as well. In the case of the free toroidal
stomatocytes and the contact-circle solutions we have fixed the initial value of ψ. For the contact-area solutions it
was more practical to scan pz instead. The value of ψ0 can then be obtained by deriving Eq. (A3a), equating it with
Eq. (A3d) and inserting the values of the other parameters in the resulting equation.
In all cases the values of ψ, ρ, z, and the corresponding conjugate momenta at s = s are determined by the boundary
ψ0, and ψ0. The Hamilton equations (A3) can then be integrated with a fourth-order Runge-
conditions for given α,
Kutta method. The integration is stopped when the curve returns to the spherical confinement. A bisection on either
α,
ψ0 or ψ0 (depending on the case, see Tab. II again) determines the correct profiles for which ψ(¯s) = 5π
Note that we have to fix pressure and surface tension for this method. Area and volume can only be calculated
for a known profile a posteriori. The equations A = 2πρ and V = πρ2 sin ψ have to be integrated from s = s to ¯s
to find the contribution to area and volume of the free part of the membrane. Adding the result to the contribution
of the part in contact with the container, Ac = −4π cos α and Vc = 2
3 π(−3 cos α + cos3 α), finally yields the values
for the whole membrane. These allow to assemble the obtained shapes in the geometrical phase diagrams (Figs. 2, 4,
and 6). The Runge-Kutta integration is performed with an integration step-size h = 10−4 in reduced units implying
a numerical error of 10−3 of the values of a and vr.
2 − α.
Appendix B: Conformal transformations of confined axisymmetric solutions
We consider the conformal transformation of an axisymmetric surface X given by Eq. (8):
¯X =
XX2 + Λ
XX2 + Λ2
(1 − Λ2) + Λ , with Λ = cst ∈ R3 ,
12
yielding a deformed surface ¯X which is only axisymmetric when Λ is parallel to the symmetry axis. The tangent
vectors of the transformed surface are given by (see chapter 20 of Ref. [46])
¯ea = ∂a ¯X =
(1 − Λ2)
X 0 + Λ2
a − 2[(X 0 + Λ) · e0
X 0 + Λ2
e0
a](X 0 + Λ)
(cid:19)
with X 0 = X
X 2 and e0
a = ∂aX 0 = (ea − 2(X · ea)X)/X2. The metric of the transformed surface follows as
¯gab =
(1 − Λ2)2
X 0 + Λ4 g0
ab =
(1 − Λ2)2
XX2 + Λ4 X4
gab .
We orient the normal towards the outside of the vesicle. The transformed normal is then given by:
¯n = −n0 +
2[n0 · (X 0 + Λ)](X 0 + Λ)
X 0 + Λ2
,
where n0 = −n + 2(n·X)X
X2
projecting it onto the normal
. The extrinsic curvature tensor ¯Kab = − ¯n · ∂a¯eb can be obtained by calculating ∂a¯eb and
¯Kab =
(1 − Λ2)
X 0 + Λ2
−K 0
ab +
ab[n0 · (X 0 + Λ)]
2g0
X 0 + Λ2
.
(B4)
When the coordinate lines on the original surface before the transformation coincide with the lines of curvature, first
and second fundamental form are diagonal. These lines will be lines of curvature on the transformed surface as well
and the principal curvatures follow as (i ∈ {1, 2}):
(cid:18)
(cid:26)
(cid:26)
(B1)
(B2)
(B3)
(B5)
(B6)
(B7)
(B8)
(cid:27)
(cid:27)
.
¯Ki =
¯Kii
¯gii
=
X 0 + Λ2
(1 − Λ2)
−K 0
i +
2[n0 · (X 0 + Λ)]
X 0 + Λ2
where K 0
i = −X2Ki + 2 n · X. The derivative is given by
(cid:8)−X 0 + Λ2∂aK 0
∂a ¯Ki =
1
(1 − Λ2)
i + 2(X 0 + Λ) · (K b 0
a e0
b − K 0
i e0
a)(cid:9) ,
where we have used the Weingarten equations in the second step.
To describe the original axisymmetric surface, we choose the coordinates (s, ϕ) . The coordinate lines are lines of
curvature with principal curvatures K1 = K⊥ = ψ and K2 = K(cid:107) = sin ψ/ρ. For each portion of the surface which is
in contact with the confining sphere (including the circular contact line(s)) one has K⊥ = K(cid:107) = 1. Following Eq. (B5)
the transformed principal curvatures at the contact lines can be obtained very easily. On the sphere one has X2 = 1,
n = X and thus K 0
i = 1. Moreover, X 0 = X, n0 = n, and one finds that the local contact curvature condition at
the contact line is also fulfilled for the transformed shape:
(cid:2)−X + Λ2 + 2n · (X + Λ)(cid:3) =
(cid:2)−1 − 2X · Λ − Λ2 + 2(1 + n · Λ)(cid:3) = 1 .
¯Ki =
1
(1 − Λ2)
1
(1 − Λ2)
The tangent and normal vectors on the contact line are given by e0
a = ea, n0 = n, and
(cid:18)
ea − 2(Λ · ea)(n + Λ)
n + Λ2
(cid:19)
¯ea =
(1 − Λ2)
n + Λ2
¯n = −n +
2(1 + n · Λ)(n + Λ)
n + Λ2
.
13
b −
i )ea = 0. Therefore, the last term in Eq. (B6) vanishes, and one obtains for the derivative of the
n+Λ4 gab. Taking into account that K b
a are diagonal, K b 0
a as well as K b 0
a e0
The metric is diagonal with ¯gab = (1−Λ2)2
a − K 0
K 0
i e0
principal curvatures
a = (K 0
∂a ¯Ki =
−n + Λ2∂aK 0
i
(1 − Λ2)
n + Λ2
(1 − Λ2)
=
∂aKi .
(B9)
To determine the external forces due to the confinement, consider the stress tensor of the fluid membrane (normalized
by κ) [47]
(cid:35)
(cid:41)
(cid:34)
(cid:35)
(cid:40)
(X · n) + σ
gab
eb −
(X · ea)
n .
(B10)
(cid:34) P
2
Projection onto the unit vector l = laea perpendicular to the contact line yields the local force density:
f a =
K(K ab − 1
2
Kgab) −
(cid:40)
(cid:34)
1
2
1
2
(cid:34) P
(K 2⊥ − K 2(cid:107) ) −
(cid:18)
(cid:19)
2
ψ2 − sin2 ψ
ρ2
−
(cid:32) P
2
(X · n) + σ
(cid:35)(cid:41)
(cid:34)
(cid:33)(cid:35)
l −
laf a =
=
∇aK − P
2
(cid:35)
(X · l)
n
∇⊥K − P
2
(cid:34)
(X · n) + σ
es −
ψ +
ψ cos ψ
ρ
− cos ψ sin ψ
ρ2
− P
2
(X · l)
(cid:35)
n ,
(B11)
where the second line only holds for an axisymmetric surface. At the circular contact line: laf a = −(
For the conformally transformed surface one obtains (see Refs. [48 -- 50] for further examples of how stresses change
under conformal transformations):
2 + σ) es − ψ n.
P
(cid:33)(cid:18) es(1 + Λ2) − 2Λ(Λ · es) + 2Λ × eϕ)
(cid:19)
n + Λ2
(cid:20)−n(1 − Λ2) − 2Λ(1 + n · Λ)
(cid:21)
(1 − Λ2)
+ ψ
.
(B12)
(cid:32) P
2
¯la ¯f a
= −
+ σ
To analyze the balance of forces let us consider a conformal transformation along the z axis: Λ = Λzz. The
projections of the normal and tangent vectors onto the basis vectors are given by
ρ · es = −z · n = cos ψ
ρ · n = z · es = sin ψ .
and thus, with n + Λ = sin ψρ + (Λz − cos ψ)z one obtains the projections
(cid:32) P
(cid:32) P
2
2
(cid:33)(cid:18) cos ψ(1 + Λ2
(cid:33)(cid:18) sin ψ(1 − Λ2
(cid:19)
z) − 2Λz
(cid:19)
1 − 2Λz cos ψ + Λ2
z
z)
1 − 2Λz cos ψ + Λ2
z
+ σ
+ σ
ρ · ¯la ¯f a
= −
z · ¯la ¯f a
= −
− ψ sin ψ ,
+
ψ
(1 − Λ2
z)
and
(cid:2)cos ψ(1 + Λ2
z) − 2Λz
(B13)
(B14)
(B15)
(B16)
(cid:3) .
The axisymmetric solutions we consider are symmetric with respect to the xy plane as well. The two circular contact
lines at ψ+ = 90◦ + γ and ψ− = 90◦ − γ are transformed into two circles of different length. To calculate the total
external force let us determine the new positions and lengths of the contact lines. Before the transformation we have
X = (sin ψ, 0,− cos ψ)T. After the transformation:
¯X = λ sin ψ ρ + [λ(Λz − cos ψ) + Λz] z ,
where λ := (1 − Λ2
arctan [λ(Λz − cos ψ) + Λz)/(λ sin ψ)]. The lengths of the contact lines are thus: ¯L = 2π sin ¯ψ = 2πλ sin ψ.
z)/(1 − 2Λz cos ψ + Λ2
z).
One obtains for the new contact angles:
The horizontal components of the force will equilibrate each other due to the axisymmetry. The vanishing of the
force in the vertical direction yields a condition which can be used to find Λz as a function of the known parameters
P , σ, and ψ:
(cid:18) −C cos γ(1 − Λ2
0 = z · (2πλ+ sin ψ+(¯la ¯f a
z)2
(1 + 2Λz sin γ + Λ2
z)2
⇔ 0 =
(cid:19)
)+ + 2πλ− sin ψ−(¯la ¯f a
(cid:2)− sin γ(1 + Λ2
z) − 2Λz
(1 + 2Λz sin γ + Λ2
z)
)−)
(cid:3)
+
(cid:18) −C cos γ(1 − Λ2
z)2
(1 − 2Λz sin γ + Λ2
z)2
(cid:19)
+
+
(cid:2)sin γ(1 + Λ2
z) − 2Λz
(1 − 2Λz sin γ + Λ2
z)
(cid:3)
,
(B17)
¯ψ = 90◦ +
where we have defined C :=
simplifies to
(cid:17)
(cid:16) P
2 +σ
ψ
. For γ = 0, i.e., the solutions which are only confined in one circle, this condition
14
0 = C(1 − Λ2
z)2 + 2Λz(1 + Λ2
z) .
(B18)
A similar equation can be obtained for Λ = Λxx. We observe that the total force acting on the membrane due to
the container is only equilibrated for special values of Λ. This implies that the conformal transformation defined in
Eq. (8) does not yield an equilibrated surface in general. However, we have shown in Sec. V that these surfaces can
be used as initial conditions for subsequent finite element simulations in order to obtain equilibrium shapes.
Appendix C: Finite element simulations
In the simulations the fluid membrane is discretized by a set of triangles with N ∼ 4000 nodes using subdivision
finite elements [51, 52]. The position of the surface in R3 is parametrized with the local coordinates s1 and s2 and
interpolated by
xh(s1, s2) =
xaN a(s1, s2) ,
(C1)
N(cid:88)
a=1
¯a)2(cid:105)
√
(
a − √
µA
2
(cid:90) (cid:104)
E =
2H 2 +
where xa is the position of node a and N a are the Loop subdivision trial functions [53]. We search for equilibrium
solutions of the energy functional
ds1ds2 +
(V − ¯V )2 ,
µV
2
(C2)
where the constants are set to µA = 105 and µV = 5 · 104 to ensure that the membrane adopts the prescribed target
values for area and volume to a numerical error of about 10−3. Note that the area constraint is recast in a local form
(local area constraint method ), in which
a is the surface Jacobian such that A =(cid:82) √
a ds1ds2 [51, 52].
√
The energy can be expressed in terms of the nodal positions xa to determine the gradient with respect to xa. This
yields the corresponding nodal forces f M
a due to the elasticity of the membrane (for more details see appendix B of
Ref. [15]). The contact between the membrane and the spherical container is modeled by a harmonic force which is
only applied to the nodes which leave the container. The force acting on such a node a is given by
(C3)
where k = 1.5 · 106 is the stiffness constant, da the penetration depth, and n the inward-pointing normal of the
container. The total force at node a is the sum of all contributions
f C
a = kd2
an ,
f a = f M
a + f C
a .
(C4)
To find an equilibrium solution, we integrate these nodal forces f a in time according to Newton's equations of motion
until a balance of forces is reached, i.e., until f a = 0 (for more details we refer again to appendix B of Ref. [15]).
All our input meshes are obtained from the axisymmetric Runge-Kutta solutions (see appendix A). Every cross
section of such a solution, which contains the symmetry axis, yields the same profile regardless of the azimuthal angle
φ. In order to obtain a three-dimensional mesh we discretize this profile in a way which does not depend on φ. We
first choose the angular stepsize ∆φ, i.e., the angle between two adjacent planes. Then we establish a meridional
stepsize, i.e., the distance between two successive points in the same plane. For the first point, we calculate the
distance between this point and the corresponding first point in the adjacent plane d1→2 = ρ1∆φ. This distance is
taken as the stepsize between the first and the second point of the same plane. We repeat this operation for all the
points i, di→i+1 = ρi∆φ until we reach the first point again to close the profile. We obtain a mesh composed of squares
with edges that are the bonds between two successive points of the same plane and the bonds between two analogous
points (of same arc length) in two adjacent planes. Dividing these squares into two we obtain triangles that are almost
equilateral. Moreover, they have the advantageous property that they are smaller close to the symmetry axis where
the curvature becomes larger. This property is important in order to increase the precision without increasing the
number of the nodes dramatically. For the non-axisymmetric meshes we apply a conformal transformation to each
node of the axisymmetric mesh while keeping the same connectivities. In this way we preserve the proportionality
between the curvature and the size of the triangles.
15
[1] Eggers J 1997 Nonlinear dynamics and breakup of free-surface flows Rev. Mod. Phys. 69, 865.
[2] Isenberg C 1992 The science of soap films and soap bubbles (Dover, Mineola, NY).
[3] Goldstein RE, Moffatt HK, Pesci AI and Ricca RL 2010 Soap-film Mobius strip changes topology with a twist singularity
Proc. Natl. Acad. Sci. USA 107 21979.
[4] Wang JC 1996 DNA topoisomerases Annu. Rev. Biochem. 65 635.
[5] Clementi C, Nymeyer H and Onuchic JN 2000 Topological and energetic factors: What determines the structural details of
the transition state ensemble and "en-route" intermediates for protein folding? An investigation for small globular proteins
J. Mol. Biol. 298 937.
[6] Voeltz GK, Rolls MM and Rapoport T A 2002 Structural organization of the endoplasmic reticulum EMBO Rep. 3 944.
[7] Mannella CA 2006 Structure and dynamics of the mitochondrial inner membrane cristae BBA - Mol. Cell. Res. 1763 542.
[8] Terasaki M et al 2013 Stacked endoplasmic reticulum sheets are connected by helicoidal membrane motifs Cell 154 285.
[9] Guven J, Huber G and Valencia DM 2014 Terasaki Spiral Ramps in the Rough Endoplasmic Reticulum Phys. Rev. Lett.
113 188101.
[10] Fletcher DA and Mullins RD 2010 Cell mechanics and the cytoskeleton Nature 463 485.
[11] Lin LC-L and Brown FLH 2005 Dynamic simulations of membranes with cytoskeletal interactions Phys. Rev. E 72 011910.
[12] Amoasii L et al 2013 Myotubularin and PtdIns3P remodel the sarcoplasmic reticulum in muscle in vivo J. Cell Sci. 126
1806.
[13] Almsherqi ZA, Landh T, Kohlwein SD and Deng Y 2009 Cubic membranes: The missing dimension of cell membrane
organization Int. Rev. Cell Mol. Biol. 274 275.
[14] Kahraman O, Stoop N and Muller MM 2012 Morphogenesis of membrane invaginations in spherical confinement EPL 97
68008.
[15] Kahraman O, Stoop N and Muller MM 2012 Fluid membrane vesicles in confinement NJP 14 095021.
[16] Sakashita A, Imai M and Noguchi H 2014 Morphological variation of a lipid vesicle confined in a spherical vesicle Phys.
Rev. E 89 040701(R).
[17] De Pascalis R, Napoli G and Turzi SS 2014 Growth-induced blisters in a circular tube Physica D 283 1.
[18] Rim JE, Purohit PK and Klug WS 2014 Mechanical collapse of confined fluid membrane vesicles Biomech. Model. Mechan.
13 1277.
[19] Guven J, Santiago A and Vazquez-Montejo P 2013 Confining spheres within hyperspheres J. Phys. A 46 135201.
[20] John JB et al 2005 The Mitochondrial Inner Membrane Protein Mitofilin Controls Cristae Morphology Mol. Biol. Cell 16
1543.
[21] Ou-Yang ZC 1990 Anchor ring-vesicle membranes Phys. Rev. A 41 4517.
[22] Seifert U 1990 Shape transformations of free, toroidal and bound vesicles Journal de Physique, Colloque 51 339.
[23] Seifert U 1991 Vesicles of toroidal topology Phys. Rev. Lett. 66 2404.
[24] Mutz M and Bensimon D 1991 Observation of toroidal vesicles Phys. Rev. A 43 4525.
[25] Fourcade B, Mutz M and Bensimon D 1992 Experimental and theoretical study of toroidal vesicles Phys. Rev. Lett. 68
2551.
[26] Fourcade B 1992 Theoretical results on toroidal vesicles J. Phys. II France 2 1705.
[27] Julicher F, Seifert U and Lipowsky R 1993 Phase diagrams and shape transformations of toroidal vesicles J. Phys. II
France 3 1681.
[28] Michalet X and Bensimon D 1995 Vesicles of Toroidal Topology: Observed Morphology and Shape Transformation J.
Phys. II France 5 263.
[29] Sakashita A, Imai M and Noguchi H 2015 Shape transformations of toroidal vesicles Soft Matter 11 193.
[30] Canham PB 1970 The Minimum Energy of Bending as a Possible Explanation of the Biconcave Shape of the Human Red
Blood Cell J. Theoret. Biol. 26 61.
[31] Helfrich W 1973 Elastic Properties of Lipid Bilayers: Theory and Possible Experiments Z. Naturforsch. 28 c 693.
[32] Evans EA 1974 Bending resistance and chemically induced moments in membrane bilayers Biophys. J. 14 923.
[33] Kreyszig E 1991 Differential Geometry (Dover, Mineola, NY)
[34] Hu M, Briguglio JJ and Deserno M 2012 Determining the Gaussian Curvature Modulus of Lipid Membranes in Simulations
Biophys. J. 102 1403.
[35] Jahn R and Scheller RH 2006 SNAREs -- engines for membrane fusion Nat. Rev. Mol. Cell Biol. 7 631.
[36] Marques FC and Neves A 2014 Min-Max theory and the Willmore conjecture Ann. Math. 179 683.
[37] Kl´eman M 1977 Energetics of the focal conics of smectic phases J. Phys. France 38 1511.
[38] Abramowitz M and Stegun IA 1970 Handbook of Mathematical Functions 9th edn. (Dover, Mineola, NY).
[39] Seifert U, Berndl K and Lipowsky R 1991 Shape transformation of vesicles: Phase diagram for spontaneous-curvature and
bilayer-coupling models Phys. Rev. A 44 1182.
[40] Golubovi´c L 1994 Passages and droplets in lamellar fluid membrane phases Phys. Rev. E 50 R2419.
[41] Morse DC 1994 Topological instabilities and phase behavior of fluid membranes Phys. Rev. E 50 R2423.
16
[42] Gompper G and Kroll DM 2000 Statistical mechanics of membranes: Freezing, undulation and topology fluctuations J.
Phys.: Condens. Matter 12 A29.
[43] Helfrich W 1978 Steric Interaction of Membranes in Multilayer Systems Z. Naturforsch. 33 a 305.
[44] Julicher F and Seifert U 1994 Shape equations for axisymmetric vesicles: A clarification Phys. Rev. E 49 4728.
[45] Numerical Recipes in C 1992 ed W. H. Press et al (Cambridge University Press, Cambridge, UK).
[46] Gray A, Abbena E and Salamon S 2006 Modern Differential Geometry of Curves and Surfaces with Mathematica
(cid:114)
3rd
edn. (Chapman & Hall/CRC, Boca Raton, FL).
[47] Guven J 2006 Laplace pressure as a surface stress in fluid vesicles J. Phys. A: Math. Gen. 39 3771.
[48] Castro-Villarreal P and Guven J 2007 Axially symmetric membranes with polar tethers J. Phys. A 40 4273.
[49] Castro-Villarreal P and Guven J 2007 Inverted catenoid as a fluid membrane with two points pulled together Phys. Rev.
E 76 011922.
[50] Guven J and V´azquez-Montejo P 2013 Force dipoles and stable local defects on fluid vesicles Phys. Rev. E 87 042710.
[51] Feng F and Klug WS 2006 Finite element modeling of lipid bilayer membranes J. Comput. Phys. 220 394.
[52] Ma L and Klug WS 2008 Viscous regularization and r-adaptive remeshing for finite element analysis of lipid membrane
mechanics J. Comput. Phys. 227 5816.
[53] Cirak F, Ortiz M and Schroder P 2000 Subdivision surfaces: a new paradigm for thin-shell finite-element analysis Int. J.
Numer. Methods Eng. 47 2039.
|
1002.0954 | 1 | 1002 | 2010-02-04T10:18:00 | Exciton Dynamics in Photosynthetic Complexes: Excitation by Coherent and Incoherent Light | [
"physics.bio-ph",
"physics.chem-ph"
] | In this paper we consider dynamics of a molecular system subjected to external pumping by a light source. Within a completely quantum mechanical treatment, we derive a general formula, which enables to asses effects of different light properties on the photo-induced dynamics of a molecular system. We show that once the properties of light are known in terms of certain two-point correlation function, the only information needed to reconstruct the system dynamics is the reduced evolution superoperator. The later quantity is in principle accessible through ultrafast non-linear spectroscopy. Considering a direct excitation of a small molecular antenna by incoherent light we find that excitation of coherences is possible due to overlap of homogeneous line shapes associated with different excitonic states. In Markov and secular approximations, the amount of coherence is significant only under fast relaxation, and both the populations and coherences between exciton states become static at long time. We also study the case when the excitation of a photosynthetic complex is mediated by a mesoscopic system. We find that such case can be treated by the same formalism with a special correlation function characterizing ultrafast fluctuations of the mesoscopic system. We discuss bacterial chlorosom as an example of such a mesoscopic mediator and propose that the properties of energy transferring chromophore-protein complexes might be specially tuned for the fluctuation properties of their associated antennae. | physics.bio-ph | physics |
Exciton Dynamics in Photosynthetic Complexes: Excitation by Coherent and
Incoherent Light
Tomáš Mančal1 and Leonas Valkunas2,3
1Charles University in Prague, Faculty of Mathematics and Physics,
Ke Karlovu 5, CZ-121 16 Prague 2, Czech Republic
2Institute of Physics, Savanoriu Avenue 231, 02300 Vilnius, Lithuania and
3Department of Theoretical Physics, Faculty of Physics of Vilnius University,
Sauletekio Avenue 9, build. 3, 10222 Vilnius Lithuania
In this paper we consider dynamics of a molecular system subjected to external pumping by a
light source. Within a completely quantum mechanical treatment, we derive a general formula,
which enables to asses effects of different light properties on the photo-induced dynamics of a molec-
ular system. We show that once the properties of light are known in terms of certain two-point
correlation function, the only information needed to reconstruct the system dynamics is the reduced
evolution superoperator. The later quantity is in principle accessible through ultrafast non-linear
spectroscopy. Considering a direct excitation of a small molecular antenna by incoherent light we
find that excitation of coherences is possible due to overlap of homogeneous line shapes associated
with different excitonic states.
In Markov and secular approximations, the amount of coherence
is significant only under fast relaxation, and both the populations and coherences between exciton
states become static at long time. We also study the case when the excitation of a photosynthetic
complex is mediated by a mesoscopic system. We find that such case can be treated by the same
formalism with a special correlation function characterizing ultrafast fluctuations of the mesoscopic
system. We discuss bacterial chlorosom as an example of such a mesoscopic mediator and propose
that the properties of energy transferring chromophore-protein complexes might be specially tuned
for the fluctuation properties of their associated antennae.
I.
INTRODUCTION
In recent years, primary processes in photosynthesis
have received a renewed interest from a broader physi-
cal community thanks to experimental observation of co-
herent energy transfer in some photosynthetic systems.
The ground breaking coherent two-dimensional electronic
spectroscopy (2D-ES) experiment of Engel et al. [1] has
led to new appreciations of the role that may be played by
coherent dynamics in excitation energy transfer (EET),
and of the quantum mechanical nature of photosynthetic
systems in general [2]. Special theoretical effort has been
made to understand the role of noise [3–7] in the dynam-
ics of excitation energy transfer, and the role of coher-
ence [8–12] in excitonicaly coupled systems. On the ex-
perimental front, the method of coherent 2D-ES [13, 14]
has established itself as a tool opening new window into
the details of energy transfer dynamics in photosynthetic
[15–20], and other molecular systems [3, 21, 22]. Coher-
ent effects have been now reported in different molecular
systems, often biologically relevant [22, 23] - a general-
ity that asks for a search of the possible evolutionary
advantage underlying their abundance in photosynthetic
pigment-protein complexes.
The principle pigment molecules responsible for the
primary processes of photosynthesis are chlorophylls
(Chls) and bacteriochlorophylls (BChls) [24, 25]. They
are involved in accumulation of light energy via the exci-
tation energy transfer to specific pigment-protein com-
plexes - reaction centers. Spectral variability of pho-
tosynthetic light-harvesting pigment-protein complexes
arises either from excitonic interactions between pigment
molecules or from their interactions with protein sur-
rounding. Both these interactions are the main factors
determining the excitation dynamics in light-harvesting
[26]. Excitonic aggregates are subject to interaction with
two types of environments, and they provide means of
transferring energy from one environment to another.
First of these environments, the radiation, is under nat-
ural conditions at much higher temperature than the
second environment, the protein scaffold and indeed the
photosynthetic chemical machinery as a whole. The ex-
cess of photons on suitable wavelength in the radiational
environment is used to excite spatially extended antenna
systems that concentrate excitation energy to the reac-
tion center, which in turn drives charge transfer processes
across cellular membranes to create the transmembrane
potential and the pH gradient [24].
Non-equilibrium processes occurring in photosynthetic
systems during light harvesting are conveniently de-
scribed by reduced density matrix (RDM) theory [26–
28] which has an advantage of being applicable to dis-
ordered statistical ensembles that the experiments often
deal with. However, with recent 2D experiments that
enable us to distinguish the homogeneous and inhomoge-
neous spectral broadening, and with the progress in single
molecular spectroscopy [29] we can gain insight into the
time evolution characteristic to single molecules interact-
ing with their environment [30, 31]. This fact enables us
to return to the wavefunction formalism and to look at
light harvesting from the point of view which takes the
superposition principle of quantum mechanics seriously.
It has been show that such an approach yields many in-
teresting insights into the emergence of the classical prop-
erties of molecular system from their underlying quantum
mechanical nature [32, 33]. As the light-harvesting pro-
cesses seem to operate on the interface between classical
and quantum worlds it seems appropriate to look at them
from the point of view of the decoherence program of Zeh,
Zurek and others [34, 35].
The process of light harvesting could then be describes
as follows. First, the system is in an "equilibrium" initial
state Ψ0i characterized by the excitonic ground state
gi, the state of protein (phonon) environment ΦP i cor-
responding to this electronic ground state and some state
of light Ξ0i, i.e.
Ψ0i = giΦBiΞ0i.
(1)
The light-harvesting occurs when the state of light is such
that the time evolution of the system leads to population
of higher excited states eni of photosynthetic antenna.
These states are formed from excited states of Chls and
other chromophores, such as carotenoids [26]. We denote
these combined excited states as excitons. In the first ap-
proximation, photosynthetic antenna remains in the ex-
cited state until the excitation energy is transferred to the
reaction center. This happens much faster than compet-
ing process of spontaneous emission which can therefore
be neglected in our discussion. When the interaction of
the antenna with light is switched on, the change occur-
ring in the ground state portion of the total state vector
after the passage of time ∆t is
Ψ0i → α∆tgiΦBiΞ0i +Xn
β(n)
∆t eniΦBiΞ′i
(2)
The subsequent time evolution of the excited state por-
tion of the state vector is independent of the ground state
part, and we can thus look at it separately. Because we
neglect spontaneous emission, any excitation to higher
excited state, as well as transitions between exciton states
due to the light, the state vector Ξ′i remains approxi-
mately unentangled with excitons and the protein bath
for the rest of the energy transfer process. It can there-
fore be omitted. The initial state for the energy transfer
process thus reads
Ψe(t0)i =Xn
β(n)(t0)eniΦBi,
(3)
where we omitted the lower index ∆t. If the basis of the
states eni is chosen so that the molecular Hamiltonian
is diagonal, the energy transfer occurs only due to inter-
action of excitons with their surrounding environment.
This interaction leads to an entanglement of excitons and
the environment
Ψe(t)i =Xn
β(n)(t)eniΦ(n)
B (t)i.
(4)
After a sufficiently long time the environment state vec-
tors corresponding to different electronic state diverge
2
maximally and the reduced density matrix becomes di-
agonal in some basis, i.e.
ρ(t) = trB{Ψe(t)ihΨe(t)}
β(n)(t)(β(m)(t))∗hΦ(m)
B (t)Φ(n)
B (t)i.
(5)
=Xmn
Often, to a good approximation, such preferred basis is
the one in which the electronic Hamiltonian is diagonal,
the so-called excitonic basis. However, notable correc-
tions to this rule are predicted even for weak system-bath
coupling [12, 36].
The final state of the energy transfer is the one in which
just reaction centers are populated
Ψe(t)i =Xk
β(RCk)(t)eRCk iΦ(RCk)
B
(t)i.
(6)
The last step of the energy transfer, from the antenna
to the reaction center is often slower than typical trans-
fer times between antenna complexes, and so the final
state is well localized on the reaction center, and coher-
ences between individual reaction centers are unlikely to
survive.
It is clear from the above discussion, that decoherence
during the energy transfer in the antenna is determined
by the evolution of the environmental degrees of free-
dom (DOF). The decoherence from the rest of the sys-
tem might be required for the localization of the energy
in the reaction center, but there is no obvious reason for
fast decoherence during the initial steps of energy transfer
in the antenna, apart from the fact that a bath formed
by a completely random disordered environment would
lead to just such fast decoherence. It has been suggested
before that the protein environment might play a more
active role in steering and protecting electronic excita-
tion [1, 23] and controlling the decoherence might be one
of the possible pathways to more robust EET.
There is however one important caveat in the above
scheme. The initial condition, Eq. (3), has been intro-
duced artificially into Eq. (2) as a result of an interac-
tion occurring during some short time interval ∆t. If the
system is continuously pumped, individual contributions
similar to Eq. (3) will interfere, possibly disabling any
effect of cooperative involvement of the bath. It is even
more important to consider the question what are the
effects of natural sun light [37], i.e. whether the coher-
ent scenario outlined above is plausible for the photosyn-
thetic system in vivo, or not. This depends strongly on
the nature of the excitation process, whether it occurs in
discrete independent jumps of the kind described by Eq.
(2), or continuously over a long period of uncertainty in-
terval of the photon arrival. The former view is usually
held in support of the relevance of ultrafast spectroscopic
finding for in vivo function of the photosynthetic systems
[11]. Below we derive a general formula which enables us
to describe all these regimes by a unified formalism, and
also enables us to place the observables of ultrafast coher-
ent spectroscopy in perspective with the dynamics under
natural conditions. In a somewhat extended form our re-
sult is also applied to another case cited in support of the
utility of coherent dynamics in photosynthetic systems,
a case where a small photosynthetic complex is excited
through another, possibly mesoscopic, antenna [11].
The paper is organized as follows. Next section in-
troduces a rather general model of photosynthetic aggre-
gate. In Section III we discuss the dynamics of a system
excited by coherent pulsed light and the observables of
the ultra-fast non-linear spectroscopy. Section IV is con-
cerned with the excitation of a photosynthetic system by
the light from a general source. Implications of the theory
for excitation by thermal and coherent light, as well as
excitation mediated by mesoscopic system are discussed
in Section V.
II. HAMILTONIAN OF A MODEL
PHOTOSYNTHETIC SYSTEM
In this section we briefly review the excitonic model
that was very successfully applied to model the spec-
troscopic properties of Chl- and Bchl- based light har-
vesting chromophore–protein complexes (see e.g.
[16]).
We assume N monomers with ground states gni, excited
states eni, n = 1, . . . , N , and with electronic transition
energies εn. These monomers are interacting with the
phonon bath of protein DOF so that the Hamiltonian of
the monomer reads
Hn = (T + Vg)gnihgn + (εn + T + Ve)enihen.
(7)
Here, T is the kinetic energy operator of the bath, and
Vg and Ve are the potential energy operators of the bath
when the system is in the electronic ground- and excited
states, respectively. We set the ground state electronic
energy to zero for conveniency. The Hamiltonian, Eq.
(7) can be split into the pure bath, pure electronic and
the interaction terms so that
Hn = (T + Vg) ⊗ In
M
+ In
B ⊗ (εn + hVe − Vgieq)enihen
}
Hn
M
{z
}
+ (Ve − Vg − hVe − Vgieq) ⊗ enihen
.
(8)
Hn
M −B
{z
}
Here, IB is the unity operator on the bath Hilbert space
and IM is the unity operator on the Hilbert space of the
electronic states. The equlibrium average hVe − Vgieq
of the potential energy operators was added to the elec-
tronic energy so that the interaction term is zero for the
system in equilibrium.
In
chromophore–protein
such
monomers are coupled by resonance coupling. The
complexes many
Hn
B
{z
3
whole complex can be described by means of collective
states including the ground state
gi =
one excitation states
⊗gni,
N
Yn=1
(9)
¯eai =
a−1
Yn=1
⊗gni ⊗ eai ⊗
N
Ym=a+1
⊗gmi,
(10)
and states containing higher number of excitations. For
the sake of brevity we now stop writing the symbol of
the direct product ⊗ and the unity operators In
B etc. ex-
plicitely. The total Hamiltonian of the complex including
resonance interaction is then defined as
HB + HM + HM −B =
H n
B +
H n
M
n
Xn=1
N
Xn=1
Xn=1
n
+
N
Xn6=m
Jnm¯enih¯em +
H n
M −B .
(11)
If the system-bath interaction with bath is weak, the re-
ferred basis into which the electronic system relaxes due
to interaction with the bath is, to a good approximation,
the one in which the electronic part of the Hamiltonian
is diagonal. Let us denote these states as eni. They are
usually termed excitons and they represent certain linear
combination of the collective states ¯eni where excitations
are localized on individual chromophore molecules. One
of the most important characteristics of this model is that
it does not include direct relaxation of the electronic ex-
cited states to the ground states due to electron-phonon
coupling. This is well satisfied by Chls and BChls on the
ultrafast time scale of which light harvesting processes
occur.
III. EXCITATION BY COHERENT PULSED
LIGHT AND NON-LINEAR SPECTROSCOPY
Let us now consider experimental methods which pro-
vide information about time evolution of excited states of
photosynthetic systems. Because of the timescale of EET
processes, spectroscopy with ultrashort time resolution is
a necessary tool. The interaction of the pulsed coherent
light with the photosynthetic system is well described in
semi-classical approximation [38]. Electric field of the
light is then considered as an external parameter of the
system Hamiltonian. Electronic DOF can be prepared
very fast in an excited state, not affecting, to a good ap-
proximation, the bath DOF. Thus, in an experiment with
an ideal time resolution, we would have the system pre-
pared in the excited state, Eq. (3). The time evolution
of the system is governed by the Schrödinger equation
∂
∂t
Ψe(t)i = −
i
(HB + HM + HM −B)Ψe(t)i
4
(18)
+ δ(t)Ψe(t0)i,
(12)
The matrix elements of the superoperator read
with initial condition Ψe(t)i = 0 for t < t0. The last
term in Eq.
(12) describes the ultrafast event of the
molecule–radiation interaction. Formal solution of this
equation reads Ψe(t)i = UB(t)UM (t)UM −B(t)ψe(t0)i,
where we defined evolution operators UB(t), UM (t) of
the bath and the molecule, respectively, as
UB(t) = Θ(t − t0) exp{−
i
HB(t − t0)},
(13)
U (e)
abcd(t) = haUM (t)UM −B(t)ci . . .
× hdU †
M −B(t)U †
M (t)bi,
where the dots . . . denote where an operator on which
U (e)(t) acts has to be inserted. The reduced evolution
operator ¯U (e)(t) defined as
¯U (e)(t) = trB{U (e)(t)},
(19)
UM (t) = Θ(t − t0) exp{−
i
HM (t − t0)},
(14)
contains information about the evolution of the RDM
only.
and the remaining interaction evolution operator as
UM −B(t) = Θ(t − t0) exp{−
t
i
t0
dτ U †
B(τ )U †
M (τ )
× HM −BUM (τ )UB(τ )},
(15)
After excitation, the process of energy transfer proceeds
according to the description presented in Introduction
and can be experimentally monitored.
A. Evolution superoperator
Matrix elements of the RDM of the molecule, which
holds the information about the population probabilities
and the amount of coherence between electronic states
are given by expectation value of projectors enihem,
ρnm(t) = hψe(t)emihenψe(t)i
= trB{henψe(t)ihψe(t)emi}
B. Non-linear spectroscopy
In non-linear spectroscopy, coherent laser light is used
to investigate the dynamics of molecular systems by ap-
plying special sequences of pulses. Some pulses act to in-
duce non-equilibrium dynamics (pump), and other pulses
act to monitor (probe) the evolution after the pump. One
of the most advanced of these methods, coherent 2D-ES
[13, 39] measures the response of a system to three pulses
traveling in different directions k1, k2 and k3. The detec-
tion is arranged in such a way (measuring in the direction
−k1 + k2 + k3) that the signal is predominantly of the
third order, with contributions of one order per pulse [38].
Let us denote delays between the first two pulses by τ and
the delay between the second and the third pulse by T .
If the pulses are ideally short, the signal is composed of
two kinds of contribution. First, contribution that in-
volves population of the excited state corresponding to
the density operator
W (e)
k2,k1
(t, τ ) = ψ(k2)
e
(t)ihψ(k1)
e
(t + τ ),
(20)
and second, contribution that involves evolution in the
ground state
β(a)(β(b))∗aihb
W (g)
k2,k1
(t, τ ) = ΨBigihψ(k2,k1)
g
(t, t + τ ).
(21)
= hentrB{UM (t)UM −B(t)Xab
ρ0
{z
}
× ΦBihΦB
U †
M −B(t)U †
M (t)}emi
(16)
Weq
{z
}
This can be rewritten by defining an evolution superop-
erator U(t) which acts on initial density matrix ρ0Weq,
i.e.
ρnm(t) = trB{henW (t)emi}
= henU (e)(t)ρ0Weqemi,
(17)
Here, we denote the pulses acting on the state vector
by their corresponding wave vector in the upper index,
and the excited state or ground state bands by the lower
index g and e, respectively. For these statistical operators
we can define evolution superoperators U (e)(t), U (g)(t),
U (eg)(t) and U (ge)(t) in analogy with Eqs. (17) and (18),
so that
(t, τ ) = U (e)(t)U (ge)(τ )ΨBihΨBµgihgµ,
(22)
W (e)
k2,k1
and
W (g)
k2,k1
(t, τ ) = U (g)(t)U (ge)(τ )ΨBihΨBµ2gihg.
(23)
The superoperator U (eg)(t) is the evolution superopera-
tor of a coherence projector Pn enihg and analogically
for U (ge)(t). After a delay T the third ultrafast pulse is
applied and the non-linear signal is recorded. The signal
corresponds to indirectly to non-linear polarization of the
sample, and is usually measured in frequency domain
E(3)(ω, T, τ ) ≈ i tr{µU (eg)(ω)µ
× (U (e)(T )U (ge)(τ )µρgµ + U (g)(T )U (ge)(τ )µ2ρg)Weq}.
(24)
Here, we denoted
U (eg)(ω) =
∞
0
dt eiωtU (eg)(t),
(25)
and ρg = gihg. In 2D coherent spectroscopy, the signal
is in addition Fourier transformed along the time delay
τ , so that the spectrum is defined as
S2D(ωt, T, ωτ ) =
∞
−∞
dτ e−iωτ τ E(3)(ωt, T, τ ).
(26)
The spectrum defined in this way has a suitable inter-
pretation of an absorption – absorption and absorption
– stimulated emission correlation plot, with a different
waiting times T between the two events. The 2D spec-
trum is in practice measured with finite pulses, and the
measured time domain signal is thus a triple convolution
of the responses to a delta pulse excitation, with the ac-
tual finite pulses [40].
From this rough sketch of the principles and the infor-
mation content of the coherent 2D spectroscopy it should
be clear that 2D spectroscopy is aimed at disentangling
the dynamics of the system during the time delay T . In
the so-called Markov approximation, when the dynamics
in time intervals τ , T and t is assumed separable, and
the bath is assumed stationary, the ground state evolu-
tion during interval T can be neglected. Then 2D mea-
surement essentially accesses the reduced evolution su-
peroperator, Eq. (19) and possibly also the more general
superoperator
¯U (e)(t, τ ) = trB{U (e)(t)U (ge)(τ )Weq }.
(27)
We will show below that this superoperator, together
with the light properties, determines the way in which
the molecule is excited in a general case, even at illumi-
nation by natural light.
IV. EXCITATION BY LIGHT
5
+HM −B+HM −R+HM −S+HB−S+HB−R+HR−S. (28)
We have divided the system into a molecule (HM ), its
environment or bath (HB), the radiation (HR) and the
light emitting body (LEB) which produces it, e.g. Sun
or laser medium (HS).
It seems reasonable to neglect
a direct interaction between the molecule (together with
its environment) and the molecules of the LEB. Conse-
quently, the terms HM −S and HB−S can be disregarded.
To make the treatment simpler we can also neglect the
interaction between radiation and the molecular environ-
ment, HB−R. The assumption is that the energy of the
molecular transition that is used to harvest light for pho-
tosynthetic purposes is much larger than any of the tran-
sitions in this environment and the two regions of the
light spectrum can thus be treated separately. One can
also assume that the part of radiation spectrum which
would interact with the bath is simply filtered out, and
the environment is kept at certain temperature by other
means.
A. Radiation entangled with the light emitting
body
An important special case is the one in which the ra-
diation and the LEB is in equilibrium with each other so
that the radiation is described by the canonical equilib-
rium density matrix
W (eq)
R =Xλq
Nλq
ωq
kB T
e−
Zλq
NλqihNλq.
(29)
Here, Nλqi is the N -photon state of the radiation mode
with polarization vector eλ and wave-vector q. As we
have already noted above, the statistical concept of den-
sity matrix will be replaced here with the concept of en-
tangled states, so that we can describe the whole system
by its state vector. Thus, we introduce a state vector
Ξ(t)i =Xλq XNλq
cN λq(t)NλqiφN λq(t)i,
(30)
in which the light is fully entangled with the states
φN λq(t)i of the LEB . The LEB states have to fulfill
the condition
hφN λq(t)φN ′λ′q′(t)i = δλλ′ δqq′ δNλqN ′
λq
,
(31)
so that when the total density matrix of the LEB and ra-
diation is averaged over the states of the body, we obtain
Eq. (29). W (eq)
is recovered provided that
R
In order to account for general light properties we will
consider the problem fully quantum mechanically, and
assume only deterministic evolution of the system wave-
function. The Hamiltonian of the system reads
cN λq(t)2 =
Nλq
ωq
kB T
e−
Zλq
.
(32)
In the absence of the light absorbing body, the evolu-
tion of the state Ξ(t)i is governed by the Hamiltonian
H = HM + HB + HR + HS
HL = HR + HS + HR−S ,
(33)
and
UL(t) = Θ(t − t0)
× exp(cid:26)−
i
(HS + HR + HR−S)(t − t0)(cid:27)
(34)
is the corresponding evolution operator.
B. Equation of motion
For the subsequent treatment of the system dynam-
ics, we introduce the interaction picture with respect to
Hamiltonian operators HM , HB and HL,
H (I)(t) = HM −B(t) + HM −R(t),
(35)
where
6
HM −R(t) is not on the far right of the expression are
equal to zero. Eq. (40) therefore simplifies into a series
Ψ(I)(t)i = Ψ0i −
t
i
t0
dτ HM −R(τ )Ψ0i
−
1
2
t
t
t0
dτ
t0
dτ ′HM −B(τ )HM −R(τ ′)Ψ0i + . . . . (42)
Now we introduce a projector Pe that excludes the exci-
tonic ground state gi
Pe =Xn
enihen.
(43)
Applying this projector to Eq. (42) has only the effect
of eliminating the first term of the series.
Introducing
abbreviations
HM −B(t) = U †
M (t)U †
B(t)HM −B UB(t)UM (t),
and
HM −R(t) = U †
M (t)U †
L(t)HM −RUL(t)UM (t).
(36)
(37)
and
S(t)i = −
t
i
t0
dτ HM −R(τ )Ψ0i
(44)
Equation of motion for the total state vector in the in-
teraction picture
Ψ(I)
e (t)i = PeΨ(I)(t)i,
(45)
Ψ(I)(t)i = U †
M (t)U †
B(t)U †
L(t)Ψ(t)i
thus reads
∂
∂t
Ψ(I)(t)i =
−
i
(HM −B(t) + HM −R(t)) Ψ(I)(t)i.
(39)
The solution of Eq. (39) can be found formally as
Ψ(I)(t)i = exp+n −
t
i
t0
dτ (HM −B (τ )
we can write
(38)
Ψ(I)
e (t)i = S(t)i −
t
i
t0
dτ HM −B(τ )S(τ )i
−
1
2
t
t0
dτ
τ
t0
dτ ′HM −B(τ )HM −B (τ ′)S(τ ′)i+. . . . (46)
It is possible to verify easily that this series is a solution
of the equation
∂
∂t
Ψ(I)
e (t)i = −
i
HM −B(t)Ψ(I)
e (t)i +
∂
∂t
S(t)i,
(47)
with initial condition Ψ(I)
e (t0)i = 0.
+ HM −R(τ ))oΨ0i.
(40)
C. Pumping source term
We will assume that the system is initially in the state
Ψ0i of Eq. (1). With this choice we have
HM −B(t)Ψ0i = 0.
(41)
Further in this paper, we will assume weak interaction
with the radiation, so that it can be described by lin-
ear theory. Thus, we need to collect all terms in the
expansion of Eq. (40) which include one occurrence of
HM −R(t). Thanks to Eq. (41), however, all terms where
Eq. (47) is an equation of motion for the excited states
of an excitonic aggregate pumped by a source term
S′(t)i =
∂
∂t
S(t)i = −
i
HM −R(t)Ψ0i.
(48)
Hamiltonian HM −R will be assumed in the dipole ap-
proximation, i.e.
HM −R ≈ −µ · ET (r),
(49)
where µ is the transition dipole moment operator of the
aggregate
Since Hamiltonian HS commutes with the radiation op-
erators and
7
dnenihg + h.c.,
(50)
U †
R(t)aλqUR(t) = e−iωq taλq,
µ =Xn
and ET is the operator of the (transversal) electric field
of the radiation
ET (r) = −iXλq (cid:16)eλ−qf−q(r)a†
λq
we have
and
λq(t) = a†
a†
λq(t)eiωq t,
aλq(t) = aλq(t)e−iωq t.
(59)
(60)
(61)
(62)
(63)
with
− eλqfq(r)aλq(cid:17),
fq(r) =r ωq
2ǫ0Ω
eiq·r.
Here, Ω is a quantization volume.
We consider a molecule much smaller than the wave-
length of the light, so that eiq·r is constant in the volume
of the molecule. The origin of the coordinates can thus
be conveniently put into the molecule yielding eiq·r ≈ 1.
The interaction Hamiltonian in Eq. (48) then reads
HM −R(t) = iXλq
µλq(t)fλ−q(0)a†
λq(t)
− iµλq(t)fλq(0)aλq(t),
(53)
where the creation and annihilation operators of the field
are in the interaction picture with respect to Hamiltonian
HL, i.e.
a†
λq(t) = U †
L(t)a†
λqUL(t),
aλq(t) = U †
L(t)aλqUL(t).
(54)
(55)
The transition dipole moment operator projected on the
polarization vector of a mode λq appears in the interac-
tion picture with respect to Hamiltonian HM ,
µλq(t) = U †
M (t)µ · eλqUM (t).
(56)
The evolution operator UL(t), Eq. (34), can be rewritten
as
UL(t) = US(t)UR(t)UR−S(t),
(57)
where
UR−S(t) = Θ(t − t0) exp+n −
t
i
t0
dτ U †
S(τ )U †
R(τ )
(51)
Here, we introduced slow oscillating envelops
(52)
and
a†
λq(t) = U †
R−S(t)a†
λqUR−S(t),
aλq(t) = U †
R−S(t)aλqUR−S(t).
Inserting these expressions into Eq. (53) we can distin-
guish two terms associated with the transition from the
ground state gi to an excited state eai with respective
phase factors ei(ωag −ωq )t and ei(ωag +ωq )t. While the first
one will lead to a resonance excitation around ωq ≈ ωag,
the later term is oscillating fast and will therefore con-
tribute very little compared to the former one. Thus we
drop the fast oscillating part, and obtain the source term
in the form
S′(t)i =
1
Xλq
µλq(t)fλq
× aλq(t)giΦBiΞ0i.
(64)
Using this form of the source term, we can find state into
which the molecule is weakly driven by any type of light.
D. Excited state dynamics under pumping
So far we have treated the problem systematically us-
ing the wavefunction approach. The time evolution of
the system wavefunction is governed by Eq.
(39). To
find the probabilities of creating population on and co-
herence between certain excitonic levels eai we solve Eq.
(39) formally,
Ψ(I)
e (t)i =
t
t0
dτ UM −B(t − τ )S′(τ )i.
(65)
Here, we used the fact that Ψe(t0)i = 0. Now let us
evaluate matrix element Pab(t) = hΨe(t)PabΨe(t)i of a
projector
× HR−SUR(τ )US(τ )o.
(58)
Pab = eaiheb,
(66)
which gives the probability of finding the molecule in
state eai if a = b, or characterizes the amount of coher-
ence between states eai and ebi if a 6= b. Note that we
have removed the interaction picture, Eq. (38). We have
Pab(t) =
1
2
t
t0
dτ
t
t0
dτ ′ Xλq,λ′q′
fλq(fλ′q′)∗
×hΞ0a†
λq(τ )aλ′ q′(τ ′)Ξ0i
8
In so called secular and Markov approximations (see
e.g. Ref. [27]) matrix elements of the evolution superop-
erator governing the coherences take a very simple form.
First, it is possible to separate the two time arguments
in the superoperator ¯U (e)(t, τ ) so that
¯U (e)(t, τ ) = ¯U (e)(t) ¯U (eg)(τ ).
(71)
Since each coherence is independent of the population
dynamics and of other coherences, the one-argument su-
peroperator elements read
× heb ¯U (e)(t − τ, τ − τ ′)ρ0
λq,λ′q′eai,
(67)
and
¯U (e)
abab(t) = e−iωabt−(Γa+Γb)t,
where the evolution superoperator ¯U (e)(t − τ, τ − τ ′) has
been defined in Eq. (27).
In Eq. (67), the light is represented by a first order
correlation function
I (1)
λq,λ′q(τ, τ ′) = (fλq(fλ′q′)∗)−1
¯U (eg)
agag(t) = e−iωag t−Γat.
Here the dephasing rate
Γa = γp +
1
2
Ka
(72)
(73)
(74)
× hΞ0a†
λq(τ )aλ′ q′(τ ′)Ξ0i
(68)
(see e. g. Ref.
properties. We also denoted
[41]), which comprises all its relevant
ρ0
λq,λ′q′ =
1
2
µλqgihgµλ′q′ .
(69)
The quantities Pab(t) are the matrix elements of the
RDM (Pab(t) = hebρ(t)eai) of the system which reads
ρ(t) =
t
t0
dτ
t
t0
dτ ′ ¯U (e)(t − τ, τ − τ ′)
λq,λ′q′I (1)
ρ0
λq,λ′q(τ, τ ′).
(70)
× Xλq,λ′q′
For a weakly driven system, Eq. (70) has a very wide
range of applicability. We will discuss its application to
thermal light and pulsed coherent light in the following
section.
comprises the pure dephasing rate γp and the rate Ka
of depopulation, i.e.
the sum of transition rates from
state eai to other states. A simplified treatment of the
populations is possible for the states that are only de-
populated, i.e. no contributions to the population can
be attributed to the transfer from other levels. They are
found at the top of the energetic funnel of the antenna.
For these states we have
¯U (e)
aaaa(t) = e−Kat.
(75)
Eqs. (72) to (75) neglect all coherence transfer effects, as
well as possible coupling between the dynamics of popu-
lation and coherence.
A. Excitation of coherences by thermal light
For an equilibrium thermal light the correlation func-
tion I (1)
λq,λ′q′ (τ, τ ′) depends only on the difference of the
times τ and τ ′. As discussed above, Ξ0i represents the
equilibrium of the system described by Hamiltonian HL.
The equilibrium density matrix is stationary, i.e.
UL(t)Ξ0ihΞ0U †
L(t) = Ξ0ihΞ0,
(76)
V. DISCUSSION
so we can write
Thorough discussion of excitation dynamics of molecu-
lar systems excited by incoherent light was made in Ref.
[37]. Molecular systems were considered without the bath
effect which is however significant for light harvesting.
Eq. (70) contains reduced evolution superoperator of the
molecular system so that the state of the system created
by the incident light depends on its reduced dynamics. It
is not possible to consider a general case of such dynam-
ics analytically, and we will therefore commit ourselves
to some simple cases.
I (1)
λq,λ′q(τ, τ ′) = fλq−2hΞ0a†
λq(τ − τ ′)aλq(0)Ξ0i
×eiωq(τ −τ ′)δλλ′ δqq′
≡ fλq−2 Iλq(τ − τ ′)δλλ′ δqq′.
It can be shown that
Iλq(−t) = I ∗
λq(t).
(77)
(78)
Assuming some simple form of a light correlation func-
tion, e.g.
Iλq(t) = I 0
λqe− t
τd
+iωq t,
(79)
we obtain for the populations
ρaa(t) = 2ReXλq
I 0
λq[ρλq]aa
t
t0
dτ
τ
t0
dτ ′e−Ka(t−τ )
× e−Γa(τ −τ ′)− τ −τ ′
τd
−i(ωag −ωq)(τ −τ ′)
(80)
Here, [ρ]ab ≡ heaρebi. We utilized Eq. (79) and the fact
that, by definition (see Eqs. (68) and (69)), the time τ
corresponds to the action of the dipole moment operator
from left, whereas time τ ′ corresponds to the same action
from the right. At long times t − t0 → ∞ this yields
ρ∞
aa =Xλq
2
Ka
)I 0
(Γa + 1
τd
λq]aa
(ωag − ωq)2 + (Γa + 1
τd
λq[ρ0
.
)2
(81)
However, neglecting the influence of environment as in
Ref. [37] yields
ρlong
aa (t − t0) =Xλq
τ −1
d 2I 0
λq[ρ0
λq]aa(t − t0)
(ωag − ωq)2 + 1
τ 2
d
,
(82)
which grows linearly with time.
For coherences we have
9
such a case fast enough to prevent averaging over many
such spikes. Eqs. (81) and (83) with large Ka describe
just such a situation. The RDM created by incoherent
light resembles in certain sense the one created by ultra-
fast pulses; it represents a linear combination of excitons.
The coherences in Eq. (83) are however static at long
times.
In our demonstration we concentrated on a simple
model assuming both Markov and secular approxima-
tions to be valid. The presence or absence of coherences
has no significance in such a case, and more involved the-
ories of the RDM dynamics [8, 9, 12] have to be used to
investigate the role of coherences in energy transfer pro-
cesses by Eq. (70).
B. Coherent pulsed light
In derivation of Eq.
(70) we assumed certain initial
state Ξ0i of the system composed of the light and its
source. The condition that the light is in a stationary
state, fully entangled with its source, has only been used
to simplify the correlation function I (1)
λq,λ′ q′(τ, τ ′) for the
case of the thermal light. In a general case Ξ0i will not
represent an equilibrium state.
It can indeed describe
even systems such as a laser producing coherent Gaussian
light pulses with some carrier frequency ω0 and a width
parameter ∆. If we in addition assume that the light is
described by a single polarization, and that we consider
the dynamics after one such pulse centered at time t = τ0,
the light is described as
λq[ρ0
I 0
λq]ab
1
iωab + (Γa + Γb)
Xλq,λ′q′
λqλ′ q′(τ, τ ′) = I0e− (τ −τ0 )2
I (1)
∆2 − (τ ′ −τ0)2
∆2
.
(84)
1
Coherence element created by such light reads
ρba(t) = eiωabt
t
t
t0
dτ
t0
(83)
dτ ′e−i(ωag−ωq )τ ei(ωbg−ωq )τ ′
ρ∞
ab = 2Xλq
×h
i(ωag − ωq) + Γa + 1
τb
1
+
−i(ωbg − ωq) + Γb + 1
τbi,
which turns into Eq. (81) for a = b (with additional as-
sumption Ka = 2Γa). In case of no dephasing, the first
fraction in Eq. (83) yields a delta function δ(ωab) [37].
Thus, for slow or non-existent relaxation due to inter-
action with environment, the system is excited predomi-
nantly into a state represented by a diagonal RDM, as all
coherence terms are negligible compared to the linearly
growing population. For fast relaxation, the coherences
may be of the same order of magnitude as the popula-
tions.
The case of very fast relaxation is particularly interest-
ing. It was suggested previously that coherent dynamics
can be relevant for the in vivo case, because the fluctuat-
ing light from the Sun corresponds to a train of ultrafast
spikes [11]. The relaxation of the antenna must be in
× e−Γa(t−τ )−Γb(t−τ ′)I0e− (τ −τ0 )2
∆2 − (τ ′ −τ0 )2
∆2
ρ0
ba,
(85)
2 hebµgihgµeai. In the limit of ultrashort
ba = 1
where ρ0
pulses when e− (τ −τ0 )2
pure state at τ0, which then dephases as
∆2 → αδ(τ − τ0) the pulse creates a
ρba(t) = Θ(t − τ0)e−(Γa+Γb)(t−τ0)
× eiωab(t−τ0)ρ0
baI0α2.
(86)
In case of a finite pulse and no dephasing our results
coincides with those found in Ref. [37].
C. Mediated excitation
at long times is in analogy with Eq. (70)
10
The major difference between excitation by the ther-
mal light and a coherence pulse is in the occurrence of a
sudden event which populates a nearly pure state of the
excited state band. Clearly, a single molecule interacting
with an ideal continuum of radiation modes in equilib-
rium does not experience such sudden events. Rather, its
interaction with light corresponds to a continuous pump-
ing, and the suddenness of the photon arrival is the conse-
quence of our ability to register only classical outcomes.
In order to register them we have to interact with the
system and become entangled with it. Our experience is
that macroscopic systems interacting with low intensity
light can be used to detect single photons, and certain
more or less definite times can be attributed to their ar-
rivals.
Interaction of a photon with a macroscopic de-
tector yields a temporal localization of the arrival event.
A mesoscopic system may play a role of such a detector
(mediator) that provides its fluctuations to be harvested
by dedicated nano sized antenna. Green photosynthetic
bacteria, from which the photosynthetic complex FMO
was isolated, collect light mainly by means of so-called
chlorosoms [24, 42]. The chlorosom is a self-assembled
aggregate of ∼ 105 BChls and carotenoids with very lit-
tle protein. The typical dimensions the chlorosom are of
the order of 100 nm [42]. It does not seem to be organized
as an energy funnel [43, 44], and the energy transfer time
between its main body and the base plate to which FMO
complexes are attached is of the order of 120 ps [45], i.e
rather slow. The excitation in such a mesoscopic system
may have enough time to become localized through in-
teraction with the large number of the systems DOF and
arrive at the FMO complex in a particle like, i.e. also
temporally localized fashion.
In this section, we will generalize our result, Eq. (70),
for a case when the excitation of the photosynthetic sys-
tems occurs by transfer from another system. We will
therefore assume that our molecule does not interact di-
rectly with light, but is pumped in a similar fashion by
another system. The source term, Eq. (64), is then gen-
eralized as
S′(t)i =
i
A(t)giΦBi
× Xn
αn(t)ξniφn(t)i! .
ρ(t) =
t
t
t0
dτ
t0
dτ ′U (e)(t − τ, τ − τ ′)Xnn′
α∗
n(τ )αn′ (τ )
× hξnA(τ )A(τ ′)ξn′ ihφn(τ )φn′ (τ ′)i.
(89)
The complicated two-point correlation function in Eq.
(89) results from the pumping of the mediator similarly
to the direct pumping of the molecule in Eq. (70). A
mesoscopic system especially when excited will, however,
always exhibit fluctuations which will prevent the corre-
lation function from having a simple smooth dependence
without recurrences. Such recurrences can temporally lo-
calize the excitation events of the molecule. In such an
excitation regime, when coherent dynamics from differ-
ent excitation times do not interfere, optimization of the
FMO's energy channeling capability for case of initially
coherent states would be an advantage.
D. Outlook
More research into specific forms of both the light cor-
relation function for different situation that may occur in
vivo, and the analogical interaction of systems like FMO
with mesoscopic antennae is clearly needed. Ultrafast
spectroscopic experiments play a pivotal role in this re-
search by yielding information about the system response
to the light. To conclude on the utility of coherent dy-
namics for the function of the photosynthetic system is,
however, only possible by taking into account the prop-
erties of light at the natural conditions, for which the re-
sults of this paper provide means. If the coherent dynam-
ics observed in some photosynthetic chromophore-protein
complexes has a significance for their light-harvesting ef-
ficiency, and these systems evolved to optimize it for their
corresponding ecological situation,
it can be expected
that the properties of at least some parts of the pho-
tosynthetic machinery would be tuned to the fluctuation
properties of their source of excitation. For plants and
some bacteria this may be the Sun light, others like FMO
complexes could be expected to be tuned to the proper-
ties of their associated chlorosoms.
(87)
VI. CONCLUSIONS
Here, A = Pα,n eniξgihξαhg + h.c. is the molecule–
mediator interaction Hamiltonian and the time depen-
dence results from the interaction picture
A(t) = U †
M (t)U †
A(t)AUA(t)UM (t).
(88)
We denoted the ground and excited states of the mediator
by ξgi and ξni, respectively. The state of the molecule
In this paper we have discussed dynamics of a molecu-
lar system subject to external pumping by a light source.
In particular we have considered excitation by thermal
light, by coherent pulsed light and an excitation through
a mesoscopic antenna. With a completely quantum me-
chanical treatment, we have derived a general formula
which enables us to study the effect of different light
properties on the photo-induced dynamics of a molecular
systems. This formula naturally contains the system–
environment interaction contribution to the excitation
process which enters via appearance of the reduced den-
sity matrix dynamics. We show that once the properties
of light are known in terms of a certain two-point correla-
tion function, the only information needed to reconstruct
the systems dynamics is the reduced evolution superop-
erator, which is in principle accessible through ultrafast
non-linear spectroscopy. This conclusion applies to any
type of light and makes thus the results of ultrafast spec-
troscopic experiments universally relevant. Considering
a direct excitation of a small molecular antenna we found
that excitation of coherences is possible due to overlap of
homogeneous line shapes associated with different exci-
tonic states. These coherences are however static and
correspond to a change of the preferred basis set into
which the system relaxes from the one defined by the
bath only, to the one defined by the action of both the
light and the bath. When an excitation of a photosyn-
thetic complex mediated by a larger, possibly mesoscopic,
system is considered, the complex can harvest fluctua-
11
tions originating from the non-equilibrium state of the
mediator. Fluctuations of the mesoscopic system such as
chlorosoms may time localize excitation events of the en-
ergy channeling complex, and to excite adjacent energy
channeling complex coherently. It is likely that in such
a case the properties of energy channeling complexes like
the well-know Fenna-Mathews-Olson complex would be
specially tuned to the fluctuation properties of their as-
sociated chlorosoms.
Acknowledgments
This work was partially supported by the Czech
nr.
Science Foundation (GACR)
206/09/0375, by the Ministry of Education, Youth and
Sports of the Czech Republic through grant KONTAKT
ME899 and the research plan MSM0021620835, and by
the Agency of the International Science and Technology
in Lithuania.
through grant.
[5] A. Olaya-Castro, C. F. Lee, F. F. Olsen, and N. F. John-
[24] R. E. Blankenship, Molecular Mechanisms of Photosyn-
[1] G. S. Engel, T. R. Calhoun, E. L. Read, T. K. Ahn,
T. Mančal, Y. C. Cheng, R. E. Blankenship, and G. R.
Fleming, Nature 446, 782 (2007).
[2] G. D. Scholes, J. Phys. Chem. Lett. 1, 2 (2010).
[3] S. Mukamel and D. Abramavicius, Chem. Rev. 104, 2073
[4] M. Plenio and S. F. Huelga, New J. Phys. 10, 113019
(2004).
(2008).
son, Phys. Rev. B 78, 085115 (2008).
[6] P. Rebentrost, M. Mohseni, I. Kassal, S. Lloyd, and
A. Aspuru-Guzik, New J. Phys. 11, 033003 (2009).
[7] M. Mohseni, P. Rebentrost, S. Lloyd, and A. Aspuru-
Guzik, J. Chem. Phys. 129, 174106 (2008).
[8] A. Ishizaki and G. R. Fleming, J. Chem. Phys. 130,
[9] A. Ishizaki and G. R. Fleming, J. Chem. Phys. 130,
234110 (2009).
234111 (2009).
[10] A. Ishizaki and G. R. Fleming, Proc. Natl. Acad. Sci.
[11] Y.-C. Cheng and G. R. Fleming, Annu. Rev. Phys. Chem.
USA 106, 17255 (2009).
60, 241 (2009).
[12] J. Olšina and T. Mančal, J. Mol. Mod. (2010), submitted.
[13] D. M. Jonas, Annu. Rev. Phys. Chem. 54, 1352 (2003).
[14] M. H. Cho, Chem. Rev. 108, 1331 (2008).
[15] T. Brixner, J. Stenger, H. M. Vaswani, M. Cho, R. E.
Blankenship, and G. R. Fleming, Nature 434, 625 (2005).
[16] M. H. Cho, H. M. Vaswani, T. Brixner, J. Stenger, and
G. R. Fleming, J. Phys. Chem. B 109, 10542 (2005).
[17] T. R. Calhoun, N. S. Ginsberg, G. S. Schlau-Cohen, Y.-
C. Cheng, M. Ballottari, R. Bassi, and G. R. Fleming, J.
Phys. Chem. B 113, 16291 (2009).
[18] G. S. Schlau-Cohen, T. R. Calhoun, N. S. Ginsberg, E. L.
Read, M. Ballottari, R. Bassi, R. van Grondelle, and
G. R. Fleming, J. Phys. Chem. B 113, 15352 (2009).
[19] N. S. Ginsberg, Y.-C. Cheng, and G. R. Fleming, Acc.
Chem. Res. 42, 1352 (2009).
[20] D. Abramavicius, B. Palmieri, D. V. Voronie, F. Šanda,
and S. Mukamel, Chem. Rev. 109, 2350 (2009).
[21] F. Milota, J. Sperling, A. Nemeth, T. Mančal, and H. F.
Kauffmann, Acc. Chem. Res. 42, 1364 (2009).
[22] E. Collini and G. D. Scholes, Science 323, 369 (2009).
[23] H. Lee, Y.-C. Cheng, and G. R. Fleming, Science 316,
1462 (2007).
thesis (Blackwell Science, Oxford, 2002).
[25] B. Grimm, R. J. Porra, W. Rüdiger, and H. Scheer, eds.,
Chlorophyls and Bacteriochlorophylls (Springer, Dor-
drecht, 2006).
[26] H. van Amerongen, L. Valkunas, and R. van Gron-
delle, Photosynthetic Excitons (World Scientific, Singa-
pore, 2000).
[27] V. May and O. Kühn, Charge and Energy Transfer Dy-
namics in Molecular Systems (Wiley-VCH, Weinheim,
2004).
[28] B. Brüggemann, D. Tsivlin, and V. May, in Quantum
Dynamics of Complex Molecular Systems, edited by D. A.
Micha and I. Burghardt (Springer, Berlin, 2007).
[29] E. Barkai, F. Brown, M. Orrit, and H. Yang, eds., Theory
and Evaluatin of Single-Molecule Signals (World Scien-
tific, New Jersey, 2008).
[30] L. Valkunas, J. Janusonis, D. Rutkauskas, and R. van
Grondelle, J. Lumin. 127, 269 (2007).
[31] J. Janusonis, L. Valkunas, D. Rutskauskas, and R. van
Grondelle, Biolphys. J. 94, 1348 (2008).
[32] M. Schlosshauer, Decoherence and the Quantum-to-
classical Transition (Springer, Berlin, 2007).
[33] E. Joos, H. D. Zeh, C. Kiefer, D. Giulini, J. Kupsch, and
I.-O.Stamatescu, Decoherence and the Appearance of a
Classical World in Quantum Theory (Springer, Berlin,
1996).
[34] H. D. Zeh, Found. Phys. 1, 69 (1970).
[35] W. H. Zurek, Phys. Rev. D 26, 1862 (1982).
[36] E. Geva, E. Rosenman, and D. Tannor, J. Chem. Phys.
[37] X.-P. Jiang and P. Brumer, J. Chem. Phys. 94, 5833
113, 1380 (2000).
(2009).
[38] S. Mukamel, Principles of nonlinear spekroscopy (Oxford
University Press, Oxford, 1995).
[39] M. Cho, Two-dimensional Optical Spectroscopy (CRC
Press, Boca Raton, 2009).
[40] T. Brixner, T. Mančal, I. V. Stiopkin, and G. R. Fleming,
J. Chem. Phys. 121, 4221 (2004).
[41] R. Loundon, Quantum Theory of Light (Oxford Univer-
sity Press, Oxford, 2000), 3rd ed.
12
[42] N.-U. Frigaard and D. A. Bryant, in Complex Intracel-
lular Structires in Prokaryotes, edited by J. M. Shively
(Springer, Berlin, 2006).
[43] T. Nozawa, K. Ohtomo, M. Suzuki, H. Nakagawa,
Y. Shikama, H. Konami, and Z.-Y. Wang, Photosynth.
Res. 41, 211 (1994).
[44] J. Pšenčík, T. P. Ikonen, P. Laurinmäki, M. C. Merckel,
S. J. Butcher, R. E. Serimaa, and R. Tuma, Biolphys. J.
87, 1165 (2004).
[45] J. Pšenčík, Y.-Z. Ma, J. B. Arellano, J. Hála, and T. Gill-
bro, Biolphys. J. 84, 1161 (2003).
|
1911.07022 | 1 | 1911 | 2019-11-16T12:53:22 | Biophysical characterization of DNA origami nanostructures reveals inaccessibility to intercalation binding sites | [
"physics.bio-ph"
] | Intercalation of drug molecules into synthetic DNA nanostructures formed through self-assembled origami has been postulated as a valuable future method for targeted drug delivery. This is due to the excellent biocompatibility of synthetic DNA nanostructures, and high potential for flexible programmability including facile drug release into or near to target cells. Such favourable properties may enable high initial loading and efficient release for a predictable number of drug molecules per nanostructure carrier, important for efficient delivery of safe and effective drug doses to minimise non-specific release away from target cells. However, basic questions remain as to how intercalation-mediated loading depends on the DNA carrier structure. Here we use the interaction of dyes YOYO-1 and acridine orange with a tightly-packed 2D DNA origami tile as a simple model system to investigate intercalation-mediated loading. We employed multiple biophysical techniques including single-molecule fluorescence microscopy, atomic force microscopy, gel electrophoresis and controllable damage using low temperature plasma on synthetic DNA origami samples. Our results indicate that not all potential DNA binding sites are accessible for dye intercalation, which has implications for future DNA nanostructures designed for targeted drug delivery. | physics.bio-ph | physics | Biophysical characterization of DNA origami
nanostructures reveals inaccessibility to
intercalation binding sites
Helen L . Miller1,2, Sonia Contera3, Adam J.M. Wollman1,4,5, Adam Hirst6, Katherine
E. Dunn7,8, Sandra Schröter6, Deborah O'Connell6, Mark C. Leake1,4
1 Department of Physics, University of York, Heslington, York, YO10 5DD, United Kingdom.
2 Current address: Department of Physics, Clarendon Laboratory, University of Oxford, Oxford, OX1
3PU, United Kingdom.
3 Department of Physics, Clarendon Laboratory, University of Oxford, Oxford, OX1 3PU, United
Kingdom.
4 Department of Biology, University of York, Heslington, York, YO10 5NG, United Kingdom.
5 Current address: Biosciences Institute, Newcastle University, NE1 7RU, United Kingdom.
6York Plasma Institute, Department of Physics, University of York, Heslington, York, YO10 5DQ,
United Kingdom.
7Department of Electronics, University of York, Heslington, York, YO10 5DD, United Kingdom.
8 Current address: School of Engineering, Institute for Bioengineering, University of Edinburgh
Faraday Building, King's Buildings, Colin Maclaurin Road, Edinburgh, EH9 3DW, United Kingdom.
E-mail: [email protected]
Abstract
Intercalation of drug molecules into synthetic DNA nanostructures formed through self-
assembled origami has been postulated as a valuable future method for targeted drug delivery.
This is due to the excellent biocompatibility of synthetic DNA nanostructures, and high
potential for flexible programmability including facile drug release into or near to target cells.
Such favourable properties may enable high initial loading and efficient release for a
predictable number of drug molecules per nanostructure carrier, important for efficient
delivery of safe and effective drug doses to minimise non-specific release away from target
cells. However, basic questions remain as to how intercalation-mediated loading depends on
the DNA carrier structure. Here we use the interaction of dyes YOYO-1 and acridine orange
with a tightly-packed 2D DNA origami tile as a simple model system to investigate
intercalation-mediated loading. We employed multiple biophysical techniques including
single-molecule fluorescence microscopy, atomic force microscopy, gel electrophoresis and
controllable damage using low temperature plasma on synthetic DNA origami samples. Our
results indicate that not all potential DNA binding sites are accessible for dye intercalation,
which has implications for future DNA nanostructures designed for targeted drug delivery.
Keywords: DNA origami, DNA damage, YOYO-1, Acridine orange, Low temperature plasma, Single-molecule microscopy
Introduction
Over the past decade important progress has been
made in the practical use of DNA nanostructures
for targeted therapeutic drug delivery [1 -- 5], in
particular for intercalating molecules [6 -- 9],
including the anti-cancer drug doxorubicin (DOX).
To deliver high doses of a drug it is desirable to
have a high initial loading into the DNA carrier
nanostructure, high uptake to the desired target
area, and a high release rate in the vicinity of a
target cell.
DNA nanostructures loaded with the leukaemia
drug Daunorubicin [10] have been shown to
reduce the viability of target cells more than the
same amount of free compound, but questions
about the optimum design of the DNA
nanostructure carrier for drug loading and cellular
uptake remain. Many previous studies have used
DNA origami with cage-like structures, but the use
of compact structures with intercalated drug
molecules could enable higher drug loading,
leading to more efficient delivery of the active
agent to the target cells.
Well-defined DNA structures instead of linear
configurations might be more robust to damage
mechanisms encountered in the body. For instance,
a recent study showed that tetrahedral DNA
nanostructures did not cause significant
impairment to native physiology [11], triggering
interest in DNA structures with an internal column
for drug transportation. Another recent work [12]
has shown that large (~MDa) DNA nanoparticles
with a high external surface area-to-volume ratio
are preferentially taken up in mammalian cell lines
relevant to therapeutic drug delivery. Zeng et al.
[9] found that rigid 3D DNA origami shapes are
more readily taken up by cells, and exhibit
sustained drug release, compared to more flexible
2D DNA structures.
Here, we investigate the interactions of
intercalating molecules with a dense DNA origami
structure using the well-studied YOYO-1 and
acridine orange dyes. Previous work with the
predominantly bis-intercalating YOYO-1 [13] has
focussed on interactions with linear DNA (see for
example [14 -- 17]), whilst acridine orange is often
used to determine whether DNA is single or
double-stranded [18]. Acridine orange binds to
double stranded DNA (dsDNA) via an
intercalative mode [19,20] and to single-stranded
DNA (ssDNA) via electrostatic binding to
phosphates [21], producing green or red
fluorescence respectively (DNA binding modes are
schematically illustrated in figure 1a).
The DNA origami tile used in this work is adapted
from the work of Rothemund [22]; the original
design uses the viral ssDNA of m13mp18ss as a
template and 216 staple strands of 32 nt in length
to fold it into a rectangle approximately
70 x 100 nm in lateral dimensions with a central
seam, as can be seen in the AFM images in figure
1d, e. For this work the DNA origami tile has been
modified to include four strands with a 5' 4-T
linker and biotin for surface attachment which all
protrude on the same side (figure 1b, and
supplementary information) and the addition of 4-
T loops to 24 of the staples to inhibit edge
stacking, as has previously been described for
other DNA origami shapes [22]). Three of the
original staples were replaced with 4 alternative
staples, one of which is extended (supplementary
information), for the purposes of an unrelated
experiment that will not be discussed here.
We have used gel electrophoresis, fluorescence
microscopy and atomic force microscopy (AFM)
to investigate the loading of intercalating
molecules into DNA origami with a compact and
highly linked structure. To controllably damage
the DNA tiles we used low temperature plasma
(LTP, figure 1c), to investigate how damage to the
DNA structure changes the available intercalation
sites. LTP, a partially ionized gas and electrically
neutral state of matter, is known to induce DNA
damage via single- and double-strand breaks [23 --
25] in a treatment time-dependent [23,26] manner.
LTPs have thus been developed for various
therapeutics [24,25]. The LTP components
interacting with the DNA origami tiles in this case
are reactive neutrals and photons (no charged
particles or electric fields).
Our findings indicate that the physical properties
of DNA origami carriers must be considered to
ensure optimal drug loading and have implications
2
for the design of DNA origami systems for
delivery of intercalated drugs and dose control.
Materials and Methods
1.1 Reagents
Phosphate buffer (PB), pH7: 39 parts 0.2M
monobasic sodium phosphate monohydrate and 61
parts 0.2M anhydrous dibasic sodium phosphate.
Purification buffer: 10mM Tris-HCl, 1mM EDTA,
50mM NaCl, 10mM MgCl2. Synthesis buffer:
1xTAE buffer (40 mM Tris-acetate and 1 mM
EDTA, pH 8.3), 12.5mM magnesium acetate.
1.2 DNA origami
Single-stranded m13mp18ss scaffold DNA (New
England BioLabs) was mixed with approximately
100 x excess of the non-biotinylated staple strands
(Integrated DNA Technologies, Inc.) in synthesis
buffer and annealed from 95ºC to room
temperature, 20C, at a cooling rate of 1ºC per
minute. After assembly, biotinylated strands were
added at approximately 100 times the scaffold
concentration and the mixture was hybridised
overnight in synthesis buffer at room temperature.
The individual staples and their complements used
in gel electrophoresis experiments were annealed
by heating to 95C and cooling to room
temperature at a rate of 1C per minute to form a
10 µM solution in synthesis buffer.
1.3 Purifying Origami
Excess staples were removed from the DNA
origami using high resolution Sephacryl S-300
media (GE Healthcare) packed into filtration
columns [27,28]; Full details of the preparation of
the filtration media and packing of the spin
columns are given in the supplementary methods.
1.4 DNA origami surface immobilization for
microscopy
Coverslips (thickness 0.13-0.17mm, MNJ-350-
020H, Menzel Gläser) were functionalised as
follows for fluorescence microscopy of
immobilised DNA origami. Coverslips were
plasma cleaned (Harrick PDC-32G) for 5 minutes
then soaked in BSA-biotin (0.5 mg/ml in PB) for 1
hour by pipetting 150-200 µl of liquid onto each
coverslip. Each coverslip was rinsed twice with PB
and air dried (after this stage in the preparation
protocol dried coverslips can be stored at room
temperature in Petri dishes sealed with Parafilm M
Figure 1: Experimental methods. (a)
Schematic diagram of DNA binding
modes. (b) Schematic diagram
showing the method of DNA
origami immobilization. (c) Image
of the low temperature plasma
source used to induce DNA damage
(d,e) AM-AFM topography images
of undamaged DNA origami tiles in
liquid.
3
(Bemis Company, Inc.) for later use as
appropriate). At all times the
cleaned/functionalised side of the coverslip was
maintained upwards such that the non-
functionalised side was in contact with the Petri
dish or equivalent surface. Tunnel slides for
microscopy were created by laying two parallel
strips of Scotch double sided tape (product
number: 70071395118, 3M) spaced 2-4 mm apart
on a slide and placing a functionalized coverslip on
top. The coverslip was tapped down with a pipette
tip avoiding the area to be imaged and the excess
tape removed. The tunnel was hydrated with 20 µl
PB then incubated with 10 µl of 1 mg/ml avidin
(pH7) for 1 hour. Excess avidin was removed by
washing with 20 µl purification buffer (using a
kimwipe and capillary action, without letting air
bubbles through) before the addition of 7.5 µl of
purified biotinylated origami in purification buffer
(concentration approximately 0.8 nM), incubated
for 5 minutes and washed with 20 µl of
purification buffer. Fluorescent dye labelling was
performed by flowing 10 µl of YOYO-1 diluted
1:199 in purification buffer and incubating for
5 minutes. Excess dye was removed by washing
with 100 µl purification buffer. All incubation
steps were performed at room temperature in a
humidity chamber.
1.5 Low temperature plasma treatment of DNA
origami
Samples of approximately 0.8 nM purified DNA
origami
(estimated based on pre-purification
concentration) were treated with LTP for 1 minute.
Samples of 20 µl were treated in the lids of 500 ml
Eppendorf tubes, without the tube attached, and
were then placed centrally [26] and vertically
10 mm underneath
low
temperature plasma source was a prototype of the
COST Reference Microplasma Jet [29] (1 mm
electrode gap and width, 30 mm length) and
operated using an applied 13.56 MHz peak-to-peak
voltage of 450-500 V, with a feed gas of 1 slm
the electrodes. The
(standard litre per minute) helium and a 0.5%
admixture of oxygen.
1.6 Fluorescence microscopy
Fluorescence microscopy was performed on a
bespoke optical imaging system, built around a
commercial microscope body (Nikon Eclipse Ti-S)
as previously described [30]. A supercontinuum
laser (Fianium, SC-400-6, Fianium Ltd.) coupled
to an acousto-optic tunable filter (AOTF) set to
45% at a central wavelength of 491 nm was used
for illumination (see supplementary figure 1 and
supplementary note), giving a power density at the
sample of 775 Wcm-2. A 475/50 nm wavelength
bandpass filter was used to provide blue light
excitation. The beam was de-expanded to generate
narrowfield epifluorescence illumination [31] with
FWHM of laser excitation at the sample of
approximately 12 µm. A 515 nm wavelength
dichroic mirror and 535/15 nm wavelength
emission filter were used in the filter set. A 100x
imTIRF objective
objective
NA 1.49) was used with a further 2
times
magnification to give a total magnification of
120 nm pixel-1 at the emCCD camera detector used
(Andor
iXonEM+ DU 860 camera, Andor
Technology Ltd).
Images were acquired at 10 ms exposure times
(corresponding to a frame time of 10.06 ms), at
pre-amplifier and EM gains of 4.6 and 300
respectively.
lens (Nikon oil
1.7 Gel electrophoresis
50ml 1% agarose (analytical grade, Promega) gels
pre-stained with SYBR safe (Invitrogen) or post-
stained with acridine orange were run horizontally
at 100V (6.7Vcm−1) in TAE buffer, for 30 minutes
and imaged via an automated gel imaging system
(ChemiDoc MP Imaging System, Bio-Rad
Laboratories). DNA samples were prepared in 6 µl
volumes with 0.6 µl of 10x running buffer, 0.75 µl
4
filtered 80% glycerol (0.45 µm diameter pore
syringe filter), and 1µl of 6X purple loading dye
(for pre-stained gels) or 1 µl of ultrapure water (for
post-stained gels). The rest of the sample volume
comprised 50-100 ng DNA or DNA diluted in
ultrapure water.
For SYBR safe pre-stained gels, 5 µl of 10,000x
concentrated SYBR Safe was mixed via gentle
swirling into the agarose before casting. For
acridine orange post-staining gels were run in TAE
buffer without stain. The post-staining protocol
was adapted from McMaster and Carmichael [18];
gels were incubated at room temperature with
rocking frequency 0.25 Hz for 30 minutes in 50 ml
of 30 µg/ml acridine orange in TAE, to completely
submerge the gel. The buffer was exchanged for 1
hour destain in 50 ml TAE under rocking
incubation at 0.25 Hz.
Images were analysed to determine ratios of red to
green fluorescence (RG ratio) and standard
deviation using custom written Matlab routines
(available at http://single-molecule-
biophysics.org/ ). Each band to be analysed was
made into a sub-image in both the red and green
channels. Both images were morphologically
opened using a disk of radius 2 pixels, thresholded
using Otsu's method [32], morphologically opened
with a disk of radius 1 pixel to remove small
artefactual segmentation holes and then dilated to
produce masks of the band areas. These masks
were correlated to determine the overlapping area,
which was used to segment each band (see
supplementary figure 2). The band standard
deviation was calculated by multiplying the
number of pixels in the band by the per pixel
standard deviation. In some images dust in the
original image had to be removed before
quantification: a 15 pixel horizontal line profile
was taken through pixels seen to contain dust and
fitted using a 5th order polynomial, with zero
weight for the dust affected pixels. The residuals
for the dust affected pixels were removed from
their intensity values before calculation of the red
to green fluorescence ratio. All pre- and post-
correction images can be seen in supplementary
figure 3.
1.8 Particle Tracking and determination of single
dye Intensity
A bespoke single-molecule precise tracking and
quantification software called ADEMS code [16]
was used to evaluate the fluorescence intensity of
the tiles over time. As DNA origami tiles are
immobilised by biotin conjugation the detected
spots were linked into fluorescent trajectories in
the code based on their location in the kinetic
series of image data. Images were corrected for
non-uniform illumination using a profile of the
beam created by raster-scanning 200 nm yellow-
green fluorescent beads (F8811, ThermoFisher
Scientific) through the focal volume. Objects
within a 40 pixel radius of the beam centre were
used to ensure a flat field of illumination,
minimising the effect of noise. Objects containing
more than 20 localizations in the first 100 image
frame intensities were fitted using an exponential
function of the form:
𝐼(𝑡) = 𝐼0𝑒−𝐵𝑡 + 𝐶
Where C is an offset to account for the presence of
photoblinking behaviour of YOYO-1 at long times
[16], B is the characteristic photobleaching time
and I0 is the height of the photobleaching
exponential above the photoblinking behaviour.
The initial intensity was taken as C+I0; any
saturated image frames were excluded from the
analysis with the exponential function fitted to the
remaining points.
The single-molecule YOYO-1 fluorescence
intensity was determined using intensity vs. time
traces for DNA origami tiles sparsely labelled with
YOYO-1. Multiple analysis techniques were used
to determine a consensus single-molecule intensity
(see supplementary methods and supplementary
figure 4): peak positions in the histogram of
fluorescence intensity of origami over time; peaks
5
in the pairwise intensity of the same data, fast
Fourier transforms of the pairwise intensity, and
Chung-Kennedy filtering [33,34] of individual
traces with two window sizes to allow for the
photobleaching and photoblinking effects of
YOYO-1 were all used to determine the single
YOYO-1 intensity of approximately 370 detector
counts (range 356-380).
1.9 Calculation of stoichiometries
Kernel density estimates [35] of the initial
intensity distribution of single tiles divided by the
characteristic intensity of a YOYO-1 molecule
were used to determine mean stoichiometries. The
locations of peaks were determined objectively
using the Matlab findpeaks.m function. The full
width half maximum (FWHM) of the first peak in
the kernel density estimate was fitted using a
Gaussian distribution to generate the mean
stoichiometry (i.e. number of detected dye
molecules present within each detected fluorescent
spot), with the sigma width used as an estimate for
the error on the mean stoichiometry. In the case
that a second peak was identified by the
findpeaks.m function within the FWHM of the first
peak the data was truncated to include only the
first peak.
1.10 Atomic force microscopy
AFM imaging was performed in amplitude
modulation (AM) mode in fluid with a MFP-3D
(Oxford Instruments Asylum Research Co., Ltd,
UK) using silicon nitride RC800 cantilever tips
(Olympus) with nominal spring constant 0.4 Nm−1.
5 µl of DNA origami tiles were incubated at room
temperature in purification buffer on freshly
cleaved mica for 15 minutes. Images were taken in
purification buffer at room temperature at a scan
speed of 1 Hz.
Before performing AFM of LTP damaged DNA
origami tiles we verified that no further damage to
DNA occurs after LTP treatment is removed
(supplementary methods, supplementary figures
5,6).
Results
Fluorescence measurements of acridine orange
bound to DNA origami indicate a non-intercalating
binding mode
Acridine orange fluoresces red when bound to
single stranded DNA via electrostatic binding to
phosphates [21], and green when bound to double
stranded DNA via an intercalative mode [19,20].
To investigate the accessibility of the DNA
origami structure various single- and double-
stranded DNA constructs and the DNA origami
tiles were subjected to agarose gel electrophoresis
and post-stained with acridine orange (figure 2 and
supplementary figure 7).
M13mp18ss (the viral DNA plasmid used as the
DNA origami tile backbone) is single stranded and
appeared red in the merged fluorescence image
(figure 2a), producing a red to green fluorescence
intensity ratio (RG ratio, see methods section 1.7)
of 21.54 ± 4.53 (± standard deviation; figure 2b)
when imaged in red and green fluorescence using a
gel imager (ChemiDoc MP Imaging System, Bio-
Rad Laboratories). The double stranded pUC19
plasmid appeared green in the fluorescence images
and gave an RG ratio of 0.33 ± 0.02. The
undamaged DNA origami tile monomer showed
intermediate behaviour, appearing neither red nor
green in the merged fluorescence images and
producing an RG ratio of 5.10 ± 0.83. A table of
RG ratios for different DNA constructs is given in
table 1 and shown in figure 2b.
DNA sample
Double-
(ds) or
single- (ss)
stranded
ss
ds
Red to green
fluorescence
intensity ratio
(± SD)
21.54 ± 4.53
0.32 ± 0.01
m13mp18ss
3kbp band; 2-log
ladder
pUC19
ds
0.33 ± 0.02
Origami Monomer ds
5.10 ± 0.83
Origami Dimer
ds
8.93 ± 2.02
Table 1: Red to green fluorescence intensity ratios of DNA
constructs run in agarose gel electrophoresis and post-stained
with acridine orange, as shown in figure 2d, with standard
errors (SE).
6
As further controls, two single stranded DNA
staples from Rothemund's original DNA origami
tile (r3t10f and r3t22f; see table 2) were also run in
the gel, along with their complements and
duplexes made by annealing the staples and
complements (figure 2c). According to predictions
from the NUPACK structure prediction software
[34], staple r3t10f has a low probability of forming
secondary structure whilst r3t22f has a high
probability of forming secondary structure.
Therefore r3t10f was expected to be mostly single-
stranded, while r3t22f was expected to be mostly
double-stranded.
Staple
name
r3t10f
r3t10f
comple
ment
r3t22f
r3t22f
comple
ment
Sequence
5'−ATTATTTAACCCAGCTACAAT
TTTCAAGAACG−3'
5'−CGTTCTTGAAAATTGTAGCTG
GGTTAAATAAT−3'
5'−AGGCGGTCATTAGTCTTTAAT
GCGCAATATTA−3'
5'−TAATATTGCGCATTAAAGACT
AATGACCGCCT−3
Table 2: Oligonucleotides used in gel electrophoresis.
Staple r3t10f is expected to have single stranded
behaviour and fluoresces red, as does its
complement. The high mobility of these staples
allows them to diffuse beyond their wells, and
r3t10f and its complement are seen to hybridise in
the overlapping region, indicated by the green
fluorescent region between these two bands (figure
2c, white arrow). The annealed duplex of these
bands fluoresces green indicating double stranded
behaviour. As the non-annealed duplex occurs at
the same position on the gel as the single strands it
is likely that this was formed due to diffusion of
the low molecular weight single stranded staples
during staining, after the removal of the
electrophoretic voltage.
Figure 2: Acridine orange interaction with DNA origami
imaged via gel electrophoresis. (a) Green, red and merged
fluorescence intensity images of DNA samples in
electrophoretic gel. Purified DNA origami bands are faint.
(b) Red to green fluorescence (RG) ratios with standard
deviations of bands from the gel shown in part (a): single
stranded species are shown in red, double stranded in blue
and the intermediate origami results in black. (c) Green, red
and merged fluorescence intensity images of DNA staple
samples in electrophoretic gel. White arrow indicates
hybridisation of r3t10f and r3t10f complement without
annealing. All gel images are from the same gel shown in full
in supplementary figure 7. All images in a colour channel are
shown at the same contrast levels.
7
Staple r3t22f, expected to have high self-
complementarity and to form secondary structure,
fluoresces green, indicating that it is indeed double
stranded, whilst its complement is single stranded
and its duplex is double stranded as predicted. For
this staple no hybridisation with its complement is
seen in the region between the bands, consistent
with the expectation that annealing is required to
remove secondary structure and allow the two
staple strands to hybridise.
The RG ratios for the individual staples run as
controls and related complexes were not evaluated
as their high mobility in the gel produced diffuse
bands that extended beyond the original lanes and
were of lower contrast than the higher molecular
weight bands.
All control samples with known single- or double-
stranded behaviour showed an RG ratio consistent
with the predicted binding mode for this structure,
whilst the DNA origami tile monomer and dimer
showed an RG ratio intermediate between the
intercalative and phosphate-binding modes.
The DNA origami tile is seen to be well-formed
into a double helical structure in AFM images (see
figure 1d, e). However, the intercalating DNA
binding mode of acridine orange is not observed to
be dominant upon post staining of an agarose gel.
The RG ratio is between the values observed for
ssDNA or dsDNA, which may indicate that both
modes of binding are present. Other possible
explanations are concentration-dependent binding
effects of acridine orange [36] but this is unlikely
for the concentration ranges used in this work and
would be inconsistent with the observed results for
controls of known strandedness. There is a region
of the M13mp18ss scaffold which is left unstapled
in the DNA origami synthesis, but this region is
likely to contain large amounts of secondary
structure [22] and be double stranded, so
phosphate binding of acridine orange to this region
is unlikely to explain this result. The t-loops added
to prevent end-stacking result in the addition of
116 single bases per DNA origami tile, a small
number compared to the expected 6480 base pairs.
The t-loops may be responsible for the observed
single-stranded binding mode, but cannot explain
the lack of fluorescence from the double-stranded
binding mode.
Gel electrophoresis indicates that acridine orange
mostly binds predominantly in its single-stranded
DNA binding mode to the double-stranded DNA
origami tile despite the DNA origami tile being
well formed; one possible explanation is that steric
effects of DNA packing may reduce the
accessibility of the DNA intercalation sites.
Single-molecule optical microscopy of the
intercalating dye YOYO-1 labelled DNA indicates
a saturating stoichiometry consistent with steric
exclusion of intercalation
To test the accessibility of the DNA origami tile
helices to intercalation we used the well-studied
bis-intercalator YOYO-1. This dye has a reported
maximum intercalative labelling of one YOYO-1
molecule to every four base pairs [13,37],
corresponding to one moiety per two base pairs. A
higher loading of one YOYO-1 every three base
pairs [38] has been reported, consistent with recent
studies using ethidium bromide, which challenge
the idea that intercalating dyes completely exclude
intercalation into neighbouring sites [39].
For the DNA origami tile used in this work, the
number of intercalating YOYO-1 molecules per
tile can be estimated using fluorescence
microscopy in the saturating dye regime; the initial
fluorescence intensity of a DNA origami tile can
be divided by the total summed pixel fluorescence
intensity (i.e. the observed brightness) of a single
YOYO-1 molecule (see methods and
supplementary methods) to estimate the number of
dye molecules loaded.
The number of dye molecules required to reach the
saturating regime of YOYO-1 intercalation was
investigated using lambda DNA (supplementary
figure 8, supplementary methods) and was found
to saturate between 0.13 and 1.3 YOYO-1
molecules per base pair. Using these ratios as a
guide, but considering the evidence for reduced
intercalation into the DNA origami tile, a range of
8
concentrations, 0.05-5µM, was tested for
fluorescence intensity saturation on the undamaged
DNA origami tile (figure 3, supplementary
figure 9, table 3).
The saturating YOYO-1 concentration was found
to be approximately 5µM; at this concentration the
mean occupancy of intercalated YOYO-1 (±
standard deviation) was found to be 67±25
molecules per DNA origami tile. The DNA
origami tile contains approximately 6480 base
pairs so we would expect a saturating
concentration of roughly 1620 molecules,
assuming maximum loading. Alternatively, if only
the outward facing half of the outermost helices
were accessible for dye binding we would expect a
loading of ~68 molecules, given the reported
maximum intercalative labelling of one YOYO-1
molecules per 4 base pairs.
The structure of this DNA origami tile is known to
be tightly packed, with measured helix separations
of 0.9-1.2 nm [22], whilst the size of a YOYO-1
analogue, TOTO, identical except for a sulphur
atom in place of an oxygen, is larger than this
separation, at around 2 nm [40], and YOYO-1 is
known to require more room than TOTO in
intercalation sites [41]. Whilst intercalation of
YOYO-1 to DNA is known to perturb the structure
of single DNA helices with each bis-intercalation
unwinding the helix by 106º [41] and causes an
extension of 0.68 ± 0.04 nm [42], it is not clear
whether this effect or the precise positioning of
DNA helix crossovers in this DNA origami tile
would have more influence on the DNA origami
structure: non- B form structure imposed in a DNA
origami tile is known to change the binding
properties of minor groove binding dyes [43].
The low intercalation of YOYO-1 into the DNA
origami tile combined with the non-intercalative
binding of acridine orange to the DNA origami tile
suggests that the inner helices of this DNA origami
tile are inaccessible to intercalation.
Figure 3: Single-molecule microscopy of the intercalating
dye YOYO-1 with DNA origami. (a) Fluorescence
microscopy images of DNA origami labelled with different
concentrations of YOYO-1, before correction for
illumination profile, with contrast levels used for display.
These images are shown at the same contrast levels in
supplementary figure 9. Scale bar 1 µm. (b) Kernel density
estimate of the stoichiometry of YOYO-1 labelling on DNA
origami tiles. (c) Peak stoichiometry (± standard deviation
from Gaussian fit) as a function of YOYO-1 concentration
from the data in (b).
YOYO-1
concentration
(µM)
Number of YOYO-1
molecules intercalated
(±SD of Gaussian)
Number
of DNA
origami
tiles
169
67
326
452
10±8
17±7
61±28
67±25
0.05
0.5
1
5
Table 3: YOYO-1 binding to DNA origami in different
concentrations, rounded to whole numbers. Number of
YOYO-1 molecules and standard deviation (SD).
Inducing low temperature plasma damage to the
DNA origami increases the number of YOYO-1
molecules per DNA origami tile observed by
fluorescence microscopy, consistent with exposing
more of the DNA structure.
If the inner DNA helices of this DNA origami tile
are inaccessible to YOYO-1 intercalation due to
steric hindrance, then reducing the structure to
increase helix separation should increase dye
intercalation.
9
To test this hypothesis low temperature plasma
was used to induce single- and double-strand
breaks in the DNA. An increase in the
stoichiometry measured by single-molecule
counting of YOYO-1 labelled DNA when DNA is
damaged in this way would indicate that more
structure becomes accessible after damage. If the
entire structure is already accessible, no increase in
stoichiometry would be seen.
Previous work with LTP has shown that small
levels of DNA damage are typically inflicted over
timescales of a few seconds to a few minutes
[23,25]. The 60 s duration of treatment was
determined to be a suitable LTP dose to inflict
damage on the DNA origami tiles whilst
maintaining some structure (supplementary
methods, supplementary figures 10,11).
DNA origami tiles which had been treated with
low temperature plasma for 60 s and labelled with
YOYO-1 showed a higher intensity than untreated
DNA origami tiles in a fluorescence microscopy
assay (see figure 4a, b, table 4). This observation
indicates that more of the DNA origami structure
is likely to be accessible to intercalation of
YOYO-1.
AFM images of the 60 s LTP treated DNA origami
tiles (figure 4c), reveal that many of the damaged
DNA origami tiles are fractured or incomplete.
Few holes in the DNA origami are seen, implying
that single staples are not removed during damage,
as omission of single staples has previously been
shown to produce small holes in the DNA tiles
[22].
Number of
DNA origami
tiles
Number of
YOYO-1
molecules
intercalated
67±25
DNA Origami tile
state
452
224
Undamaged
60 s LTP treatment 108±12
Table 4: Number of YOYO-1 molecules bound to DNA
origami tiles at a 5 µM YOYO-1 concentration in different
LTP treatment states, rounded to whole numbers and
standard deviation.
Together, these fluorescence microscopy and AFM
results indicate that LTP does break the DNA
origami tiles and creates more YOYO-1 accessible
base pairs within a single tile, supporting the idea
that intercalation is strongly affected by steric
interactions.
Figure 4: LTP damages the DNA origami tiles allowing
greater YOYO-1 binding. (a) Representative fluorescence
images of YOYO-1 binding to undamaged and 60s LTP
treated DNA origami tiles. Scale bar 1 µm (b) Kernel density
estimate of fluorescence intensity of YOYO-1 emission of
DNA origami tiles before and after LTP treatment. (c) AFM
images of 60s LTP treated DNA origami tiles.
Conclusions
Taken together, our results show that the inner
helices of the DNA origami tile used in this work
are likely to be inaccesible to intercalating binding,
but that higher loading can be achieved when the
tiles are damaged to expose more DNA helices; we
suggest that intercalation is prevented in the
undamaged tile by either steric or electrostatic
effects due to the proximity of the helices, but this
may depend on intercalator size, which was not
investigated here. DNA origami nanostructures are
promising candidates for use as drug-delivery
vehicles for mediating targeted delivery due to
10
their high biocompatibility; factors affecting drug
loading, such as those outlined here, have
important ramifications for dose control. Recent
work has shown that large DNA origami tiles with
a high external surface area-to-volume ratio are the
most easily taken up into cells, but our results
suggest that these DNA origami structures may be
sub-optimal for drug delivery due to unavailability
of drug binding sites, and open, mesh-like
structures [44,45] might be better suited for
intercalating drug delivery due to a higher
proportion of the bases being intercalator
accessible.
Acknowledgements
We are grateful for discussions with Dr Steve
Johnson (Department of Electronics, University of
York) concerning surface chemistry
immobilisation methods, and for use of resources
in the Bio-Inspired Technology Laboratory. MCL
was supported by the BBSRC (grants
BB/N006453/1, BB/P000746/1 and
BB/R001235/1) and the EPSRC (grants
EP/T002166/1). KD's work on this paper was part-
funded by the EPSRC Platform Grant
EP/K040820/1.
References
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Linko V, Ora A, Kostiainen MA. DNA Nanostructures as
Smart Drug-Delivery Vehicles and Molecular Devices.
Trends Biotechnol 2015;33:586 -- 94.
doi:10.1016/j.tibtech.2015.08.001.
Ruan W, Zheng M, An Y, Liu Y, Lovejoy DB, Hao M, et
al. DNA nanoclew templated spherical nucleic acids for
siRNA delivery. Chem Commun 2018;54:3609 -- 12.
doi:10.1039/C7CC09257A.
Hu Y, Chen Z, Zhang H, Li M, Hou Z, Luo X, et al.
Development of DNA tetrahedron-based drug delivery
system. Drug Deliv 2017;24:1295 -- 301.
doi:10.1080/10717544.2017.1373166.
Crawford R, Erben CM, Periz J, Hall LM, Brown T,
Turberfield AJ, et al. Non-covalent Single Transcription
Factor Encapsulation Inside a DNA Cage. Angew Chemie
Int Ed 2013;52:2284 -- 8. doi:10.1002/anie.201207914.
Jiang Q, Liu S, Liu J, Wang Z-G, Ding B. Rationally
Designed DNA-Origami Nanomaterials for Drug Delivery
In Vivo. Adv Mater 2018:1804785.
doi:10.1002/adma.201804785.
Zhang Q, Jiang Q, Li N, Dai L, Liu Q, Song L, et al. DNA
Origami as an In Vivo Drug Delivery Vehicle for Cancer
Therapy. ACS Nano 2014;8:6633 -- 43.
[1]
[2]
[3]
[4]
[5]
[6]
doi:10.1021/nn502058j.
Zhuang X, Ma X, Xue X, Jiang Q, Song L, Dai L, et al. A
Photosensitizer-Loaded DNA Origami Nanosystem for
Photodynamic Therapy. ACS Nano 2016;10:3486 -- 95.
doi:10.1021/acsnano.5b07671.
Zhao Y-X, Shaw A, Zeng X, Benson E, Nyström AM,
Högberg B. DNA Origami Delivery System for Cancer
Therapy with Tunable Release Properties. ACS Nano
2012;6:8684 -- 91. doi:10.1021/nn3022662.
Zeng Y, Liu J, Yang S, Liu W, Xu L, Wang R. Time-lapse
live cell imaging to monitor doxorubicin release from
DNA origami nanostructures. J Mater Chem B
2018;6:1605 -- 12. doi:10.1039/C7TB03223D.
Halley PD, Lucas CR, McWilliams EM, Webber MJ,
Patton RA, Kural C, et al. Daunorubicin-Loaded DNA
Origami Nanostructures Circumvent Drug-Resistance
Mechanisms in a Leukemia Model. Small 2016;12:308 -- 20.
doi:10.1002/smll.201502118.
Xia K, Kong H, Cui Y, Ren N, Li Q, Ma J, et al.
Systematic Study in Mammalian Cells Showing No
Adverse Response to Tetrahedral DNA Nanostructure.
ACS Appl Mater Interfaces 2018;10:15442 -- 8.
doi:10.1021/acsami.8b02626.
Bastings MMC, Anastassacos FM, Ponnuswamy N, Leifer
FG, Cuneo G, Lin C, et al. Modulation of the Cellular
Uptake of DNA Origami through Control over Mass and
Shape. Nano Lett 2018;18:3557 -- 64.
doi:10.1021/acs.nanolett.8b00660.
Larsson A, Carlsson C, Jonsson M, Albinsson B.
Characterization of the Binding of the Fluorescent Dyes
YO and YOYO to DNA by Polarized Light Spectroscopy.
J Am Chem Soc 1994;116:8459 -- 65.
doi:10.1021/ja00098a004.
Persson F, Bingen P, Staudt T, Engelhardt J, Tegenfeldt
JO, Hell SW. Fluorescence nanoscopy of single DNA
molecules by using stimulated emission depletion (STED).
Angew Chem Int Ed Engl 2011;50:5581 -- 3.
doi:10.1002/anie.201100371.
Flors C, Ravarani CNJ, Dryden DTF. Super-resolution
imaging of DNA labelled with intercalating dyes.
Chemphyschem 2009;10:2201 -- 4.
doi:10.1002/cphc.200900384.
[16] Miller H, Zhou Z, Wollman AJM, Leake MC.
Superresolution imaging of single DNA molecules using
stochastic photoblinking of minor groove and intercalating
dyes. Methods 2015;88:81 -- 8.
doi:10.1016/j.ymeth.2015.01.010.
[17] Miller H, Wollman AJM, Leake MC. Designing a Single-
Molecule Biophysics Tool for Characterising DNA
Damage for Techniques that Kill Infectious Pathogens
Through DNA Damage Effects. Adv Exp Med Biol
2016;915:115 -- 27. doi:10.1007/978-3-319-32189-9_9.
[18] McMaster GK, Carmichael GG. Analysis of single- and
[19]
[20]
[21]
double-stranded nucleic acids on polyacrylamide and
agarose gels by using glyoxal and acridine orange. Proc
Natl Acad Sci U S A 1977;74:4835 -- 8.
Lerman LS. Structural considerations in the interaction of
DNA and acridines. J Mol Biol 1961;3:18 -- IN14.
doi:10.1016/S0022-2836(61)80004-1.
Lerman LS. The structure of the DNA-acridine complex.
Proc Natl Acad Sci U S A 1963;49:94 -- 102.
Blake A, Peacocke AR. The interaction of aminoacridines
with nucleic acids. Biopolymers 1968;6:1225 -- 53.
doi:10.1002/bip.1968.360060902.
11
[22]
[23]
[24]
[25]
[26]
Rothemund PWK. Folding DNA to create nanoscale
shapes and patterns. Nature 2006;440:297 -- 302.
doi:10.1038/nature04586.
O'Connell D, Cox LJ, Hyland WB, McMahon SJ, Reuter
S, Graham WG, et al. Cold atmospheric pressure plasma
jet interactions with plasmid DNA. Appl Phys Lett
2011;98:43701. doi:10.1063/1.3521502.
Packer JR, Hirst AM, Droop AP, Adamson R, Simms MS,
Mann VM, et al. Notch signalling is a potential resistance
mechanism of progenitor cells within patient‐derived
prostate cultures following ROS‐inducing treatments.
FEBS Lett 2019:1873 -- 3468.13589. doi:10.1002/1873-
3468.13589.
Hirst AM, Simms MS, Mann VM, Maitland NJ, O'Connell
D, Frame FM. Low-temperature plasma treatment induces
DNA damage leading to necrotic cell death in primary
prostate epithelial cells. Br J Cancer 2015;112:1536 -- 45.
doi:10.1038/bjc.2015.113.
Privat-Maldonado A, O'Connell D, Welch E, Vann R, van
der Woude MW. Spatial Dependence of DNA Damage in
Bacteria due to Low-Temperature Plasma Application as
Assessed at the Single Cell Level. Sci Rep 2016;6:35646.
doi:10.1038/srep35646.
[27] Wickham SFJ, Endo M, Katsuda Y, Hidaka K, Bath J,
[28]
[29]
Sugiyama H, et al. Direct observation of stepwise
movement of a synthetic molecular transporter. Nat
Nanotechnol 2011;6:166 -- 9. doi:10.1038/nnano.2010.284.
Dunn KE, Dannenberg F, Ouldridge TE, Kwiatkowska M,
Turberfield AJ, Bath J. Guiding the folding pathway of
DNA origami. Nature 2015. doi:10.1038/nature14860.
Golda J, Held J, Redeker B, Konkowski M, Beijer P,
Sobota A, et al. Concepts and characteristics of the "COST
Reference Microplasma Jet." J Phys D Appl Phys
2016;49:84003. doi:10.1088/0022-3727/49/8/084003.
[30] Miller H, Cosgrove J, Wollman AJM, Taylor E, Zhou Z,
O'Toole PJ, et al. High-Speed Single-Molecule Tracking
of CXCL13 in the B-Follicle. Front Immunol 2018;9:1073.
doi:10.3389/FIMMU.2018.01073.
[31] Wollman AJM, Leake MC. Single-Molecule Narrow-Field
[32]
[33]
[34]
Microscopy of Protein -- DNA Binding Dynamics in
Glucose Signal Transduction of Live Yeast Cells, 2016, p.
5 -- 15. doi:10.1007/978-1-4939-3631-1_2.
Otsu N. A Threshold Selection Method from Gray-Level
Histograms. IEEE Trans Syst Man Cybern 1979;9:62 -- 6.
Chung SH, Kennedy RA. Forward-backward non-linear
filtering technique for extracting small biological signals
from noise. J Neurosci Methods 1991;40:71 -- 86.
Leake MC, Wilson D, Bullard B, Simmons RM. The
elasticity of single kettin molecules using a two-bead laser-
tweezers assay. FEBS Lett 2003;535:55 -- 60.
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
doi:10.1016/S0014-5793(02)03857-7.
Leake MC. Analytical tools for single-molecule
fluorescence imaging in cellulo. Phys Chem Chem Phys
2014;16:12635 -- 47. doi:10.1039/c4cp00219a.
Sayed M, Krishnamurthy B, Pal H, Hazra P, Rajule RN,
Satam VS, et al. Unraveling multiple binding modes of
acridine orange to DNA using a multispectroscopic
approach. Phys Chem Chem Phys 2016;18:24642 -- 53.
doi:10.1039/C6CP03716J.
Benvin AL, Creeger Y, Fisher GW, Ballou B, Waggoner
AS, Armitage BA. Fluorescent DNA nanotags:
supramolecular fluorescent labels based on intercalating
dye arrays assembled on nanostructured DNA templates. J
Am Chem Soc 2007;129:2025 -- 34. doi:10.1021/ja066354t.
Reuter M, Dryden DTF. The kinetics of YOYO-1
intercalation into single molecules of double-stranded
DNA. Biochem Biophys Res Commun 2010;403:225 -- 9.
doi:10.1016/j.bbrc.2010.11.015.
Dikic J, Seidel R. Anticooperative Binding Governs the
Mechanics of Ethidium-Complexed DNA. Biophys J
2019;116:1394 -- 405. doi:10.1016/j.bpj.2019.03.005.
Spielmann HP, Wemmer DE, Jacobsen JP. Solution
structure of a DNA complex with the fluorescent bis-
intercalator TOTO determined by NMR spectroscopy.
Biochemistry 1995;34:8542 -- 53.
Johansen F, Jacobsen JP. 1 H NMR Studies of the Bis-
Intercalation of a Homodimeric Oxazole Yellow Dye in
DNA Oligonucleotides. J Biomol Struct Dyn 1998;16:205 --
22. doi:10.1080/07391102.1998.10508240.
Biebricher AS, Heller I, Roijmans RFH, Hoekstra TP,
Peterman EJG, Wuite GJL. The impact of DNA
intercalators on DNA and DNA-processing enzymes
elucidated through force-dependent binding kinetics. Nat
Commun 2015;6:7304. doi:10.1038/ncomms8304.
Kollmann F, Ramakrishnan S, Shen B, Grundmeier G,
Kostiainen MA, Linko V, et al. Superstructure-Dependent
Loading of DNA Origami Nanostructures with a Groove-
Binding Drug. ACS Omega 2018;3:9441 -- 8.
doi:10.1021/acsomega.8b00934.
Benson E, Mohammed A, Gardell J, Masich S, Czeizler E,
Orponen P, et al. DNA rendering of polyhedral meshes at
the nanoscale. Nature 2015;523:441 -- 4.
doi:10.1038/nature14586.
Benson E, Lolaico M, Tarasov Y, Gådin A, Högberg B.
Evolutionary Refinement of DNA Nanostructures Using
Coarse-Grained Molecular Dynamics Simulations. ACS
Nano 2019:acsnano.9b03473.
doi:10.1021/acsnano.9b03473.
12
|
1312.0897 | 1 | 1312 | 2013-12-03T18:30:22 | Multiscale approach to the physics of radiation damage with ions | [
"physics.bio-ph"
] | The multiscale approach to the assessment of biodamage resulting upon irradiation of biological media with ions is reviewed, explained and compared to other approaches. The processes of ion propagation in the medium concurrent with ionization and excitation of molecules, transport of secondary products, dynamics of the medium, and biological damage take place on a number of different temporal, spatial and energy scales. The multiscale approach, a physical phenomenon-based analysis of the scenario that leads to radiation damage, has been designed to consider all relevant effects on a variety of scales and develop an approach to the quantitative assessment of biological damage as a result of irradiation with ions. This paper explains the scenario of radiation damage with ions, overviews its major parts, and applies the multiscale approach to different experimental conditions. On the basis of this experience, the recipe for application of the multiscale approach is formulated. The recipe leads to the calculation of relative biological effectiveness. | physics.bio-ph | physics | Multiscale approach to the physics of radiation damage with ions
Eugene Surdutovich1,2 and Andrey V. Solov'yov2,3
1Department of Physics, Oakland University, Rochester, Michigan 48309, USA
2Frankfurt Institute for Advanced Studies, Ruth-Moufang-Str. 1, 60438 Frankfurt am Main, Germany
3On leave from A.F. Ioffe Physical Technical Institute, St. Petersburg, Russian Federation
(Dated: May 15, 2018)
The multiscale approach to the assessment of biodamage resulting upon irradiation of biological
media with ions is reviewed, explained and compared to other approaches. The processes of ion
propagation in the medium concurrent with ionization and excitation of molecules, transport of
secondary products, dynamics of the medium, and biological damage take place on a number of
different temporal, spatial and energy scales. The multiscale approach, a physical phenomenon-based
analysis of the scenario that leads to radiation damage, has been designed to consider all relevant
effects on a variety of scales and develop an approach to the quantitative assessment of biological
damage as a result of irradiation with ions. This paper explains the scenario of radiation damage
with ions, overviews its major parts, and applies the multiscale approach to different experimental
conditions. On the basis of this experience, the recipe for application of the multiscale approach is
formulated. The recipe leads to the calculation of relative biological effectiveness.
PACS numbers:
I.
INTRODUCTION
The physics and chemistry of radiation damage caused
by irradiation with protons and heavier ions has recently
become a subject of intense interest because of the use
of ion beams in cancer therapy [1–5]. Ion-beam cancer
therapy (IBCT) was first realised in the 1950s as proton-
beam therapy after being suggested by R. Wilson in 1946
because of the favourable shape of the depth-dose distri-
bution due to the fundamental difference in the energy
deposition profile between massive projectiles and mass-
less photons. This shape is characterised by the Bragg
peak, which is a sharp maximum in the linear energy
transfer (LET) of ions at the end of their trajectories.
Due to this key feature, IBCT allows a delivery of high
doses into tumours, maximising cancer cell destruction,
and simultaneously minimising the radiation damage to
surrounding healthy tissue. The effectiveness of radia-
tion with ions depends on the choice of ions; it can be
described by three factors: the peak value of LET, the
proximal plateau value of LET, and the size of a tail distal
to the peak. Since the LET is proportional to the square
of charge of the projectile, ions heavier than protons are
expected to be more effective; however, the increase of
LET in the plateau region and the increasing size of the
tail hinder the usage of heavier ions and, as a result, car-
bon ions, besides protons, are the most clinically used
modality [3, 4]. Because of its high costs, there are only
43 centres for proton beam therapy in 16 counties around
the world1. More proton centres are under construction.
Although heavy ion therapy was adopted in the 1990s,
there are only four clinical centres (in Germany, Italy,
and Japan) where carbon ions are used [7].
The Bragg peak occurs because the inelastic cross sec-
tions of interactions of projectiles with the molecules of
the medium increase up to the maximum values as the
speed of the projectile decreases. As a result, the depo-
sition of destructive energy to the tissue per unit length
of the ion's path is maximised within 1 mm of the ion's
trajectory. The location of the Bragg peak depends on
the initial energy of the ions. Typical depths for car-
bon ions (in liquid water representing tissue) range from
about 2.5 to 28 cm as the initial energy ranges from 100
to 430 MeV/nucleon [4, 8–12]. Hence, a deeply-seated
tumour can be scanned with a well-focused pencil beam
of ions with minimal lateral scattering.
Over the past 20 years, technological and clinical ad-
vances of IBCT have developed more rapidly than the
understanding of radiation damage with ions. Although
an empirical approach has produced exciting results for
thousands of patients thus far, many questions concern-
ing the mechanisms involved in radiation damage with
ions remain open and the fundamental quantitative sci-
entific knowledge of the involved physical, chemical, and
biological effects is, to a significant extent, missing. In-
deed, the series of works that elucidated the importance
of low-energy (below ionisation threshold) electrons ap-
peared in ca. 2000, while the treatment of patients at
GSI2 started in 1997. The dominant molecular mecha-
nism of a double strand break (DSB), the most impor-
tant DNA lesion [13, 14], still remains unknown. Even
the significance of the relation of DNA damage (including
DSBs) compared to the damage of other cellular compo-
nents to the cell death is not entirely clear. This list
can be continued. Besides IBCT, the mechanisms of bio-
damage due to irradiation with heavy ions have attracted
1 As of August 2013 [6].
2 Gesellschaft fur Schwerionenforschung, Darmstadt, Germany
3
1
0
2
c
e
D
3
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
7
9
8
0
.
2
1
3
1
:
v
i
X
r
a
attention in regards to radioprotection from galactic cos-
mic rays, especially during potential long-term space mis-
sions [4].
Over many decades of using radiation with photons,
vast data relating the radiation damage to deposited dose
were accumulated. These data are currently used to de-
scribe the biological damage due to ions [4]. Nonetheless,
there are substantial qualitative and quantitative differ-
ences between the effects of ions and photons on tissue.
The first difference is in the localisation of the dose dis-
tribution for ions distinguished from the mostly uniform
dose distribution for photons. This feature reveals itself
longitudinally (along the ion's path) as the Bragg peak.
Radially (with respect to the ion's path) it shows up as
the sharply decreasing (within several tens of nm) radial
dose distribution, while the average distance between ad-
jacent ions in clinically used beams are several hundreds
of nm.
The second difference is a consequence of the first.
Secondary particles such as electrons, free radicals, etc.,
produced as a result of the interaction (ionisation and
excitation) of ions with the medium, emerge at the loca-
tion of the Bragg peak in much larger number densities
than those produced by photons, and their distribution
is also non-uniform. These secondary particles largely
cause the biological damage, and in order to assess the
damage, it is important to distinguish the biological ef-
fects of the locally deposited dose and the local number
density of secondary particles. In other words, the (ra-
dial) dose is not the only characteristic that determines
the biological damage. For instance, clustered damage,
more lethal than isolated damage, can be caused by sev-
eral low-energy electrons, which are not associated with
a large dose deposition. This qualitatively and quantita-
tively changes the effect of the radiation [1, 4, 15].
There are also differences in the chemical interactions
related to a different balance between free electrons, free
radicals, and other agents for ions versus photons. These
differences, for example, affect the resistivity of cells to
radiation and thus are quite important for the assess-
ment of radiation damage. Finally, the Bragg peak leads
to thermomechanical effects, which stem from the non-
uniformity of the radial dose deposition.
One of the most important questions in the foundation
of science devoted to radiation damage with ions is the
question about molecular mechanisms leading to DNA
damage, or more generally, biodamage. While "whether
the biodamage leads to cell death?" is a biological ques-
tion, the question about the mechanisms of biomolecu-
lar damage belongs to the realms of physics and chem-
istry. The role of low energy (sub-15 eV) electrons has
been especially emphasised in Refs. [16–19]. A number of
quantum effects, such as dissociative electron attachment
(DEA), formation of electronic and phononic polarons,
are discussed in the context of the interaction of these
electrons with biomolecules. DEA is deemed to be the
leading mechanism for DNA single strand breaks (SSBs)
at low energies, while a number of ideas, including the
2
action of Auger electrons, in relation to the mechanism of
double strand breaks (DSBs) has been suggested [18, 20].
The Auger effect along with intermolecular Coulombic
decay (ICD) are discussed not only in relation to the
mechanism of DSBs, but also as important channels for
production of secondary electrons, especially in the pres-
ence of nanoparticles as sensitizers [21]. Still more un-
derstanding is needed for the interaction of electrons of
higher energies.
This paper is devoted to the overview of the main ideas
of the multiscale approach to the physics of radiation
damage that has the goal of developing knowledge about
biodamage at the nanoscale and molecular level and find-
ing the relation between the characteristics of incident
particles and the resultant biological damage [1, 22]. This
approach is unique in distinguishing essential phenomena
relevant to radiation damage at a given time, space, or
energy scale and assessing the resultant damage based
on these effects. The significance of understanding the
fundamental mechanisms of radiation damage in order
to exploit this knowledge for practical applications has
inspired the European COST Action [23], which sup-
ports collaborations of physicists, chemists, and biolo-
gists, studying these phenomena both theoretically and
experimentally.
The multiscale approach was formulated and then
elaborated upon, as different aspects of the scenario were
add-ed in a series of works [11, 12, 15, 22, 24–30].
Its
name emphasizes the fact that important interactions in-
volved in the scenario happen on a variety of temporal,
spatial, and energy scales. These scales are schematically
shown in Fig. 1. From the very beginning, the approach
was formulated as phenomenon-based and was aimed at
elucidating the physical, chemical, or biological effects
that are important or dominating on each scale in time,
space, and energy. The practical goal of the multiscale
approach is the calculation of relative biological effective-
ness (RBE) [3, 4, 31, 32], one of the key integral charac-
teristics of the effect of ions compared to that of photons.
The RBE is defined as a ratio of doses of photons and
different projectiles leading to the same biological effect,
such as killing a given percentage of cells in an irradi-
ated region. This is why the calculation of RBE is so
important. Other characteristics, such as the oxygen en-
hancement ratio (OER), which compares the biological
action of given projectiles to that at different aerobic or
hypoxic conditions of irradiated targets.
This paper is organised in the following way.
In
Sec. II, the scenario of radiation damage with ions is
described. Section III is devoted to the ion's transport
in the medium. Section IV describes the applications
of the random walk approach to the electron transport
relevant for biodamage. Examples of DNA damage cal-
culations are discussed there as well. These calculations
are compared to experiments with plasmid DNA and foci
studies. In Sec. V thermomechanical effects are explored.
Section VI is devoted to the evaluation of the probabil-
ity for an irradiated cell to survive based on the calcu-
3
FIG. 1: (Colour online) Features, processes, and disciplines, associated with radiation therapy shown in a space – time diagram,
which shows approximate temporal and spatial scales of the phenomena. The history from ionization/exciation to biological
effects on the cellular level are shown in the main figure and features of ion propagation are shown in the inset.
lation of clustered DNA damage. This is followed by
the recipe for the assessment of radiation damage with
ions using the multiscale approach starting from obtain-
ing and analysing the LET dependence and ending by
obtaining the RBE. The discussion is followed by conclu-
sions.
II. MULTISCALE SCENARIO OF RADIATION
DAMAGE
Radiation damage due to ionizing radiation is initi-
ated by the ions incident on tissue. Initially, they have
energy ranging from a few to hundreds of MeV. In the
process of propagation through tissue they lose their en-
ergy in the processes of ionization, excitation, nuclear
fragmentation, etc. Most of the energy loss of the ion is
transferred to tissue3. Naturally, radiation damage is as-
sociated with this transferred energy, and the dose (i.e.,
deposited energy density) is a common indicator for the
assessment of the damage [1, 4, 31]. The profile of the
dose deposition along the ion's path is characterised with
a plateau followed by a sharp Bragg peak. The position
of this peak depends on the initial energy of the ion and
marks the location of the maximum radiation damage.
In the process of radiation therapy, a tumour is being
"scanned" with the Bragg peak4 in order to deposit a
large dose to the target and spare healthy tissues sur-
3 The only part that is not transferred is emitted as radiation.
This part, in the case of ions interacting with tissue, is deemed
to be insignificant.
4 This scanning produces the so-called spread-out Bragg peak
(SOBP).
rounding it.
However, the deposition of large doses in the vicinity of
the Bragg peak does not explain how the radiation dam-
age occurs, since projectiles themselves only interact with
a few biomolecules along their trajectory and this direct
damage is only a small fraction of the overall damage.
It is commonly understood that the secondary electrons
and free radicals produced in the processes of ionization
and excitation of the medium with ions are largely re-
sponsible for the vast portion of the biodamage.
Secondary electrons are produced during a rather short
time of 10−18 − 10−17 s following the ion's passage. The
energy spectrum of these electrons has been extensively
discussed in the literature [11, 12, 33–35] and the main re-
sult (relevant for this discussion) is that most secondary
electrons have energy below 50 eV (more than 80% for an
ion energy5 of 0.3 MeV/u) and only a few (less than 10%
for 0.3 MeV/u-ions) have energy higher than 100 eV.
Moreover, this is true for a very large range of ion en-
ergy. This has several important consequences. First,
the ranges of propagation of these electrons in tissue are
rather small, around 10 nm [36]. Second, the angular
distribution of their velocities as they are ejected from
their original host, and as they scatter further, is largely
uniform [37]; this allows one to consider their transport
using a random walk approach [15, 20, 22, 38, 39].
The next time scale 10−16 − 10−15 s corresponds to
the propagation of secondary electrons in tissue. These
electrons (which start with about 45-50 eV energy) are
called ballistic.
In liquid water, the mean free paths
of elastically scattered and ionizing 50-eV electrons are
about 0.43 and 3.5 nm, respectively [37]. This means that
they ionize a molecule after about seven elastic collisions,
while the probability of second ionization is small [11].
Thus, the secondary electrons are losing most of their
energy within first 20 collisions and this happens within
1-1.5 nm of the ion's path [29]. After that they continue
propagating, elastically scattering with the molecules of
the medium until they get bound or solvated electrons
are formed. It is important to notice that these low en-
ergy electrons remain important agents for biodamage
since they can attach to biomolecules like DNA causing
dissociation [19, 40]. The solvated electrons may play an
important role in the damage scenario as well [14, 41, 42].
Additionally, the energy lost by electrons during the
previous stage in the processes of ionization, excitation
and electron-phonon interaction is transferred to the
medi-um. As a result of this relaxation, the medium
within about a 1 − 1.5-nm cylinder (for ions not heav-
ier than iron) around the ion's path becomes very
hot [26, 29]. This cylinder is referred to as the hot
cylinder. The pressure inside this cylinder increases by
a factor of about 103 compared to the pressure in the
4
medium outside the cylinder. This pressure builds up by
about 10−14 − 10−13 s and it is a source of a cylindrical
shock wave [28]. This shock wave propagates through
the medium for about 10−13 − 10−11 s. Its relevance to
the biodamage is as follows. If the shock wave is strong
enough (the strength depends on the distance from the
ion's path and the LET), it may inflict damage directly
by breaking covalent bonds in a DNA molecule [29]. Be-
sides, the radial collective motion that takes place dur-
ing this time is instrumental in propagating the highly
reactive species such as hydroxyl radicals, just formed
solvated electrons, etc. to a larger radial distance (up to
tens of nm) thus increasing the area of an ion's impact.
The assessment of the primary damage to DNA
molecu-les and other parts of cells due to the above effects
is done within the multiscale approach. This damage
happens within 10−5 s from the ion's passage and consists
of various lesions on DNA and other biomolecules. Some
of these lesions may be repaired by the living system, but
some may not and the latter may lead to cell death. The
scenario described above is illustrated in Fig. 2.
III. PROPAGATION OF IONS IN TISSUE AND
PRIMARY IONIZATION OF THE MEDIUM
A. The main characteristics of ion's propagation in
the medium
The scenario starts with the traverse of an ion through
tissue. Ions enter the medium with a sub-relativistic en-
ergy (for therapy, the carbon ion energy ranges through
100–420 MeV/nucleon and the proton energy can be up
to 250 MeV, while the ions of galactic cosmic rays are
much more energetic). Then, the ions lose energy prop-
agating in the tissue. This process is described by the
stopping power, S, of the medium, equal to −dE/dx,
where E is the kinetic energy of the ion and x is the
longitudinal coordinate. For projectiles such as protons
or heavier ions, there is not much difference between the
location of the energy loss by projectiles and that ab-
sorbed by the medium longitudinally, i.e., along the ion's
path6. Therefore, the linear energy transfer (LET), i.e.,
the energy absorbed by the medium per unit length of the
projectiles's trajectory becomes similar to the stopping
power. Hence, the terms "LET" and "stopping power"
are used synonymously. The energy loss occurs due to
ionization of the medium, nuclear fragmentation in col-
lisions with nuclei, excitations of the medium, etc. The
LET profile for ions is characterized by a plateau fol-
lowed by the sharp Bragg peak, where the LET reaches
its maximum. The tail is caused by the energy loss of
5 This value corresponds to the kinetic energy of ions near the
Bragg peak.
6 This is so because the energy is mostly transferred to electrons
and other secondary particles, whose longitudinal ranges are
many times smaller than the characteristic scale of x.
5
FIG. 2: (Colour online) The scenario of biological damage with ions.
Ion propagation ends with a Bragg peak, shown in
the top right corner. A segment of the track at the Bragg peak is shown in more detail. Secondary electrons and radicals
propagate away from the ion's path damaging biomolecules (central circle). They transfer the energy to the medium within
the hot cylinder. This results in the rapid temperature and pressure increase inside this cylinder. The shock wave (shown in
the expanding cylinder) due to this pressure increase may damage biomolecules by stress (left circle), but it also effectively
propagates reactive species, such as radicals and solvated electrons to larger distances (right circle). A living cell responds to all
shown DNA damage by creating foci, in which enzymes attempt to repair the induced lesions. If these efforts are unsuccessful,
the cell dies; an apoptotic cell is shown in the lower right corner.
the lighter products of nuclear fragmentation, such as
protons, neutrons, α-particles, etc.
The behaviour of the LET is explained by features of
inelastic cross sections of the projectile in the medium.
The Bragg peak in the stopping power of massive charged
particles is described by the Bethe - Bloch formula [43–
45].
− dE
dx
=
4πnez2e4
mV 2
ln
2mV 2
(cid:104)I(cid:105)(1 − β2)
− β2
(cid:20)
(cid:21)
,
(1)
where m and e are the mass and charge of electron, V
is the velocity of the projectile, β = V /c (c is the speed
of light in vacuum), ze is the charge of projectile, ne is
the number density of electrons in the target, and (cid:104)I(cid:105) is
the mean excitation energy of its molecules.
This formula provides the dependence of the stopping
power on the energy of the ion and practically depends
on a single parameter, the mean excitation energy. This
parameter for liquid water is chosen empirically some-
where between 70 and 80 eV [10, 46]. The use of such a
non-physical parameter is sufficient for the calculations of
the position of the Bragg peak and its shape, and Eq. (1)
is used in many Monte Carlo (MC) simulations [10] for
that purpose. This parameter, however, hides all phys-
ical processes such as ionization and excitation of the
medium, even though these same processes are important
for the understanding of the scenario of radiation dam-
age. Therefore, it is better to use a different approach,
which uncovers the physics integrated in the empirical
parameter. In Refs. [11, 12, 22], the singly-differentiated
(with respect to the secondary electron energy) ionization
cross sections of water molecules in the medium has been
employed as a physical input. This allowed not only de-
scribing the features of the Bragg peak, but also obtain-
ing the energy spectrum of secondary electrons, which
are very much involved in subsequent radiation damage.
B. Singly-differentiated cross sections of ionization
The total ionization cross section, σt, differentiated
with respect to secondary electron kinetic energy, W ,
i.e., singly-differentiated cross section (SDCS) is the
main quantity in our analysis. Besides the kinetic en-
ergy of secondary electrons and the properties of water
molecules, the SDCS depends on the velocity V of the
projectile and its charge, ze.
1. Calculation of the SDCS using a parametric
semiempirical approach
In Refs. [11, 12, 47], the semi-empirical Rudd's expres-
sion for the calculation of SDCS has been used. This
analytic expression, containing a number of parameters,
is a combination of the experimental data and calcu-
lations within the plane-wave Born approximation and
other theoretical models [48]. Since this model was de-
veloped for non-relativistic protons, it had to be modified
to include heavier ions at relativistic velocities. The orig-
inal SDCS is given in the following form [48]:
= z2(cid:88)
i
dσt
dW
(cid:18) I0
(cid:19)2 ×
4πa0Ni
Ii
Ii
(2)
F1(vi) + F2(vi)ωi
(1 + ωi)3 (1 + exp(α(ωi − ωmax
i
,
)/vi))
where the sum is taken over the electron shells of the
water molecule, a0 = 0.0529 nm is the Bohr radius, I0 =
13.6 eV, Ni is the shell occupancy, Ii is the ionization
potential of the shell, ωi = W/Ii is the dimensionless
normalised kinetic energy of the ejected electron, vi is
the dimensionless normalised projectile velocity given by
(cid:115)
vi =
mV 2
2Ii
.
(3)
6
(cid:113) 2E
(cid:113) m
When V (cid:28) c, V =
M (where M is the mass of a
. When V approaches
projectile), and, hence vi =
c, the definition of vi, given by (3), holds, however, the
projectile's velocity V is given by βc, where β2 = 1 −
1/γ2 = 1 − (M c2/(M c2 + E))2, and γ is the Lorentz
factor of the projectile.
E
Ii
M
Functions F1 and F2 in (2) are given by
F1(v) = A1
ln(1 + v2)
B1/v2 + v2 +
C1vD1
1 + E1vD1+4 ,
(4)
and
F2(v) = C2vD2
A2v2 + B2
C2vD2+4 + A2v2 + B2
.
(5)
The fitting parameters A1 ... E1, A2 ... D2, and α depend
on the medium.
In Ref. [48], they are given for water
vapour. The comparison of positions of Bragg peaks for
different initial carbon ion energies with those measured
in experiments provided sufficient material for refitting
of these parameters for liquid water medium [12]. These
parameters are listed in Table I [12]. The cut-off energy
ωmax is given by
ωmax
i
= 4v2
i − 2vi − I0
4Ii
,
(6)
where the first term on the right-hand side represents the
free-electron limit, the second term represents a correc-
tion due to electron binding, and the third term gives
the correct dependence of the SDCS for vi (cid:28) 1 [48]. For
vi (cid:29) 1, Eq. (2) should asymptotically approach the rela-
tivistic Bethe - Bloch formula (1). This is accomplished
when F1, given by (4), is replaced by the following ex-
pression,
F1(v) = A1
1−β2 ) − β2
ln( 1+v2
B1/v2 + v2 +
C1vD1
1 + E1vD1+4 .
(7)
Indeed, the asymptotic behaviour of (7) at v (cid:29) 1 is given
by
A1
v2
v2
ln(
(cid:20)
dx ∼ (cid:80)
(cid:21)
1 − β2 ) − β2
(cid:82) (W + Ii) dσt
,
which, after being substituted to Eq. (2) and the un-
derstanding that dE
dW dW , leads to
Eq.
(1). The correction of Eq. (7) reveals itself as an
increase of the cross section at high energies.
i
2. Calculations of SDCS based on the energy-loss function
An alternative method has been used in Ref.
[30],
where the dielectric formalism based on the experimen-
tal measurements of the energy-loss function (ELF) of
the target medium, Im (−1/(E, q)), where (E, q) is the
TABLE I: Fitting parameters and ionization energies for three outer and two inner shells of water molecules in a liquid water
environment [12].
7
Shells
Outer: 1b1, 3a1, 1b2 10.79, 13.39, 16.05
Inner: 2a1, 1a1
32.3, 539.0
Ionization energies (eV) A1 B1 C1 D1 E1 A2 B2 C2 D2 α
1.02 82 0.5 −0.78 0.38 1.07 14.5 0.61 0.04 0.64
3.0 1.1 1.3 1.0 0.0 0.66
1.25 0.5 1.0
1.0
complex dielectric function, and q and E are the momen-
tum and energy transferred in the electronic excitation,
respectively [49, 50]. This formalism allows obtaining the
SDCS not only for liquid water but for a real biological
medium containing sugars amino acids, etc. If the ELF is
experimentally known, many-body interactions and tar-
get physical state effects are naturally included in these
calculations.
According to that formalism, the macroscopic (nonrel-
ativistic) SDCS for ionization of the electronic shell i is
given by,
dσi(W,E)
M z2
−1
(cid:20)
(cid:21)
dW
(q, Ii + W )
√
2M (
q−
E − E). Equation (8) can be
where q± =
used for different charged projectiles by properly taking
into account their charge state, or for electrons by intro-
ducing an exchange term in the integrand and imposing
the correct integration limits.
i
(cid:90) q+
E ± √
E
e2
=
nπ2
√
dq
q
Im
, (8)
Since Eq. (8) requires the contribution of each elec-
tronic shell of the target to its ELF, and the latter is
usually measured for all the excitations and ionizations
of the electronic system in the optical limit (q = 0), the
algorithm for obtaining the data at q > 0 and splitting
this ELF into different electronic shells is needed in ad-
dition to the experimentally measured ELF.
The optical ELF bioorganic condensed compounds and
liquid water are rather similar and can be parameterized
with a single-Drude function [51]
Im
−1
(q = 0,E)
a(Z)E
(E 2 − b(Z)2)2 + c(Z)2E 2
=
,
(9)
(cid:20)
(cid:21)
where a(Z), b(Z), and c(Z) are the functions of the mean
atomic number of the target Z, corresponding to the
height, position, and width [51]. While b(Z) and c(Z)
are parametric functions, a(Z) is obtained by imposing
the f-sum rule [52], linked to the number of electrons in
the target, Z, also accounting for the contribution from
the inner shells, as explained in Ref. [51]. Using this
approach the ELF of an arbitrary bioorganic compound
can be estimated, even in the case where no experimen-
tal data exist. A wide variety of extension algorithms for
extrapolation of optical-ELF to q > 0 are available due
to extensive research [53]. In Ref. [30] a simple quadratic
dispersion relation introduced by Ritchie and Howie [54],
with its parameters for liquid water [53], has been used.
The issue of splitting of the ELF into contributions
from different shells has been studied for liquid wa-
ter [55, 56], providing parameterizations of the ELF split
in ionization and excitation arising from each different
shell. In Ref. [30] a specially designed approximation has
been applied to split the ELF for biomolecules. In order
to describe the outer-shell ionization of biomolecules, the
mean value of their binding energies, ¯I, is calculated7. It
is then assumed that the outer-shell electrons will be ion-
ized if the transferred energy satisfies E > ¯I. Then, the
ejected electron energy is W = E − ¯I. In Ref. [30], SDCSs
are calculated and compared with other calculations and
experiments for protons interacting with water, adenine,
and benzene.
The total ionization cross sections (TICS) can also
be estimated for different biomolecules relevant for
IBCT [30]. For example, in Fig. 3a [30] the macroscopic
TICSs are calculated for proton impact in five represen-
tative biological materials relevant for cancer therapy:
liquid water, dry DNA (C20H27N7O13P2), protein, lipid,
and the cell nucleus. Their atomic compositions and den-
sities can be found in the ICRU Report 46 [60] and other
sources, and a reasonable value of their mean binding
energies can be estimated from the values of their molec-
ular components, such as the water molecule, DNA bases
and backbone, and amino acids [55, 57, 58]. The exper-
imental data for water vapour [61–63] are also shown.
They agree well with the calculations of Ref. [30] above
100 keV, where the first Born approximation is applica-
ble without further corrections. From these results, it is
plausible that all the biological targets different from wa-
ter have a larger ionization probability than water. One
can also see that the TICS of a cell nucleus is only slightly
larger than that of liquid water, and that protein has a
slightly larger TICS than the rest of the biomaterials.
In Fig. 3b [30] the microscopic TICS per molecule
for proton impact in the DNA molecular components,
such as adenine, cytosine, guanine, thymine, and sugar-
phosphate backbone are shown. Their atomic composi-
tion can be easily found in the literature, and their mean
binding energies were estimated from quantum chemistry
calculations [57, 58]. Also shown are experimental data
at high energies for adenine [64], which are in excellent
agreement with the predictions of Ref. [30]. This method
allows one to estimate the ionization probability of each
7 The relevant data are available for some biological molecules,
such as the DNA bases and the sugar-phosphate backbone [57]
and some amino acids [58] and others [59].
8
targets, for which data are lacking, both experimentally
and theoretically. This model can be easily extended to
ions heavier than protons, in different charge states, as
well as to electron impact ionization, by introducing ap-
propriate corrections, such as the description of the elec-
tronic structure of the ion, or exchange and relativistic
corrections for electrons.
C. The position of the Bragg peak
The stopping cross section, defined as
σst =
(W + Ii)
dσt,i
dW
dW ,
(10)
(cid:90) ∞
(cid:88)
0
i
where the sum is taken over all electrons of the target,
gives the average energy lost by a projectile in a single
collision, which can be further translated into energy loss
within an ion's trajectory segment, dx:
= −nσst(E) .
dE
dx
(11)
This quantity is known as the stopping power [31, 46]. As
was discussed above in Sec. III A, for ions this quantity
is similar to the linear energy transfer (LET).
The LET found from Eq. (11) is a function of the ki-
netic energy of the ion rather than the ion's position
along the path in the medium. The dependence of LET
(and other quantities) on this position, however, is more
suitable for cancer therapy applications. Integrating in-
verse LET, given by (11), yields
(cid:90) E0
E
x(E) =
dE(cid:48)
dE(cid:48)/dx ,
(12)
where E0 is the initial energy of the projectile. We ob-
tain the correspondence between the position of the ion
along the path and its energy. This allows one to obtain
all quantities of interest in terms of x rather than E. The
depth dependence of the average LET as a function of x is
shown in Fig. 4. The calculations of the LET include the
effects that were discussed above, such as SDCS calcu-
lated using semi-empirical parameterization (2), modified
for relativistic energies (7) with the use of the effective
charge described below in Sec. III D. The effect of en-
ergy straggling due to multiple ion scattering, described
in the Sec. III E is also taken into account. This effect
explains why the height of the Bragg peak decreases with
the increasing initial energy of ions and thus increasing
depths of the corresponding Bragg peaks. The contri-
bution of non-ionization processes, such as excitation of
neutral molecules, are also included in these calculations.
In order to accomplish this, the excitation cross sections
for proton projectiles [66] were scaled using the ratio of
the effective charges for carbon and proton at a given
energy E.
In Fig. 4, our calculated LET is compared with the ex-
FIG. 3: (Colour online) (a) Calculated macroscopic TICS for
proton impact in liquid water, DNA, protein, lipid, and cell
nucleus. (b) Calculated microscopic TICS for proton impact
in the DNA components adenine, cytosine, guanine, thymine,
and sugar-phosphate backbone. Symbols represent experi-
mental data [30]. N is the molecular density of the target.
constituent of the DNA molecule, which gives important
information on the sensitivity of each one to radiation
damage. According to these results, the DNA backbone
is the most probable part of the DNA to be ionized by
proton impact (a similar behaviour was previously ob-
served for electron impact in Ref. [57]; also, recent theo-
retical estimates [65] point towards sugar-phosphate C-O
bond cleavage due to interaction with low energy elec-
trons) and, between bases, adenine and guanine are the
most sensitive to proton impact ionization. This fact
could have important implications in the DNA damage,
since it seems that single or double strand breaks could
be more probable than base damage, or that regions of
the DNA with a bigger concentration of adenine or gua-
nine would be more likely damaged by radiation than
other parts of the genome, attending to direct ionization
effects.
Much more information can be obtained with this me-
thod, such as the number of emitted electrons, the av-
erage energy of electrons, SDCS and TICS for other bi-
ological targets and projectiles. This model, using little
input information and physically motivated approxima-
tions, can provide useful information about the ion im-
pact ionization of a huge number of relevant biological
9
their velocity has been suggested by Barkas [68], where
the following empirical formula for the effective charge,
zef f , is introduced,
zef f = z(1 − exp(−125βz−2/3)) ,
(13)
where z is the charge of the stripped ion. This formula
is a result of studies of energy loss of ions in emulsions.
More detailed descriptions of charge transfer effects have
became available recently [69]. These studies allow one
to not only estimate the effective charge of the ion, but
also find its fluctuations. These fluctuations are impor-
tant since LET increases proportionally as z2 and if LET
becomes large enough, qualitative differences related to
thermomechanical effects may become substantial (see
Sec. V below).
Regardless of the method of the calculation of the ef-
fective charge, in order to find the stopping power and
estimate the secondary electron spectra (in the first ap-
proximation) z in Eq. (2) should be replaced by an effec-
tive charge zef f which decreases with decreasing energy
making the ionization cross section effectively smaller. In
Refs. [11, 12] the parameterization (13) was used. The
effective charge given by this expression slowly changes
at high projectile velocity, but rapidly decreases in the
vicinity of the Bragg peak. As a result, charge transfer
significantly affects the height of the Bragg peak, and
only slightly shifts its position towards the projectile's
entrance. This happens because the stopping cross sec-
tion as a function of velocity has a sharp peak as velocity
decreases. At the same time σst is proportional to z2
ef f .
If the latter decreases with decreasing V , the Bragg peak
shifts towards the direction of the beam's entrance to the
tissue. For instance, with the account for charge transfer,
for carbon ions the Bragg peak occurs at E = 0.3 MeV/u
rather than at E = 0.1 MeV/u.
E. The effect of ion scattering
It will become clear below, in Sec. IV C 3 and Sec. VI B,
that tracks of ions emerging from clinically used accelera-
tors do not interfere, i.e., the effects of a single ion do not
spread far enough to reach the area affected by adjacent
ions. Therefore, it is usually sufficient to study a single
ion interacting with tissue and then combine these ef-
fects relating the action of the beam with the dose. Even
though the Bragg peak is a feature of every ion's LET,
each peak cannot be observed separately. Since each of
the projectiles in the beam experiences its own multi-
ple scattering sequence, peaks for different ions occur at
a slightly different spatial location and only the Bragg
peak, averaged over the whole beam, is observed exper-
imentally. Therefore, in order to compare the shape of
the Bragg peak with experiments, the whole ion beam
should be considered.
In Ref. [11], the Bragg peak for an ion beam was ob-
tained via introduction of the energy-loss straggling due
FIG. 4: (Colour online) The dependence of the LET on depth
with the Bragg peak, plateau, and tail for carbon ions in liq-
uid water. The calculations (solid line) are done for ions with
the initial energy of 330 MeV/u and with use of Eqs. (10) and
(11). Experimental results [8] for the same energy are shown
with dots. The dashed line depicts the LET dependence with-
out the effect of energy straggling. Two almost coinciding
curves in the inset show the agreement between the analytical
calculations and MC simulations [10] for 420 MeV/u carbon
ion projectiles with straggling being included.
perimental results [8]. As can be seen from the figure, the
experimental dots at the Bragg peak are systematically
lower than the calculated curve, the difference being due
to in the nuclear fragmentation component, which has
not been included in the analytical calculations.
It is
feasible to include it, as has been done in Ref. [67] for
protons, if the appropriate fragmentation cross sections
are known.
As confirmed by MCHIT MC simulations [10], nuclear
fragmentation reactions become important for heavy-
nuclei beams and deeply-located tumours. For example,
both experimental data [8] and MCHIT calculations [10]
indicate that more than 40% of primary 200 MeV/u
12C6+ nuclei undergo fragmentation before they reach
the Bragg peak position, and this fraction exceeds 70%
for a 400 MeV/u 12C6+ beam. As a result of nuclear
reactions the beam is attenuated. New projectiles such
as protons, neutrons, and α-particles are formed. Since
these particles are lighter than the incident ions, after
fragmentation they carry a larger portion of the energy
and their penetration depths are larger than that of the
original ions [8]. This results in a tail after the Bragg
peak also seen in Fig. 4.
D. Charge transfer effect
The incident ions are usually stripped of all electrons,
but as they slow down they pick electrons off and their
charge reduces. The dependence of the charge of ions on
(cid:28) dE
dx
(cid:29)
(x)
to ion scattering. The energy straggling, described by a
semi-analytical model [70], is given by
=
(cid:90) x0
(cid:20)
− (x(cid:48) − x)2
2λ2
str
(14)
(cid:21)
dx(cid:48) ,
(x(cid:48)) exp
dE
dx
√
1
λstr
2π
0
where x0 is a maximum penetration depth of the pro-
jectile and λstr = 0.8 mm is the longitudinal-straggling
standard deviation computed by Hollmark et al. [71] for
a carbon ion of that range of energy. The Bragg peak
shown in Fig. 4 was calculated using Eq. (15).
F. Energy spectra of secondary electrons
The most important effect that takes place during the
propagation of the ion in tissue is the ionization of the
medium. This is how, when, and where the secondary
electrons, the key player in the scenario of radiation dam-
age, are produced. The information, required for the un-
derstanding of phenomena related to secondary electrons,
is the number of electrons produced per unit length of
the ion's trajectory and their energy distribution. This
section is devoted to the analysis of the electron energy
distributions obtained from ionization cross sections dis-
cussed above.
The emission of electrons in collisions of protons with
atoms and molecules has been under theoretical and ex-
perimental investigation for decades [35, 48, 72, 73]. The
quantity of interest is the probability to produce Ne sec-
ondary electrons with kinetic energy W , in the interval
dW , emitted from a segment ∆x of the trajectory of a sin-
gle ion at the depth x corresponding to the kinetic energy
of the ion, E. This quantity is proportional to the singly-
differentiated cross ionization section (SDCS)8, discussed
in Sec. III B.
dNe(W, E)
dW
= n∆x
dσt
dW
.
(15)
where n is the number density of molecules of the medium
(for water at standard conditions n ≈ 3.3 × 1022cm−3).
Equation (15) relates the energy spectrum of secondary
electrons to the SDCS regardless of the method, by which
the latter are obtained.
One important characteristics that can be obtained
from the SDCS is the average energy of the secondary
electrons, (cid:104)W(cid:105), which is given by
(cid:90) ∞
0
(cid:104)W(cid:105)(E) =
1
σt
W
dσt
dW
dW .
(16)
The dependence of the average energy of electrons on
8 The SDCS are integrated over full solid angle of electron emission
10
FIG. 5: Average energy of secondary electrons produced as
the result of impact ionization as a function of kinetic energy
(per nucleon) of 12C6+ ions.
the energy of the projectile, given by the result of inte-
gration (16) for liquid water medium is shown in Fig. 5.
Notice, that this figure is different from similar figures
of Refs. [11, 22], where the calculations were done with
parameters for water vapour. This dependence indicates
that the energy of secondary electrons is somewhere be-
low 50 eV for the whole range of the ion's energy and
it levels out as the the energy of projectiles increases.
There are several consequences from this. First, since
the dependence of (cid:104)W(cid:105) on the ion's energy E on a rel-
evant range of projectile energies (0.3 – 400 MeV/u) is
weak for the large range of the ion's energy, the number
of produced secondary electrons is largely proportional to
the value of LET, more precisely to the electronic com-
ponent of the LET, Se, that excludes nuclear stopping.
Indeed, if the ion is destroyed in a nuclear collision, ion-
ization due to its debris should be discussed instead; if
it survives then its ionizing capabilities do not change
too much, unless it slows down considerably; then, its
stopping power may change correspondingly. Second,
the expression for (cid:104)W(cid:105) is independent of the charge of
the projectile, e.g., the difference between, say, protons
and iron ions is in their values of Se, i.e., in the number
of secondary electrons, but not in their relative energy
spectra. Therefore, the difference between the effects of
these different ions will be in the number of secondary
electrons produced by these ions per unit length of path.
Third, most of the secondary electrons are capable of ion-
izing just one or two water molecules; thus, there is no
significant avalanche ionization effect [11]. This can be
explained by a simple estimate. Since the average energy
of secondary electrons in the vicinity of the Bragg peak
is about 40 eV (somewhat below this value), the maxi-
mum average energy that can be transferred to the next
generation secondary electron is just (40− Ii)/2, which is
about 15 eV for the outermost electrons, an energy barely
enough to cause further ionization. Finally, what is of
crucial importance for the consideration of the next scale
of electron propagation is that at sub-50 eV energies, the
electrons' cross sections are nearly isotropic [37, 74] and
it is possible to use the random walk approximation in
order to describe their transport [15, 22, 38, 39]. This
transport is described in the next section.
IV. RANDOM WALK APPROXIMATION FOR
THE DESCRIPTION OF THE SECONDARY
ELECTRON TRANSPORT
The next stage of the scenario is related to secondary
electrons ejected from the molecules of the medium as a
result of ionization. As has been discussed above, most
of these electrons have energies below 50 eV. They are
called ballistic electrons until their energy becomes suf-
ficiently small and coupling with phonons, recombina-
tion, and other quantum processes start dominating their
transport. While the electrons are ballistic, their inter-
actions with molecules can be described as a sequence of
elastic and inelastic collisions. Many works, by and large
using MC simulations, describe the transport of ballistic
electrons. They are known as track structure codes [37].
Some of them describe chemical reactions in the medium
including production of radicals and their propagation.
However, regardless of how sophisticated these codes are,
they do not contain the whole physical picture as will be
shown below. In this section, a rather simple analytical
approach is applied to the description of the propagation
of ballistic electrons and its results are compared to MC
simulations. It is also demonstrated how to make sense
of radiation damage based on these calculations.
The main mechanism of radiation damage by ballis-
tic electrons is inelastic collisions with targets. A target
in this discussion is a biomolecule, such as DNA. There-
fore, the probability of biodamage is a combination of
the number of electrons (or other secondary particles)
colliding with a given segment of a biomolecule and the
probability of a certain inelastic process on impact. The
first part is described by the fluence of electrons or other
particles on the target. Fluence is the integral of the
flux of particles (the number of particles hitting a part
of the target's surface per unit time) over the entire time
after the ion's passage and over the surface of the tar-
get. In general, the fluence depends on the distance of
the target from the ion's path and its geometrical orien-
tation. It will be shown that this part can be calculated
analytically with accuracy, sufficient for understanding
the scenario of radiation damage. The second part, i.e.,
the probability of a certain inelastic process on impact,
is more difficult to assess mainly because of the diverse
variety of possible processes. However, there are plenty
of data that allows one to make reasonable quantitative
estimates for this probability.
Let us start with the calculation of fluence for a num-
ber of relevant configurations. It will be shown that im-
11
portant characteristics of the track structure such as ra-
dial dose can also be calculated via fluence. The random
walk approach [75] used for these problems allows one
to make simple analytical calculations of fluence. The
main requirement for the use of this approach is that the
elastic and inelastic scattering of secondary electrons is
isotropic. The anisotropy in the angular dependence of
the cross sections for sub-50-eV electrons appears to be
insignificant [37]. As was noted above, more than 80%
of secondary electrons satisfy this condition and only for
less than 10% of δ-electrons with energies higher than
100 eV is this condition violated significantly. The ef-
fects of δ-electrons will be considered in Sec. IV E.
In
Sec. IV D, the transport and effects of radicals, whose
role in radiation damage is quite substantial, will be dis-
cussed.
Sections IV A–IV C are devoted to the transport of
sub-50-eV electrons. Moreover, unless specifically stated
to the contrary, these secondary electrons are produced
by carbon ions in the vicinity of the Bragg peak in liquid
water. At this part of the ion's trajectory, while a 0.3-
MeV/u carbon ion passes 1 µm along the path, a typical
radius within which the secondary electrons propagate is
about 1 nm [28, 29]. This allows one to assume that the
electron diffusion is cylindrically symmetric with respect
to the ion's path. The electronic component of the LET,
Se, remains nearly constant along this 1 µm of ion's path
described by the coordinate ζ. Therefore, the number of
ejected secondary electrons per unit length dNe
is inde-
dζ
pendent of ζ. A typical elastic mean free path of sub-50-
eV electrons l ranges between 0.1 and 0.45 nm [37, 74].
Since the scale along the Bragg peak is measured in tens
of µm, while the radial scale is only tens of nm, therefore
one can assume ζ to be ranging from −∞ to +∞.
A. Calculation of the fluence of secondary electrons
In the three-dimensional axially symmetric propaga-
tion of ballistic electrons from the axis, the key differen-
tial quantity is the flux of secondary electrons originating
from a segment dζ of the ion's path through an area dA
located at a distance ρ from the ion's path, as is shown in
Fig. 6. Vector r connects the element dζ with dA. This
flux is given [38, 39, 75] by the following expression:
dNd(r, t)
dt
= dA · D∇P (t, r)
= dA · Dnr
∂P (t, r)
∂r
dNe
dζ
dNe
dζ
dζ
dζ ,
(17)
where, D = ¯vl/6 is the diffusion coefficient, ¯v is the av-
erage speed of electrons, nr is a unit vector in the radial
direction (from the segment to the center of the area dA),
and
(cid:18) 3
(cid:19)3/2
(cid:18)
(cid:19)
P (t, r) =
2π¯vtl
exp
− 3r2
2¯vtl
(18)
calculated as the integral over the surface of the target:
(cid:90)
F(ρ) =
12
(cid:90) ∞
(cid:19)
r/l
dk
dNA(r) =
(cid:18) 3
2πkl2
A
× r
2k
dA · nr
(cid:18)
dζ
dNe
dζ
− 3r2
2kl2 − γk
exp
.
(20)
(cid:90)
(cid:19)3/2
A
Strictly speaking, the fluence given by Eq. (20) depends
on more variables than just the distance between the tar-
get and the ion's path. These variables include the elas-
tic mean free path of secondary electrons and more ge-
ometrical parameters. The mean free path corresponds
to some energy between zero and 50 eV and thus en-
ergy averaging is achieved. After this averaging, the en-
ergy of electrons is assumed to be constant. In different
works [15, 20, 22, 38, 39] this averaging was done accord-
ing to the particular physical problem. However, it is
the dependence on the distance ρ, kept in Eq. (20), that
remains important for calculations of radiation damage.
The application of this method to specific geometries that
were considered in some of these works are demonstrated
below.
B. Calculation of the radial dose
The radial dose is an important quantity in the physics
of IBCT since the dose distribution around the ion path
is highly non-uniform. Starting from the works of Katz
et al. [76–78] the radial dose has been used for the as-
sessment of radiation damage with ions. Since then the
radial dose has been calculated since then using MC
simulations [79–81]. In Ref. [38], it was shown that the
radial dose, i.e., the locally absorbed energy density as
a function of the distance from the ion's path, ρ, can be
calculated analytically using the random walk approach.
This calculation is based on the application of Eq. (20)
to the simplest geometry, where the target is a cylinder of
radius ρ and length δ, coaxial with the ion's path, shown
in the inset of Fig. 7. In Sec. A 0 a it is shown how to
calculate the number of secondary electrons, Fδ(ρ), inci-
dent on such a surface. Then, the number of ionization
events in a shell between ρ and ρ + dρ is proportional to
the number of secondary electrons (of a given energy) in-
cident on the inner cylindrical surface multiplied by the
probability of ionization per electron (effective area over
the total area). This is equal to the number of water
molecules inside the volume (number density times vol-
ume n2πρδdρ) multiplied by the ionization cross section
σ and divided by the total area of the cylindrical shell
(2πρδ)
dN = Fδ(ρ)
nσ2πρδdρ
2πρδ
= Fδ(ρ)nσdρ .
(21)
The energy deposited in this shell is equal to the product
of this number of events and the average energy per event
FIG. 6: (Colour online) Geometry for the general calculation
of fluence through a segment of surface dA. The ion path is
along the axis.
is the probability density to observe a randomly walking
electron at a time t and a distance r from the electron's
origin.
The next step in the calculation of fluence is the in-
tegration of Eq. (17) over time. In order to do this, we
change variables from t to the number of steps by sec-
ondary electrons k using ¯vt = kl. We rewrite Eq. (17),
substituting (18), and switching from variable t to k as
dNA(r) =
(cid:90) dNd((cid:126)r, t)
(cid:18) 3
dt
× r
2k
2πkl2
(cid:90) ∞
(cid:19)
r/l
dk
. (19)
dt = dA · nr
(cid:19)3/2
(cid:18)
dζ
dNe
dζ
− 3r2
2kl2 − γk
exp
An attenuation exponential factor e−γk is introduced in
order to take into account electrons falling out from the
random walk. The coefficient γ is equal to the ratio of the
cross section of processes in which electrons stop being
ballistic to the total cross section. An example of such
a process is an inelastic collision of an electron with a
water molecule after which the energy of the electron
drops below a certain excitation or ionization threshold
related to the molecules of the medium. This does not
completely inactivate it as an agent of radiation damage
since it may attach itself to a molecule and bring about its
dissociation, but such electrons vanish from the picture
of radial dose delivery.
The integration over k in Eq. (19) is carried out from
the minimal number of steps necessary to reach a dis-
tance r to infinity. After that, the fluence, F(ρ), can be
in Sec. V D), which increase the volume around the ion's
path where the energy is absorbed and thus decrease the
radial dose.
13
C. Targeting a twist of DNA with secondary
electrons
The first analytical calculation of biodamage using a
random walk approach was done in Ref. [22], where the
dependence of the fluence through a twist of DNA, which
was represented as a cylinder of size corresponding to
one twist of a DNA molecule (radius of 1.15 nm and
length of 3.4 nm), was calculated. A choice of a twist of a
DNA molecule as a target is related to the types of DNA
damage, such as single and double strand breaks (SSB
and DSB) which are widely discussed in the literature [13,
14, 82]. The DSB is a severe lesion, which can still be
repaired, but its contribution to the probability of cell
death is significant. The DSB is defined as two SSBs of
the opposite strands within 10 base pairs of each other,
i.e., within a single twist of a DNA molecule.
The probability of an SSB or a DSB in a given twist
is related to the fluence of secondary electrons produced
by the passing ion. Therefore, the first problem is to cal-
culate the fluence of these electrons through a cylinder
enwraping the twist. This cylinder may be arbitrarily
oriented with respect to the ion's path. A perpendicu-
larly (and symmetrically) oriented cylinder is shown in
the inset of Fig. 8.
In Appendix A, it is shown how to apply Eq. (20) to
different orientations of a cylindrical target. With expres-
sions for r2 and nr · dA, the integrations over ζ and the
area of cylinder (20) give the fluence through the cylinder
in two limiting cases of different orientation. The results
are shown in Fig. 8, where, the fluence through a perpen-
dicular cylinder is compared with MC simulations [39].
The fluence through a parallel cylinder is larger than that
of a perpendicular one by about 20%. This allows one
to average the fluence through the cylinder enwraping
the DNA twist over its orientation with respect to the
ion's path. This fluence can be extrapolated to the re-
gion 0 < ρ < a , where a = 1.15 nm is the radius of
the cylinder enwraping the twist in order to estimate the
probabilities of DNA damage in the "whole" space with-
out limitations.
1. Calculation of the number of SSBs per single ion
An estimate of the number of SSBs per unit length
of the ion's trajectory can be obtained assuming that
this number is proportional to the number of secondary
electrons incident on a given twist of a DNA molecule.
For example, for a straight segment of length dζ of the
FIG. 7: (Colour online) The normalised radial dose deter-
mined using the random walk approximation (solid line) com-
pared with the results of Ref. [79] for 1-MeV protons (dots).
These dots are digitized from the solid line and thus only rep-
resent a fragment of the data. These calculations were done
with l = 0.15 nm and γ = 0.0006 [38]. In the inset, the geom-
etry for the calculation of radial dose. The ion path is along
the axis. Secondary electrons propagate radially and the en-
ergy is deposited in the coaxial cylindrical shell of length δ,
inner radius ρ, and outer radius ρ + dρ.
¯W ,
dE = ¯WFδ(ρ)nσdρ .
(22)
Finally, the radial dose is the volume density of the de-
posited energy, i.e., Eq. (22) divided by the volume of
the shell 2πρδdρ:
D(ρ) = ¯W
Fδ(ρ)nσdρ
2πρδdρ
= ¯W nσ
dNe
dζ
Q(ρ/l, γ) ,
(23)
where the function Q(ρ/l, γ) is defined by Eq. (A2) in
the Appendix.
Thus, the radial dose due to ions propagating in a
medium in the vicinity of the Bragg peak, obtained us-
ing a random walk approximation, is given by Eq. (23).
This dependence is studied analytically in some special
cases [38]. It compares reasonably well with the MC sim-
ulations of Ref. [79] at small and moderate distances from
the ion's path, as shown in Fig. 7. The radial dose cal-
culated using the random walk as well as that obtained
using MC simulations corresponds to the radial dose in
a static medium where the effects of relaxation of the
deposited energy are not included. This corresponds to
the dose distribution up to 10−14 s when this relaxation
takes place and leads to collective flow effects (discussed
14
electrons ejected in the primary ionization with projec-
tiles and does not include ionizations due to secondary
electrons. Since the low-energy electrons produced in the
latter are important for biodamage, the number dNe
dζ and,
therefore, both the fluence and φ are underestimated by
a factor of about two [11].
The probability of the production of a SSB by an elec-
tron incident on a twist of a DNA molecule, ΓSSB, ap-
pears in Eq. (24) as well as in (25). This probability can
be estimated as the cross section for breaking an impor-
tant covalent bond that leads to a SSB multiplied by the
number of such bonds in a single DNA twist and divided
by the lateral area of this DNA segment, represented
above by a cylinder. However, the cross section for break-
ing a covalent bond is energy-dependent and the energy
of secondary electrons varies from zero to about 50 eV.
At low energies (below the ionization threshold) the cross
section is deemed to be that of DEA, i.e., resonant attach-
ment of the secondary electron to the molecule (forma-
tion of temporary negative ion) followed by dissociation
(SSB). At higher energies of impact electrons, the cross
sections contributing to ΓSSB are defined by the ioniza-
tion cross sections provided that the formation of a cation
leads to a strand break. These processes are being stud-
ied theoretically and experimentally [19, 42, 65, 83–85].
Their typical cross sections vary, but the cross section for
a SSB as a consequence of DEA for about 1-eV electrons
can be up to 10 nm2 [86, 87] per plasmid DNA, which
can be converted to 3 × 10−2 nm2 per single twist and
therefore ΓSSB = 10−3. For higher energy electrons, ion-
ization of a DNA molecule does not necessarily lead to a
SSB and many pathways are being discussed. Neverthe-
less, reported SSB yields at higher electron energies are
of the same order (if not higher) as those for low-energy
electrons [19], which once again gives ΓSSB ≈ 10−3.
ion's path, the number of SSBs is given by the integral,
dNSSB
dζ
= ΓSSB
F(ρ)nt2πρdρ ,
(24)
(cid:90) ∞
FIG. 8: (Colour online) Fluences of secondary electrons pro-
duced by a single 12C6+ ion in the vicinity of a Bragg peak
through a cylinder enwraping a DNA twist are shown with re-
spect to the distance of the cylinder from the ion's path. Two
different orientations (parallel and perpendicular) are shown
as well as MC simulations for the perpendicular case. In the
inset, the geometry for the calculation of fluence through a
cylinder enwraping a DNA twist is shown. In this figure, the
cylinder is perpendicular to the ion's trajectory and symmet-
ric with respect to the plane of incidence.
0
2. Calculation of the number of DSBs
where ΓSSB is the probability that an electron incident
on a DNA twist induces a SSB and nt is the number
density of DNA twists (i.e., cylinders). Since the spatial
dependence of nt is unknown, it is reasonable (in the
first approximation) to assume that it is constant. The
fluence F(ρ) for carbon ions at the Bragg peak, obtained
in Sec. IV C, can be substituted in the integral (24). This
gives us an estimate of
where φ = (cid:82) ∞
dNSSB
dζ
= ΓSSBntφ ,
(25)
0 F(ρ)2πρdρ = 1.1 × 103 nm2. The value
of φ is obtained using a simple diffusion model that con-
tains two parameters, the mean free path l (assumed to
be the same for all electrons) and the ratio of elastic
and inelastic cross sections γ. The third input in this
value is dNe
dζ . This number can be calculated using ion-
ization cross sections discussed in Sec. III B. However,
the number calculated from Eq. (25) only includes the
The estimate of the number of DSBs is more ambigu-
ous than that of SSBs. This is mainly due to the lack of
understanding of the mechanism of producing this lesion.
Many works [16, 18] suggest that a DSB is the result of
the action of a single electron that dissociatively attaches
to a DNA molecule. The dissociative attachment is con-
sidered to be an important pathway of SSBs at very low
energies and in about one out of five such incidences, a
DSB takes place due to the interactions with the debris
of a SSB [18]. Alternatively, DSBs can be due to two
separate SSBs on opposite strands. This may be possi-
ble if the number density of secondary electrons is high
enough. It is also possible that double ionization events
play a significant role [20]. Such events create a high
local number density of low energy electrons at a con-
siderable distance from the ion's path and if this occurs
in the vicinity of a DNA twist, at least two of the three
electrons involved in a double ionization event may be
incident on the same twist. This depends on the values
of the cross sections for double ionization. The probabil-
ity of ICD-effects on DNA molecule and water molecules
adjacent to it may also be an important factor [20, 88].
Regardless of the pathway for DSBs, for a given ion in
a given medium, the ratio of yields of DSBs and SSBs
(per unit length of ion trajectory) is fixed and dose inde-
pendent unless tracks of different ions interact. Indeed,
each ion's track is determined by the type of ion and an
increased dose just means an increase in the density of ion
tracks. Only after some critical value of dose is reached,
do the tracks start overlapping. Only then can the de-
pendencies of yields of SSBs and DSBs on dose become
not proportional to each other. These conditions are not
being observed in the analysed experiments or in ther-
apy9, however, if laser-driven ion beams are used [89],
track interaction effects may become important.
Therefore, the DSB yield can be calculated as a sum of
two terms, the first of which represents the events where
SSBs are converted to DSBs and the second accounts for
DSBs due to separate electrons. In order to calculate the
second term, the average number of SSBs per twist, N ,
can be introduced as10
N = ΓSSBF(ρ) .
(26)
Then, the probability of a DSB due to two separate elec-
trons in this twist is given by 1
term in the DSB yield is given by the integration over
the volume similar to Eq. (24). Thus, the estimate for
DSBs is given by
2N 2 exp [−N ]. The second
(cid:90) ∞
(cid:90) ∞
0
dNDSB
dζ
= λΓSSBnt
+
nt
2
0
F(ρ)2πρdρ
N 2 exp [−N ]2πρdρ,
(27)
where N is given by Eq. (26) and λ is a fraction of SSBs
converted to DSBs, i.e., the number of DSBs due to the
action of a single electron.
At this point the phenomenon-based approach can be
related to experiments. If real tissue is irradiated, one
can only find the percentage of cells surviving.
If this
value is measured as a function of dose, the survival curve
is obtained as a result. Many interactions on sub-cellular,
cellular, or even at the organismic level may affect the
survival curve. In in vitro experiments on cell cultures,
elimination of some of these interactions allows, e.g., syn-
chronizing cell cycles, control over the environment, etc.
Still, there are no direct ways of relating cell death to,
e.g., DSBs produced by secondary electrons. Therefore,
the comparison with experiments on DNA molecules ir-
radiated with ions is the most appropriate.
3. Comparison with experiments on plasmid DNA
15
Of all the experiments investigating DNA molecules ir-
radiated with ions, the study of plasmid DNA is the most
valuable, since there are reasonably reliable ways to dis-
tinguish the intact molecules from those with a SSB and
from those with a DSB. Another important feature is
that the effects of DNA damage observed in these exper-
iments are not affected by the biological effects of repair
that take place in living cells. This allows for a more
pure comparison.
An undisturbed plasmid is a closed loop of a super-
coiled DNA molecule [90]. This loop contains a given
number of base pairs, e.g., in experiments described in
Ref. [91] plasmid DNA pBR322 irradiated with carbon
ions contains 4361 bp. The characteristic size of this
molecule is about 100 nm. If such a molecule experiences
a SSB, it becomes "circular" or just a loop without the
supercoil structure. A DSB makes the plasmid "linear"
since both of its strand are broken. These structural con-
formations can be distinguished using electrophoresis or
high-performance liquid chromatography [19, 91]. This
allows the measuring of SSB and DSB yields experimen-
tally. In one of the experiments described in Ref. [91],
plasmid DNA was dissolved in a 600 mmol/l solution of
mannitol in water. Mannitol serves as a radical scav-
enger so their contribution to DNA damage may be ne-
glected. This is adequate for the theoretical treatment
(Sections IV C 1 and IV C 2), which only includes sec-
ondary electrons.
The results of experiments of Ref. [91] are shown in
Fig. 9 with dots. They represent the probabilities for
two outcomes after an irradiation with carbon ions at the
spread-out Bragg peak. The first outcome (open squares)
is for the plasmid to become open circular (not super-
coiled), associated with a SSB. There is a reported prob-
lem with the quality of the data resulting in the probabil-
ity corresponding to SSBs not starting from zero at a zero
dose [91]. This means that some plasmids are either not
supercoiled to begin with or appear as such in the elec-
trophoresis. This probability remains elevated by about
the same value throughout the dose range. In order to
compare these data with our calculations, the zero-level
probability of the SSB yield was subtracted in order to
"clean" the data. These data points are shown with filled
squares. The second outcome is for the plasmid to be-
come linear, associated with a DSB and is shown with
filled circles. These probabilities (filled squares and cir-
cles) are monotonically increase with dose with the SSB
dependance being slightly non-linear. In order to explain
these data using the multiscale approach, let us start with
the dose dependance.
9 In this section only effects of secondary electrons are discussed.
The situation may be different when radicals are included, see
Sec. IV B.
10 This number is a part of the integrand of Eq. (24).
When a beam of carbon ions is incident on the plasmid
solution, there is a dose-dependent probability that ν ions
will traverse through a plasmid. This probability is given
16
where ¯xp is the average length of an ion's path through
the plasmid. The subscript "e" indicates that this yield
is only due to secondary electrons. Each term of this
sum is a product of the number of SSBs per unit length
of trajectory of a single ion, the length of this trajectory
through the plasmid and the number of ions traversing
the plasmid. The length of a trajectory, ¯xp, is equal
to the average chord length of a sphere, representing a
plasmid, which is about 0.78 of its diameter. If exactly ν
ions pass through the plasmid, this length is multiplied
by ν. This is the first term in the sum of Eq. (30). Then
the factor Pν gives the probability that ν ions are passing
through the target. Hence, the whole sum multiplied by
¯xp determines the average length of tracks through the
plasmid.
The sum in Eq. (30) does not include interactions of
different ions that could occur if trajectories of two or
more ions are so close that the same twist of a DNA
molecule could be hit with electrons originating from the
different ions. The probability of such an interference can
be estimated. Since the range of 50-eV electrons in liquid
water is about 10 nm, the two ion's trajectories must be
within 20 nm, for the interference to occur. Then the
estimate is obtained from Eqs. (28) and (29) with ν = 2
and Ap = π × 102 nm2. For the maximal dose of 300 Gy
used in Ref. [91] the resulting probability is 5 × 10−6.
This number is very small compared to the probability
that one ion will pass through the plasmid at this dose
(equal to 0.3) or even that two ions will pass through
it (equal to 0.02). Therefore, the interference term in
Eq. (30) can be neglected.
in the sum (cid:80)∞
The only term of Eq. (30) that depends on dose is Pν,
therefore the dose dependence of the yield is contained
ν=1 νPν. This dependence is not unique
for the yield of SSBs. The same sum appears in all cal-
culations, provided that the damage due to each ion is
localised in its track and the tracks do not interfere. The
dependence of this sum on dose is asymptotically expo-
nential at large values of Nion. This means that on a
semi-logarithmic plot the dose dependence will be asymp-
totically a straight line. This will be seen below in the
analysis of survival curves in Sec. VI B.
The numbers relevant to the experiments of Ref. [91],
such as Ap = 7.8 × 103 nm2 and ¯Se = 189 eV/nm, sub-
stituted to Eq. (29) give Nion = 2.6 × 10−4d with the
dose in Gy. This means that even at the highest dose of
300 Gy used in Ref. [91] Nion (cid:28) 1. However, Ref. [91]
gives the dose dependence of the probability of a SSB per
plasmid rather than yield. This probability is given by
Poisson statistics,
PSSB,e = YSSB,e exp [−YSSB,e]
SSB,e exp [−YSSB,e] ,
Y 2
+
1
2
(31)
FIG. 9: (Colour online) Probabilities for SSBs and DSBs in-
duced in plasmid DNA by secondary electrons as a function
of dose. Dots correspond to experiments [91]: open squares to
the original SSBs, filled squares to the "cleaned" SSBs, and
filled circles to DSBs. Calculated probabilities are shown with
lines. Solid line corresponds to the probability of SSBs calcu-
lated using Eq. (30). The dashed line depicts the probability
for DSBs calculated using Eq. (32).
by the Poisson distribution:
Pν =
N ν
ion
ν!
exp [−Nion] ,
(28)
where Nion is the average number of ions passing through
the cross sectional area of a plasmid, Ap ≈ 7.8× 103 nm2
[90]. The average number of ions passing through this
area is equal to the ratio of this area to the average area
per ion. The average area per ion, A, can be calculated
if a uniform distribution of ions in the beam is assumed.
Then the dose is equal to the LET (which is associated
with the average for the Bragg peak stopping power due
to ionization processes, ¯Se) divided by the average area
per ion, i.e., d = ¯SeA . Then, Nion is given by:
Nion =
ApA =
Ap
¯Se
d .
(29)
In Ref. [91], the average LET over the spread-out Bragg
peak, ¯Se is 189 ± 15 eV/nm. This includes energy strag-
gling effect along the ion's trajectory. Then, the number
of SSBs that are likely to be induced in a plasmid, i.e.,
SSB yield per plasmid is given by the sum,
YSSB,e =
dNSSB
dζ
¯xp
νPν ,
(30)
where the first term corresponds to a single SSB in the
plasmid DNA and the second term corresponds to two
SSBs on the same strand. The fit of Eq. (31) to the
∞(cid:88)
ν=1
dζ
¯xp = 0.12. If ¯xp ≈ 75 nm, dNSSB
probability of the SSB dependence on dose, shown in
dζ ≈
Fig. 9, gives dNSSB
1.6 µm−1, then comparing this with Eq. (25) and taking
nt = 5.6× 10−2 nm−3, we obtain an estimate for ΓSSB =
1.9 × 10−3, which is larger than the the value estimated
in Sec. IV C 1 on the basis of experimental results by the
factor of 1.9.
Now the comparison for DSBs can be made. Similar
to Eq. (30), the number of DSBs induced in a plasmid (a
DSB yield per plasmid) is given by the sum,
∞(cid:88)
ν=1
YDSB,e =
dNDSB
dζ
¯xp
νPν ,
(32)
and the probability of a DSB per plasmid is given by,
PDSB = YDSB,e exp [−YDSB,e] .
(33)
dζ
A fit of Eq. (33) to the probability of the DSB depen-
dence on dose (for Ref. [91]) gives dNDSB
¯xp = 0.015. The
substitution of ¯xp ≈ 75 nm gives dNDSB
dζ ≈ 0.2 µm−1.
Then comparing this with Eq. (27) and taking nt =
5.6 × 10−2 nm−3, we obtain an estimate for λ = 0.15,
which is in reasonable agreement with the values between
0.1 and 0.2 for different electron energies [16, 18, 19].
Thus, the comparison of our model for the effect of sec-
ondary electrons with the results of Ref. [91] for a plas-
mid DNA solution in the presence of radical scavengers
is reasonable.
Some comments regarding these calculations should be
made. First, as has been noted in Sec. IV C 1, the number
of secondary electrons is underestimated. This happens
because in our calculations only the electrons ejected by
ions were included, missing those ejected in the process of
secondary ionization by electrons. The correction for this
number will increase fluence, but will not affect the dose
dependence. Since the actual fluence will then be larger
(by the factor of about two [11]), ΓSSB will be smaller (by
the same factor, i.e., closer to 10−3 (see Sec. IV C 1). The
second issue is that the treatment of a supercoiled plas-
mid as an object with uniformly distributed chromatin
may be a little far-fetched. Also, if a plasmid suffers a
single strand break, its size increases by a factor larger
than two and then it may be a target for another ion.
Nevertheless, the comparison that was just made is quite
reasonable and encouraging for further steps in the as-
sessment of radiation damage.
D. Damage of plasmid DNA in the presence of free
radicals
17
direct mechanism involving secondary electrons are so
important, multiple experiments [91] indicate that the
damage due to radicals exceeds that due to direct elec-
trons.
The damage done by radicals can be calculated in the
same fashion as the damage due to electrons if their flu-
ence, Fr, and the probability of producing a strand break
on impact with a radical, Γr, are known. However, the
analysis of damage in a plasmid DNA solution in pure
water, where the action of radicals is not abated (studied
in Ref. [91]) shows that the picture of the dose depen-
dence is quite different from the one for secondary elec-
trons that was discussed above. The main difference of
this picture is a strong dependence on dose for the same
conditions as in the experiment with mannitol, compare
Figs. 9 and 10. In Eq. (30) the probability of a SSB per
segment of an ion's trajectory, dNSSB
, is independent of
dose. The dependence on dose comes from the probabil-
ity Pν that a certain number of ions traverse through the
plasmid. This probability does not change in the case of
radicals while the dose dependence does.
dζ
In Fig. 10, measured SSB and DSB probabilities de-
pending on dose are shown with dots. It is obvious that
in this case the curves are not proportional to each other.
This means that the interference term, absent in Eq. (30),
plays an important role in the case of radicals. One way
to explain this phenomenon is to infer that the radicals
are distributed much more broadly than secondary elec-
trons. Since a uniform distribution of ions in the beam is
assumed, it is reasonable to assume (in the first approxi-
mation) that radicals are also distributed uniformly, such
that their number density is proportional to the dose.
Then the probability that they inflict a SSB on a plas-
mid is
1
2
N 2
r exp [−Nr] ,
PSSB,r = Nr exp [−Nr] +
(34)
where Nr = ΓrFr is the average number of SSBs due to
a given number density of radicals per plasmid. The sec-
ond term includes the events when two SSBs take place
on the same strand or are too far from each other to
cause a DSB. This, however, is not sufficient since sec-
ondary electrons are still present in the experimental re-
sults shown in Fig. 10. Then, the probability of a SSB is
given by
PSSB = PSSB,r(1 − PSSB,e)
+(1 − PSSB,r)PSSB,e +
1
2
PSSB,rPSSB,e .
(35)
Free radicals play a very important role in DNA dam-
age [13, 14]. Their role has been especially emphasized
in the case of irradiation with photons, where they are
the main instrument of the so-called indirect damage of
DNA [92]. However, even in the context of IBCT, where
This expression can be compared with the experiment
and a fit gives the value of Nr = 0.012d, where the dose
is in Gy. The results of this comparison are shown with
a solid line in Fig. 10. They reasonably agree with the
experiment at least for doses less than 250 Gy.
electrons and radicals, i.e., the comparison of Figs. 9 and
10 is that the effect of radicals on DSBs is quite substan-
tial.
18
1. Comparison with repair foci observations
It is also possible to apply Eqs. (27) and (37) to the
observed phosphorylated histone variants H2AX, which
accumulate near DNA DSBs in cell nuclei. These accu-
mulations are called foci and their distribution along the
ion's track can be translated to the number of DSBs per
unit length of trajectory.
Reference [93] reports the foci distributions in the nu-
clei of human lung adenocarcinoma epithelial cells (A549)
after they were irradiated with carbon ions. The cross
sectional area of an A549 nucleus is 83 µm2 [94]. Given
the approximate number of base pairs in human DNA
(3.2×109) and assuming an uniform distribution of chro-
matin, the DNA twist density is nt = 6.5 × 10−4 nm−3.
The average distance between the projections of foci on
the ion's trajectory observed in Ref. [93] is 2.2 µm, i.e.,
dNDSB/dζ ≈ 0.5 µm−1. The integral fluence,
(cid:90) ∞
F(ρ)2πρdρ ,
0
is in this case smaller, since the carbon ions interacting
with cells were of energy 52 MeV/u, which is far from
the Bragg peak. The value of LET (not measured in
Ref. [93]) can be calculated using the methods developed
in Sec. III. It is about 50 eV/nm, so the integral fluence
(linear with respect to LET) can be estimated to be about
25% of the value used in Sec.IV C 1, i.e., about 290. Then
assuming only the action of secondary electrons ΓSSB can
be estimated from Eq. (32). It turns out to be 6.5×10−3,
which is about 3.4 times higher than our estimate from
experiments with plasmid DNA. The discrepancy can be
attributed to the unaccounted action of radicals. The
radicals are included in the analysis of cell survival in
Sec. VI B. Here, it is pertinently to give an estimate
for the number of produced radicals to be (far from the
Bragg peak) about two times larger than the number of
secondary electrons ejected by ions (if at least one radical
is produced by a secondary electron). Therefore, it is
plausible that the production of DSBs by electrons and
radicals combined can explain the number dNDSB/dζ ≈
0.5 µm−1 observed experimentally.
E. Accounting for δ-electrons
The effects due to secondary electrons with energies of
100 eV and above or the so-called δ-electrons should be
discussed separately. These particles cannot be included
in the diffusion model because their cross sections are
strongly peaked in the forward direction, their mean free
paths exceed 1 nm and they lose their energy ionizing the
FIG. 10: (Colour online) Probabilities for SSBs and DSBs in-
duced in plasmid DNA with secondary electrons and radicals
as a function of dose. Dots correspond to experiments [91]:
squares to the "cleaned" SSBs, and circles to DSBs. Cal-
culated probabilities are shown with lines. Solid line corre-
sponds to the probability of SSBs calculated using Eq. (35).
The dashed line depicts the probability for DSBs calculated
using Eq. (37).
The expression for DSBs is obtained similarly:
PDSB,r =
N 2
r exp [−Nr]
1
2
and
PDSB = PDSB,e(1 − PDSB,r − 1
2
+PDSB,r(1 − PDSB,e − 1
2
PSSB,rPSSB,e)
PSSB,rPSSB,e(1 − PDSB,e − PDSB,r) .
+
1
2
PSSB,rPSSB,e)
(36)
(37)
The results for DSBs are shown in Fig. 10 with a dashed
line. Once again, a reasonable agreement for doses less
than about 150 Gy can be observed. As the dose in-
creases further, the higher order effects that are not in-
cluded in Eqs. (34-37) contribute to the number of ob-
served DSBs and this number is thus underestimated.
The comments after previous section are still relevant,
but it is also important to add to them the discussion
about the spatial distribution of the fluence of radicals. It
has been assumed to be uniform, but it was not discussed
why it could be such. A possible mechanism that can be
much more effective than diffusion is the collective trans-
port due to shock waves to be discussed in Sec. V B. One
inference from the comparison of the results for secondary
medium and are capable of producing a number of extra
electrons and creating a cluster-damage site. In order to
estimate corrections due to δ-electrons, several quantities
pertinent to these particles need to be analysed.
The first is the mean free path. According to Ref. [37],
both the elastic and inelastic mean free paths of 100-eV
electrons are about 1 nm. If such an electron is ejected
in the most likely direction according to the binary in-
teraction model [48], about 70◦, this electron will start
losing energy within 1 nm of the ion's trajectory. Even
if it produces more electrons than a sub-50-eV electron,
they will not spread much further than them. Because
of the kinematic limit, for an ion in the Bragg peak re-
gion, energies of ejected secondary electrons are below
0.7 keV. These electrons with elastic mean free path of
about 4.5 nm are emitted in the forward direction, and it
can be shown that the maximal distance between the first
collision and the ion's path is 1.6 nm and it is reached
by the electrons of energies 400 − 500 eV. There is no
way that further transport can carry further generations
of electrons far beyond the 10 nm distance off the ion's
path. In addition, the probability of producing a 400-eV
secondary electron is only about 0.02 of that producing
a 50-eV electron. Therefore, even though δ-electrons are
not included in the random walk approach, the location
of their effect is by and large overlapped with that of
sub-50 eV secondary electrons. The number of electrons
ejected as the consequence of ionization by δ-electrons
can be estimated from energy conservation and these
electrons have already been effectively included in the
random walk, since dNe/dζ was obtained from the value
of the stopping power, Se.
Still another possibility exists for δ-electrons to affect
the discussed scenario. If a much more energetic electron,
i.e., with energy larger than 20 keV, then with the mean
free path of the order of 100 nm, it can cause damage
elsewhere. Moreover, these electrons cannot be ejected
in the Bragg peak region, since the required ion energy
must be over 9 MeV. The probability of such events is
very small; it is less than that of emitting a 50-eV electron
by a factor over 106. Therefore, this possibility is realised
so rarely that it can be neglected.
V. THERMOMECHANICAL EFFECTS
Thus far, the energy loss by incident ions, the transport
of produced secondary particles, and the radiation dam-
age induced by these particles have been discussed. The
transport, described by diffusion or MC simulations, is
that of the ballistic electrons, radicals, etc.
in a static
medium. This transport does not include the whole
physical picture because propagating secondary particles
transfer the energy further, making the medium hot and
dynamic.
Energy relaxation in the medium has been studied in
Ref. [26], where the inelastic thermal spike model was
applied to liquid water irradiated with carbon ions. This
19
model has been developed to explain track formation in
solids irradiated with heavy ions and it studies the energy
deposition to the medium by swift heavy ions through
secondary electrons [95–105]. In this model, the electron-
phonon coupling (strength of the energy transfer from
electrons to lattice atoms) is an intrinsic property of the
irradiated material.
The application of the inelastic thermal spike model
to liquid water predicted that the temperature increases
by 700-1200 K inside the hot cylinder by 10−13 s after
the ion's traverse [26]. However, within this model, only
coupled (between electrons and atoms of the medium)
thermal conductivity equations are solved, while the fur-
ther dynamics of the medium is missing. This dynamics
is the consequence of a rapid pressure increase inside the
hot cylinder around the ion's path up to 104 atm, while
the pressure outside of it is about atmospheric. Since
the medium is liquid, this pressure difference prompts
rapid expansion, resulting in a shock wave, which has
been analysed in Refs. [28, 29, 106, 107].
A. Hydrodynamic expansion on the nanometre
scale
The problem of the expansion of the medium driven by
the high pressure inside the hot cylinder is in the realm
of hydrodynamics and it has been thoroughly analyzed
in Ref. [28].
It has been shown that the expansion is
cylindrically symmetric. If the ratio of pressures inside
and outside of the hot cylinder is high enough, as hap-
pens for large values of LET, the cylindrical expansion
of the medium is described as a cylindrical shock wave,
driven by a "strong explosion" [108]. For an ideal gas,
this condition holds until about t = 1 ns, but in liquid
water the shock wave relaxes much sooner. In Ref. [106]
the molecular dynamics simulations of liquid water ex-
pansion showed that the shock wave weakens by about
0.5 ps after the ion's passage.
The hydrodynamic problem describing the strong ex-
plosion regime of the shock wave is self similar. Its so-
lution, as well as its mechanical features and limitations,
is very well described in Refs. [108–110].
In Ref. [28],
the solution for the cylindrical case has been reproduced
and analyzed in order to apply it for the nanometre-scale
dynamics of the DNA surroundings. In this section, only
the results pertinent to the further discussion of biodam-
age are presented.
The self similar flow of water and heat transfer depend
on a single variable, ξ. This variable is a dimensionless
combination of the radial distance, ρ, from the axis, i.e.,
the ion's path, the time t after the ion's passage, the en-
ergy dissipated per unit length along the axis, which is
equal to the LET per ion, Se, and the density of undis-
turbed water, = 1 g/cm3. This combination is given
by
(cid:20)
(cid:21)1/4
t
Se
,
ξ =
√
ρ
β
(38)
20
where β is a dimensionless parameter equal to 0.86 for
γ = CP /CV = 1.222 [28]. The radius and the speed of
the wave front are given by
(cid:20) Se
(cid:21)1/4
(cid:21)1/4
(cid:20) Se
(39)
,
(40)
R = ρ/ξ = β
√
t
and
u =
dR
dt
=
R
2t
=
√
β
2
t
respectively. It is also worthwhile to combine Eqs. (40)
and (39) and obtain the expression of the speed of the
front in terms of its radius R,
(cid:20) Se
(cid:21)1/2
u =
β2
2R
.
(41)
Using Eq. (41), pressure P at the wave front can be ob-
tained as
P =
2
γ + 1
u2 =
1
γ + 1
β4
2
Se
R2 .
(42)
Then, one can solve the hydrodynamic equations in or-
der to obtain the expressions for speed, pressure, and
density in the wake of the shock wave, i.e., behind the
wave front [28].
The following intriguing questions have been raised
in Refs. [26, 28]. What can such a shock wave do to
biomolecu-les such as DNA located in the region of its
propagation through the medium; can it cause biodam-
age by mechanical force? The forces acting on DNA seg-
ments were predicted to be as large as 2 nN, which is
more than enough to break a covalent bond, causing a
strand break; however, these forces are only acting for a
short time and it remained unclear whether this is suf-
ficient to cause severe damage to DNA molecules. The
other question is: how significant can the transport due
to the collective flow of this expansion be compared to
the diffusion of secondary particles?
B.
Investigations of the effects of shock waves
using simulations
There are several effects of shock waves directly or in-
directly related to biodamage or cell death. The first
effect is the direct thermomechanical damage of a DNA
molecule as a result of interaction with the shock wave,
and it can be explored using molecular dynamics (MD)
simulations [29]. The second effect is a similar rupture
of covalent bonds in water molecules leading to the extra
FIG. 11: (Artistic view, colour online) The cylindrical shock
wave front in water (on the right; the ion's path is the axis
of this cylinder, perpendicular to the figure plane) interacts
with a DNA segment on the surface of a nucleosome (on the
left). The bright dot indicates the place where interactions
occur. The medium is very dense following the wave front
and is rarefied in the wake.
production of radicals. This effect is still being studied.
The third effect is the propagation of reacting species as
a result of collective motion initiated by the shock wave.
The fourth effect, which is also under investigation, is
related to the damage of a cell membrane due to ac-
tion of the shock wave [111]. Irreversible damage of the
cell membrane may be lethal for the cell. In Ref. [29],
the effects of the interaction of the shock wave formed
in a liquid water medium following the traverse of an ion
through this medium at different values of LET have been
studied. The values of the LET were used as parameters
describing the energy propagated by the shock wave, and
the physical conditions in which the action of the shock
wave is significant or even dominant for radiation damage
assessment were analysed.
In eukaryotic cells, DNA molecules are packed into
chromatin fibers. A nucleosome, a histone-protein oc-
tamer wrapped about with a DNA double helix, is the
primary structural unit of chromatin. Therefore, the MD
simulations were focused on the interaction of the cylin-
drical shock wave originating from ion's path with a frag-
ment of a DNA molecule situated on the surface of a nu-
cleosome. The artistic picture of this interaction is shown
in Fig. 11 [29]. The ion's path is perpendicular to the pa-
per plane as well as the axis of the nucleosome disk. The
simulations were done for four values of LET, 0.9, 1.7, 4.7,
and 7.2 keV/nm, corresponding to the predicted values
of LET at the Bragg peak for carbon, neon, argon, and
iron ions, respectively [69]. Among these, carbon ions are
currently the most used ions for heavy-ion therapy. Iron
ions are important for space-mission safety assessments.
Between these are neon and argon ions, which are being
considered for medical applications; they are used in a
number of experimental studies [4].
C. Simulations of the direct shock wave effect
The estimate for the radius of the hot cylinder can be
obtained from the analysis of the diffusion of secondary
electrons from the ion's path. This radius is associated
with the average radius within which secondary electrons
lose most of their energy [20]. At a time t, a secondary
electron originating from the ion's path is most likely to
be situated at a distance from the path, equal to ¯ρ =
(cid:82) ρP (t, r)d3r, where P (t, r) is given by Eq. (18). This
integral is equal to l(cid:112)πk/6, where l is the average elastic
mean free path of electrons ejected at the ion's path and
k is an average number of elastic collisions they undergo
before they lose energy in inelastic collisions. For Se =
0.9 keV/nm, the estimate for ¯ρ is 1 nm and this was taken
to be the radius of the hot cylinder.
The simulations show a noticeable distortion due to the
shock wave at 10 ps after the expansion starts. This dis-
tortion comprises the rupture of the secondary structure
of the most exposed parts of the DNA molecule, mani-
fested by the nonnative orientation of DNA nucleotides.
Many hydrogen bonds are broken, and the bases are lo-
cated outside the DNA double helix. However, these dis-
tortions are reversible, while our main interest is in the
investigation of more permanent covalent bond breaking
events.
In order to study whether the covalent bonds in the
DNA backbone can be broken during the shock wave ac-
tion, the energy temporarily deposited to these bonds
was calculated. If this energy exceeded the binding en-
ergy of a given bond, it was assumed that thermomechan-
ical stresses in the DNA fragment were sufficiently high
to break the bond. The corresponding binding energies
are referred to as thresholds for breaking the DNA back-
bone covalent bonds; they are between 3 and 6 eV [112].
Even though the thresholds may be lower (even as low as
0.3 eV) in the environment as a consequence of the ion's
passage [41], high thresholds were kept in order to obtain
conservative estimates for the direct action of the shock
wave.
The analysis of MD simulations performed for four val-
ues of LET (0.9, 1.7, 4.7, and 7.2 keV/nm) gives the
distributions of the bond energy records. These records
can be represented by a histogram that assigns to ev-
ery interval of energy (ε, ε + δε), the number of records
corresponding to the bond energies from this interval.
For each value of LET, the bond energy distribution was
constructed. These distributions (normalised to the total
number of records Nr for each value of LET) are shown
in Fig. 12, where ln(1/NrdN/dE) is plotted vs. the cor-
responding energy interval.
21
For the most part, these distributions correspond to
Boltzmann distributions with different temperatures and
they can be fitted as
(cid:20)
(cid:21)
1
N0
dNsw
dε
=
1
N0
δNsw
δε
=
1
kBT
exp
− ε
kBT
,
(43)
where δε = 0.01 eV is the width of the energy bin, δN is
the number of records with energy between ε and ε + δε,
deposited in selected covalent bonds, the normalisation
constant N0 = 2.17 × 104 and the temperature T are
parameters; kB is the Boltzmann constant. Both param-
eters, N0 and T , are determined from the fitting of the
distributions obtained from the MD simulations.
T is the temperature corresponding to the thermal
parts of the distributions (for the four values of LET)
shown as linear fits in Fig. 12. The values of T are 870,
1130, 2580, and 3970 K, indicating that the temperature
increase above the temperature of the medium before the
interaction with the shock wave, T0 =310 K (correspond-
ing to a biological system), is directly proportional to Se,
T − T0 = αSe ,
(44)
where α = 494 K·nm·keV−1.
The average number of breaks can be estimated by
integrating Eq. (43) over energies ε, exceeding a chosen
threshold ε0:
(cid:20)
(cid:21)
Nsw =
dNsw
dε
dε = N0 exp
− ε0
kBT
.
(45)
(cid:90) ∞
ε0
Since the parameters N0 and T are fitted, this procedure
allows us to predict the number of these over-threshold
bond energy records for any value of LET. These numbers
correspond to the number of bond breaks caused by the
ion's passage.
In order to compare the number of strand breaks due
to the shock wave action, to chemical effects, the prob-
ability of a strand break based on bond-breaking events
has to be calculated. From the predicted number of
bond breaks Nsw, Poisson statistics yields the probabil-
ities for the exact number, ν, of strand breaks to occur,
P (ν) = exp (−Nsw) N ν
sw/ν!. Then, the probability Psw
of at least one strand break in a given segment of a DNA
molecule is equal to 1 − P (0), i.e.,
Psw = 1 − exp (−Nsw) .
(46)
The dependence of this probability on LET for different
thresholds is shown in the inset of Fig. 12, where this
probability is compared to the probability of producing
a single strand break owing to chemical effects in a similar
DNA segment located at the same distance from the ion's
path. The probability of SSBs due to chemical effects,
Pch, is estimated using the argument of Section IV C 1
22
D. Transport of reactive species by the radial
collective flow
The study done in Ref. [28] suggests that a consider-
able collective radial flow emerges from the hot cylinder
region of medium. The maximal mass flux density car-
ried by the cylindrical shock wave is given by f u, where
f = γ+1
γ−1 is the matter density on the wave front. This
expression is proportional to u and its substitution from
√
Eq. (41) yields that the mass flux is proportional to the
Se. This flux density is inversely proportional to radius
ρ and is linear with respect to the
Se. It sharply drops
to zero in the wake of the wave along with the density. A
sharp rarefaction of the volume in the wake of the wave
follows from the results of Ref. [28]. This is the effect
of cavitation on a nanometer scale and due to this ef-
fect the water molecules of the hot cylinder along with
all reactive species formed in this cylinder are pushed
out by the radial flow. Such a mechanism of propaga-
tion of reactive species, formed within the hot cylinder,
is competitive with the diffusion mechanism, studied in
MC simulations done using track structure codes [37].
√
Intriguingly, the cylindrical shock wave accomplishes
the transfer of reactive species such as hydroxyl and sol-
vated electrons, which play important roles in chemical
DNA damage [14, 19, 41] much more effectively than the
diffusion mechanism. Indeed, the time at which the wave
front reaches a radius ρ can be derived from Eq. (39)
as it is equal to (ρ2/β2)(cid:112)/Se. This time has to be
to ((cid:112)/Se)D/β2. For all relevant species, the diffusion
above ratio is less than 10−3/(cid:112)Se(keV/nm), which is
compared to diffusion times, which can be estimated for
different reactive species as ρ2/D, where ρ is the dis-
tance from the ion's path and D is the corresponding
diffusion coefficient. The ratio of these times is equal
coefficient is less than 10−4 cm2/s [113]. Therefore, the
much less than unity even for protons. For instance, for
carbon ion projectiles, the wave front reaches 5 nm from
the path in 2.8 ps after the ion's traverse, while hydroxyl
radicals reach the same distance via the diffusion mecha-
nism in about 9 ns, a more than 3000 times longer time.
In fact, the lifetime of hydroxyl free radicals is shorter
than 5 ns [14, 31, 113], therefore the shock wave trans-
port may be the only means to deliver hydroxyl radicals
to distances farther than 3.5 nm of the ion's path.
The collective flow is expected to play a significant
role in the transport of reactive species at values of LET
that are large enough to produce a shock wave, even if
this wave is not sufficiently strong to cause covalent bond
ruptures. The analysis shows that even at small values
of LET, typical for the plateau region in the LET de-
pendence on depth (well before the Bragg peak), a shock
wave is formed; however it damps and becomes acoustic
at radii under 10 nm. At Se = 0.9 keV/nm shock waves
propagate further than 10 nm.
Thus, the effects following the local heating of the
medium in the vicinity of an ion's path are quite strik-
FIG. 12: (Colour online) The dependence of the logarithm of
the normalised number of the covalent bond energy records
for the selected DNA backbone region per 0.01 eV energy in-
terval on the bond energy for four values of LET: 0.9, 1.7, 4.7,
and 7.2 keV/nm, corresponding to the Bragg peak values for
ions of carbon, neon, argon, and iron, respectively. Straight
lines correspond to the fits of these distributions. In the inset,
the dependence of the probability of producing at least one
SSB in a 3-base-pair segment of a DNA molecule located be-
tween 1.5 and 2.2 nm from the ion's path, on LET. The shock
wave probability lines correspond to the estimates done using
Eqs. (45-46).
and is given by,
Pch = κ
Se
Se,0
,
(47)
where κ = (1.1 ± 0.5) × 10−3 and Se,0 = 0.9 keV/nm.
The inset of Fig. 12 predicts that for a given threshold,
the shock wave breaking effect starts at a certain criti-
cal value of LET. After that, the probability of direct
breaking increases with increasing LET steeply, readily
overcoming chemical effects that include interactions of
DNA molecules with free radicals, secondary electrons,
solvated electrons, etc. The inset of Fig. 12 indicates
that bond breaking due to the shock wave mechanism
starts (for the 3-eV threshold) at Se ≈ 4 keV/nm and
by 5 keV/nm it becomes the dominant effect in radia-
tion damage. Two smaller thresholds of 2 and 2.5 eV
are shown for comparison. This means that for heavier
than Ar ions propagating in tissue, the bond breaking in
DNA molecules located within about 2 nm of the ion's
path will primarily be due to the direct effect of shock
waves. The radius of dominance of this effect increases
with further increasing LET [29].
ing. The MD simulations of a shock wave on a nanome-
ter scale,
initiated by an ion propagating in tissue-
like medium, demonstrate that such a wave generates
stresses, capable of breaking covalent bonds in a back-
bone of a DNA molecule located within 1.5 nm from the
ion's path when the LET exceeds 4 keV/nm and this
becomes the dominating effect of strand breaking at11
Se (cid:38) 5 keV/nm. The LET of ∼ 4 − 5 keV/nm corre-
sponds to the Bragg peak values for ions close to Ar and
heavier in liquid water. Besides the dramatic effects at
such high values of LET, it was found that weaker shock
waves produced by carbon ions or even protons transport
the highly reactive species, hostile to DNA molecules,
much more effectively than diffusion.
The notion of thermomechanical effects represents a
paradigm shift in the understanding of radiation damage
due to ions and requires re-evaluation of the relative bi-
ological effectiveness. This is due to the collective trans-
port effects for all ions and direct covalent bond breaking
by shock waves for ions heavier than argon. These ef-
fects will also have to be considered for high-density ion
beams, irradiation with intensive laser fields, and other
conditions prone to causing high gradients of tempera-
ture and pressure on a nanometer scale.
VI. ESTIMATION OF RADIO-BIOLOGICAL
EFFECTS
The essence of results obtained in sections IV C 1
and IV C 2 is that for a given ion, the numbers of SSBs
and DSBs per unit length of the ion's path can be calcu-
lated. Or, alternatively, for a given DNA twist, the prob-
ability of the above lesions can be calculated. However,
those calculations are still far from predicting whether
the cell containing a given segment of DNA molecule will
die or survive. This question is largely in the realm of
biology, because of a variety of biological mechanisms,
which are activated following the creation of a lesion.
Nearby proteins are engaged in DNA repair and may or
may not be successful. Such an activity is marked by
the appearance of the so called foci that can be observed
experimentally [93, 114, 115]. These protein foci remain
visible until the repair is finished. If a lesion cannot be
repaired the cell containing this DNA molecule is likely to
die. There is a plethora of biological studies directed at
determining the probabilities of a successful DNA repair
depending on the extent of the damage.
It is established that a simple SSB is most likely to
be fixed within minutes after this lesion is produced.
DSBs can also be fixed, however, with smaller proba-
bility and there is also a chance that its repair (e.g., the
11 These values correspond to conservative estimates (ε0 =
3 eV) [29]. They may be much lower if the actual thresholds
appear to be smaller [41].
23
non-homolo-gous end joining (NHEJ) type of DSB re-
pair [116]) may not be successful. The probability of
repair is even smaller for multiply-damaged sites also
known as clustered DNA lesion or complex DNA dam-
age. A clustered DNA lesion is defined as the number of
DNA lesions, such as DSBs, SSBs, abasic sites, damaged
bases, etc., that occur within about two helical turns of
a DNA molecule so that, when repair mechanisms are
engaged, they treat a cluster of several of these lesions as
a single damage site [82, 117–121]. Let us start our dis-
cussion with the analysis of this type of damage in order
to arrive at a prediction of cell death/survival caused by
biodamage of a certain complexity that can be quantified
by the formalism described above.
A. Assessment of the complex DNA damage
When a DSB is induced due to secondary electrons,
as discussed in Sections IV C 2 and IV C 3, there is a
substantial probability (between 0.1 and 0.2 for plasmid
DNA) for a DSB to occur as a result of the interaction of
the molecule with a single electron. However, it is diffi-
cult to expect a clustered DNA damage site to be caused
by a single electron or another secondary particle, since
the distance between lesions in such a site can be too
large (more than 5 nm). Therefore, in Refs. [15, 27],
the complexity of DNA damage has been quantified by
defining a cluster of damage as a damaged portion of a
DNA molecule by several independent agents, such as
secondary electrons or radicals. Then, it is reasonable to
expect that the probability that the electrons or radicals
induce clustered damage is related to the fluence of these
agents on a given DNA segment in the same sense as the
probabilities of other types of lesions, such as SSBs or
DSBs, as discussed in Sections IV C 2 and IV C 3.
Therefore, it is natural to start with the calculation
of the number of clustered damage sites, produced by an
ion, per unit length of its trajectory, dNC
dζ , similar to what
was done for SSBs and DSBs in Eqs. (24) and (27). DNA
molecules are on the surface of nucleosomes and the latter
is modelled as a cylinder of radius 5.75 nm. Then, an
element of its lateral surface that enwraps two twists of a
DNA molecule serves as a target for secondary electrons
and radicals. This segment of the surface is 2.3 nm wide
(along the axis of the cylinder) and 6.8 nm long (along the
cylinder's circumference). For definiteness, this cylinder
is taken to be perpendicular to the ion's path, its axis to
be at a distance ρ from the path, and the cylinder to be
symmetric with respect to the plane containing the path
and vector ρ.
First, the fluence of secondary electrons on such a tar-
get, Fc, using Eq. (20) has to be calculated. The geome-
try for this problem is similar to that for the calculation
of the fluence on a cylinder perpendicular to the ion's
path, considered in Sec. IV C. The details for arranging
the integration are given in the Appendix. The integra-
tion gives Fc(ρ), which when multiplied by ΓSSB gives
24
where ns is the number density of sites, gives the num-
ber of clustered damage sites per unit length of the ion's
trajectory. Equation (49) can be numerically integrated
similar to Eqs. (24) and (27). For ΓSSB = 2 × 10−3 (ap-
proximately corresponding to that found in Sec. IV C 3,
0 Pc(ρ)2πρdρ = 3.2×10−2 nm2 (these calculations
ψ =(cid:82) ∞
are done for carbon ions near the Bragg peak). How-
ever, it is very inconvenient to to deal with the intro-
duced number density ns. If one notices that each nucle-
osome geometrically contains about 5.3 targets for clus-
tered damage sites, ns, can be exchanged for the num-
ber density of nucleosomes nn. With this consideration,
Eq. (49) becomes,
(cid:90) ∞
0
dNc
dζ
= (5.3nn)
Pc(ρ)2πρdρ = nnΨ,
(50)
where Ψ = 0.17 nm2 and nn is the number density of nu-
cleosomes in cell nucleus. This number density depends
on the type of cells and, in principle, on the cell cycle. In
our estimates it is assumed that nn is uniform through-
out the cell nucleus.
In Sec. VI B, Eq. (50) is related
to the dose and thus, the survival curves leading to the
calculation of the RBE are obtained.
However, the estimate given by Eq. (50) does not in-
clude the effect of radicals, which may be significant for
the overall assessment of radiation damage. In order to
include it, the ρ-dependent distribution of the probabil-
ity of inducing a SSB by radicals, i.e., Nr, introduced in
Sec. IV D, is needed. Since this distribution is not known,
one can start with a uniform distribution of radicals as
was done for the in vitro experiment in Sec. IV D, but
only within a certain radius from the ion's path. This
implies that the reactive species, formed in the nearest
proximity to the path, are transported by the shock wave
and their number density is nearly uniform inside the
cylinder that enwraps the decayed shock wave. The es-
timate of this radius for carbon ions at the Bragg peak
is 10 nm (see Sec. V D). Then Nr(ρ) can be added to
Nc(ρ) and the sum can be substituted into Eq. (49) for
ρ < 10 nm. Of course, in reality the boundary of radi-
cal propagation beyond the estimated radius will not be
sharp (as shown if Fig. 13) because of diffusion, but the
estimate of the softness of the boundary is not essential.
The value of Nr(ρ) may be affected by environmental
conditions in the tissue. As is known, the fixation of
damage due to radicals depends on the presence of oxy-
gen at the damage site [14, 31]. This means that even
before the enzymatic repair mechanisms are engaged, the
radical-induced damage may be fixed if oxygen is not
present, i.e., in hypoxic conditions. Then, the value of
Nr(ρ) is effectively reduced. The study of such a reduc-
tion for different concentrations of oxygen leads to the
calculation of the oxygen enhancement ratio (OER), an-
other important parameter for optimization of IBCT.
FIG. 13: (Colour online) The dependence of probabilities for
complex damage to be induced by secondary electrons on the
distance from the nucleosome to the ion's path. The solid
line represents the degrees of complexity larger than one; the
dashed line represents those larger than two. The dotted line
shows the inclusion of the effect of radicals uniformly dis-
tributed inside a 10-nm cylinder. This curve is plotted with
parameters discussed in Sec. VI B.
the average number of SSBs for a given DNA segment,
Nc = ΓSSBFc(ρ). Then it is assumed, as was suggested
in Refs. [15, 27, 39], that the degree of complexity of
damage is given by the number of agents that cause it12.
The sum of probabilities,
Pc(ρ) =
N ν
c
ν!
exp [−Nc] ,
(48)
∞(cid:88)
ν=2
gives the probability, Pc(ρ), that the damage complex-
ity at a given site is larger than or equal to two. This
probability is shown in Fig. 13.
The next step is the integration of this probability over
the volume of the cell nucleus with the number density of
such sites. More precisely, the integration over ρ is done
from zero to infinity, since Fc(ρ) rapidly decreases with ρ
and one does not have to worry about reaching the limits
of the cell nucleus. This integral,
(cid:90) ∞
0
dNc
dζ
= ns
Pc(ρ)2πρdρ = nsψ ,
(49)
12 In the first approximation, it is assumed that all lesions compris-
ing a clustered damage site occur with the same probability as a
SSB.
B. Obtaining the survival curves
A survival curve is the dependence of the probability
of cell survival on the absorbed dose of radiation. On one
side, it relates the goal with the means, i.e., it predicts the
dose that is necessary in order to achieve cell deactivation
with a desired probability. On the other side, it allows
comparing different modalities (photons, protons, heav-
ier ions, etc.) and thus allows one to optimize the choice
of therapy. This comparison is achieved via the calcula-
tion of the ratio of doses of different projectiles necessary
to achieve the same probability of cell survival. The ratio
of the dose due to photons to that for other projectiles is
called the relative biological effectiveness (RBE).
The assessment of RBE for ions, from the point of view
of the multiscale approach, starts from the calculation of
survival curves for a given type of cell irradiated with a
given type of ion. This means that for a given type of
cell and a given dose the probability of cell survival (or
death) has to be calculated. In the previous section, the
probability of cell death was related to the probability of
inducing a DNA lesion of a given complexity, so that it
is unlikely to be repaired with proteins. In this section,
the accomplishments of previous section are applied to
the calculation of survival curves.
The conditions of
irradiation are the same as in
Sec. VI A, i.e., the doses are small and the ion tracks
are not going to interfere. Then, for a given type of cell
and a given dose, the number of ions that traverse a cell
nucleus can be calculated. This, similar to our experience
with plasmid DNA in Sec. IV C 3, gives us the dose de-
pendence. The average number of complex DNA lesions
in the cell nucleus is given by the following expression,
similar to Eqs. (30) and (32),
∞(cid:88)
ν=1
Yc =
dNc
dζ
¯xnc
νPν(d) ,
(51)
where Pν(d), given by Eq. (28) is the probability that
ν ions traverse the cell nucleus and ¯xnc is the average
distance of the ion's traverse through the cell nucleus.
The average number of traversing ions, Nion, is given by
Eq. (29) with Ap replaced with the cross section of the
cell nucleus. The value of dNc
dζ
is taken from Sec. VI A.
It is worthwhile to apply this method to the calcula-
tion of the survival curve for A549 cells irradiated with
α-particles at ¯Se = 115 eV/nm, studied in Ref. [122].
The nucleosome number density is estimated to be 2.2 ×
10−4 nm−3. The average number of ions traversing such
a nucleus gives Nion ≈ 4d, where the dose is in Gy. This
is a much larger number than that in the case of plas-
mids even for doses not exceeding 2 Gy. This means that
a goodly number of terms in Eq. (51) has to be retained.
25
FIG. 14: (Colour online) Survival curves for A549 cells irradi-
ated with α particles. Solid line is calculated using Eq. (53).
The dots represent the experimental data [122].
death, Πd is given by
Πd = 1 − exp [−Yc] .
(52)
This means that the probability is unity less the proba-
bility of zero clustered damage sites occurring in the cell
nucleus, which is given by the second term in Eq. (52).
Finally, the probability of cell survival is given by unity
less the probability of cell death, i.e., by that second term
of Eq. (52):
Πsurv = 1 − Πd = exp [−Yc] .
(53)
This probability depends on dose and this dependence,
shown in Fig. 14, is the survival curve for A459 cells ir-
radiated with α-particles with ¯Se = 115 eV/nm. This
curve is compared to the survival curves for the same
cells in the same conditions reported in Ref. [122]. The
dose dependence is very close to exponential (nearly a
straight line in a semi-logarithmic plot); it is determined
by Eq. (51) and is universal for the radiation conditions
considered in this work, i.e., where the interaction be-
tween ion tracks is absent. This corresponds to a large
variety of observed survival curves for different cells and
projectiles [122–124]. From the model point of view this
corresponds to the so-called single-hit model described
in Ref. [31]. Enzymatic repair may affect the behaviour
of this dependence only at very low doses, which may
explain a slight discrepancy seen in Fig. 14.
Equation (51) gives the number of clustered damage
sites per cell nucleus. Since each site of this kind is as-
sumed to be lethal for the cell, the probability of cell
The calculated survival curve depends on the following
numbers, which were either used as parameters or were
determined from the comparison with experiments: the
number of secondary electrons produced by the ion per
unit length, dN/dζ, calculated in Sec. III F and depen-
dent on LET, the probability for an electron incident on
DNA to induce a SSB per one DNA twist, ΓSSB, the
fluence of radicals on a DNA twist, Fr, or the number of
produced radicals and the average radius of their prop-
agation from the ion's trajectory that allow one to esti-
mate it, the probability for radicals to induce SSBs, Γr,
the size of the nucleosome and their number density, nn
(assumed to be uniform) in a given cell, and the lethal
degree of the lesion complexity. Some of the parameters
in this list can be found from the literature, estimated
or calculated (rather accurately) for given media, cells,
and projectiles. Some of them remain unknown for now,
but further research may clarify their values. It is note-
worthy that a complicated problem of the calculation of
the RBE can be reduced to the search of several micro-
scopic parameters that can be determined theoretically
or experimentally.
C. The recipe for obtaining the RBE
In this section, the multiscale approach is summarised
in a recipe for a phenomenon-based assessment of radi-
ation damage that results in the calculation of survival
curves and RBE. Let us imagine that a certain type of
cell is irradiated with certain ions. Here are the steps nec-
essary for finding the location of damage and the RBE
in the irradiated region.
First, it is desirable to know the composition of the
medium. Cross sections of ionization, excitation, and
nuclear fragmentation will affect the shape of the LET
curve. Section III B gives recipes for determining some
of these cross sections for water and more complex media.
Second, using Eqs. (10–15), the LET dependence on en-
ergy and longitudinal coordinate can be obtained. This
gives the location of the Bragg peak, its height, and other
features of the LET curve. The conditions related to sec-
ondary particles should be assessed: the energy spectrum
of secondary electrons, their average energy, and other
features provide the grounds for inference on what meth-
ods can be used for the calculation of their transport
and energy transfer.
If the value of the LET (at least
at some section of the ion's propagation) is higher than
4 keV/nm, shock wave effects may dominate the scenario
of biodamage.
Third, the cells should be thoroughly investigated. In
this work, the cell nucleus was by and large discussed as a
target, however there could be conditions in which other
parts of a cell, such as the cell membrane, cytoplasm, mi-
tochondria, and other organella are targets, whose dam-
age may be lethal to the cell. If the cell nucleus' DNA is
the target, it is important to know how it is distributed.
Any information on the structure of chromatin, size of
nucleosomes, their number density, etc., is important for
the description of the target.
Fourth, as soon as the target is described it is impor-
26
tant to calculate the fluence of secondary particles, such
as secondary electrons and radicals, on this target. The
random walk has been used in this and other works re-
lated to the multiscale approach to describe transport
of electrons, but it can be calculated using more sophis-
ticated methods. As for radicals, they are carried by
the collective flow of the shock wave and diffuse through
the medium. More research is needed to describe their
transport. The damage probability due to radicals de-
pends on their production and transport, both of which
depend on the LET. High temperatures inside the hot
cylinder and consequent shock waves contribute to these
processes. The ultimate effect of radicals depends on the
hypoxic/aerobic conditions in the medium. If the effec-
tiveness is known, the OER can be determined for the
given conditions.
Fifth, the average number of DNA lesions of interest
(that could be lethal) per unit length of the ion's trajec-
tory should be calculated. This implies the integration
and averaging of damaging effects in the radial direc-
tion with respect to the ion's trajectory. Then, knowing
the size of the cell nucleus (or other target), the dose
dependence using Eqs. (29) and (51) can be determined.
The survival curve can then be calculated using Eq. (53).
The comparison of the survival curve with that for x-
rays gives the RBE for a given location. This location
is described with the value of the LET and the depth
coordinate that corresponds to it. This means that these
calculations predict the RBE (and OER) at the Bragg
peak, plateau, and the tail of the LET-depth dependence.
Then, if the tissue can be scanned to produce a spread-
out Bragg peak, the calculations can be superimposed.
Enzymatic repair mechanisms play an important role
for the overall damage assessment. They may be included
in several places. For instance, their effectiveness deter-
mines the definition of "lethal" damage, e.g., a DSB or
degree of complexity (size of a clustered damage site).
Probabilities such as ΓSSB and Γr can also be adjusted
or calculated using quantum mechanical methods.
D. Multiscale approach vs. other models for the
assessment of radiation damage with ions
The expertise for the assessment of radiation damage
historically comes from the times when x-rays (photons)
were used as the only projectiles. For x-rays, the dose
distribution is practically uniform. Therefore, it is not
accidental that the dose has been chosen as the main
parameter for prediction of radiation damage. Treatment
plans had to deliver certain doses to certain locations in
order to achieve the desired results.13 A vast majority,
if not all, of other existing models that "calculate" the
13 The optimisation related to reducing dose deposition in healthy
regions and treatment partitioning is left aside.
survival curves are based on an empiric formula,
− ln Πs = αd + βd2 ,
(54)
where α and β are coefficients. Several features of these
curves have been discussed, one of which is the ratio of
α/β. If this ratio is large, the survival curve is "steep"
and more like a straight line14; if it is small then it is a
"shouldered" curve. A series of models suggested since
1955 [31, 125–130] provided a phenomenological expla-
nation to this dependence and developed approaches to
the calculation of RBE. The coefficients α and β depend
on the kind of cells, on the cell cycle, on the access of
oxygen to the irradiated cells and other factors.
For many practical purposes, an experimentally ob-
tained curve, given by Eq. (54) is sufficient information
for the evaluation of radiation damage, and it has been
used for many years for treatment planning and opti-
mization. Atomic or molecular interactions are not men-
tioned in those models; these and more information are
hidden in the purely empirical coefficients α and β. Since
the dose distribution is uniform, it is possible to solve
all practical problems without atomic/molecular physics,
since it brings up too many difficult questions involv-
ing interactions with biomolecules that seem irrelevant
as compared to the biological unknowns related to repair
mechanisms.
Particle projectiles change this picture. As was shown
above, the dose distribution around each particle's path
is highly nonuniform. The track structure and the con-
sequent damage are much more complicated. A solution
to this problem was suggested by the Katz approach in
which the radial dose distribution is calculated and re-
lated to the inactivation of sub-cell-nucleus targets [76–
78, 131]. The quality factor of radiation was introduced
in order to relate the survival curve parameters to a given
type of radiation, differentiating between track types, in-
activation modes, the structural complexity of targets,
etc. The eventual goal of the Katz model was to calcu-
late the RBE. Nevertheless, the biological relation of the
radial dose distribution with the cell survival probability
was done based on the survival curves for x-rays, without
analyzing particular physical processes, i.e., the empiric
coefficients α and β remain central to this approach.
The Local Effect Model (LEM), developed at GSI, cal-
culates the RBE assuming that the biological effect of
radiation is entirely determined by the spatial distribu-
tion of the radial dose inside the cell nucleus. It relates
the response of biological systems, following ion irradi-
ation, to the corresponding response after x-ray irradia-
tion [4, 132]. Corrections for the quality of damage was
included in a later version of the LEM [4]. This model
operates on the schematic level using Eq. (54) with em-
14 For example, for the survival curve, shown in Fig. 14, α = 1 and
β = 0, which is typical for cells irradiated with ions.
27
pirical coefficients α and β. The LEM solves technical
problems related to the optimization of treatments, leav-
ing no place for ab initio approaches and physical, chem-
ical, or biological effects in general; even a consideration
of DNA lesions such as DSBs is beyond the scope of the
LEM [2, 4].
The calculation of survival curves shown in Sec. VI B
demonstrates a new way of relating the physical param-
eters with biological outcomes thus fulfilling the goal of
the multiscale approach and predicting the RBE. The
multiscale approach is unique in relating the biological
consequences of radiation to the actual physical, chemi-
cal, and biological effects.
VII. CONCLUSIONS AND OUTLOOK
The multiscale approach to the assessment of radia-
tion damage with ions has been reviewed. It was demon-
strated that the main difference from other approaches is
the in-depth focusing on physical effects, and, therefore,
our approach is referred to as a phenomenon-based ap-
proach. The state of the art of this approach is discussed
and some techniques of calculations are demonstrated.
The main advantages of the multiscale approach follow
from its architecture, its fundamentality, and its versa-
tility. The approach evaluates the relative contributions
and significance of a variety of phenomena; it elucidates a
complex multiscale scenario in sufficient detail and has a
solid predictive power. It is structurally simple and inclu-
sive, and allows for modifications and extensions by in-
cluding new effects on different scales and improvements
on the way.
There are several areas in which major developments
are expected. First, as it has been shown, the empiri-
cal models for calculation of survival curves can be im-
proved using the multiscale approach. This can be imple-
mented in the optimization codes for clinical purposes.
The curves and radiation strategy will depend on the
kind of cells and type of radiation. Moreover, the modal-
ity can also be optimized for particular cases. In order
to achieve this, a more thorough comparison with ex-
periments should be established and parameters such as
ΓSSB should be tuned. The criteria for cell death should
also be understood in contact with biologists. Since the
radicals play a significant role in radiation damage, the
aerobic/hypoxyc conditions of the target will determine
the degree of inclusion of radicals in the calculations.
The second area of development is related to the mod-
ification of the medium. The use of nanoparticles such
as gold nanoparticles (GNP) as sensitizers has been dis-
cussed both theoretically and experimentally [21, 133],
in order to boost the production of secondary electrons
near the target and thus increase the RBE. The use of
nanoparticles is considered for different modalities. Such
a modification of the medium should be feasible within
the multiscale approach. The relevant cross sections of
secondary electron production and their energy spectrum
will define their effect on nearby biomolecules.
a. Fluence for the radial dose calculation
28
In order to adjust Eq. (19) for the calculation of the
radial dose, dA is chosen to be an element of the surface
of a cylinder of radius ρ, coaxial to the ion's path, dA =
dAnρ, where nρ is a unit vector in the radial direction
toward the element. Then, the square of the distance
from any point on the ion path to the area element is
given by r2 = ρ2+ζ 2, where ρ is the radius of the cylinder
from the ion path and ζ is the coordinate along the ion's
path and nρ · nr =
. Finally, if the element is a
belt of radius ρ and width δ, dA = 2πρδ, and the number
of secondary electrons incident on this belt is given by the
integral of Eq. (19) over the whole ζ-axis:
ρ√
ρ2+ζ2
Nδ(ρ) = 2πρδ
× r
2k
= 2πρδ
× exp
= 2πρδ
dN
dζ
(cid:18) 3
(cid:90) ∞
2πkl2
ρ
2k
(cid:18)
ρ/l
dk
dN
dζ
− 3ρ2
2kl2 − γk
(cid:90) ∞
(cid:20)
dN
dζ
× exp
dk
ρ/l
(cid:90) ∞
(cid:19)3/2
−∞
dk
(cid:21)
(cid:90) ∞
r/l
dζ
ρ(cid:112)ρ2 + ζ 2
(cid:20)
− 3r2
2kl2 − γk
(cid:18) 3
(cid:19)3/2
(cid:19)(cid:90) √
2πkl2
exp
k2l2−ρ2
√
k2l2−ρ2
dζ exp
(cid:21)
(cid:20)
− 3ζ 2
2kl2
−
3ρ
4πk2l2
(cid:21)
(cid:20) 3
(cid:18)
(cid:19)(cid:21)
k − ρ2
kl2
− 3ρ2
2kl2 − γk
erf
= 2πρδ
Q(ρ/l, γ) , (A1)
2
dN
dζ
The third area is the modification of modality.
Ions
heavier than carbon require a better understanding
of thermomechanical effects discussed above, since the
shock waves initiated in the Bragg peak area would be
more pronounced. The use of these ions may not neces-
sarily be therapeutic. Rather, the understanding of the
mechanisms of radiation damage at very high values of
LET will help the assessment of the hazards of exposure
to such ions during space missions or elsewhere. Also,
the targets may not necessarily be biological, e.g., the
assessment of radiation damage of electronics or other
equipment can be done in a similar fashion. Another
aspect, which can also be regarded as a modality modifi-
cation, is the series of effects related to irradiation with
ion beams produced by high-power lasers [89]. In these
conditions, the beam is much more dense and the tracks
substantially interfere. The application of the multiscale
approach for the calculation of survival curves may be
especially beneficial in this case.
The fourth area is related to the change of the field of
science related to radiation damage with ions. The devel-
opment of the multiscale approach is shifting a paradigm
as in the case with thermomechanical damage. However,
in some cases, where analytical methods turned out to
be successful, a new understanding was obtained. Some
common terms, such as dose, are shown to have limita-
tions when describing radiation damage with ions. This
means that the science of radiation damage is evolving.
The future development of the multiscale approach will
make a worthwhile tool for the assessment of radiation
damage on the molecular level. While there is more work
to be done to make it practical, its fundamental basis and
depth related to atomic/molecular physics is becoming
more and more evident.
Acknowledgements
×
where
Q(ρ/l, γ) =
(cid:90) ∞
ρ/l
3ρ
4πl2
dk
k2 exp
(cid:20)
(cid:21)
erf
(cid:20) 3
2
(cid:18)
k − ρ2
kl2
(cid:19)(cid:21)
(A2)
− 3ρ2
2kl2 − γk
We are grateful to J. S. Payson and B. Roth who crit-
ically read the manuscript, R. Garcia-Molina, M. Niklas,
I. M. Solovyeva, I. A. Solov'yov, P. de Vera for the assis-
tance with figures, important advice, and insight, Center
for Scientific Computing of Goethe University, and the
support of COST Action MP1002 "Nano-scale insights
in ion beam cancer therapy."
has been introduced. Equation (A1) with (A2) provide
the general expression for the number of secondary elec-
trons incident on a cylindrical belt coaxial with the ion's
path.
b. Cylinders enwraping a DNA twist
Appendix A: Calculations of fluence for different
geometries
Here it is shown how to calculate the fluence of sec-
ondary electrons applied to different geometries.
The geometry for the calculations of the fluence
through a cylinder enwraping a DNA twist is shown in
the inset of Fig. 8. The vector r from the point of ori-
gin of a secondary electron on the ion's path to a point
on the surface of the cylinder enwraping a DNA twist is
given by,
r = (a cos ϕ − ρ)i + (a sin ϕ − ζ sin α)j
+(z − z0 − ζ cos α)k ,
(A3)
c. DNA on the surface of a nucleosome for complex
damage calculation
29
where z is the coordinate along the cylinder. Equation
(19) has to be integrated over the length of the path ζ
and the area of the cylinder. It is reasonable to compare
two limiting cases of the cylinder: that of the cylinder
being parallel and perpendicular to the path for z0 = 0.
The perpendicular case corresponds to α = π/2 and r2
and the dot product of Eq. (19) with (A3) are as follows:
r2 = (a cos ϕ − ρ)2 + (a sin ϕ − ζ)2 + z2,
dA · nr = adϕdznA · nr
and
(a cos ϕ − ρ) cos ϕ + (a sin ϕ − ζ) sin ϕ
= adϕdz
. (A4)
r
r
The parallel case corresponds to α = 0 and the similar
expressions are given by
r2 = (a cos ϕ − ρ)2 + (a sin ϕ)2 + (z − ζ)2 , and
dA · nr = adϕdznA · nr
(a cos ϕ − ρ) cos ϕ + (a sin ϕ − ζ) sin ϕ
= adϕdz
.
(A5)
In the parallel case, one has to also include the bases of
the cylinder, which correspond to:
r2 = (r(cid:48) cos ϕ − ρ)2 + (r(cid:48) sin ϕ)2 + (z − ζ)2 , and
dA · nr = r(cid:48)dr(cid:48)dϕnA · nr
= r(cid:48)dr(cid:48)dϕ
(r(cid:48) cos ϕ − ρ) cos ϕ + (r(cid:48) sin ϕ − ζ) sin ϕ
. (A6)
r(cid:48)
In the case of the calculation of complex damage the
electrons are incident on a cylindrical surface of a nu-
cleosome. The corresponding r2 and nr · nA are given
by
r2 = (an cos ϕ − ρ)2 + (an sin ϕ − ζ)2 + z2,
dAn · nr = andϕdznA · nr
and
(an cos ϕ − ρ) cos ϕ + (a sin ϕ − ζ) sin ϕ
r
= andϕdz
(A7)
and the integration is done over z in the limits from −1.15
nm to +1.15 and over ϕ from
max
(cid:104)−ψ, arctan ζ
(cid:104)
(cid:105)
(cid:105)
to
an
ζ2+ρ2
, where ψ is the ratio of
ρ − arccos
ρ + arccos
an
ζ2+ρ2
min
ψ, arctan ζ
the length of a twist (3.4 nm) to the radius of a nucleo-
some (5.75 nm) [15, 134].
[1] E. Surdutovich and A. Solov'yov, J. Phys.: Conf. Ser.
[12] E. Scifoni, E. Surdutovich, and A. Solovyov, Phys Rev.
373, 012001 (2012).
E 81, 021903 (2010).
[2] I. Baccarelli, F. Gianturco, E. Scifoni, A. Solov'yov, and
[13] A. Chatterjee and W. R. Holley, Adv. Radiat. Biol. 17,
E. Surdutovich, Eur. Phys. J. D 60, 1 (2010).
181 (1993).
[3] U. Amaldi and G. Kraft, J. Radiat. Res. 48, A27 (2007).
[4] D. Schardt, T. Elsasser, and D. Schulz-Ertner, Rev.
[14] C. von Sonntag, The chemical basis of radiation biology
(Taylor & Francis, London, 1987).
Mod. Phys. 82, 383 (2010).
[15] E. Surdutovich, D. C. Gallagher, and A. V. Solov'yov,
[5] M. Durante and J. Loeffler, Nat. Rev. Clin. Oncol. 7,
Phys. Rev. E 84, 051918 (2011).
37 (2010).
[6] Proton therapy – wikipedia, accessed on 09/2013, URL
http://en.wikipedia.org/wiki/Proton{\_}therapy.
[7] Particle therapy – wikipedia, accessed on 09/2013,
http://en.wikipedia.org/wiki/Particle{\_
URL
}therapy.
[8] E. Haettner, H. Iwase, and D. Schardt, Rad. Protec.
Dosim. 122, 485 (2006).
[16] L. Sanche, Eur. Phys. J. D 35, 367 (2005).
[17] B. Boudaıffa, P. Cloutier, D. Hunting, M. A. Huels, and
L. Sanche, Science 287, 1658 (2000).
[18] M. A. Huels, B. Boudaıffa, P. Cloutier, D. Hunting, and
L. Sanche, JACS 125, 4467 (2003).
[19] L. Sanche, in Radical and radical ion reactivity in nu-
cleic acid chemistry, edited by M. Greenberg (J. Wiley
& Sons, Inc., New York, 2010), p. 239.
[9] L. Sihver, D. Schardt, and T. Kanai, Jpn. J. Med. Phys.
[20] E. Surdutovich and A. V. Solov'yov, Eur. Phys. J. D
18, 1 (1998).
66, 206 (2012).
[10] I. Pshenichnov, I. Mishustin, and W. Greiner, Nucl.
Inst. Meth. B 266, 1094 (2008).
[11] E. Surdutovich, O. Obolensky, E. Scifoni, I. Pshenich-
nov, I. Mishustin, A. Solov'yov, and W. Greiner, Eur.
Phys. J. D 51, 63 (2009).
[21] S. McMahon, W. Hyland, M. Muir, J. Coulter, S. Jain,
K. Butterworth, G. Schettino, G. Dickson, A. Hounsell,
J. O'Sullivan, et al., Sci. Rep. 1, 18 (2011).
[22] A. Solov'yov, E. Surdutovich, E. Scifoni, I. Mishustin,
and W. Greiner, Phys. Rev. E79, 011909 (2009).
30
[23] Cost nano-ibct – nanoscale insights into ion beam
cancer therapy, accessed on 01/2013, URL http:
//www.cost.eu/domains{\textunderscore}actions/
mpns/Actions/nano-ibct/.
[24] E. Surdutovich and A. Solov'yov, Europhys. News
40/2, 21 (2009).
[52] M. Altarelli and D. Smith, Phys. Rev. B 9, 1290 (1974).
[53] R. Garcia-Molina, I. Abril, I. Kyriakou, and D. Em-
fietzoglou, in Radiation Damage in Biomolecular Sys-
tems, edited by G. G. G´omez-Tejedor and M. C. Fuss
(Springer, Dordrecht, 2012), p. Chap. 15.
[54] R. H. Ritchie and A. Howie, Philos. Mag. 36, 436
[25] E. Surdutovich, E. Scifoni, , and A. Solov'yov, Mutat.
(1977).
Res. 704, 206 (2010).
[55] M. Dingfelder, D. Hantke, M. Inokuti, and H. Paretzke,
[26] M. Toulemonde, E. Surdutovich, and A. Solov'yov,
Radiat. Phys. Chem. 53, 1 (1999).
Phys. Rev. E 80, 031913 (2009).
[27] E. Surdutovich, A. Yakubovich, and A. Solov'yov, Eur.
[56] D. Emfietzoglou, Radiat. Phys. Chem. 66, 373 (2003).
[57] P. Bernhardt and H. G. Paretzke, Int. J. Mass Spectrom.
Phys. J. D 60, 101 (2010).
223-224, 599 (2003).
[28] E. Surdutovich and A. Solov'yov, Phys. Rev. E 82,
[58] A. Peudon, S. Edel, and M. Terrisol, Radiat. Prot.
051915 (2010).
[29] E. Surdutovich, A. V. Yakubovich, and A. V. Solov'yov,
Sci. Rep. 3, 1289 (2013).
[30] P. de Vera, R. Garcia-Molina, I. Abril, and A. V.
Solov'yov, Phys. Rev. Lett. 110, 148104 (2013).
[31] E. L. Alpen, Radiation Biophysics (Academic Press,
San Diego, London, Boston, New York, Sydney, Tokyo,
Toronto, 1998).
[32] E. J. Hall and A. J. Giaccia, Radiobiology for Radiolo-
gist (Lippincott Williams & Wilkins, Philadelphia, Bal-
timore, New York, London, 2012).
Dosim. 122, 128 (2006).
[59] Y.-K. Kim et al., Electron-impact ionization cross sec-
tion for ionization and excitation database (version
3.0) (2004), URL http://www.nist.gov/pml/data/
ionization/index.cfm.
[60] D. R. White, R. V. Griffith, and I. J. Wilson, Photon,
Electron, Proton and Neutron Interaction Data for Body
Tissues (International Commission on Radiation Units
and Measurements (ICRU 46), Bethesda, MD, 1992).
[61] W. E. Wilson, J. H. Miller, L. H. Toburen, and S. T.
Manson, J. Chem. Phys. 80, 5631 (1984).
[33] S. Pimblott and L. Siebbeles, Nucl. Inst. Meth. B 194,
[62] M. Rudd, T. Goffe, R. DuBois, and L. Toburen, Phys.
237 (2002).
Rev. A31, 492 (1985).
[34] S. Pimblott, J. LaVerne, and A. Mozumder, J. Phys.
[63] M. A. Bolorizadeh and M. E. Rudd, Phys. Rev. A 33,
Chem 100, 8595 (1996).
888 (1986).
[35] S. Pimblott and J. LaVerne, Rad. Phys. Chem. 76, 1244
[64] Y. Iriki, Y. Kikuchi, M. Imai, and A. Itoh, Phys. Rev.
(2007).
A 84, 052719 (2011).
[36] J. Meesungnoen, J.-P. Jay-Gerin, A. Filali-Mouhim, and
S. Mankhetkorn, Radiat. Res. 158, 657 (2002).
[65] J. Simons, Adv. Quantum Chem. 52, 171 (2007).
[66] M. Dingfelder, M. Inokuti, and H. Paretzke, Rad. Phys.
[37] H. Nikjoo, S. Uehara, D. Emfietzoglou, and F. A. Cu-
Chem. 59, 255 (2000).
cinotta, Radiat. Meas. 41, 1052 (2006).
[38] E. Surdutovich and A. V. Solov'yov, Eur. Phys. J. D
66, 245 (2012).
[39] M. Bug, E. Surdutovich, H. Rabus, A. B. Rosenfeld,
and A. V. Solov'yov, Eur. Phys. J. D 66, 291 (2012).
[40] Y. Park, Z. Li, P. Cloutier, L. Sanche, and J. Wagner,
Radiat. Res. 175, 240 (2011).
[67] R. Garcia-Molina, I. Abril, P. de Vera, I. Kyriakou,
and D. Emfietzoglou, J. Phys.: Conf. Ser. 373, 012015
(2012).
[68] W. H. Barkas, Nuclear Research Emulsions I. Tech-
niques and Theory, vol. 1 (Academic Press, New York,
London, 1963).
[69] G. Schiwietz and P. L. Grande, Nucl. Instr. Meth. B
[41] M. Smyth and J. Kohanoff, J. Am. Chem. Soc. 134,
175-177, 125 (2001).
9122 (2012).
[42] D. Becker, A. Adhikary, and M. Sevilla, in Charged Par-
ticle and Photon Interactions with Matter Recent Ad-
vances, Applications, and Interfaces (CRC Press, Tay-
lor & Francis, Boca Raton, 2010).
[43] H. Bethe, Ann. Phys. 397, 325 (1930).
[44] F. Bloch, Z. Phys. A: Hadrons Nucl. 81, 363 (1933).
[45] F. Bloch, Ann. Phys. 408, 285 (1933).
[46] I. Abril, R. Garcia-Molina, C. Denton, I. Kyriakou, and
[70] P. Kundrat, Phys. Med. Biol. 52, 6813 (2007).
[71] M. Hollmark, J. Uhrdin, D. Belkic, I. Gudowska, and
A. Brahme, Phys. Med. Biol. 49, 3247 (2004).
[72] M. Inokuti, Rev. Mod. Phys. 43, 297 (1971).
[73] H. Schmidt-Bocking, L. Schmidt, T. Weber, V. Mergel,
O. Jagutzki, A. Czasch, S. Hagmann, R. Doerner,
Y. Demkov, T. Jahnke, et al., Radiat. Phys. Chem. 71,
627 (2004).
[74] C. Tung, T. Chao, H. Hsieh, and W. Chan, Nucl. Inst.
D. Emfietzoglou, Radiat. Res. 175, 247 (2011).
Meth B 262, 231 (2007).
[47] O. Obolensky, E. Surdutovich,
I. Pshenichnov,
I. Mishustin, A. Solov'yov, and W. Greiner, Nucl. Inst.
Meth. B 266, 1623 (2008).
[48] M. E. Rudd, Y.-K. Kim, D. H. Madison, and T. Gay,
[75] S. Chandrasekhar, Rev. Mod. Phys. 15, 1 (1943).
[76] J. J. Butts and R. Katz, Radiat. Res. 30, 855 (1967).
[77] R. Katz, B. Ackerson, M. Homayoonfar, and S. C.
Sharma, Radiat. Res. 47, 402 (1971).
Rev. Mod. Phys. 64, 441 (1992).
[78] M. Korcyl and M. Walig´orski, Int. J. Radiat. Biol. 85,
[49] L. Landau, E. Lifshitz, and L. Pitaevskii, Electrody-
namics of Continuous Media, Second Edition: Volume
8 (Butterworth-Heinemann, Burlington, 1984).
1101 (2009).
[79] M. Waligorski, R. Hamm, and R. Katz, Nucl. Tracks
Radiat. Meas. 11, 309 (1986).
[50] J. Lindhard and K. Dan, Vidensk. Selsk. Mat. Fys.
[80] F. A. Cucinotta, R. Katz, and J. W. Wilson, Radiat.
Medd. 28, 8 (1954).
[51] Z. Tan, Y. Xia, M. Zhao, X. Liu, F. Li, B. Huang, and
Y. Ji, Nucl. Instrum. Methods Phys. Res. B 222, 27
(2004).
Environ. Biophys. 37, 259 (1998).
[81] I. Plante and F. Cucinotta, Radiat. Environ. Biophys.
49, 5 (2010).
[82] J. Ward, Radiat. Res. 142, 362 (1995).
31
[83] I. I. Fabrikant, S. Caprasecca, G. A. Gallup, and J. D.
Gorfinkiel, J. Chem. Phys. 136, 184301 (2012).
[84] D. Becker and M. Sevilla, in Advances in Radiation Bi-
ology, edited by J. Lett (Acad. Press, 1993), vol. 17, pp.
121–180.
[85] F. A. Gianturco, F. Sebastianelli, R. R. Lucchese,
I. Baccarelli, and N. Sanna, J. Chem. Phys. 128, 174302
(2008).
[86] L. Sanche, in Radiation Damage in Biomolecular Sys-
tems, edited by G. Garcia and M. C. Fuss (Springer,
2012).
[87] R. Panajotovic, F. Martin, P. Cloutier, D. Hunting, and
L. Sanche, Radiat. Res. 165, 452 (2006).
[88] M. Mucke, M. Braune, S. Barth, M. Forstel, T. Lis-
chke, V. Ulrich, T. Arion, U. Becker, A. Bradshaw, and
U. Hergenhahn, Nat. Phys. 6, 143 (2010).
[89] S. S. Bulanov, A. Brantov, V. Y. Bychenkov,
V. Chvykov, G. Kalinchenko,
T. Matsuoka,
P. Rousseau, S. Reed, V. Yanovsky, K. Krushel-
nick, et al., Med. Phys. 35, 1770 (2008).
J. Stoquert, F. Studer, and M. Toulemonde, Phys. Rev.
B48, 920 (1993).
[106] A. V. Yakubovich, E. Surdutovich, and A. V. Solov'yov,
AIP Conf. Proc. 1344, 230 (2011).
[107] A. V. Yakubovich, E. Surdutovich, and A. V. Solov'yov,
Nucl. Instr. Meth. B 279, 135 (2012).
[108] L. Landau and E. Lifshitz, Fluid dynamics, Second Edi-
tion: Volume 6 (Reed-Elsevier, Oxford, Boston, Johan-
nesburg, 1987).
[109] Y. Zeldovich and Y. Raiser, Physics of Shock Waves and
High-Temperature Hydrodynamic Phenomena (Volume
1) (Oxford, New York, 1966).
[110] G. Chernyj, Gas dynamics (Nauka, Moscow, 1994).
[111] I. A. Solov'yov, A. V. Yakubovich, E. Surdutovich, and
A. V. Solov'yov, Unpublished (2013).
[112] K. Range, M. J. McGrath, X. Lopez, and D. M. York,
J. Am. Chem. Soc. 126, 1654 (2004).
[113] J. LaVerne, Radiat. Phys. Chem. 34, 135 (1989).
[114] B. Jakob, M. Scholz, and G. Taucher-Scholz, Radiat.
Res. 159, 676 (2003).
[90] J. Adamcik, J.-H. Jeon, K. J. Karczewski, R. Metzler,
[115] F. Tobias, M. Durante, G. Taucher-Scholz, and
and G. Dietler, Soft Matt. 8, 8651 (2012).
B. Jakob, Mutat. Res. 704, 54 (2010).
[91] H. M. Dang, M. J. V. Goethem, E. R. V. D. Graaf,
S. Brandenburg, R. Hoekstra, and T. Schlatholter, Eur.
Phys. J. D 63, 359 (2011).
[92] M. Sevilla, D. Becker, and Y. Razskazovskii, Nucleonika
42, 283 (1997).
[93] M. Niklas, A. Abdollahi, M. Akselrod, J. Debus,
O. Jakel, and S. Greilich, Int. J. Radiat. Oncol. p. in
press (2013).
[94] R. de Jiang, H. Shen, and Y.-J. Piao, Roman. J. Mor-
phol. Embryol. 51(4), 663 (2010).
[95] M. Toulemonde, C. Dufour, A. Meftah, and E. Paumier,
[116] W. P. Roos and B. Kaina, Trends Mol. Med. 12, 440
(2006).
[117] J. Ward, Prog. Nucleic Acid. Res. Mol. biol. 35, 95
(1988).
[118] D. T. Goodhead, Int. J. Radiat. Biol. 65, 7 (1994).
[119] S. Malyarchuk, R. Castore, and L. Harrison, DNA Re-
pair 8, 1343 (2009).
[120] S. Malyarchuk, R. Castore, and L. Harrison, Nucleic
Acids Res. 36, 4872 (2008).
[121] E. Sage and L. Harrison, Mutat. Res. 711, 123 (2011).
[122] A. C. Heuskin, C. Michiels, and S. Lucas, Phys. Med.
Nucl. Inst. Meth. B 166-167, 903 (2000).
Biol. 58, 6495 (2013).
[96] M. Toulemonde, C. Trautmann, E. Balanzat, K. Hjort,
[123] M. Scholz, A. Kellerer, W. Kraft-Weyrather, and
and A. Weidinger, Nucl. Inst. Meth. B 216, 1 (2004).
G. Kraft, Radiat. Environ. Biophys. 36, 59 (1997).
[97] M. Skupinski, M. Toulemonde, M. Lindeberg, and
[124] W. K. Weyrather, S. Ritter, M. Scholz, and G. Kraft,
K. Hjort, Nucl. Inst. Meth. B 240, 681 (2005).
Int. J. Rad. Biol. 75, 1357 (1999).
[98] F. Pawlak, C. Dufour, A. Laurent, E. Paumier,
J. Perri`ere, J. P. Stoquert, and M. Toulemonde, Nucl.
Inst. Meth. B 151, 140 (1999).
[99] M. Toulemonde, W. Assmann, C. Dufour, A. Meftah,
F. Studer, and C. Trautmann, Mat. Fys. Medd. 52, 263
(2006).
[100] H. Dammak, D. Lesueur, A. Dunlop, P. Legrand, and
J. Morillo, Radiat. Eff. Defect. Sol. 126, 111 (1993).
[101] H. D. Mieskes, W. Assmann, F. Gruner, H. Kucal, Z. G.
Wang, and M. Toulemonde, Phys. Rev. B 67, 155414
(2003).
[102] A. Meftah, M. Djebara, N. Khalfaoui, and M. Toule-
[125] R. Hawkins, Int. J. Radiat. Biol. 69, 739 (1996).
[126] R. Hawkins, Radiat. Res. 172, 761 (2009).
[127] M. C. Frese, V. K. Yu, R. D. Stewart, and D. J. Carlson,
Int. J. Radiat. Oncol. 83, 442 (2012).
[128] D. Goodhead, J. Thacker, and R. Cox, Int. J. Radiat.
Biol. 63, 543 (1993).
[129] H. Nikjoo, C. Bolton, R. Watanabe, M. Terrisol,
P.O'Neill, and D. Goodhead, Radiat. Prot. Dosim. 99,
77 (2002).
[130] D. Goodhead, Radiat. Prot. Dosim. 122, 3 (2006).
[131] F. Cucinotta, H. Nikjoo, and D. Goodhead, Radiat. En-
viron. Biophys. 38, 81 (1999).
monde, Nucl. Instr. Meth. B 146, 431 (1998).
[103] V. Katin, Y. Martinenko, and Y. Yavlinskii, Sov. Techn.
[132] M. Scholz and G. Kraft, Adv. Space Res. 18, 5 (1996).
[133] Y. Zheng, D. J. Hunting, P. Ayotte, and L. Sanche,
Phys. Lett. 13, 276 (1987).
Radiat. Res. 169, 19 (2008).
[104] M. Toulemonde, W. Assmann, C. Trautmann, and
[134] M. Depken and H. Schiessel, Biophys. J. 96, 777 (2009).
F. Gruner, Phys. Rev. Lett. 88, 057602 (2002).
[105] A. Meftah, F. Brisard, J. Costantini, M. Hage-Ali,
|
1906.07269 | 1 | 1906 | 2019-06-17T20:57:24 | Variational implicit-solvent predictions of the dry-wet transition pathways for ligand-receptor binding and unbinding kinetics | [
"physics.bio-ph",
"cond-mat.soft",
"physics.chem-ph"
] | Ligand-receptor binding and unbinding are fundamental biomolecular processes and particularly essential to drug efficacy. Environmental water fluctuations, however, impact the corresponding thermodynamics and kinetics and thereby challenge theoretical descriptions. Here, we devise a holistic, implicit-solvent, multi-method approach to predict the (un)binding kinetics for a generic ligand-pocket model. We use the variational implicit-solvent model (VISM) to calculate the solute-solvent interfacial structures and the corresponding free energies, and combine the VISM with the string method to obtain the minimum energy paths and transition states between the various metastable ('dry' and 'wet') hydration states. The resulting dry-wet transition rates are then used in a spatially-dependent multi-state continuous-time Markov chain Brownian dynamics simulations, and the related Fokker-Planck equation calculations, of the ligand stochastic motion, providing the mean first-passage times for binding and unbinding. We find the hydration transitions to significantly slow down the binding process, in semi-quantitative agreement with existing explicit-water simulations, but significantly accelerate the unbinding process. Moreover, our methods allow the characterization of non-equilibrium hydration states of pocket and ligand during the ligand movement, for which we find substantial memory and hysteresis effects for binding versus unbinding. Our study thus provides a significant step forward towards efficient, physics-based interpretation and predictions of the complex kinetics in realistic ligand-receptor systems. | physics.bio-ph | physics |
Variational implicit-solvent predictions of the
dry-wet transition pathways for ligand-receptor
binding and unbinding kinetics
Shenggao Zhoua, R. Gregor Weissb, Li-Tien Chengc, Joachim Dzubiellad, J. Andrew McCammone,1, and Bo Lic,1
aDepartment of Mathematics and Mathematical Center for Interdiscipline Research, Soochow University, 1 Shizi Street, Suzhou 215006, Jiangsu, China; bLaboratory of
Physical Chemistry, ETH Zürich, Vladimir-Prelog-Weg 2, CH-8093 Zürich, Switzerland; and Institut für Physik, Humboldt-Universität zu Berlin, Newtonstrasse 15, D-12489
Berlin, Germany; cDepartment of Mathematics, University of California, San Diego, 9500 Gilman Drive, La Jolla, California 92093-0112, USA; dPhysikalisches Institut,
Albert-Ludwigs-Universität Freiburg, Hermann-Herder-Strasse 3, 79104 Freiburg, Germany; and Research Group Simulations of Energy Materials (EE-GSEM),
Helmholtz-Zentrum Berlin, Hahn-Meitner-Platz 1, 14109, Berlin, Germany; eDepartment of Chemistry and Biochemistry, Department of Pharmacology, University of California,
San Diego, 9500 Gilman Drive, La Jolla, California 92093-0365, USA
This manuscript was compiled on June 19, 2019
Ligand-receptor binding and unbinding are fundamental biomolec-
ular processes and particularly essential to drug efficacy. Environ-
mental water fluctuations, however, impact the corresponding ther-
modynamics and kinetics and thereby challenge theoretical descrip-
tions. Here, we devise a holistic, implicit-solvent, multi-method ap-
proach to predict the (un)binding kinetics for a generic ligand-pocket
model. We use the variational implicit-solvent model (VISM) to calcu-
late the solute-solvent interfacial structures and the corresponding
free energies, and combine the VISM with the string method to ob-
tain the minimum energy paths and transition states between the
various metastable ("dry" and "wet") hydration states. The result-
ing dry-wet transition rates are then used in a spatially-dependent
multi-state continuous-time Markov chain Brownian dynamics simu-
lations, and the related Fokker -- Planck equation calculations, of the
ligand stochastic motion, providing the mean first-passage times for
binding and unbinding. We find the hydration transitions to signif-
icantly slow down the binding process, in semi-quantitative agree-
ment with existing explicit-water simulations, but significantly accel-
erate the unbinding process. Moreover, our methods allow the char-
acterization of non-equilibrium hydration states of pocket and ligand
during the ligand movement, for which we find substantial memory
and hysteresis effects for binding versus unbinding. Our study thus
provides a significant step forward towards efficient, physics-based
interpretation and predictions of the complex kinetics in realistic
ligand-receptor systems.
Ligand-receptor binding/unbinding kinetics dry-wet transitions vari-
ational implicit-solvent model level-set method string method
D RAFT
metastable "dry" or "wet" hydration states of the binding
site, separated by an energetic barrier which is on the order
of kBT (17). Such a moderate energetic hurdle facilitates
repeated condensation and evaporation of water in the pocket
region, leading to large collective hydration fluctuations (18).
In general, the dewetting of local regions generates strong
hydrophobic forces in molecular association and dissociation
(6, 7, 19, 20). In particular, it has been demonstrated that
the dry-wet transitions are a precursor of the ligand-receptor
binding and unbinding (17, 21, 22). Besides being the origin for
the thermodynamically driven forces, water fluctuations also
modify the friction and kinetics of associating hydrophobic
molecules (23 -- 27), slowing down the binding kinetics and
giving rise to local non-Markovian effects (18, 27).
While water plays a critical role in molecular recognition,
efficient modeling of water is rather challenging due to an over-
whelming number of solvent degrees of freedom, many-body
effects, and the multi-scale nature of molecular interactions.
Explicit-water molecular dynamics (MD) simulations have
been the main tool in most of the existing studies of the kinet-
ics of ligand-receptor binding and unbinding (18, 22, 25, 26, 28 --
The complex process of ligand-receptor binding and unbind-
ing in aqueous environment is fundamental to biological
function. Understanding the thermodynamics and kinetics of
such processes has far-reaching practical significance, partic-
ularly in rational drug design (1, 2). Water is a key player
in ligand-receptor binding and unbinding, and in molecular
recognition in general (3, 4). In particular, it has been well
established that hydrophobic interactions can drive the associ-
ation and dissociation of biological molecules (5 -- 8).
Hydration contributes significantly to the ligand-receptor
binding free energy, determining the thermodynamic stability
of the bound unit (9, 10). Recent experimental and theoretical
studies have indicated that the kinetics of ligand-receptor
binding and unbinding is crucial for drug effectiveness and
efficacy (2, 11, 12). Often, a ligand binds to a hydrophobic
pocket on the surface of a receptor molecule (13 -- 16). Water
molecules fluctuate around such an apolar pocket, leading to
Significance Statement
The kinetics of ligand-receptor (un)binding -- how fast a ligand
binds into and resides in a receptor -- cannot be inferred solely
from the binding affinity which describes the thermodynamic
stability of the bound complex. A bottleneck in understanding
such kinetics, which is critical to drug efficacy, lies in the mod-
eling of the collective water fluctuations in apolar confinement.
We develop a new theoretical approach that couples a varia-
tional implicit-solvent model with the string method to describe
the dry-wet transition pathways, which then serve as input for
the ligand multi-state Brownian dynamics. Without explicit de-
scriptions of individual water molecule, our theory predicts the
key thermodynamic and kinetic properties of unbinding and
binding, the latter in quantitative agreement with explicit-water
molecular dynamics simulations.
JD, JAM, and BL designed research. SZ, RGW, and LTC performed research. SZ, LTC, and BL
developed numerical methods and LTC wrote the initial level-set code. SZ, RGW, and JD analyzed
computational results. SZ, RGW, JD, JAM, and BL wrote the paper.
The authors declare no conflict of interest.
1To whom correspondence should be addressed.
[email protected]
E-mail:
[email protected] or
www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX
PNAS June 19, 2019
vol. XXX no. XX 1 -- 6
Fig. 1. (A) A schematic of the ligand (blue sphere), explicit water, and the pocket of a
concave wall. (B) 1s-dry: The VISM surface (blue line) is a single surface enclosing all
the wall atoms and also the ligand atom, hence a dry state of the pocket. (C) 2s-dry:
The VISM surface has two disjoint components, one enclosing all the wall atoms with
a dry pocket, and one enclosing the ligand. (D) 2s-wet: The VISM surface has two
components, tightly wrapping up the wall and ligand, respectively, with no space for
water, hence a wet pocket.
three VISM surfaces. In addition to 1s-dry and 2s-wet, the
third one is 2s-dry; cf. Fig. 1 (C). Once the ligand is away from
the pocket with z > 8 Å, there are only two VISM surfaces:
2s-dry and 2s-wet.
33). While explicitly tracking water molecules, MD simulations
are still limited to systems of relatively small sizes and events
of relatively short time scales. In particular, slow and rare wa-
ter fluctuations and large ligand residence times in the pocket
still challenge the prediction of unbinding times.
In this work, we develop a holistic, multi-method, implicit-
solvent approach to study the kinetics of ligand-receptor bind-
ing and unbinding in a generic pocket-ligand model exactly
as studied previously by explicit-water MD simulations (18),
focusing on the effect of solvent fluctuations and multiple
hydration states on such processes.
Our approach is based on the variational implicit-solvent
model (VISM) that we have developed in recent years (34 -- 38).
In VISM, one minimizes a solvation free-energy functional of
solute-solvent interfaces to determine a stable, equilibrium
conformation, and to provide an approximation of the solva-
tion free energy. The functional couples the solute surface
energy, solute-solvent van der Waals (vdW) dispersive inter-
actions, and electrostatics. This theory resembles that of
Lum -- Chandler -- Weeks (39) [cf. also (40, 41)], and is different
from the existing SAS (solvent-accessible surface) type models.
We have designed and implemented a robust level-set method
to numerically minimize the VISM functional with arbitrary
3D geometry (36 -- 38, 42).
Here, for our model ligand-pocket system, we use our level-
set VISM to obtain different hydration states and their solva-
tion free energies, and use the VISM-string method (43, 44)
to find the minimum energy paths connecting such states and
the corresponding transition rates. Such rates are then used in
our continuous-time Markov chain Brownian dynamics simula-
tions, and the related Fokker -- Planck equation calculations, of
the ligand stochastic motion to obtain the mean first-passage
times for the ligand binding and unbinding. We compare our
results with existing explicit-water MD simulations.
The model ligand-receptor system. The generic pocket-ligand
model (45) consists of a hemispherical pocket and a methane-
like molecule; cf. Fig. 1 (A). The pocket, with the radius R = 8
Å and centered at (0, 0, 0), is embedded in a rectangular wall,
composed of apolar atoms aligned in a hexagonal close-packed
grid of lattice constant 1.25 Å. The wall surface is oriented
in xy-plane. The ligand, a single neutral Lennard-Jones (LJ)
sphere, is placed along the pocket symmetry axis, the z-axis,
which is taken to be the reaction coordinate. Fig. 1 (B) -- (D)
depict the cross sections of all the possible VISM surfaces, i.e.,
the stable solute-solvent interfaces separating the solute region
Ωm and solvent region Ωw, representing different hydration
states for a fixed position of ligand.
D RAFT
V (z) = −kBT ln
Fig. 2. (A) Solvation free energies of different VISM surfaces vs. the ligand location.
(B) The equilibrium PMF.
Fig. 2 (B) shows the equilibrium PMF, defined as
(cid:18)X
(cid:19)
−G[Γ(z)]/kBT
e
+ U0(z) + V∞,
[1]
Γ(z)
VISM solvation free energy at Γ(z), and U0(z) =P
where Γ(z) runs over all the VISM surfaces with G[Γ(z)] the
ULJ(ri−
rz) with rz the ligand position vector, ri running through all
the wall atoms, and ULJ(r) a 12 -- 6 LJ potential. The constant
V∞ is chosen so that V (∞) = 0. The PMF agrees well with
the result from MD simulations (17, 46, 47).
ri
Results and Analysis
Multiple hydration states and the potential of mean force
(PMF). We use our level-set method to minimize the VISM
solvation free-energy functional (cf. Eq. [2] in Theory and
Methods) and obtain a VISM surface. By choosing differ-
ent initial solute-solvent interfaces, we obtain different VISM
surfaces describing different hydration states; cf. Fig. 1.
Fig. 2 (A) shows the solvation free energies for different
VISM surfaces against the reaction coordinate z. For z < −0.5
Å, there is only one VISM surface, 1s-dry; cf. Fig. 1 (B). In
addition to 1s-dry, a second VISM surface, 2s-wet, appears for
−0.5 < z < 5 Å; cf. Fig. 1 (D). For 5 < z < 8 Å, there are
Dry-wet transition paths and energy barriers. At a fixed reac-
tion coordinate z with multiple hydration states, we use our
level-set VISM coupled with the string method to calculate
the minimum energy paths (MEPs) that connect these states,
and the corresponding transition states, energy barriers, and
ultimately the transition rates. A string or path here consists
of a family of solute-solvent interfaces, and each point of a
string, which is an interface in our case, is called an image.
In Fig. 3, we display the solvation free energies of images on
MEPs that connect the three hydration states, 1s-dry, 2s-dry,
and 2s-wet, at z = 6 Å. There are two MEPs connecting 1s-dry
(marked (I)) and 2s-dry (marked (IV)). One of them passes
2 www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX
Zhou et al.
(C) 2s-dry(B) 1s-dryz0zBindingUnbinding0 n⌦w⌦mMeri(a)(b)(A)(D) 2s-wetz0 ⌦w⌦mriz0 ⌦w⌦mriz0 ⌦w⌦mri-202468101214SolvationEnergy(kBT)292296300304(A)1s-dry2s-dry2s-wetz(8A)-202468101214PMF(kBT)-6-4-202(B) through the axisymmetric transition state marked (III), and
the other passes through the axiasymmetric transition state
marked (II). Here, symmetry or asymmetry refers to that of
the 3D conformation of the VISM surface. Energy barriers
in the transition from the state 1s-dry to 2s-dry along the
two transition paths are estimated to be 1.09 kBT and 0.52
kBT , respectively. Only one MEP is found to connect 2s-dry
(marked (IV)) and 2s-wet (marked (VI)), and the correspond-
ing transition state (marked (V)) is also found. The MEP
from 1s-dry to 2s-wet always passes through the state 2s-dry.
Fig. 4. Transition energy barriers vs. the reaction coordinate z with −0.5 ≤ z ≤ 4 Å
(top) and 5 ≤ z ≤ 12 Å (middle and bottom). Sym or Asym stands for a MEP with
an axisymmetric or axiasymmetric transition state.
D RAFT
Fig. 3. Solvation free energies of images on MEPs that connect the hydration states
1s-dry (I), 2s-dry (IV), and 2s-wet (VI) (shown in the bottom) with transition states (II),
(III), and (V) (shown on top) and the transition energy barriers for z = 6 Å. In the
middle plots, the horizontal axis is the string parameter α.
Fig. 4 summarizes all the energy barriers in the transitions
from one hydration state to another for each reaction coordi-
nate z. For 0 ≤ z ≤ 4 Å shown in the top of Fig. 4, there are
only two hydration states: 1s-dry and 2s-wet. The 1s-dry has
a lower free energy; cf. Fig. 2 (A), and hence the barrier in
the wetting transition from 1s-dry to 2s-wet (shown in red) is
higher than that in the dewetting transition from 2s-wet to
1s-dry (shown in blue). The dewetting barrier first increases as
the ligand approaches the entrance of the pocket (from z = 4
to z = 1 Å), and then decreases after the ligand enters the
pocket (from z = 1 to z = −0.5 Å). This is because that more
attractive solute-solvent vdW interaction is lost in dewetting
as the ligand-pocket distance reduces from z = 4 to z = 1 Å,
and that the decrease in interfacial energy outweighs the vdW
contribution to the solvation free energy as the distance further
reduces from z = 1 to z = −0.5 Å. Our predictions agree well
with those by the explicit-water MD simulations (17).
For 5 ≤ z ≤ 8 Å, there are three hydration states 1s-dry,
2s-wet, and 2s-dry; cf. Fig. 2 (A). In the middle of Fig. 4, we
plot for z in this range the energy barriers along the MEPs,
both axisymmetric and axiasymmetric, connecting the two
states 1s-dry and 2s-dry; cf. Fig. 3. Note that, as the ligand
approaches the pocket, the solute-solvent interfacial energy
changes rapidly, and hence the barrier in the transition from
1s-dry to 2s-dry increases quickly, while the barrier in the
reverse transition decreases quickly.
In the bottom of Fig. 4, we plot energy barriers for tran-
sitions between the states 2s-dry and 2s-wet in the range
5 ≤ z ≤ 12 Å; cf. Fig. 2 (A). As the ligand-pocket distance
increases, the barrier for the wetting transition (marked red)
first increases, since the newly created solvent region with
attractive solute-solvent vdW interaction decreases. It then
reaches a plateau after the distance is greater than 7 Å. The
pocket dewetting barrier (marked blue) is slightly larger when
the ligand is close to the pocket, since contributions of solute-
solvent vdW interaction are lost during the pocket dewetting.
Kinetics of binding and unbinding. We perform continuous-
time Markov chain (CTMC) Brownian dynamics (BD) simu-
lations and solve the related Fokker -- Planck equation (FPE)
calculations for the ligand stochastic motion with the pocket
dry-wet fluctuations; see Theory and Methods. For compar-
ison, we also perform the usual BD simulations and FPE
calculations without including such fluctuations.
Fig. 5 (A) and (B) show the mean first-passage times (MF-
PTs) for the binding and unbinding, respectively. Note that
the BD simulations and FPE calculations agree with each
other perfectly for both binding and unbinding, without and
with the pocket dry-wet fluctuations, respectively. This vali-
dates mutually the accuracy of our numerical schemes. Note
also that the binding/unbinding MFPT increases/decreases
monotonically as the ligand-pocket distance increases, due to
elongated/shortened ligand travel.
In Fig. 5 (A), we see that the MFPT for binding is very
small if z < −0.5 Å. This is because the ligand diffusion
constant Din inside the pocket is large and the PMF is highly
attractive; cf. Fig. 2 (B). As the initial position z increases
from 0 Å to 5 Å, the difference between the two MFPTs with
and without the pocket dry-wet fluctuations increases from
nearly 0 ps to 100 ps. Such an increasing difference results from
the existence of the hydration state 2s-wet in this range, and
the solvation free energy of this state increases as the ligand
moves from z = 5 Å to z = 0 Å; cf. Fig. 2 (A). The pocket
dry-wet fluctuations thus decelerate considerably the ligand-
pocket association. Such deceleration has been explained by
the reduced diffusivity of the ligand in the vicinity of pocket
entrance due to the slow solvent fluctuations (18).
Our predictions of the MFPT for binding, with the dry-wet
fluctuations included, agree very well with the explicit-water
MD simulations (18), improving significantly over those with-
out such fluctuations. Note that our model predicts somewhat
shorter binding times than the MD simulations for 1 < z < 6 Å.
Zhou et al.
PNAS June 19, 2019
vol. XXX no. XX 3
α00.20.40.60.811.21.41.61.82SolvationEnergy(kBT)2993003013023031.09 kBT0.52 kBT0.70 kBT1s-dry2s-dry2s-wetTransition StateTransition StateTransition State(I)(I)(II)(II)(III)(III)(IV)(IV)(V)(V)(VI)(VI)-0.500.511.522.533.54048122s-wet to 1s-dry1s-dry to 2s-wet55.566.577.58EnergyBarrier(kBT)0246Sym 1s-dry to 2s-drySym 2s-dry to 1s-dryAsym 1s-dry to 2s-dryAsym 2s-dry to 1s-dryz(A)567891011120.60.91.22s-wet to 2s-dry2s-dry to 2s-wet When the ligand is far away, there are only two VISM
surfaces, 2s-dry and 2s-wet, cf. Fig. 2 (A). For such a case, our
BD simulations predict the probability 32% of a wet pocket
(i.e., χp = 0.32 for large z) in the binding and unbinding pro-
cesses. This is perfectly consistent with the equilibrium prob-
ability e−G[Γ2s−wet]/kBT /(e−G[Γ2s−dry]/kBT + e−G[Γ2s−wet]/kBT )
predicted by our VISM theory. We observe that the pocket
hydration peaks at the entrance of the pocket in binding, agree-
ing well with MD simulations (17, 18), where it was argued
that stronger pocket hydration is induced by the penetration of
the ligand solvation shell. When the ligand enters the pocket
the latter becomes dry as anticipated.
In comparison, the maximum pocket hydration for un-
binding is shifted a bit away from the pocket. This kinetic
asymmetry or "translational mismatch" can be explained as
well by the asymmetric hydration states of the ligand, see
Fig. 5 (E), which exits the pocket without a complete solva-
tion shell. This behavior is reminiscent of a hysteresis, that is,
the hydration states during the ligand passage depend on the
history of the ligand, i.e., where it comes from.
The standard deviations of pocket hydration shown in
Fig. 5 (D) depict that the dry-wet fluctuations have local
maxima close to the pocket entrance (z ' 3− 5 Å) and behave
also significantly different for binding and unbinding. The
corresponding standard deviations of ligand hydration shown
in Fig. 5 (F) show massively unstable hydration (i.e., large
peaks) close to the pocket entrance, while inside and far away
from the pocket the fluctuations are zero, indicating a very
stable (de)hydration state. Again the peaks are at different
locations for binding versus unbinding, reflecting the hysteresis
and memory of dry-wet transitions during ligand passage.
Conclusions
We have developed an implicit-solvent approach, coupling our
VISM, the string method, and multi-state CTMC BD simu-
lations, for studying the kinetics of ligand-receptor binding
and unbinding, particularly the influence of collective solvent
fluctuations on such processes. Without any explicit descrip-
tions of individual water molecules, our predictions of the
MFPT for the binding process, which is decelerated by the
solvent fluctuations around the pocket, agree very well with
the less efficient explicit-water MD simulations. Moreover,
we find surprisingly that the solvent fluctuations accelerate
the ligand unbinding from the pocket, which involves a much
larger timescale and is thus more challenging for explicit-water
MD simulations (26, 30). Importantly, our implicit-solvent ap-
proach indicates that the water effects are controlled by a few
key physical parameters and mechanisms, such as polymodal
nano-capillarity based on surface tension of the solute-solvent
interface and the coupling of the random interface forces to
the ligand's diffusive motion.
Our approach provides a promising new direction in ef-
ficiently probing the kinetics, and thermodynamics, of the
association and dissociation of complex ligand-receptor sys-
tems, which have been studied mostly using enhanced sampling
techniques (18, 25, 26, 28, 30, 32). Our next step is to extend
our approach for more realistic systems with general reaction
coordinates and different techniques for sampling transition
paths (48, 49). Our VISM can treat efficiently the electro-
static interactions using the Poisson -- Boltzmann theory (38).
To account for the flexibility of the ligand and receptor in their
D RAFT
Fig. 5. The MFPT for: (A) the binding of ligand that starts from zinit = z and reaches
the pocket at zL = −4 Å; and (B) the unbinding of ligand that starts from zinit = z
and reaches zR = 15.5 Å, predicted by: BD simulations without (BD No SolFlt) and
with (BD With SolFlt) the dry-wet fluctuations; and FPE calculations without (FP No
SolFlt) and with (FP With SolFlt) the dry-wet fluctuations, respectively. Note that the
time unit on the vertical axis in (B) is ns while that in (A) is ps. The MFPT obtained
by explicit-water MD simulations (MD) (18) is also shown in (A). (C) -- (F) The mean
values and standard deviations of the pocket and ligand hydration states χp(z) and
χl(z), respectively, against the ligand location z during the nonequilibrium binding
process from the BD simulations starting at zinit = 6 Å (cf. (C) and (E)) and the
unbinding process starting at zinit = −2 Å (cf. (D) and (F)).
In this region, the hydration fluctuations are maximal, and
this visible but relatively small (when compared to the MFPT
from the farthest distance) discrepancy reflects some of the
approximations of our implicit-solvent theory and the model
reduction on just a few states.
Fig. 5 (B) shows that the timescale for unbinding is signifi-
cantly larger than that of the binding, by nearly three orders
of magnitude. Without the pocket dry-wet fluctuations, the
unbinding MFPT is constant for z < 4 Å and decreases lin-
early for z > 4 Å. Note that the MFPT for binding in this case
also starts to increase significantly at z = 4 Å; cf. Fig. 5 (A).
With the pocket dry-wet fluctuations, the unbinding MFPT
is much smaller, since the solvation free energy of the 2s-wet
state is higher when the ligand is closer to the pocket (cf.
Fig. 2 (A)), favoring the ligand unbinding. In this case, the
MFPT remains constant up to z = 2 Å and then decays almost
linearly. This suggests that the wetting transitions occur if
z > 2 Å. Note from Fig. 5 (A) that the binding MFPT starts
increasing rapidly also around z = 2 Å.
We now study the interesting hydration of the pocket
and ligand individually during the non-equilibrium bind-
ing/unbinding processes. For this, we define a pocket hy-
dration parameter to be χp(z) = 0 or 1 if the pocket is dry or
wet, respectively. Analogously, we set for the ligand χl(z) = 0
or 1 if the ligand is dry or wet, respectively. The values 0 and
1 of these ligand-position dependent random variables χp(z)
and χl(z) are defined by the three hydration states 1s-dry,
2s-dry, and 2s-wet (cf. Fig. 1 (B) -- (D)) as follows:
χp(z) = 0 and χl(z) = 0 for a 1s-dry VISM surface;
χp(z) = 0 and χl(z) = 1 for a 2s-dry VISM surface;
χp(z) = 1 and χl(z) = 1 for a 2s-wet VISM surface.
Fig. 5 (C) -- (F) show the mean values, hχp(z)i and hχl(z)i,
and the standard deviations, σ[χp(z)] and σ[χl(z)], during the
binding and the unbinding processes, respectively.
4 www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX
Zhou et al.
-4-202468101214Binding MFPT (ps)0100200300400500(A)BD No SolFltFP No SolFltBD With SolFltFP With SolFltMD-4-202468101214Unbinding MFPT (ns)020406080100120(B)BD No SolFltFP No SolFltBD With SolFltFP With SolFltz(A)-4-202468101214<χp(z)>00.20.40.60.81(C)BindingUnbindingz(A)-4-202468101214σ[χp(z)]00.20.4(D)BindingUnbindingz(A)-4-202468101214<χl(z)>00.20.40.60.81(E)BindingUnbindingz(A)-4-202468101214σ[χl(z)]00.20.4(F)BindingUnbinding binding and unbinding, we shall expand our solvation model
to include the solute molecular mechanical interactions (50).
1s-dry, 2s-dry, and 2s-wet (cf. Fig. 1) as the states 0, 1, and 2,
respectively. We define for each i ∈ {0, 1, 2} the potential
Theory and Methods
Variational implicit-solvent model (VISM). We consider the sol-
vation of solute molecules, with all the solute atomic positions
r1, . . . , rN, in an aqueous solvent that is treated implicitly
as a continuum. (For our model ligand-pocket system, the
solute atoms include those of the concave wall and the single
atom of the ligand; cf. Fig. 1.) A solute-solvent interface Γ
is a closed surface that encloses all the solute atoms but no
solvent molecules. The interior and exterior of Γ are the solute
and solvent regions, denoted Ωm and Ωw, respectively. We
introduce the VISM solvation free-energy functional (34, 35):
G[Γ] = ∆P vol (Ωm) +
γ dS + ρ0
U(r) dV + Ge[Γ].
[2]
Z
Γ
Z
Ωw
Vi(z) = Gi(z) + U0(z),
[3]
where Gi(z) is the solvation free energy of the ith state at z
(cf. Fig 2 (A)) and U0(z) is the ligand-pocket vdW interaction
potential defined below Eq. [1]. We set Vi(z) = 0 if the ith
state does not exist at z.
With the energy barriers summarized in Fig. 4, we can
calculate for each z the rate Rij = Rij(z) of the transition
from one state i to another j. If a MEP from i to j passes
through another state k (cf. Fig. 3), then we set Rij(z) = 0.
If there is only one MEP connecting i and j (see, e.g., z < 4
in Fig. 2), then Rij = R0e−Bij(z)/kBT with Bij(z) the energy
barrier from i to j and R0 a constant prefactor, describing the
intrinsic time scale of water dynamics in the pocket. Finally,
if there are two MEPs (axisymmetric and axiasymmetric)
connecting i and j, we use the same formula but with Bij an
effective barrier. For instance, consider i and j the states (I)
and (IV) in Fig. 3, respectively. The two transition states are II
and III, respectively. We set Bij(z) = BI,IV(z) = p(GII− GI)+
(1−p)(GIII−GI), where p = e−(GII−GI)/kBT /(e−(GII−GI)/kBT +
e−(GIII−GI)/kBT ) and GA is the VISM solvation free energy
at state A ∈ {I, II, III}. To determine the prefactor R0, we
calculate the equilibrium (i.e., the large z limit) energy barriers
Bdw and Bwd in the pocket dry-wet and wet-dry transitions,
respectively, and equate [R0(e−Bdw/kBT + e−Bwd/kBT )]−1 with
the time scale for the relaxation of water fluctuation of 10 ps
as predicted by explicit-water MD simulations (18). See SI for
discussions on the sensitivity of the results on R0.
Continuous-time Markov chain (CTMC) Brownian dynamics
(BD) simulations and the mean first-passage time (MFPT). To
include explicitly the dry-wet fluctuations, we introduce a
position-dependent, multi-state, random variable η = η(z):
η(z) = i (i ∈ {0, 1, 2}) if the system is in the ith hydration state
when the ligand is located at z, with the transition rates Rij(z)
given above. We define the potential Vfluc(η, z) = Vi(z) (cf.
Eq. [3]) if η(z) = i. (52). The random position z = z(t) = zt
of the ligand is now determined by our CTMC BD simulations
in which we solve the stochastic differential equation
(cid:21)
dt +p2D(zt) dξt.
∂Vfluc(η(zt), zt)
+ D
0(zt)
∂z
Here, the partial derivative of Vfluc is with respect to its second
variable, D(z) is an effective diffusion coefficient that smoothly
interpolates the diffusion coefficients Din and Dout inside and
outside the pocket, respectively, and ξt is the standard Brow-
nian motion. Solutions to this equation are constrained by
zt ∈ [zL, zR] for some zL and zR. For the simulation of a bind-
ing process, we reset the value of zt to be 2zR − zt if zt ≥ zR,
and we stop the simulation if zt ≤ zL. For the simulation of an
unbinding process, we reset the value of z(t) to be zL if zt ≤ zL,
and we stop the simulation if zt ≥ zR. The distribution of η(z0)
for an initial ligand position z0 is set based on the equilibrium
j=0 e−Gj /kBT (i = 0, 1, 2), where Gi
is the solvation free energy of the ith hydration state at z0.
We run our CTMC BD simulation for the ligand starting
at a position z0 = zinit, and record the time at which the
ligand reaches zL (or zR) for the first time for a binding (or
unbinding) simulation. We run simulations for 3, 000 times
and average these times to obtain the corresponding MFPTs.
probabilities e−Gi/kBT /P2
bulk solvent (i.e., water) density, and U(r) =PN
Here, ∆P is the difference of pressures across the interface
Γ, γ is the solute-solvent interface surface tension, ρ0 is the
i=1 Ui(r−ri)
with each Ui a standard 12 -- 6 LJ potential. We take γ =
γ0(1 − 2τ H), where γ0 is the surface tension for a planar
interface, τ is the curvature correction coefficient often known
as the Tolman length (51), and H is the local mean curvature.
The last term Ge[Γ] is the electrostatic part of the solvation
free energy, which we will not include in this study.
Minimizing the functional Eq. [2] among all the solute-
solvent interfaces Γ determines a stable, equilibrium, solute-
solvent interface, called a VISM surface, and the corresponding
solvation free energy. A VISM surface is termed dry, represent-
ing a dry hydration state, if it loosely wraps up all the solute
atoms with enough space for a few solvent molecules, or wet,
representing a wet hydration state, if it tightly wraps up all
the solute atoms without extra space for a solvent molecule.
D RAFT
(cid:20)
− D(zt)
kBT
dzt =
Implementation by the level-set method. Beginning with an
initially guessed solute-solvent interface, our level-set method
evolves the interface step by step in the steepest descent direc-
tion until a VISM surface is reached. Different initial surfaces
may lead to different final VISM surfaces. See Supporting
Information (SI) for more details of implementation.
The level-set VISM-string method for minimum energy paths
(MEPs). Let us fix all the solute atomic positions and assume
that Γ0 and Γ1 are two VISM surfaces (e.g., dry and wet
surfaces). We apply the string method (43, 44) to find a MEP
that connects Γ0 and Γ1. A string or path here is a family
of solute-solvent interfaces {Γα}α∈[0,1] that connects the two
states Γ0 and Γ1. Such a string is a MEP, if it is orthogonal to
the level surfaces of the VISM free-energy functional. To find a
MEP connecting Γ0 and Γ1, we select some initial images (i.e.,
points of a string), and then update them iteratively to reach
a MEP. Different initial images may lead to different MEPs.
Once a MEP is found, we can then find a saddle point on
the MEP. Alternatively, we can fix one of the VISM surfaces,
select some initial images, and allow the last image to climb
up to reach a saddle point, and then find the MEP connecting
the two VISM surfaces passing the saddle point. We refer to
SI for more details on our implementation of the method.
Consider now our ligand-pocket system; cf. Fig. 1. For any
reaction coordinate z, we label all the three hydration states
Zhou et al.
PNAS June 19, 2019
vol. XXX no. XX 5
Fokker -- Planck equations (FPE) and the MFPT. The probabil-
ity densities Pi = Pi(z, t) for the ligand at location z at time
t with the system in the ith hydration state are determined
by the generalized FPEs (25, 52):
+ 1
∂z
kBT
Rji(z)Pj −
io
(cid:18) X
n
+ X
h ∂Pi
∂Pi
∂t
= ∂
∂z
Rij(z)
Pi
0
i (z)Pi
V
(cid:19)
D(z)
0≤j≤2,j6=i
0≤j≤2,j6=i
double integral ofP2
for i = 0, 1, 2, where Vi is defined in Eq. [3]. These equations
are solved for zL < z < zR, with the boundary conditions
Pi(zL, t) = 0 and ∂zPi(zR, t) = 0 for binding, and ∂zPi(zL, t)+
(1/kBT )V 0
i (zL)Pi(zL, t) = 0 and Pi(zR, t) = 0 for unbinding,
respectively. The initial conditions are Pi(z, 0) = δ(z − zinit)
if the ligand is initially at zinit. We obtain the MFPT as the
i=0 Pi(z, t) over (z, t) ∈ [zL, zR] × [0,∞).
Parameters. We set the temperature T = 298 K, bulk water
density ρ0 = 0.033 Å−3, the solute-water surface tension
constant γ0 = 0.143 kBT /Å2 (kB is the Boltzmann constant),
and the Tolman length τ = 0.8 Å. We set ∆P vol (Ωm) = 0
as it is relatively very small. The LJ parameters for the wall
particles, ligand, and water are εwall = 0.000967 kBT and
σwall = 4.152 Å, εligand = 0.5 kBT and σligand = 3.73 Å, and
εwater = 0.26 kBT and σwater = 3.154 Å, respectively. The
interaction LJ parameters are determined by the Lorentz --
Berthelot mixing rules. The prefactor R0 = 0.13 ps−1. The
diffusion constants are Dout = 0.26 Å2/ps (18), and Din =
1 Å2/ps. The cut-off position distinguishing the inside and
outside of the pocket is zc = −0.5 Å. BD simulations and FPE
calculations are done for zL ≤ z ≤ zR with zL = −4 Å and
zR = 15.5 Å.
SZ was supported in part by NSF of
ACKNOWLEDGMENTS.
Jiangsu Province, China, through grant BK20160302, NSFC through
grant NSFC 21773165 and NSFC 11601361, and Soochow University
through a start-up grant Q410700415. RGW and JD thank the
DFG for financial support. JD also acknowledges funding from
the ERC within the Consolidator Grant with Project No. 646659 --
NANOREACTOR. Work in the McCammon group is supported in
part by NIH, NBCR, and SDSC. LTC and BL were supported in
part by the NSF through the grant DMS-1620487. SZ thanks Dr.
Yanan Zhang for helpful discussions on the string method.
13. Young T, Abel R, Kim B, Berne B, Friesner R (2007) Motifs for molecular recognition exploiting
hydrophobic enclosure in protein-ligand binding. Proc. Natl. Acad. Sci. USA 104(3):808 -- 813.
14. Qvist J, Davidovic M, Hamelberg D, Halle B (2008) A dry ligand-binding cavity in a solvated
protein. Proc. Natl. Acad. Sci. USA 105(17):6296 -- 6301.
15. Wang L, Berne BJ, Friesner RA (2011) Ligand binding to protein-binding pockets with wet
and dry regions. Proc. Natl. Acad. Sci. USA 108:1326 -- 1330.
16. Beuming T, et al. (2012) Thermodynamic analysis of water molecules at the surface of pro-
teins and applications to binding site prediction and characterization. Proteins 80:871 -- 883.
17. Setny P, et al. (2009) Dewetting-controlled binding of ligands to hydrophobic pockets. Phys.
Rev. Lett. 103:187801.
18. Setny P, Baron R, Kekenes-Huskey P, McCammon JA, Dzubiella J (2013) Solvent fluctuations
in hydrophobic cavity-ligand binding kinetics. Proc. Natl. Acad. Sci., USA 110:1197 -- 1202.
19. Huang X, Margulis CJ, Berne BJ (2003) Dewetting-induced collapse of hydrophobic particles.
Proc. Natl. Acad. Sci., USA 100(21):11953 -- 11958.
20. Liu P, Huang X, Zhou R, Berne BJ (2005) Observation of dewetting transition in the collapse
of the melittin tetramer. Nature 437:159 -- 162.
21. Baron R, Setny P, McCammon JA (2010) Water in cavity-ligand recognition. J. Am. Chem.
Soc. 132:12091 -- 12097.
22. Weiss RG, Setny P, Dzubiella J (2017) Principles for tuning hydrophobic ligand-receptor bind-
ing kinetics. J. Chem. Theory Comput. 13(6):3012 -- 3019.
23. Morrone JA, Li J, Berne BJ (2012) Interplay between hydrodynamics and the free energy
surface in the assembly of nanoscale hydrophobes. J. Phys. Chem. B 116(1):378 -- 389.
24. Li J, Morrone JA, Berne BJ (2012) Are hydrodynamic interactions important in the kinetics of
hydrophobic collapose? J. Phys. Chem. B 116:378 -- 389.
25. Mondal J, Morrone J, Berne BJ (2013) How hydrophobic drying forces impact the kinetics of
molecular recognition. Proc. Natl. Acad. Sci., USA 110(33):13277 -- 13282.
26. Tiwary P, Mondal J, Morrone JA, Berne BJ (2015) Role of water and steric constraints in the
kinetics of cavity-ligand unbinding. Proc. Natl. Acad. Sci., USA 112(39):12015 -- 12019.
27. Weiss RG, Setny P, Dzubiella J (2016) Solvent fluctuations induce non-Markovian kinetics in
hydrophobic pocket-ligand binding. J. Phys. Chem. B 120(33):8127 -- 8136.
28. Dror RO, et al. (2011) Pathway and mechanism of drug binding to G-protein-coupled recep-
tors. Proc. Natl Acad. Sci. USA 108(32):13118 -- 13123.
29. Plattner N, Noé F (2015) Protein conformational plasticity and complex ligand-binding kinetics
explored by atomistic simulations and Markov models. Nat. Commun. 6:7653.
30. Tiwary P, Limongelli V, Salvalaglio M, Parrinello M (2015) Kinetics of protein-ligand unbinding:
Predicting pathways, rates, and rate-limiting steps. Proc. Natl. Acad. Sci., USA 112(5):E386 --
E391.
31. Jagger BR, Lee CT, Amaro RE (2018) Quantitative ranking of ligand binding kinetics with a
multiscale milestoning simulation approach. J. Phys. Chem. Lett. 9(17):4941 -- 4948.
32. Miao Y, Huang Y, Walker RC, McCammon JA, Chang C (2018) Ligand binding pathways and
conformational transitions of the HIV protease. Biochemistry 57:1533 -- 1541.
33. Bruce NJ, Ganotra GK, Kokh DB, Sadiq SK, Wade RC (2018) New approaches for computing
ligand-receptor binding kinetics. Curr. Opin. Struct. Biol. 49:1 -- 10.
34. Dzubiella J, Swanson JMJ, McCammon JA (2006) Coupling hydrophobicity, dispersion, and
electrostatics in continuum solvent models. Phys. Rev. Lett. 96:087802.
35. Dzubiella J, Swanson JMJ, McCammon JA (2006) Coupling nonpolar and polar solvation free
energies in implicit solvent models. J. Chem. Phys. 124:084905.
36. Cheng LT, Dzubiella J, McCammon JA, Li B (2007) Application of the level-set method to the
implicit solvation of nonpolar molecules. J. Chem. Phys. 127:084503.
37. Wang Z, et al. (2012) Level-set variational implicit solvation with the Coulomb-field approxi-
mation. J. Chem. Theory Comput. 8:386 -- 397.
38. Zhou S, Cheng LT, Dzubiella J, Li B, McCammon JA (2014) Variational implicit solvation with
Poisson -- Boltzmann theory. J. Chem. Theory Comput. 10(4):1454 -- 1467.
39. Lum K, Chandler D, Weeks JD (1999) Hydrophobicity at small and large length scales. J.
Phys. Chem. B 103:4570 -- 4577.
40. Ramirez R, Borgis D (2005) Density functional theory of solvation and its relation to implicit
solvent models. J. Phys. Chem. B 109:6754 -- 6763.
41. Bates PW, Chen Z, Sun YH, Wei GW, Zhao S (2009) Geometric and potential driving forma-
tion and evolution of biomolecular surfaces. J. Math. Biol. 59:193 -- 231.
D RAFT
1. Babine RE, Bender SL (1997) Molecular recognition of protein-ligand complexes: Applica-
tions to drug design. Chem. Rev. 97(5):1359 -- 1472.
2. Pan AC, Borhani DW, Dror RO, Shaw DE (2013) Molecular determinants of drug -- receptor
binding kinetics. Drug Discov. Today 18:668 -- 673.
3. Levy Y, Onuchic J (2006) Water mediation in protein folding and molecular recognition. Ann.
Rev. Biophys. Biomol. Struct. 35:389 -- 415.
4. Baron R, McCammon JA (2013) Molecular recognition and ligand association. Annu. Rev.
Phys. Chem. 64:151 -- 175.
5. Hummer G, Garde S, Garcia S, Pratt LR (2000) New perspectives on hydrophobic effects.
Chem. Phys. 258:349 -- 370.
6. Chandler D (2005) Interfaces and the driving force of hydrophobic assembly. Nature 437:640 --
647.
7. Berne BJ, Weeks JD, Zhou R (2009) Dewetting and hydrophobic interaction in physical and
biological systems. Annu. Rev. Phys. Chem. 60:85 -- 103.
8. Ben-Amotz D (2016) Water-mediated hydrophobic interactions. Annu. Rev. Phys. Chem.
67:617 -- 638.
9. Gilson MK, Given JA, Bush BL, McCammon JA (1997) The statistical-thermodynamic basis
for computation of binding affinities: A critical review. Biophys. J. 72:1047 -- 1069.
10. Gilson MK, Zhou HX (2007) Calculation of protein-ligand binding affinities. Annu. Rev. Bio-
phys. Biomol. Struct. 36:21 -- 42.
11. Bernetti M, Cavalliab A, Mollica L (2017) Protein -- ligand (un)binding kinetics as a new
paradigm for drug discovery at the crossroad between experiments and modelling. Med.
Chem. Commun. 8:534 -- 550.
12. Schuetz DA, et al. (2017) Kinetics for drug discovery: An industry-driven effort to target drug
residence time. Drug Discov. Today 22:896 -- 911.
42. Cheng LT, Li B, Wang Z (2010) Level-set minimization of potential controlled Hadwiger valua-
tions for molecular solvation. J. Comput. Phys. 229:8497 -- 8510.
43. E W, Ren W, Vanden-Eijnden E (2007) Simplified and improved string method for computing
the minimum energy paths in barrier-crossing events. J. Chem. Phys. 126:164103.
44. Ren W, Vanden-Eijnden E (2013) A climbing string method for saddle point search. J. Chem.
Phys. 138:134105.
45. Setny P, Geller M (2006) Water properties inside nanoscopic hydrophobic pockets. J. Chem.
Phys. 125:144717.
46. Setny P (2007) Water properties and potential of mean force for hydrophobic interactions
J. Chem. Phys.
of methane and nanoscopic pockets studied by computer simulations.
127:054505.
47. Cheng LT, et al. (2009) Interfaces and hydrophobic interactions in receptor-ligand systems: A
level-set variational implicit solvent approach. J. Chem. Phys. 131:144102.
48. Maragliano L, Fischer A, Vanden-Eijnden E, Ciccotti G (2006) String method in collective vari-
ables: Minimum free energy paths and isocommittor surfaces. J. Chem. Phys. 125:024106.
Identification of collective variables and metastable
49. Sittel F, Stock G (2018) Perspective:
states of protein dynamics. J. Chem. Phys. 149:150901.
50. Cheng LT, et al. (2009) Coupling the level-set method with molecular mechanics for variational
implicit solvation of nonpolar molecules. J. Chem. Theory Comput. 5:257 -- 266.
51. Tolman RC (1949) The effect of droplet size on surface tension. J. Chem. Phys. 17:333 -- 337.
52. Doering CR, Gadoua JC (1992) Resonant activation over a fluctuating barrier. Phys. Rev.
Lett. 69(16):2318 -- 2321.
6 www.pnas.org/cgi/doi/10.1073/pnas.XXXXXXXXXX
Zhou et al.
|
1710.01954 | 2 | 1710 | 2018-03-19T15:11:26 | Actuated rheology of magnetic micro-swimmers suspensions : emergence of motor and brake states | [
"physics.bio-ph",
"cond-mat.soft",
"physics.flu-dyn"
] | We study the effect of magnetic field on the rheology of magnetic micro-swimmers suspensions. We use a model of a dilute suspension under simple shear and subjected to a constant magnetic field. Particle shear stress is obtained for both pusher and puller types of micro-swimmers. In the limit of low shear rate, the rheology exhibits a constant shear stress, called actuated stress, which only depends on the swimming activity of the particles. This stress is induced by the magnetic field and can be positive (brake state) or negative (motor state). In the limit of low magnetic fields, a scaling relation of the motor-brake effect is derived as a function of the dimensionless parameters of the model. In this case, the shear stress is an affine function of the shear rate. The possibilities offered by such an active system to control the rheological response of a uid are finally discussed. | physics.bio-ph | physics | Actuated rheology of magnetic micro-swimmers suspensions: emergence of motor and
brake states
Benoit Vincenti1, Carine Douarche2, Eric Clement1
1 Laboratoire de Physique et M´ecanique des Milieux H´et´erog`enes (PMMH),
CNRS, ESPCI Paris, PSL Research University, Sorbonne Universit´e,
Universit´e Paris Diderot, 10 rue Vauquelin, 75005 Paris, France
2 Laboratoire de Physique des Solides, CNRS, Paris-Sud University,
Paris-Saclay University, UMR 8502 91405 Orsay Cedex, France.
(Dated: March 20, 2018)
We study the effect of magnetic field on the rheology of magnetic micro-swimmers suspensions.
We use a model of a dilute suspension under simple shear and subjected to a constant magnetic
field. Particle shear stress is obtained for both pusher and puller types of micro-swimmers. In the
limit of low shear rate, the rheology exhibits a constant shear stress, called actuated stress, which
only depends on the swimming activity of the particles. This stress is induced by the magnetic field
and can be positive (brake state) or negative (motor state). In the limit of low magnetic fields, a
scaling relation of the motor-brake effect is derived as a function of the dimensionless parameters
of the model. In this case, the shear stress is an affine function of the shear rate. The possibilities
offered by such an active system to control the rheological response of a fluid are finally discussed.
I. INTRODUCTION
Many micro-organisms are able to move autonomously in fluids at a very low Reynolds number [1] and recently,
micron-size artificial particulate systems were designed to insure self-propulsion using either chemical [2, 3], magnetic
excitations [4, 5] or even the mixing of biological material with mechanical parts [6, 7]. The hydrodynamics of
suspensions laden with such self-propelled objects is currently the focus of many fundamental studies [8, 9] and it has
been found that original macroscopic constitutive properties can stem from the swimming activity of the suspended
particles [10–21, 41]. According to the intrinsic nature of the propulsive mechanism, one can observe specific increases
[14] (puller swimmers) or decreases (pusher swimmers) [13, 20, 21] of the viscosity with the swimmer concentration.
Recent theory predicted an additional "swimming pressure" contribution that will eventually contribute, at low shear
rate, to lower the viscosity for both types of swimmers [22]. Furthermore, it was found experimentally that in
an intermediate range of concentrations, the macroscopic viscosity may even cancel in analogy with the superfluid
transition [21, 23, 24] of quantum liquids. These recent experimental results and the models proposed to account for
them are reported in a review of fluid mechanics [43].
In nature, some strains of bacteria are able to synthesize and assemble linear arrays of nano-magnets and have
developed a biological sensitivity to the magnetic field direction [25, 26]. Such magnetotactic bacteria are able to
move preferentially to one of the magnetic poles and are called accordingly north-seekers (NS) or south-seekers (SS).
Recently, these suspensions were found to exhibit complex collective behaviors under flow and magnetic field [27].
In this article, we consider suspensions of motile elongated particles bearing an intrinsic magnetic moment along
their swimming direction.
In the simple magnetotactic model we present here, we do not consider any biological
feedback on the swimming direction in response to the magnetic field. The magnetotactic sensitivity is only due
to a passive alignment in the direction of the field. We are interested in understanding how the application of an
external magnetic field can modify the macroscopic rheology of the suspension. The suspension is subjected to a
simple shear and a constant magnetic field is applied at a given orientation with respect to the flow direction. First,
the swimming orientation distribution is computed in the framework of a Fokker-Planck equation that includes a
stochastic disorientation process. Then, in this framework, we compute the particle-borne shear stress and establish,
for any type of swimmer (pusher or puller), the emergence of new rheological states induced by the magnetic field
and imputable only to the swimming activity of the particles. We finally discuss these results in the perspective of
reproducing these specific states with magnetotactic bacteria or artificial micro-swimmers and using them to control
the flow.
8
1
0
2
r
a
M
9
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
2
v
4
5
9
1
0
.
0
1
7
1
:
v
i
X
r
a
2
0.04
0.08
0.12
0.16
0.05
0.15
0.25
0.1
0.2
0.3
FIG. 1: Left : 3D parameterization of a bacterium in spherical coordinates (θ,φ). The magnetic field B = Bb is contained in
the (x,y) plane and its orientation in this plane is given by the angle α. Right : 3D representation of the orientation distribution
function for P em = 1, α = 45◦ and P eH = 10−4 (a), P eH = 4 (b) and P eH = 10 (c).
II. MODEL OF MAGNETIC MICRO-SWIMMERS IN A SIMPLE SHEAR FLOW
The active magnetic model we use consists of rod-shaped particles bearing a magnetic moment m = mp pointing in
the swimming direction p (NS). The swimmer is described as an ellipsoidal slender rod of aspect ratio r = L/a ≫ 1,
where L is its total length and a its equatorial diameter. The swimming mechanism can either be of the pusher or
puller type and its active hydrodynamic field is simplified as a force dipole of strength ǫσ0 [1], where σ0 is a positive
quantity and ǫ = 1 for pullers, −1 for pushers. We will only deal with dilute suspensions of number density n such
that na2L ≪ 1.
A simple shear flow characterized by a velocity v = vx x and a shear rate γ = ∂vx
∂y is applied to the suspension. The
magnetic field B = Bb is oriented at an angle α from the flow direction x (see figure 1). Importantly, because the
shear and magnetic fields are spatially homogeneous, the orientation and spatial degrees of freedom are not coupled i.e.
the orientation of each particle does not depend on space. In particular, only reorientation processes are important.
These processes are described by a kinetic equation of the Jeffery-Bretherton type [28, 29] that includes a magnetic
part due to the torque m × B [30]. Thus, the kinetic equation governing p is :
p =
dp
dt
= (I − pp)(βE + Ω) · p + Ωm × p ,
(1)
where the first term in the right-hand side of the equation stands for the flow contribution. I is the identity tensor,
E = (1/2)(cid:0)∇v + (∇v)T(cid:1) is the strain-rate tensor and Ω = (1/2)(cid:0)∇v − (∇v)T(cid:1) is the vorticity tensor. β = (r2 −
1)/(r2 + 1) ≃ 1 is the Bretherton parameter which will set to β = 1 from now on. Ωm = (mB/ξr) p × b is the
rotation vector of the bacterium towards the magnetic field direction. The magnetic moment m of the particle
relaxes towards the direction of the magnetic field with a characteristic time ω−1
m = ξr/(mB), where ξr is the
rotational friction coefficient of the particle which can be computed in the framework of the slender body theory :
ξr = πη0L3/ (3 ln (2L/a)).
Another source of swimming disorientation arises from a rotational diffusion term characterized by a coefficient
Dr which represents either a Brownian noise or the effect of a run and tumble process characterizing the bacterium
motility. Under these hypothesis, the steady-state orientation distribution of the particle Ψ(θ, φ) is solution of the
following Fokker-Planck equation :
∇s · ( pΨ) = Dr∇2
sΨ ,
(2)
where ∇s is the gradient operator on the unit sphere. The particle orientation is parameterized in spherical coordi-
nates : θ is the azimuthal angle while φ is the meridian angle (see figure 1).
Equation (2) contains three non-dimensional parameters : the hydrodynamic rotational Peclet number P eH = γ/Dr ;
the ratio of the rotational diffusion time to the magnetic relaxation time, P em = ωm/Dr, which we call magnetic Peclet
number ; and α, the magnetic field orientation. A detailed expansion of equation (2) in terms of these parameters
reads :
∇2
sΨ − P eHΓshear(Ψ) + P em [cos(α)Γmx(Ψ) + sin(α)Γmy(Ψ)] = 0
(3)
where Γshear, Γmx and Γmy are the following differential spherical operators :
Γshear(Ψ) =
sin(2φ)
2 sin θ
∂
∂θ(cid:0)sin2 θ cos θΨ(cid:1) −
Γmx(Ψ) = 2 sin θ cos φΨ − cos θ cos φ
Γmy(Ψ) = 2 sin θ sin φΨ − cos θ sin φ
∂
∂φ(cid:0)sin2 φΨ(cid:1)
sin φ
sin θ
cos φ
sin θ
∂Ψ
∂φ
∂Ψ
∂φ
+
−
∂Ψ
∂θ
∂Ψ
∂θ
3
(4)
(7)
Solving equation (3) in 3D requires numerical tools. We used an expansion of Ψ on a spherical harmonics basis (see
Strand et al.
[36] and Satoh [30] for technical details). The application of a magnetic field creates a preferential
alignment of the particles in its direction in competition with both the alignment on the flow axis due to shear
and the disorientation process due to the rotational diffusivity. For P em = 1, for which magnetic alignment is
equivalent to diffusion disorientation, and orientation angle α = 45◦, we display on figure 1 the orientation distribution
of the magnetic rod for P eH = 10−4 (flow orientation negligible compared to magnetic orientation), P eH = 4
(equivalent contributions of flow and magnetic orientations), P eH = 10 (dominant flow orientation) to show the
relative importance of the magnetic field compared to the flow orientation. While P eH becomes important compared
to P em, the maximum of the orientation distribution becomes progressively aligned along the flow direction and the
distribution becomes symmetric by rotation of π around the z-axis, due to Jeffery orbits.
Now we investigate the consequences of the swimming orientation distribution induced by the magnetic field on
the mechanical response of the suspension. Following the approach of Saintillan [32] (see also Haines et al. [31]) who
adapted the original method developed by Leal and Hinch [37] for Brownian fibers, the total dimensional stress Σ
can be expressed as a combination of both the fluid stress and the particle stress Σp :
where P is the fluid bulk pressure, ηs is the suspending fluid dynamic viscosity. The particle stress contains four
terms [33, 34, 36]:
Σ = −P I + 2ηsE + Σp ,
(5)
Σp = n
ξr
I
3
2 "< pppp > −
+3nDrξr"< pp > −
< pp ># : E
3# + ǫnσ0"< pp > −
I
I
3#
(6)
−nmB < b⊥p >
where b⊥ = (I − pp) · b is the normalized projection of the magnetic field onto the plane perpendicular to the rod.
We will focus our investigation on Σp. First, let us consider a dimensionless version of it :
eΣp = Σp/(nσ0)
The energy density nσ0 represents the maximal work per unit volume stemming from the swimming activity which is
characterized microscopically by a time scale tH = ξr/σ0 needed for the swimmer to move the fluid over its own size.
This time scale is used to define an activity number : A = 1/(DrtH ). The higher A the more directionally persistent
is the bacterial swimming. The (x,y) component of the dimensionless particle stress contains four terms and reads
(see [33, 34, 36] for details of the calculation) :
=
< p2
xp2
y >
1
2
+
(cid:18)eΣp(cid:19)xy
{z
eΣact
+ ǫ < pxpy >
P em
A
}
P eH
A
3
A
+
< pxpy >
eΣdrag
}
{z
{z
<(cid:2)pxpyby − bx(cid:0)1 − p2
eΣdiff
eΣmag
{z
}
x(cid:1)(cid:3) py >
}
(8)
4
FIG. 2: Rescaled particle shear stress eΣp, derived numerically from equation (8), as a function of the rotational Peclet number
P eH (α = 45◦, 135◦ ; P em = 1 ; A = 10 ; ǫ = 1 (puller ) and −1 (pusher )). The active contribution eΣact is color-labelled and
referred to equation (8). When P eH ≪ 1, eΣp is equal to the active stress and tends to a constant value, the actuated stress.
Depending on the sign of the actuated stress, the suspension can be turned to brake and motor states. For each of the four
graphs, the brake and motor states are interpreted by a sketch in which we show the orientation of the elongated particles
relatively to the flow direction. The red (brake) and green (motor ) lines and arrows correspond to the stream lines created by
the hydrodynamic dipole of each particle (represented by a black ellipsoid).
FIG. 3: Phase diagram in the (P em, α) space of the motor-brake effect for puller and pusher swimmers. Positive values and
negative values of the actuated stress eΣ0 are respectively red and green-labelled and correspond to brake and motor states.
The relative magnitude of the actuated stress is indicated by a logarithmic color gradient.
whereeΣdrag andeΣdiff are passive contributions and account for drag on the surface of the particle from shear flow and
diffusive process respectively; eΣact is related to the swimming activity of the particle; eΣmag represents the stress due
to the perturbation of the flow by the magnetic field driven rotation of the particle. The brackets <, > correspond
to an angular average weighted by Ψ. Note that expression (8) is only valid for β = 1. See section IV discussing the
case of β = 0.
III. THE MOTOR-BRAKE EFFECT
(cid:18)eΣp(cid:19)xy
will be denoted eΣp. We restrict the investigations to Peclet numbers such that swimming remains the
dominant contribution in the rheological response. Accordingly, in the following, we decided to fix the activity number
at a value A = 10.
5
FIG. 4: (a). Example of relation eΣp(P eH) obtained from a full numerical solution of the problem for A = 10, α = 45◦,
P em = 0.6 (symbols). The solid line is a linear fit taking into account the data up to the limit P emax
H such that the goodness of
H and the intercept eΣ0. In regime (i), the actuated stress
the fit R2 remains larger that 0.999. The slope of the curve yields P e∗
dominates. In regime (ii), the stress is essentially linear in shear rate, the actuated stress is negligible. In regime (iii), the affine
H , then
expansion is no longer valid because other stress terms contribute to the shear stress. When the calculated P e∗
we conclude that the regime eΣp = eΣ0 + ηpP eH is never reached : the stress is constant at low P eH, up to a transition to a
regime not anymore dominated by the active and drag stresses.
(b). Validity of the affine expansion of the stress (10) in the parameter space (P em, α) for A = 10. The color code corresponds
to the value of ∆ = log10(P emax
H), which is the range of validity, in decades of P eH , for which equation (10)
holds. When parameters are chosen in the regions in blue, equation (10) is no longer valid and other terms have to be taken
into account in the expansion of the shear stress.
H ) − log10(P e∗
H > P emax
determined by the active stress contribution, tends linearly to a non-zero constant (see figure 2 for puller and pusher
On figure 2, we display the behavior ofeΣp with respect to P eH for different orientations of B. An important feature
of the rheological response is that for P eH ≪ 1 and P eH ≪ P em (for any P em), the diffusive eΣdiff and magnetic
eΣmag stresses do compensate each other. Then, the particle stress eΣp is mainly a combination of the active eΣact
and the drag eΣdrag stresses. While the drag stress vanishes for P eH → 0, the particle stress eΣp, which is completely
swimmers). This constant stress will be called the actuated stress and denoted eΣ0. It is created by the swimming
activity and induced by the magnetic field. Indeed, at these low P eH, the magnetic particle is essentially oriented in
the direction of the magnetic field and both pusher and puller swimmers can increase the shearing of the fluid (motor
state) or decrease it (brake state). This is illustrated on figure 2 by the drawing of the swimmer orientations and a
sketch of the corresponding flow lines.
The intensity of the motor-brake effect relies on the value of the actuated stress eΣ0 which itself depends on the
magnitude and orientation of the magnetic field. On figure 3, we plot two phase-diagrams in the (P em, α) space for
both pusher and puller swimmers : one can notice that the larger the magnetic field, the stronger the motor-brake
effect. For a given P em, the effect is maximal for α = 45◦ and α = 135◦ for which the extensional and compression
axis of the flow created by the particles are aligned (motor ) or perpendicular (brake) to the ones of the imposed simple
shear flow. Similarly, for α = 0, 90 and 180◦, the actuated stress vanishes because the magnetic swimmers shear the
fluid in the orthogonal direction to the imposed shear.
To investigate more quantitatively the motor-brake effect, we compute an analytical asymptotic expression for the
active and drag dimensionless stresses to leading orders in P em and P eH . As mentioned above, the magnetic and
diffusive stresses do compensate each other in the limit of low P eH and P em, such that their combined contribution
vanishes in the asymptotic expansion of the stress. Note that, in other limits which are not investigated by this paper,
the contributions of these stresses do not compensate and must be considered explicitly in the total stress balance.
We restrict the expansion of the distribution function Ψ(θ, φ) to the first two-spherical harmonics :
Ψ(θ, φ) =
1
4π 2Xn=0
nXl=0
Al
nP l
n(cos θ) cos lφ +
n(cos θ) sin lφ!
Bl
nP l
2Xn=1
nXl=1
(9)
where Al
n and Bl
n are the coefficients of the expansion, solutions of the Fokker-Planck equation (3). In this expansion,
6
10
5
1
0
-1
-5
-10
-15
-20
-10
-5
0
5
10
FIG. 5: Rescaled particle shear stress eΣp/eΣ0 as a function of the rescaled Peclet number fP eH for different values of α,
for P em = 0.1, 0.4 and 1 and for both pusher and puller swimmers. The α values are indicated by the color code. Values
α = 0◦, 90◦, 180◦ are excluded because they correspond to the situation eΣ0 = 0, for which the master equation (11) is not
defined. The numerical solution displays, for a wide range of P eH and P em, a collapse onto expression (8), represented in
black solid (pushers) and doted (pullers) lines. Negative values of shear rate correspond to a gradient of flow velocity in the -y
direction, and vorticity in the z direction. By reversing the shear flow, the actuated stress keeps its sign and the motor-brake
effect is reversed (from motor to brake state and vice-versa). Inset : Rescaled actuated stress (cid:12)(cid:12)(cid:12)eΣ0/(cid:0) 1
2 sin(2α)(cid:1)(cid:12)(cid:12)(cid:12) as a function
m for P em ≪ 1 as predicted analytically. For P em ≫ 1, it
of the magnetic Peclet number P em. It exhibits a scaling in P e2
saturates.
2(θ, φ) = B2
the spherical harmonics which contribute dominantly to the stress terms < pxpy > and < p2
Ψ2
equation (3), we show that B2
particle shear stress:
2 (cos θ) sin 2φ/(4π) and Ψ0
0/(4π) = 1/(4π).
0(θ, φ) = A0
2 P 2
y > are respectively
Injecting Ψ(θ, φ) into the Fokker-Planck
xp2
m + P eH(cid:1) and we obtain an asymptotic scaling expression for the
m +(cid:18) 1
+ ǫ(cid:19) P eH(cid:21) + o(P eH , P e2
(10)
m)
A
2 = 1/12(cid:0)sin(2α)P e2
30(cid:20)ǫ sin(2α)P e2
eΣp =
1
1
A
regimes is observed at P e∗
which can be written in the form eΣp = eΣ0 + ηpP eH, where eΣ0 =
30(cid:18) 1
rescaling the particle stress by(cid:12)(cid:12)(cid:12)eΣ0(cid:12)(cid:12)(cid:12) and the hydrodynamic Peclet number by the cross-over value P e∗
+ ǫ(cid:19) is the particle-borne viscosity contribution. A cross-over between the linear and the actuated stress
(cid:12)(cid:12)(cid:12)(cid:12). A master curve for this asymptotic scaling limit can be obtained
H i.e. : fP eH =
m is the actuated stress and ηp =
H = (cid:12)(cid:12)(cid:12)(cid:12)
sin(2α)(P em)2
sin(2α)(P em)2
sin(2α)P e2
P eH/P e∗
1/A + ǫ
1/A + ǫ
ǫ
30
H = P eH /(cid:12)(cid:12)(cid:12)(cid:12)
(cid:12)(cid:12)(cid:12)(cid:12). The expression for this master curve is then :
eΣp
= ǫ sign(sin(2α)) +fP eH sign(1/A + ǫ)
eΣ0
To test the validity of the scaling expression, we propose to compare the results stemming from the full numerical
solution of the Fokker-Planck equation yielding Ψ(θ, φ) and the particle stress eΣp (equation (8)), with the analytical
expression for the master curve (equation (11)). From the computation of eΣp(P eH ) at different P em, we extract the
(11)
H using an affine fit in P eH (see figure 4(a)). We then construct the master curve of figure 5 where
each data is a symbol : we indeed observe a collapse of the full numerical solution of equation (8) onto the master
equation (11) for a wide range of parameters. This scaling law is no longer valid at large P em, beyond the validity
value ofeΣ0 and P e∗
limit of the expansion done in equation (10), which we determine by computing the deviation of eΣp(P eH ) from an
The total shear stress of the suspension (including the contributions of both the suspending fluid and the magnetic
affine fit (see figure 4(a)). The validity domain is reported figure 4(b).
micro-swimmers) is then, in its dimensional version :
7
where n = Φ/VB = 6Φ/(πa2L) is the volume density of active particles in the fluid, Φ is the particle volume fraction
and VB the volume of a single ellipsoidal particle. Injecting the asymptotic scaling (10) of eΣp into equation (12), we
obtain the dimensional total shear stress exerted on the suspension :
(12)
Σxy = ηs γ + nσ0eΣp
Σxy = ǫ
sin(2α)P e2
m
nσ0
30
Σ0
nσ0
{z
}
+(cid:18)ηs +
30Dr(cid:18) 1
{z
ηeff
A
+ ǫ(cid:19)(cid:19)
}
γ
(13)
This constitutive relation generalizes the result of Saintillan [32].
It contains the dimensional actuated stress
Σ0 ≡ nσ0eΣ0 described above and a linear dependance with the shear rate γ, defining an effective viscosity of the
suspension ηeff ≡ ∂Σxy/∂ γ. Remarkably, the magnetic field angle can be chosen so that the actuated shear stress
becomes negative for both pusher and puller swimmers. When the actuated stress dominates, the swimming power
of the bacteria transferred to the fluid induces a shear of the suspension which can be oriented in the same direction
as the imposed shear (motor state) or in the opposite direction (brake state).
An other feature of equation (13) is that ηeff is identical to the effective viscosity of non-magnetic micro-swimmers
suspensions, derived in [32] for instance. This comes from the fact that, in the range of validity of equation (13),
there is no coupling between magnetic and hydrodynamic terms in the stress. However, in other ranges of parameters,
non-trivial couplings between these terms change the effective viscosity of the suspension.
Note that, for P em ≫ 1, i.e. out from the validity limit of expression (13), the actuated stress exhibits a saturation.
Indeed, in this limit, the micro-swimmers are mainly aligned in the magnetic field direction and deliver collectively
the maximum shear allowed by their swimming energy. Then, the maximal intensity of the motor-brake effect is
reachable by achieving the limit P em ≫ 1, which is confirmed by the numerical computation of the actuated stress
(see inset of figure 5).
IV. GENERALITY OF THE MOTOR-BRAKE EFFECT
In this section, we show that the properties of the motor-brake effect are not specific to the hypothesis of our model.
First, the rheological response of SS swimmers is the same as the one of NS swimmers. By symmetry, SS corresponds
to NS after a re-orientation of the magnetic field by an angle π. Thus, if the NS swimmers are in motor state, the SS
will also be in motor state and vice-versa.
We also investigate the role played by the particle geometry. More specifically, we analyzed the case of a purely
p = (I−pp)Ω·p+Ωm×p = (Ω + Ωm)×p,
spherical particle (β = 0). In this case, the kinematic equation is changed :
where Ω is the fluid vorticity vector. The corresponding change in the Fokker-Planck equation (3) concerns the
operator Γshear(Ψ) =
. Concerning the stresses, the expressions are also different from the elongated rod case.
1
2
∂Ψ
∂φ
The dimensional particle stress then reads :
Σp =
nξrE + nσ0"< pp > −
I
3# +
5
6
nmB
2
[< p > b − b < p >]
(14)
In this stress, the contribution of diffusive processes are zero due to the symmetry of the particle. Only remain the
stress corresponding to the friction on the particle body, the active stress and the traceless magnetic stress (it is
antisymmetric here due to the spherical geometry of the particle, see [36]).
On figure 6 are some numerical results obtained for both pusher and puller spherical swimmers. The phenomenology
is the same as for elongated particles, meaning the appearence of a constant shear stress at low shear rate (or P eH ).
Moreover, the value of this constant, identical to the one of rod-shaped particles, is solely due to the active stress and
the value and the sign of this constant depends on P em, α and ǫ, in the same way as for elongated swimmers. This
shows that the motor-brake effect is general for various kinds of particle shapes. The main quantitative difference
between the case of spherical and elongated particles is the effective viscosity of the suspension which is always positive
for spherical particles, as already described before for non-magnetic swimmers [32].
8
FIG. 6: Computation of the particle shear stress and active stress for spherical micro-swimmers. The motor-brake effect is
recovered also for these kind of particles and the value of the actuated stress is identical to the one of rod-shaped particles.
V. ORDERS OF MAGNITUDE OF THE EFFECT AND CONCLUSION
In order to test whether the motor-brake effect can be significant experimentally, let us take the example of a
suspension of magnetotactic bacteria of dimensions L = 5 µm, a = 1 µm at a volume fraction of 1% (i.e. of number
density n ≃ 1016 m−3). The force dipole magnitude of an active swimmer moving with a velocity of 10 µm.s−1 is
typically σ0 = 10−18 J [42]. From experiments on magnetotactic bacteria using the Magnetospirillum gryphiswaldense
MSR-1 strain, we get values of m ≃ 10−16 A m−2 (see also [26, 38]); Dr = 1 s−1, tH = 0.1 s (typical time for the
bacteria to move over its size). The activity number A is then typically equal to 10, and for a magnetic field B = 1
mT, P em = 1. The magnetic field orientation is chosen to be at α = 135◦.
It is then possible to evaluate the
P eH below which Σxy < 0, using equation (13). We obtain P eH ≃ 1 which corresponds to γ ≃ 1 s−1. The value
obtained for P eH remains in the domain of validity of equation (13). The corresponding shear rate magnitude can
be reached by known rheometry [20, 21], meaning that this effect could indeed be observed experimentally. Note
that published experimental setups would allow to test the effect in the conditions akin to the model [39]. Moreover,
one can estimate a numerical value for the maximum shear stress available from the system at low shear rate, which
is nσ0 ∼ 10−2 − 10−1 Pa. This value needs to be compared to typical pressure loss in microfluidics : 10−1 Pa is
needed to flow water in a cylindrical channel of 1cm-length and 100µm-radius at 1nL.s−1, which is of the same order
of magnitude. This indicates that the effect could be used to control microfluidic flows.
For standard rheo-magnetic suspensions, a negative-viscosity effect was found in response to oscillatory magnetic
fields [40] or adding a constant torque on non-colloidal particles [41]. The motor-brake effect derived here is very
different conceptually. First, it relies on the activity of magnetic swimmers under a constant magnetic field. Second,
the actuation of a constant negative shear stress at low shear rate (motor state) is new in rheology. The tunability of
the motor and brake states for such suspensions could open the way to several practical applications, as direct flow
control in microfluidic devices or energy harvesting to build microscopic motorized systems.
We acknowledge the support of the ANR-2015 "BacFlow", a critical reading of the manuscript by Dr. Laurette
Tuckerman and scientific discussions with Prof. Anke Lindner.
9
[1] E.Lauga & T.R. Powers, The hydrodynamics of swimming microorganisms, Rep. Prog. Phys., 72, 096601 (2009).
[2] W. F. Paxton et al., Catalytic nanomotors: autonomous movement of striped nanorods, J. Am. Chem. Soc., 126, 13, 424
(2004).
[3] C. C. Maass, C.Krger, S. Herminghaus & C. Bahr, Swimming droplets, Annual Review of Condensed Matter Physics, 7,
171-193 (2016).
[4] R. Dreyfus et al., Microscopic artificial swimmers, Nature, 437, 862 (2005).
[5] P.Tierno, Magnetically actuated colloidal microswimmers, J. Phys. Chem. B, 112, 16525 (2008).
[6] S.Martel, Bacterial microsystems and microrobots, Biomed Microdevices, 14, 1033 (2012).
[7] Williams B. J., Anand S. V., Rajagopalan J. & Saif, M. T. A., A self-propelled biohybrid swimmer at low Reynolds number,
Nature Com., 5, 18 (2014).
[8] D.L Koch & G. Subramanian, Collective hydrodynamics of swimming microorganisms:
living fluids, Annual Review of
Fluid Mechanics, 43, 637–659 (2011).
[9] M.C Marchetti et al., Hydrodynamics of soft active matter, Rev. Mod. Phys., 85, 1143 (2013).
[10] Y. Hatwalne, S. Ramaswamy, M. Rao & R. Aditi Simha, Rheology of active-particle suspensions, Phys. Rev. Lett., 92,
118101 (2004).
[11] J. Toner, Y. Tu & S. Ramaswamy, Hydrodynamics and phases of flocks, Ann. Phys., 318, 170 (2005).
[12] X.-L. Wu et al., Particle diffusion in a quasi-two-dimensional bacterial bath, Phys. Rev. Lett., 84, 3017 (2000).
[13] S. Sokolov & I.S. Aranson, Reduction of viscosity in suspension of swimming bacteria, Phys. Rev. Lett., 103, 148101
(2009).
[14] S. Rafaı, L. Jibuti & P. Peyla, Effective viscosity of microswimmer suspensions, Phys. Rev. Lett., 104, 098102 (2010).
[15] K. C. Leptos et al., Dynamics of enhanced tracer diffusion in suspensions of swimming eukaryotic microorganisms, Phys.
Rev. Lett., 103, 198103 (2009).
[16] A. Sokolov et al., Swimming bacteria power microscopic gears, Proc. Natl. Acad. Sci. U.S.A., 107, 969 (2010).
[17] R. Di Leonardo et al., Bacterial ratchet motors, Proc. Natl. Acad. Sci. U.S.A., 107, 9541 (2010).
[18] Mino G et al, Enhanced diffusion due to active swimmers at a solid surface, Phys. Rev. Lett, 106 048102 (2011).
[19] R. Rusconi, J.S. Guasto & R. Stocker, Bacterial transport suppressed by fluid shear, Nature Physics, 10, 212-217 (2014).
[20] J. Gachelin, A. Rousselet, A. Lindner & E. Clement, Collective motion in an active suspension of Escherichia coli bacteria,
New Journal of Physics, 16, 025003 (2014).
[21] Lopez H.M., J. Gachelin, C. Douarche, H.Auradou, E. Clement, Turning bacteria suspensions into superfluids, Phys. Rev.
Lett., 115, 028301 (2015).
[22] S.C. Takatori & J.F. Brady, Superfluid Behavior of Active Suspensions from Diffusive Stretching, Phys. Rev. Lett., 118,
018003 (2017).
[23] M.E. Cates, S.M Fielding, D. Marenduzzo, E. Orlandini & J.M. Yeomans, Shearing active gels close to the isotropic-nematic
transition, Phys.Rev. Lett., 101, 068102 (2008).
[24] L. Giomi, T.B. Liverpool & M.C. Marchetti, Sheared active fluids: Thickening, thinning, and vanishing viscosity, Phys.
Rev. E, 81, 051908 (2010).
[25] R.Uebe, D.Schler, Magnetosome biogenesis in magnetotactic bacteria, Nature Reviews Microbiology, 14, 621637 (2016).
[26] M. Reufer et al., Switching of swimming modes in Magnetospirillium gryphiswaldense, The Biophysical Journal, 106, 37-46
(2014)
[27] N. Waisbord, C. T. Lef`evre, L. Bocquet, C. Ybert & C. Cottin-Bizonne, Destabilization of a flow focused suspension of
magnetotactic bacteria, Phys. Rev. Fluids, 1 (2016)
[28] G.B. Jeffery, The motion of ellipsoidal particles immersed in a viscous fluid, Proc. R. Soc. London, Ser. A, 102, 161 (1922).
[29] F.P. Bretherton, The motion of rigid particles in a shear flow at low Reynolds number, J.Fluid Mech., 14, 284 (1962).
[30] A. Satoh, Rheological Properties and Orientational Distributions of Dilute Ferromagnetic Spherocylinder Particle Disper-
sions, Journal of Colloid and Interface Science, 234, 42533 (2001).
[31] Haines B.M. et al., Three-dimensional model for the effective viscosity of bacterial suspensions, Phys. Rev. E ,80, 041922
(2009).
[32] D. Saintillan, The dilute rheology of swimming suspensions: A simple kinetic model, Exp. Mech., 50, 125 (2010).
[33] H. Brenner, Rheology of a dilute suspension of axisymmetric Brownian particles, Int. J. Multiphase Flow, 1, 195-341 (1974).
[34] Jansons KM, Determination of the constitutive equations for a magnetic fluid, J. Fluid. Mech., 137, 187-216 (1983).
[35] R. E. Rosensweig, Ferrohydrodynamics, Cambridge University Press (1985).
[36] S.R. Strand & S. Kim, Dynamics and rheology of a dilute suspension of dipolar nonspherical particles in an external field:
Part 1. Steady shear flows, Rheologica Acta, 31, 94-117 (1992).
[37] E. J. Hinch & L.G. Leal, Constitutive equations in suspension mechanics. Part 2. Approximate forms for a suspension of
rigid particles affected by Brownian rotations, Journal of Fluid Mechanics, 76, 187-208 (1976).
[38] R. Nadkarni, S. Barkley & C. Fradin, A comparison of methods to measure the magnetic moment of magnetotactic bacteria
through analysis of their trajectories in external magnetic fields, PLOS ONE, 8, 12 (2013).
[39] X. Cheng, J. H. McCoy, J. N. Israelachvili & I. Cohen, Imaging the microscopic structure of shear thinning and thickening
colloidal suspensions, Science, 333, 1276-1279 (2011).
[40] J.-C. Bacri, R. Perzynski, M. I. Schliomis & G. I. Burde, Negative-viscosity effect in a magnetic fluid, Phys. Rev. Lett., 75
2128 (1995).
[41] L. Jibuti, S. Rafai & P. Peyla, Suspensions with a tunable effective viscosity: a numerical study, Journal of Fluid Mechanics,
693, 345-366 (2012).
[42] K. Drescher, J. Dunkel, L. H. Cisneros, S. Ganguly & R. E. Goldstein, Fluid dynamics and noise in bacterial cellcell and
cellsurface scattering, PNAS, 108, 27, 1094010945 (2011).
[43] D. Saintillan, Rheology of Active Fluids, Annual Review of Fluid Mechanics, 50, 563-592 (2018).
10
|
1506.08090 | 1 | 1506 | 2015-06-26T14:20:18 | Myosin-II dependent cell contractility contributes to spontaneous nodule formation of mesothelioma cells | [
"physics.bio-ph",
"q-bio.CB"
] | We demonstrate that characteristic nodules emerge in cultures of several malignant pleural mesothelioma (MPM) cell lines. Instead of excessive local cell proliferation, the nodules arise by Myosin II-driven cell contractility. The aggregation process can be prevented or reversed by suitable pharmacological inhibitors of acto-myosin contractility. A cell-resolved elasto-plastic model of the multicellular patterning process indicates that the morphology and size of the nodules as well as the speed of their formation is determined by the mechanical tension cells exert on their neighbors, and the stability of cell-substrate adhesion complexes. A linear stability analysis of a homogenous, self-tensioned Maxwell fluid indicates the unconditional presence of a patterning instability. | physics.bio-ph | physics | Myosin-II dependent cell contractility contributes to
spontaneous nodule formation of mesothelioma cells.
Julia T´arnoki-Z´ach1, Dona Greta Isai 2, Elod M´ehes1, S´andor
Paku, Zolt´an Neufeld, Bal´azs Hegedus, Bal´azs Dome, Andras
Czirok 1,2‡ ,
1 Department of Biological Physics, Eotvos University, Budapest, Hungary
2 Department of Anatomy & Cell Biology, University of Kansas Medical Center,
Kansas City, KS, USA
E-mail: [email protected]
Abstract.
We demonstrate that characteristic nodules emerge in cultures of several malignant
pleural mesothelioma (MPM) cell lines. Instead of excessive local cell proliferation,
MPM nodules arise in culture by Myosin II-driven cell contractility. Accordingly,
the aggregation process can be prevented or reversed by suitable pharmacological
inhibitors of Myosin II activity. A cell-resolved elasto-plastic model of the multicellular
patterning process predicts that the morphology and size of the nodules as well as the
speed of their formation is determined by the mechanical tension cells exert on their
neighbors, and the stability of cell-substrate adhesion complexes. For small tension
forces nodules are slow to develop and localized at the boundary of cell free areas. In
contrast, a high intralayer tension quickly transforms all cell-covered areas into dense
clusters interconnected by multicellular strands. Model simulations also indicate that
a decreased stability of cell-substrate adhesions favors the formation of fewer, but
larger clusters. Linear stability analysis of a homogenous, self-tensioned Maxwell fluid
indicates the unconditional presence of the patterning instability.
5
1
0
2
n
u
J
6
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
0
9
0
8
0
.
6
0
5
1
:
v
i
X
r
a
‡ corresponding author
Contractile tumor nodules
1. Introduction
2
Cell contractility and forces exerted on the cell's microenvironment constitute an
important mechanism of multicellular patterning. Morphogenetic movements in
epithelia are particularly well known to utilize cell contractility:
for example, an
anisotropic contractile activity gives rise to cell intercalation, a process altering cell
connectivity/adjacency in such a way that the whole tissue elongates in one direction
while narrows along the perpendicular direction [1, 2]. Similarly, contractility-driven
constriction of the free (apical) surface of the epithelium can give rise to bending
or budding within an epithelial sheet [3]. Mesenchymal cells can also contract the
surrounding extracellular matrix (ECM): if a cell aggregate is placed on the surface of
a collagen gel, cell traction reorganizes the collagen and create bundles of ECM that
radiate away from the aggregate [4, 5]. This observation led to the development of
the mechano-chemical theory of pattern formation [6, 7], according to which cells exert
traction forces on an underlying deformable substrate and the resulting strain transports
(convects) both cells and the ECM. Furthermore, strain-oriented ECM filaments can
guide cell motility, as cells are more likely to move parallel with the orientation of
the ECM [8, 9, 10]. This mechanism was suggested to guide vascular patterning, and
endothelial cells were reported to be able to detect and respond to substrate strains
created by the traction stresses of neighboring cells [11].
Cellular contractility is also an important factor in tumor progression as it can
contribute both to local and distant spreading of malignant cells [12, 13, 14, 15]. Here
we focus on the role of cell contractility in malignant pleural mesothelioma (MPM), a
tumor arising from mesothelial cells lining the pleural cavity. This highly aggressive
disease has an incidence of 1 in 100 000 in Europe [16]. Despite recent advances in the
treatment of MPM, it has an extremely poor prognosis with almost all patients dying
from their tumor. A characteristic, pathognomonic feature of MPM is the formation of
multiple, macroscopic pleural tumor nodules, which -- due to the special two dimensional
environment -- may pinch off and contribute to the local spreading of malignant cells.
In this manuscript we demonstrate that several MPM cell lines can form nodule-
like aggregates in vitro when cultured at high cell density. Time-lapse analysis of their
formation and experiments with myosin II-specific inhibitors reveal that cell contractility
is a key factor in the nodule formation process. To interpret these experimental
findings, we propose a cell-resolved elasto-plastic model of a contractile cell layer. By
computational simulations we explore the effects of key model parameters, such as the
the stability of cell-cell and cell-substrate adhesion and the magnitude of the cell-exerted
contractile forces.
Contractile tumor nodules
3
2. Methods
2.1. Cell lines
SPC111 cells were established from human biphasic MPMs and kindly provided by
Prof. R. Stahel (University of Zurich, Switzerland). P31 cells were a kind gift from
Prof. K. Grankvist (University of Umea, Sweden). The Meso cell lines were established
by the Vienna MPM group as described recently [17]. NP3 normal mesothelial cells
were isolated from pneumothorax patients using the same protocol [18].
2.2. Culture conditions
Gelatin-coated dishes were obtained by incubating 10% gelatin solution-B (Sigma) in
PBS for 45 min at room temperature. We also prepared 1.7 mg/ml Collagen-I (Corning)
gels according to the manufacturer's instructions.
Cells were grown at 37oC in a humidified, 5% CO2, 95% air atmosphere.
The DMEM (Lonza) medium was supplemented with 10% FCS (Invitrogen) and
1% penicillin-streptomycin-amphotericin B (Lonza). MPM cells were cultured on
plastic tissue culture substrates (Greiner). NP3 cells are, however, more sensitive to
environmental factors and have a limited proliferative potential. Thus, NP3 cells were
cultured in gelatin-coated dishes, for up to three passages.
2.3. Reagents
To interfere with normal Myosin II function, we utilized Y27632, the rho kinase
(ROCK) inhibitor (Merck Millipore) and Blebbistatin (Merck Millipore), an inhibitor of
actomyosin crosslinking. Y27632 was solved in destilled water, stored in the form of a 10
mM stock solution, and used at 100 µM final concentrations in DMEM. Blebbistatin was
solved in DMSO, stored as 50 mM aliquots, and used at 40-80 µM final concentrations.
2.4. Immunostaining
SPC111 cells were seeded at confluent density (40000 cells/cm2) either on the surface
of collagen gels or on glass coverslips and were grown for 6 days. Samples were fixed
using 4% PFA for 15 min at 4oC, permeabilized with 0.25% Triton-X100 for 10 min
at 4oC, and incubated with the following antibodies: Polyclonal anti-fibronectin (1:100,
Ab2033, Millipore); polyclonal anti-beta-catenin (1:100, C2206, Sigma). After washing,
sections were incubated for 30 min with anti-rabbit Alexa-488 secondary antibody
(Life Technologies, Carlsbad, CA). Filamentous actin was stained by Phalloidin-TRITC
(P1951, Sigma). Samples were analyzed by confocal laser scanning microscopy using
the Bio-Rad MRC-1024 system (Bio-Rad, Richmond, CA).
Contractile tumor nodules
2.5. Physical sections
4
Physical cross sections were obtained from cultures grown on the surface of collagen-I
gels. Samples were embedded in Spurr's mixture. Semithin (1 µm) sections were cut
perpendicularly to the surface of the collagen and stained with 0.5% Toluidin blue.
Images were taken using Zeiss Axioskop 2 microscope equipped with a 100x objective
and coupled to an Olympus DP50 camera.
2.6. Time-lapse microscopy
Time-lapse recordings were performed on a computer-controlled Leica DM IRB inverted
microscope equipped with a Marzhauser SCAN-IM powered stage and a 10x N-PLAN
objective with 0.25 numerical aperture and 17.6 mm working distance. The microscope
was coupled to an Olympus DP70 color CCD camera. Cell cultures were kept at 37C
in humidified 5% CO2 atmosphere during imaging. Phase contrast images of cells were
collected consecutively every 10 minutes from each of the microscopic fields.
To maintain high cell density cultures for several days, we 3D printed three 6 mm
diameter mini-wells into regular 35 mm tissue culture dishes. The side wall of the wells
was formed by fused polylactic acid filaments. This setup allowed us to achive high
(106/cm2) cell density with a medium/cell ratio characteristic for culture with ten times
lower cell density (105/cm2). The 3 ml medium was replaced every 5 days. Four modified
culture dishes, each containing 3 wells, were observed by time-lapse videomicroscopy for
up to 14 days. We recorded four adjacent fields from each well, thus we collected 48
images in each imaging cycle (every 10 minutes).
2.7. Computational model
To explore the dynamics of cell contraction-driven aggregation and nodule formation,
we adapted our cell-resolved elasto-plastic computational framework [19] into a two
dimensional model. Briefly, forces distributed along the membranes of mechanically
coupled adherent cells are replaced by a single net force and a torque, simplifying the
system to a network of particles and elastic beams (Fig. 1).
In the computational model torques and shear forces arise due to relative movement
of adjacent particles. Hook's law determines the force F (cid:107) needed to uniaxially compress
is associated with changing the length of
or stretch cells (Fig. 1b). In particular, F
link l which connects particles i and j located at ri and rj, respectively:
(cid:107)
l
l = k(rj − ri − (cid:96)l).
(cid:107)
F
(1) (cid:96)l is the equilibrium length of link l and k > 0 is a model parameter,
In Eq.
characterizing the stiffness of the cytoskelon.
A link l exerts a torque if its preferred direction at particle i, ti,l, is distinct from
its actual end-to-end direction uij (Figs. 1c,d). We assume that the torque exerted on
particle i is proportional to the difference between the preferred and actual directions:
Mi,l = g(ti,l × uij),
(1)
(2)
Contractile tumor nodules
5
Figure 1. Mechanical model of multicellular clusters.
Links represent
mechanically connected cytoskeletal structures of adherent cells. Rectangular
shapes indicate cell membranes, not explicitly resolved in the model.
a:
Two particles, i and j interconnected with a link in a mechanical stress-free
configuration. Unit vectors, ti and tj, which co-rotate with the particles,
represent the preferred direction of the link.
b: Compressed cells. The
interaction of the two cells gives rise to spatially distributed forces (red arrows),
which are replaced by the repulsive net forces Fi and Fj. c: A symmetric
rotation of both particles yields torques Mi and Mj acting on particles i and j,
respectively. These torque vectors are perpendicular to the plane of the figure.
d: A lateral misalignment of the particles creates torques Mi, Mj and also
shear forces Fi, Fj acting at the particles.
where the microscopic bending rigidity g, a model parameter, can be calibrated from
macroscopic material properties of the tissue [19]. The condition for mechanical
equilibrium, together with Eqs (1) and (2) allow the calculation of the forces and torques
within the system [19].
Multicellular plasticity is modeled using rules that rearrange the network of
intercellular adhesions. The stability of intercellular adhesion complexes depends on
the tensile forces they transmit [20]. Thus, in our model the probability of removing
link l during a short time interval ∆t follows Bell's rule [21] as
pl∆t = AeF t
l /F 0∆t,
where F 0 is a threshold value, and
F t
l = Θ(F
(cid:107)
l )F
(cid:107)
l
(3)
(4)
is the tensile component of the force transmitted by the link and A is a scaling factor
which characterizes the stability of connections.
Mechanical connections can be established between two Voronoi neighbor particles,
i and j. We assume, that during a short time interval ∆t the probability of inserting a
new link is a decreasing function of the distance di,j between the particles:
(cid:19)
(cid:18)
1 − di,j
dmax
qi,j∆t = B
∆t.
(5)
Contractile tumor nodules
6
The scaling factor B represents the intensity of cellular protrusive activity devoted to
scanning the environment and the ability to form intercellular contacts. The maximal
distance cells explore for new connections is denoted by dmax.
Simulations are event-driven: using the probability distributions (3) and (5), we
generate the next event µ and waiting time τ according to the stochastic Gillespie
algorithm [22]. The waiting time until the next event is chosen from the distribution
(cid:32)(cid:88)
(cid:33)
(cid:88)
log P (τ ) = −τ
pl +
qi,j
,
(6)
where the sums are evaluated by iterating over existing links l as well as over all possible
Voronoi neighbor particle pairs i, j not connected by a link.
l
i,j
2.8. Simulation Parameters
Two dimensional initial conditions were generated by randomly positioning N = 400
particles in a square of size L = 20. The unit distance of the simulations was set to
the average cell size, d0 ≈10µm, thus the 2D cell density is 1 cell/unit area.
In the
initial condition we enforced that the distance of two adjacent particles is greater than
dmin = 0.8d0. Particles that are Voronoi neighbors are connected by links when their
distance is less than dmax = 2d0. For a mechanical stress-free initial configuration we
set the ti,l preferred link direction vectors as well as the equilibrium link lengths (cid:96)l so
that no internal forces or torques are exerted in the system. As a boundary condition,
we fixed the position and orientation of the particles located near the perimeter of the
simulation domain.
The mean waiting time between simulation events is set by parameters A and B.
We choose our time unit as 1/B ≈ 10 min, the time needed for two adjacent cells
to establish a mechanical link. We set the lifetime of an unloaded link to 1/B ≈ 1
day. Thus, according to these values, two cells pulled away by a force F 0 separate in
∼ 10 h, a characteristic value consistent with the time scale observed in our cell culture
experiments.
3. Results
3.1. MPM cells form nodules in vitro
To study the long-term behavior of mesothelioma monolayers, we cultured several,
human patient-derived MPM cell lines for up to a few weeks. In this time frame MPM
cells form macroscopic multicellular aggregates or nodules, which can be 50 µm thick and
reach a millimeter in lateral extent (Fig. 2). The time needed to generate nodules varies
across the cell lines. When SPC111 and p31 cells are seeded at confluency, aggregates
reach a macroscopic size after 3-5 days. In contrast, nodule formation in the Meso53,
Meso62 and Meso80 lines takes a few weeks. Formation of similar nodules is rather
uncommon in cultures of other epithelial cells. Most importantly, under similar culture
Contractile tumor nodules
7
Figure 2. Mesothelioma cells spontaneously form nodules in vitro. Panels a-e
depict five distinct human MPM cell lines, SPC111, Meso62, p31, Meso80 and
Meso53, after one (a, c, e) or two (b, d) weeks in culture.
In comparison,
NP3 human primary mesothelial cells remain in a monolayer and do not form
aggregates even after 17 days in vitro (f). Asterisks mark nodules, arrow
indicates strands interconnecting nodules.
conditions and duration NP3 human primary mesothelial cells remain in a monolayer
(Fig. 2f). Long-term NP3 cultures reach a quiescent state in which both the number of
cell divisions and cell deaths are substantially reduced.
Physical cross-sections of the nodules reveal densely packed cells containing
prominent, pleiomorphic nuclei.
Immunohistochemistry with antibodies against the
extracellular matrix protein fibronectin indicates the presence, but not the abundance
of ECM in the MPM nodules. Thus, the in vitro MPM nodules are dense, highly cellular
structures.
Contractile tumor nodules
8
Figure 3. Time course of spontaneous nodule formation in a culture of SPC111
cells. Panels a-e are frames from a time-lapse recording taken at seeding, 24h,
48h, 60h and 70h after seeding, respectively. The initial confluent monolayer
(a) develops cell density fluctuations (b, c), which eventually become three
dimensional nodules interconnected by multicellular strands (d-e).
f: The
aggregation process is visualized as a kymogram, where pixels along the same
line are plotted for each frame. Lines from earlier frames are located at the top
of the image. Scale bars correspond to 150 µm.
Contractile tumor nodules
9
In vitro MPM nodules are rich in stress filaments and are
Figure 4.
mechanically integrated.
a: Actin filaments are visualized by confocal
microscopy using TRITC-phalloidin (red). Nodules are interconnected with
multicellular strands that exhibit parallel bundles of stress filaments.
b:
Stress cables within multicellular strands align across cell membranes (arrows,
visualized by catenin antibodies -- green).
Time-lapse microscopy recordings of the aggregate formation process (Fig. 3) reveal
Instead, cells move
that nodules do not form by unusually active cell proliferation.
towards nodules or interconnecting strands. Kymograms (pixels along a selected line
plotted for several consecutive frames) indicate a gradual decrease of the extent of an
aggregate, a decrease which is approximately linear in in time.
3.2. Nodule formation is driven by actomyosin contractility
Visualization of actin filaments with fluorescent phalloidin (Fig. 4) reveals the abundance
of stress cables in the strands, organized into parallel bundles connecting the adjacent
nodules. Higher resolution confocal images indicate that several stress filaments reach
across cell bodies and even form structures that are continuous across cell membranes.
Both the observed cell movements and the presence of profound stress cables within
the aggregates suggest that Myosin II dependent cell contractility is an important driving
force to collect MPM cells into nodules. To test this hypothesis, we administered
drugs that interfere with normal Myosin II activity. Blebbistatin is a potent blocker
of acto-myosin contractility as it reduces the affinity of myosin heads to actin [23]. The
compound Y27632 is a specific inhibitor of Rho-associated kinase (ROCK), one of the
activators of Myosin.
Both blebbistatin (data not shown) and Y27632 (Fig. 5) could substantially hinder
or completely eliminate both the stress cables and nodule formation in cultures of
SPC111 cells. When inhibitors were administered at high concentrations (100 µM
for Y27632 and 80 µM for blebbistatin), the previously formed nodules flattened:
their height decreased by up to 50% while simultaneously extended laterally. When
the inhibitor is present from the onset of the culture, nodule formation can be fully
prevented (data not shown). The inhibitors are reversible: aggregation resumes after
Contractile tumor nodules
10
Figure 5. In vitro nodule formation requires Myosin II activity. Frames of a
time-lapse recording show the morphology of two parallel cultures, an untreated
control (a,c,e,g) and one which was transiently exposed to Y27632, a Rho-
associated kinase inhibitor (b,d,f,h). The two cultures are similar at 3 days
in vitro, at the onset of Y27632 treatment (a,b). Five hours long exposure
to Y27632 is sufficient to induce simultaneous flattening and lateral extension
of the nodules (d). Cell-dense three dimensional structures, clearly visible in
untreated control cultures (e), are largely absent after 4 days of ROCK inhibitor
treatment (f). Removal of the drug restores the contractility of MPM cells:
three days after replacing the medium nodule morphologies are similar in both
cultures (g,h).
Contractile tumor nodules
11
replacing the medium with fresh DMEM, and three days after medium replacement
nodule morphologies are similar in the unperturbed and previously myosin-blocked
cultures.
3.3. Computational model of contractile cells
To obtain a computational model of contractile cell sheets, we augmented our elasto-
plastic particle model with two new rules: one that controls equilibrium link lengths to
maintain a steady intercellular tension, and another to represent adhesions between the
cells (particles) and the culture substrate.
We assume that cells strive for a specific contractile environment which they
maintain as a homeostatic state: cells increase their contractility if tensile forces between
adjacent cells are below a target value, F ∗. In our model this is achieved by reducing
the equilibrium link length dl in Eq. (1). Conversely, when tensile forces are too strong,
the equilibrium length of the links is increased. In particular, we assume that the rate
of change in the link length is proportional to the difference between F ∗ and F t
l , the
tensile component (4) of the force transmitted by the link:
ddl
dt
l − F ∗
F t
F ∗ + µl(t),
= Cd0
(7)
where 1/C ≈ 1 h sets the temporal scale of the feedback regulation. The last term in (7)
is an uncorrelated (white) noise, representing random cell shape changes due to factors
not considered explicitly in our model. Equation 7 was integrated over a time interval
∆t, elapsed until the next stochastic event effecting particle connectivity, resulting:
Fl − F ∗
F ∗ ∆t +
√
dl(t + ∆t) = dl(t) + Cd0
D∆tξ
(8)
where ξ is a pseudo-random variable with unit standard deviation and parameter
D ≈ 1µm2/h controls the noise amplitude.
The mechanical equilibrium of surface-attached cells requires the net force exerted
by adjacent cells to be balanced by the force transmitted through the cell-substrate
adhesion complexes to the substrate. Thus, in our model for each particle i we introduce
i and angle φ0
the equilibrium position r0
i . The force (and torque) transmitted to the
substrate is given by the linear relations
and
Fi,0 = k0(ri − r0
i )
Mi,0 = g0(φi − φ0
i )ez.
(9)
(10)
As we discuss below, external forces acting on a cell may contribute to its displacement
trough complex and mostly unexplored processes. Here we assume that cells tend to
move in the direction of the net external force, hence the eqilibrium position is updated
according to
dr0
i
dt
=
αd0
F 0 Fi,0
(11)
Contractile tumor nodules
12
Figure 6. Time development of a typical simulation, at t=2.5 h (a), t=7.5 h
(b), t=13 h (c) and t=16 h (d). Blue to green colors indicate increasing tensile
stress within the links. Simulation parameters are listed in table I.
and
dφ0
i
dt
=
β
d0F 0 Mi,0.
(12)
Parameters α and β set the external force-related bias in cell movements, while the noise
in Eq. (8) gives rise to random walk-like movements.
3.4. Simulations
Simulations of the contractile monolayer (Fig. 6) reveal a pattern formation sequence
which is in several aspects analogous to the process seen in vitro (Fig. 3). First, the slight
initial inhomogeneity of the monolayer is amplified resulting in the formation of holes.
Then, cell free areas continue to grow as their boundary is unstable: links constituting
the boundary need to balance the pulling forces exerted by the bulk of the cells. Thus,
connections at boundaries experience higher tensile stress which increases both their
length and the probability of their removal. Both effects extend the area of the holes.
Contractile tumor nodules
13
Figure 7. The progress of patterning, characterized by the standard deviation
of particle density, S, as a function of time. Pattern formation is faster for
higher target force F ∗ values (a) and for higher values of α, the motion bias
towards external forces (b).
As a consequence, cell density increases in the rest of the system, and eventually cell-
covered areas appear as contractile nodes connected by linear strands. While nodules
develop in several MPM cultures (Fig. 3), the appearance of cell-free areas is observable
only in a subset of MPM lines. For example, SPC111 or p31 nodules form on top of
a basal monolayer and in such cases our model corresponds to the dynamic upper cell
layers.
During the aggregation process the spatial distribution of cells becomes more
inhomogeneous. Thus, the progress of patterning can be characterized by S, the
standard deviation of the coarse-grained particle density field. While the time course
shown in Fig. 6 is characteristic for all simulations, two model parameters have important
roles in determining both the morphology and the speed of the aggregation (Fig. 7). One
such parameter is the ratio between the steady state contractile force of the cells, F ∗,
and the Bell threshold of adhesion stability, F 0. Figure 8 compares configurations that
are at the same, late stage (S ≈ 4) of the patterning process. For small forces (F ∗ < F 0)
high cell density clusters develop at the boundary of cell free areas, similar to the in
vitro patterns observed for the Meso80 and Meso53 lines (Fig. 2).
In this case the
aggregation process is slower -- again, in accord with empirical in vitro data. For large
forces (F ∗ > F 0) links with high particle density interconnect similarly dense nodules,
reminiscent to the patterns observed in SPC111 cultures.
Model parameter α, characterizing the magnitude of external force-directed cell
displacements, sets the spatial scale of the pattern (Fig. 9). When cell-substrate
adhesions are stable (α (cid:28) 1), smaller clusters develop which are close to each other. In
contrast, when cells respond strongly to external forces (i.e., substrate adhesion is weak
or highly adaptable), fewer and larger clusters form.
Contractile tumor nodules
14
Figure 8. Morphologies characteristic for various target force values.
Simulations performed with F ∗/F 0 = 4 (a), F ∗/F0 = 2 (b), F ∗/F 0 = 1 (c), and
F ∗/F 0 = 0.5 (d) are shown at the same stage of pattern formation (S = 3.85).
For small forces F ∗ < F 0 high cell density clusters develop at the boundary
of cell free areas (arrowheads), while the particle density remains low far from
such boundaries (asterisk). In contrast, for large forces F ∗ > F 0 the particle
density is more uniformly high.
3.5. Linear stability analysis
To better understand the patterning mechanism we performed a linear stability analysis
of the spatially homogenous state. As a continuum model of the multicellular system,
we consider a viscoelastic Maxwell material which relaxes shear stresses through an
exponential decay, similar to the behavior of the particle model [19]. In one dimension
the stress distribution σ(x, t) is related to the velocity field of the tissue, v(x, t), through
the equation
∂σ
∂t
= E
∂v
∂x
− C(σ − σ∗(ρ))
(13)
Contractile tumor nodules
15
Figure 9. Morphologies characteristic for various adhesion parameter values.
Simulations performed with α = 0.3 (a), α = 0.1 (b), α = 0.03 (c), and α = 0.01
(d) are shown at the same stage of pattern formation (S = 3.5). For smaller
values of α, the size of the aggregates and the characteristic distance between
aggregates (arrows) decreases.
where E is the macroscopic elastic modulus, determined by model parameters k and g
[19], and 1/ C is the rate of macroscopic stress adjustment analogous to the parameter
C in Eq. (7). The first term on the right hand side represents elastic stress and the
second term describes the relaxation of the active contractile stress generated by the
cells to a value σ∗(ρ) that increases with the local cell density as σ∗(ρ) ∼ F ∗ρ.
Velocity is set by stress divergence as
∂σ
∂x
= αv
(14)
where the drag coefficient α is analogous to the α pararmeter in Eq. (11).
In the absence of cell death and proliferation the cell density ρ(x, t) satisfies the
conservation equation
∂
∂x
∂ρ
∂t
+
(vρ) = 0.
(15)
Contractile tumor nodules
16
Considering no-flux boundary conditions the total cell numbers are conserved so that the
average cell density ρ0 stays constant in time. Thus the spatially uniform equilibrium
solution of the above system is ρ(x, t) = ρ0, v(x, t) = 0, σ(x, t) = σ0 = σ∗(ρ0).
We can analyse the linear stability of the uniform state by considering the time
evolution of small perturbations of the form
ρ(x, t) = ρ0 + ρest+iqx
v(x, t) = vest+iqx
σ(x, t) = σ0 + σest+iqx
(16)
(17)
(18)
where s is the growth rate of a monochromatic perturbation of wavenumber q, in the
linearised system assuming that the perturbation amplitudes ρ, v, σ are small.
After substitution and neglecting higher then linear terms we obtain
Eliminating the amplitudes we obtain the following quadratic equation for the growth
rate s
sρ + iqρ0v = 0
sσ
iqσ
= i Eqv − C(σ − σ(cid:48)
= ασ
∗(ρ0)ρ)
(cid:18)
(cid:19)
− q2 σ(cid:48)
∗(ρ0) E C
= 0
C + Eq2 1
α
α
(cid:16)−b(q) ±(cid:112)b(q)2 + 4q2a
(cid:17)
(19)
(20)
(21)
(22)
(23)
(24)
s2 + s
with the solutions
s±(q) =
1
2
where
a =
σ(cid:48)
∗(ρ0) E C
α
,
b(q) = C + Eq2 1
α
Since a and b(q) are positive, one of the solutions is always negative while the other
is always positive for all wavenumbers. Thus the spatially uniform equilibrium state is
unconditionally unstable resulting in increasing cell density fluctuations. This leads to
the formation of cell aggregates, however as the perturbations grow the development of
the system is not described by the linear approximation since the neglected nonlinearities
become important.
4. Discussion
4.1. Cell displacements and external forces
As in the case of cell-cell connections, mechanical load acting on cell-substrate adhesion
complexes reduces their lifetime [20]. To relate cell movements and external forces acting
on a cell we envision the following process: when a cell-substrate connection breaks, the
same mechanical load is distributed along the remaining adhesion complexes. Thus,
each of the remaining adhesion sites transmits a larger force, their strain is increased
Contractile tumor nodules
17
leading to a small displacement of the cell body in the direction of the net external force
acting on the cell. Furthermore, when new adhesion complexes form, their equilibrium
(stress-free) configuration will correspond to the actual, slightly shifted position of the
cell. Thus, by detaching and re-attaching adhesion complexes, the cell relaxes the shear
stress between its cytoskeleton and the adhesion substrate, and moves in the direction
of the external force. In addition to this purely mechanical connection, external stress
may also effect the polarity of active cell migration [24, 25].
4.2. Contraction and aggregation
Aggregation and sorting involves acto-myosin contractility within the cortex of zebrafish
cells [26]. Here we show that cell groups may also contract through stress cables spanning
across several cells, and the resulting system self-organizes into expanding cell-free
areas and eventually into free-standing aggregates. Similar behavior also takes place
at much smaller length scales in the cytoskeleton of individual cells, where acto-myosin
contractility gives rise to f-actin bundles. For example, a similar contractile system,
but one that also includes diffusion, has been studied by [27] in the context of pattern
formation on the actomyosin cell cortex in which contractility is regulated at molecular
level. Further extensions of this problem have been described recently[28, 29].
4.3. Future treatment options
The demonstrated ability of myosin-II inhibitors to flatten mesothelioma nodules may
open a new therapeutic method. As cells within avascular, three dimensional nodules
are less exposed to systemic drugs, efficient inhibition of nodule formation could enhance
the effective drug concentration at the targeted cells. In this case, we would expect a
synergistic effect between existing anti-cancer drugs and myosin-II inhibitors.
Acknowledgements
This work was supported by the Hungarian Development Agency (KTIA AIK 12-1-2012-
0041), NIH grant GM102801, and the and the G. Harold & Leila Y. Mathers Charitable
Foundation.
References
[1] Claire Bertet, Lawrence Sulak, and Thomas Lecuit. Myosin-dependent junction remodelling
controls planar cell intercalation and axis elongation. Nature, 429(6992):667 -- 671, Jun 2004.
[2] Hisao Honda, Tatsuzo Nagai, and Masaharu Tanemura. Two different mechanisms of planar cell
intercalation leading to tissue elongation. Dev Dyn, 237(7):1826 -- 1836, Jul 2008.
[3] Adam C Martin and Bob Goldstein. Apical constriction: themes and variations on a cellular
mechanism driving morphogenesis. Development, 141(10):1987 -- 1998, May 2014.
[4] D. Stoplak and A.K. Harris. Connective tissue morphogenesis by fibroblast traction. Dev. Biol.,
90:383 -- 398, 1982.
Contractile tumor nodules
18
[5] Ravi K Sawhney and Jonathon Howard. Slow local movements of collagen fibers by fibroblasts
drive the rapid global self-organization of collagen gels. J Cell Biol, 157(6):1083 -- 1091, 2002.
[6] J.D. Murray, G.F. Oster, and A.K. Harris. A mechanical model for mesenchymal morphogenesis.
J. Math. Biol., 17:125 -- 129, 1983.
[7] G.F. Oster, J.D. Murray, and A.K. Harris. Mechanical aspects of mesenchymal morphogenesis. J
Embryol Exp Morphol, 78:83 -- 125, 1983.
[8] J. D. Murray, D Manoussaki, S. R. Lubkin, and R Vernon. A mechanical theory of in vitro
In C. D. Little, V Mironov, and E. H. Sage, editors, Vascular
vascular network formation.
morphogenesis: In vivo, in vitro, in mente., pages 223 -- 239. Birkhauser, Boston, 1998.
[9] J. D. Murray. Mathematical Biology. Springer Verlag, Berlin, 2 edition, 2003.
[10] D. Manoussaki, S. R. Lubkin, R. B. Vernon, and J. D. Murray. A mechanical model for the
formation of vascular networks in vitro. Acta Biotheor, 44(3-4):271 -- 282, 1996.
[11] Cynthia A Reinhart-King, Micah Dembo, and Daniel A Hammer.
Cell-cell mechanical
communication through compliant substrates. Biophys J, 95(12):6044 -- 6051, Dec 2008.
[12] Lekhana Bhandary, Rebecca A Whipple, Michele I Vitolo, Monica S Charpentier, Amanda E
Boggs, Kristi R Chakrabarti, Keyata N Thompson, and Stuart S Martin. Rock inhibition
promotes microtentacles that enhance reattachment of breast cancer cells. Oncotarget,
6(8):6251 -- 6266, Mar 2015.
[13] Claudia Tanja Mierke, Niko Bretz, and Peter Altevogt. Contractile forces contribute to increased
J Biol
glycosylphosphatidylinositol-anchored receptor cd24-facilitated cancer cell
Chem, 286(40):34858 -- 34871, Oct 2011.
invasion.
[14] Renaud Poincloux, Olivier Collin, Floria Lizrraga, Maryse Romao, Marcel Debray, Matthieu Piel,
and Philippe Chavrier. Contractility of the cell rear drives invasion of breast tumor cells in 3d
matrigel. Proc Natl Acad Sci U S A, 108(5):1943 -- 1948, Feb 2011.
[15] D. Krndija, H. Schmid, J-L. Eismann, U. Lother, G. Adler, F. Oswald, T. Seufferlein, and G. von
Wichert. Substrate stiffness and the receptor-type tyrosine-protein phosphatase alpha regulate
spreading of colon cancer cells through cytoskeletal contractility. Oncogene, 29(18):2724 -- 2738,
May 2010.
[16] Ro-Ting Lin, Ken Takahashi, Antti Karjalainen, Tsutomu Hoshuyama, Donald Wilson, Takashi
Kameda, Chang-Chuan Chan, Chi-Pang Wen, Sugio Furuya, Toshiaki Higashi, Lung-Chang
Chien, and Megu Ohtaki. Ecological association between asbestos-related diseases and historical
asbestos consumption: an international analysis. Lancet, 369(9564):844 -- 849, Mar 2007.
[17] Tams Garay, va Juhsz, Eszter Molnr, Maria Eisenbauer, Andrs Czirk, Barbara Dekan, Viktria Lszl,
Mir Alireza Hoda, Balzs Dme, Jzsef Tmr, Walter Klepetko, Walter Berger, and Balzs Hegeds.
Cell migration or cytokinesis and proliferation? -- revisiting the "go or grow" hypothesis in cancer
cells in vitro. Exp Cell Res, 319(20):3094 -- 3103, Dec 2013.
[18] Viktoria Laszlo, Mir Alireza Hoda, Tamas Garay, Christine Pirker, Bahil Ghanim, Thomas
Klikovits, Yawen W Dong, Anita Rozsas, Istvan Kenessey, Ildiko Szirtes, Michael Grusch, Marko
Jakopovic, Miroslav Samarzija, Luka Brcic, Izidor Kern, Ales Rozman, Helmut Popper, Sabine
Zchbauer-Mller, Gerwin Heller, Corinna Altenberger, Barbara Ziegler, Walter Klepetko, Walter
Berger, Balazs Dome, and Balazs Hegedus. Epigenetic downregulation of integrin 7 increases
migratory potential and confers poor prognosis in malignant pleural mesothelioma. J Pathol,
May 2015.
[19] Andras Czirok and Dona Greta Isai. Cell resolved, multiparticle model of plastic tissue
deformations and morphogenesis. Phys Biol, 12(1):016005, 2014.
[20] Xiaohui Zhang, Susan E Craig, Hishani Kirby, Martin J Humphries, and Vincent T Moy. Molecular
basis for the dynamic strength of the integrin alpha4beta1/vcam-1 interaction. Biophys J,
87(5):3470 -- 3478, Nov 2004.
[21] G. I. Bell. Models for the specific adhesion of cells to cells. Science, 200(4342):618 -- 627, May
1978.
[22] Daniel T. Gillespie. Exact stochastic simulation of coupled chemical reactions. The Journal of
Contractile tumor nodules
19
Physical Chemistry, 81(25):2340 -- 2361, 1977.
[23] Mih´aly Kov´acs, Judit T´oth, Csaba Het´enyi, Andr´as M´aln´asi-Csizmadia, and James R Sellers.
Mechanism of blebbistatin inhibition of myosin ii. J Biol Chem, 279(34):35557 -- 35563, Aug
2004.
[24] Dhananjay T Tambe, C. Corey Hardin, Thomas E Angelini, Kavitha Rajendran, Chan Young Park,
Xavier Serra-Picamal, Enhua H Zhou, Muhammad H Zaman, James P Butler, David A Weitz,
Jeffrey J Fredberg, and Xavier Trepat. Collective cell guidance by cooperative intercellular
forces. Nat Mater, 10(6):469 -- 475, Jun 2011.
[25] Gregory F Weber, Maureen A Bjerke, and Douglas W DeSimone. A mechanoresponsive cadherin-
keratin complex directs polarized protrusive behavior and collective cell migration. Dev Cell,
22(1):104 -- 115, Jan 2012.
[26] Jean-Lon Maitre, Hlne Berthoumieux, Simon Frederik Gabriel Krens, Guillaume Salbreux, Frank
Jlicher, Ewa Paluch, and Carl-Philipp Heisenberg. Adhesion functions in cell sorting by
mechanically coupling the cortices of adhering cells. Science, 338(6104):253 -- 256, Oct 2012.
[27] Justin S Bois, Frank Jlicher, and Stephan W Grill. Pattern formation in active fluids. Phys Rev
Lett, 106(2):028103, Jan 2011.
[28] K. V. Kumar, J. S. Bois, F. Jlicher, and S. W. Grill. Pulsatory patterns in active fluids. Physical
Review Letters, 112:208101, 2014.
[29] Thomas Moore, Selwin K Wu, Magdalene Michael, Alpha S Yap, Guillermo A Gomez, and Zoltan
Neufeld. Self-organizing actomyosin patterns on the cell cortex at epithelial cell-cell junctions.
Biophys J, 107(11):2652 -- 2661, Dec 2014.
|
1003.4372 | 1 | 1003 | 2010-03-23T10:49:29 | Aspiration of biological viscoelastic drops | [
"physics.bio-ph",
"cond-mat.soft"
] | Spherical cellular aggregates are in vitro systems to study the physical and biophysical properties of tissues. We present a novel approach to characterize the mechanical properties of cellular aggregates using micropipette aspiration technique. We observe an aspiration in two distinct regimes, a fast elastic deformation followed by a viscous flow. We develop a model based on this viscoelastic behavior to deduce the surface tension, viscosity, and elastic modulus. A major result is the increase of the surface tension with the applied force, interpreted as an effect of cellular mechanosensing. | physics.bio-ph | physics |
Aspiration of biological viscoelastic drops
Karine Guevorkian,1 Marie-Jos´ee Colbert,2 M´elanie Durth,3 Sylvie Dufour,4 and Fran¸coise Brochard-Wyart1, ∗
1 Unite Mixte de Recherche 168, Centre National de la Recherche Scientifique-Institut Curie, Paris 75248 cedex France
2Department of Physics and Astronomy, McMaster University , Ontario L8S4L8, Canada
3Ecole Polytechnique, Laboratoire d'Hydrodynamique, Palaiseau 91128 cedex, France
4Unite Mixte de Recherche 144, Centre National de la Recherche Scientifique-Institut Curie, Paris 75248 cedex, France
Spherical cellular aggregates are in vitro systems to study the physical and biophysical properties
of tissues. We present a novel approach to characterize the mechanical properties of cellular aggre-
gates using micropipette aspiration technique. We observe an aspiration in two distinct regimes, a
fast elastic deformation followed by a viscous flow. We develop a model based on this viscoelastic
behavior to deduce the surface tension, viscosity, and elastic modulus. A major result is the increase
of the surface tension with the applied force, interpreted as an effect of cellular mechanosensing.
Embryonic morphogenesis, wound healing, cancer
growth and metastasis are a few examples where the
physical laws play an important role along with genetic
cues in the functioning of a tissue. An aggregate of living
cells, used as a model tissue, behaves like a viscoelastic
liquid. Spreading and sorting are signatures of liquid-like
behavior of embryonic tissues [1, 2]. Moreover, cellular
aggregates in solution round up to form "spheroids" in
order to minimize their surface energy, similar to oil drops
in water. This is a manifestation of surface tension, which
has been related to intercellular adhesion energy [3]. In
the past, the simple analogy between liquids and tissues
has lead to valuable findings about the mechanics of em-
bryonic mutual envelopment [4], tissue spreading [5], and
cancer propagation [6]. A knowledge of the surface ten-
sion of tissues has also been essential for organ printing
in tissue engineering [7].
To measure the surface tension of cellular aggregates
and investigate the role of surface tension in cell sorting,
Steinberg and coworkers [2] introduced the parallel plate
compression apparatus, which has since been used by
other groups [8, 9]. In this method, an aggregate is sub-
jected to an imposed deformation and the surface tension
is inferred from the relaxation force, while the viscosity
of the tissue is obtained from the shape relaxation [10].
Difficulties in the evaluation of the principal radii of a
compressed aggregate and the contact angle between the
aggregate and the plate make this technique rather deli-
cate. Deformation of aggregates under centrifugal forces
is an alternative way that has been used to classify aggre-
gates of various cell types [11]. Recently this technique
has been combined with Axisymmetric drop shape anal-
ysis (ASDA) for measuring the surface tension of embry-
onic tissue [12].
In this letter, we propose the use of micropipette as-
piration technique to study the surface tension and the
mechanical properties of cellular aggregates. This tech-
nique has previously been used to evaluate the viscoelas-
tic properties of single cells [13, 14] and the stiffness of
∗Electronic address: [email protected]
tissues [15–17] at small deformations. For a Newtonian
fluid, the aspiration dynamics is governed by the Wash-
burn law, L(t) ∼ t1/2, where L(t) is the advancement of
the liquid inside the pipette [18]. For a tissue, a com-
pletely different behavior is observed due to its viscoelas-
tic properties. Under applied stress σ, a tissue responds
like an elastic solid at times shorter than a characteristic
time τ [19], and like a fluid for t > τ . This behavior
can be described by dσ/dt + σ/τ = Edǫ/dt, where ǫ is
the strain; the viscosity η, of the material is related to
its elastic modulus E, through η ≈ Eτ [20]. In the case
of parallel plate compression, ǫ is constant and the stress
relaxes to equilibrium, whereas for the case of aspiration,
σ stays constant and the tissue flows.
Spherical cellular aggregates are useful systems to
study the mechanical properties of tissues since the
adhesion energy between the subunits (cells) can be
controlled. We have used murin sarcoma (S180) cell
lines transfected to express various levels of E-cadherin
molecules at the surface of the cells [21], thereby control-
ling the intercellular adhesion energy. Here, we focus on
the most adhesive cell lines. Cells were cultured under
5% air/ 5% CO2 atmosphere in DMEM enriched with
10% calf serum (culture medium) and prepared for ag-
gregation following a procedure similar to Ryan et al.'s
[5]. Aggregates ranging from 250 µm to 400 µm in diam-
eter were obtained from 5 ml of cell suspension in CO2-
equilibrated culture medium at a concentration of 4 ×105
cells per ml in 25 ml erlenmeyer flasks, and placed in a
gyratory shaker at 75 rpm at 37◦C for 24 hours. The
flasks were pretreated with 2% dimethylchlorosilane in
chloroform and coated with silicon to prevent adhesion
of cells to the glass surface. We performed the aspiration
of the aggregates using pipettes with diameters 3-5 times
that of a single cell (40-70 µm). The pipettes were fab-
ricated by pulling borosilicate capillaries (1 mm/0.5 mm
O/I diameter) with a laser-based puller (P-2000, Sutter
Inst. Co, Novato, CA ), and sized to the desired diameter
by using a quartz tile. To prevent adhesion of the cells
to the micropipette walls, the pipettes were incubated
in 0.1 mg/ml PolyEthyleneGlycol-PolyLysin (PLL(20)-
g[3.5]-PEG(2), Surface Solution, Dubendorf Switzerland)
in HEPES solution (pH 7.3) for one hour. The obser-
2
FIG. 2: Aspiration of a viscoelastic drop. (A) Schematic pre-
sentation of an aspirated drop. (B) Creep curve showing a
L∞.
fast elastic deformation, δ, followed by a viscous flow,
(C) Modified Maxwell model. The Kelvin body accounts for
the initial elastic deformation, where k1 is the spring con-
stant related to the elasticity of the aggregate, k2 accounts
for the initial jump in L(t), and ξc is a local friction coeffi-
cient, related to the raising time of the elastic deformation.
The dashpot represents the viscous dissipation of the flowing
tissue.
retraction cycle on each aggregate to maintain the same
initial conditions. Both aspiration and retraction curves
show a fast initial deformation, followed by a slow flow
with constant velocity L∞; the transition between the
two regimes is marked by an arrow. This creep behav-
ior is a signature of viscoelastic materials. We proceeded
by considering these cell aggregates as viscoelastic liquid
drops with a surface tension γ.
− 1
The total energy of a drop aspirated inside a non-
adhesive pipette, "zero" wetting, is given by F = (4πR2+
2πRpL)γ − πR2
pL∆P [22], where R and Rp are the radii
of the drop and the pipette respectively, and ∆P is the
applied pressure, as shown schematically in Fig. 2(A).
Considering volume conservation, the aspiration force is
f = πR2
p(∆P − ∆Pc), where the critical pressure to as-
pirate, ∆Pc, relates to the surface tension through the
Laplace law: ∆Pc = 2γ( 1
R ). Note that R is not con-
Rp
stant, but can be approximated by R ≈ R0 for Rp ≪ R0.
From scaling laws the aspiration force and the elastic de-
formation at short time, δ, are related by f
= C × E δ
,
A0
Rp
where A0 = πR2
p, E is the elastic modulus, and C is a
geometrical factor C ≈ 1 for our experimental conditions
[15], leading to f ≈ πRpEδ. At long times, f is balanced
by the friction force due to the viscous flow into the orifice
[23] and the slippage of the advancing tongue on the wall
as: f = 3π2ηRp L + 2πkRpL L, where η is the viscosity
of the tissue, and k is the wall-tissue friction coefficient.
We define Lc = 3πη/2k as a characteristic length associ-
ated to the wall friction. In the limit of L > Lc we have
L ∼ t1/2, whereas for L < Lc we find L = L∞t, where
L∞ = f /3π2ηRp. We have estimated k ≈ 108 N.s/m2
from the advancement velocity of a completely aspirated
aggregate, leading to Lc ≈ η/k ≈ 2 mm (see below for η).
Therefore we can ignore the wall friction. To combine the
elastic and viscous regimes, we use the modified Maxwell
model depicted in Fig. 2(C). The total displacement L(t)
is given by:
L(t) =
f
k1 (cid:18)1 −
k2
k1 + k2
e−t/τc(cid:19) +
f
ξt
t,
(1)
where k1 = πRpE, and ξt = 3π2ηRp. The first term
FIG. 1: Micropipette aspiration of spherical cellular aggre-
gates. (A)-(C) Aspiration of an aggregate with ∆P = 1370
(14 cmH2O), R0 = 150 µm, Rp = 30 µm, scale bar is 50
µm. (D) Aspiration and (E) retraction cycles for an aggre-
gate at ∆P = 1180 Pa, with R0 = 175 µm, and Rp = 35
µm. Arrows indicate the transitions from elastic to viscous
regimes. Dotted lines are fits to the experimental curves using
the viscoelastic model (see text for details).
vation chamber consisted of a thick U-shaped Parafilm
spacer (2 cm×2 cm×5 mm), sandwiched in between two
microscope slides by gentle heating. Aggregates were
then suspended in CO2 equilibrated culture medium and
the pipette was introduced into the chamber. To pre-
vent evaporation, the open end was sealed with mineral
oil. A large range of pressures (∆P = 0.1 − 5 kPa) was
attained by vertically displacing a water reservoir, con-
nected to the pipette, with respect to the observation
chamber. Aspirated aggregates were visualized on an in-
verted microscope (Zeiss Axiovert 100) equipped with a
×20 air objective (NA 0.45). Movies of the advancement
of the aggregates inside the pipette were recorded with a
CCD camera (Luca-R, Andor, Belfast UK) with a 5-30
second interval. Cell viability in aspirated aggregates was
checked using trypan blue exclusion test. After 3 hours
of aspiration, trypan blue was added to the experimental
chamber to a final concentartion of 25%. A small num-
ber of dead cell were present at the core of the aggregate,
comparable to the aggregates at rest, but no significant
cell death was seen in the aspirated tongue.
Fig.1(A)-(C) shows snapshots of the aspiration of an
aggregate inside a pipette at a constant pressure. The
advancement of the aggregate inside the pipette is char-
acterized by tracking the displacement of the front of the
tongue with respect to the pipette tip, represented by
L(t) in Fig. 2(A). As a first approach, steps of ∆P were
applied at a time interval of 2-3 hours, in order to de-
termine the dynamics of aspiration as a function of ∆P .
However, we observed a degradation of the cells when
aggregates stayed under aspiration for over 6 hours, lim-
iting the number of steps. Consequently, we modified the
procedure and applied cycles of pressure as shown in Fig.
1(D)-(E). After each aspiration at constant pressure, the
pressure was set to zero and the retraction of the tongue
was monitored. In general, we performed one aspiration-
3
FIG. 4: Manifestation of surface tension augmentation. (A)
Successive aspiration on the same aggregate. Dashed lines
represent the pressure profile. The second aspiration shows a
smaller elastic deformation in spite of the applied ∆P being
larger . (B) Increasing and decreasing ∆P in steps. When ∆P
is decreased to 3 kPa, the aggregate stops flowing, indicating
that the surface tension has increased from its steady state
value.
exp
∆P −∆Pc(γ)
∆P −∆Pc(γ0) , and τ r = τ r
As mentioned above, the relaxation time for a vis-
coelastic material to flow is τ ≈ η/E. This characteris-
tic time can experimentally be evaluated from the creep
curve as τexp = δ/ L∞ (Fig. 2(B)). However, as can be
seen from the curves on Fig. 1(D)-(E), the retraction
of the tongue has a much faster dynamics, resulting in
τ a
exp ≫ τ r
exp. This is due to γ increasing from γ0 (elas-
tic regime) to γ (viscous regime) during the slow aspira-
tion and not relaxing during the fast retraction. Taking
these corrections into account, τ = τexp × fvisc/felastic,
∆Pc(γ)
leading to τ a = τ a
∆P .
Taking γ0 = 6 mN/m we obtain τ a = 47 ± 10 min.,
and τ r = 40 ± 7 min., resulting in an average value of
¯τ = 44 ± 7 min. We estimate an elastic modulus of
E = 3πη/¯τ ≈ 700 ± 100 Pa for these aggregates, compa-
rable to values reported for embryonic liver tissue [24].
The elastic local cell's relaxation time, τc, is one order of
magnitude smaller than the tissue relaxation times. We
systematically find τ a
c , showing that pre-stressed
c
tissue has a faster elastic response. We have character-
ized mechanical properties of tissue such as their sur-
face tension, viscosity and elasticity using micropipette
aspiration technique. We have found that the surface
tension of the aggregate is stress-dependent, suggesting
that upon the application of a permanent external force,
tissue cohesion is reinforced. Successive aspiration on
the same aggregate validates our finding. As shown in
≫ τ r
exp
FIG. 3: Viscosity and surface tension of aspirated aggregates.
(A) γ as a function of applied force R2
p∆P . Filled symbols
are obtained from the relationship between La
∞ and
open symbols are obtained from La
∞ and using the measured
value for η. The curve is presented to guide the eye. (B) Flow
velocity L∞ as a function of Rp σ, (slope equals to 1/3πη).
∞ and Lr
characterizes the elastic regime with τc = ξc(k1+k2)
being
the raising time of the elastic deformation δ, and the
second term characterizes the flow at constant velocity
L∞. The tissue relaxation time separating the elastic
and viscous regimes is given by τ = ξt/k1 = 3πη/E.
k1k2
∞/( Lr
∞ + La
∞ and Lr
The dashed lines in Fig.
1(D)-(E) are the adjust-
ment of Eq. 1 to the data with four fitting parameters:
δ = f /k1, L∞ = f /ξt, β = k2/(k1+k2), and τc. The criti-
cal pressure is deduced from ∆Pc = ∆P Lr
∞),
where La
∞ are the aspiration and the retrac-
tion flow rates, respectively. Using the values for ∆Pc,
the surface tension, γ, is derived from the Laplace law.
Fig. 3(A) shows an increase in γ as the applied force
is increased. By extrapolation, we obtain the surface
tension of the aggregate at rest, γ0 ≈ 6 mN/m, compara-
ble to previously obtained values for similar tissue types
[2, 9, 24]. We also measured directly a lower bound for
∆Pc(γ0) by finding the maximum pressure (≈ 300 − 400
Pa for Rp = 35 µm) at which the aggregate does not
penetrate into the pipette, leading to γ0 ∼ 5 − 7 mN/m.
The flow velocities of aggregates during aspiration and
retraction are shown in Fig. 3(B) as a function of the ap-
plied stress, σ, where σ = ∆P − ∆Pc for aspiration, and
σ = ∆Pc for retraction. The observed linear relationship
between L∞ and Rpσ shows that η stays constant and no
shear thinning effect is observed in the range of pressures
(1-3 kPa) applied in our experiments. The slope of the
fitted line gives η = 1.9 ± 0.3 × 105 Pa.s, comparable to
the values previously reported for aggregates of mouse
embryonal carcinoma F9 cell lines (η ≈ 2 × 105 Pa.s)
[9, 25] and various chicken embryonic tissues (η ≈ ×105
Pa.s) [24, 26]. Preliminary results on aggregates of the
same cell lines with less intercellular cohesion have shown
a similar but much faster aspiration dynamics, indicating
a smaller viscosity for these aggregates (data not shown).
In our analysis we have assumed that ∆Pc does not re-
lax in the time scale of our experiment when ∆P = 0.
This assumption is justified, since the slopes of the fast
retraction curves stay constant as seen in Fig. 1(E).
4
Fig.4(A), the elastic deformation of the second aspira-
tion in smaller, indicating a larger initial γ0. Another
direct manifestation of the reinforcement of γ is shown in
Fig.4(B), when ∆P is decreased to a few times ∆Pc(γ0),
the aggregate relaxes instead of flowing.
The reinforcement of γ is a signature of an active re-
sponse of the cells to mechanical forces [27, 28] leading to
cytoskeletal remodeling [29], which may involve stretch-
activated membrane channels [30], stress fiber polymer-
ization and tensening by Myosin II motors [31, 32], and
clustering of cadherins [33]. At the tissue level, it has
also been shown that application of an external force to
the tissue using a 20 µm micro-needle increases the tissue
tension, leading to morphogenetic movements [34]. Pro-
tein labeling and cytoskeleton modifying drugs have to be
used to better understand the reinforcement mechanism
at a the cellular level. This novel method brings com-
plementary features to the classical parallel plate com-
pression technique, since instead of relaxing to equilib-
rium, the cells flowing into the pipette are continuously
stretched. Moreover, this technique allows us to reach
much higher stresses, up to hundred times the aggregate's
Laplace pressure.
How the surface tension and the viscoelastic properties
of an aggregate depend on the properties of the subunits
and on their interconnection remains an open question.
Previous studies have measured the surface tension of ag-
gregates as a function of the level of expression of inter-
cellular binders (cadherin molecules) [3]. However, the
relationship between the adhesion energy and the sur-
face tension is still debated. We anticipate using the mi-
cropipette aspiration technique to relate the surface ten-
sion of aggregates to the cell-cell adhesion energy, which
has been previously measured by one of us [21].
Complete aspiration of aggregates inside a pipette can
also be used to apply high pressures (∼ γ/Rp) to cancer-
ous tissue and thus investigate the validity of the home-
ostatic pressure model, which predicts that metastatic
cells can only grow if the internal pressure of the ag-
gregate is below a critical "homeostatic pressure" [35].
Combined with confocal microscopy, tissue relaxation un-
der stress can be studied at microscopic level by prob-
ing the cellular rearrangements inside an aspirated ag-
gregate. Compared to more conventional methods, the
micropipette aspiration technique is easy to set up and
can be applied to in-vivo examination of biological sys-
tems, such as living tissue or drug treated tumors, and to
other complex fluids, such as viscous pastes and foams.
We would like to thank D. Cuvelier for his help with
the experimental setup, C. Clanet for useful discussions,
and J. Elgeti and D. Gonzalez-Rodriguez for their critical
reading of the manuscript. F. B. W. and S. D. would like
to thank Curie PIC program for funding. The group
belongs to the CNRS consortium CellTiss.
[1] R. Gordon et al., J. Theor. Biol. 37, 43 (1972).
[2] R. A. Foty et al., Phys. Rev. Lett. 72, 2298 (1994).
[3] R. A. Foty and M. S. Steinberg, Dev. Biol. 278, 255
(2005).
[4] R. A. Foty et al., Development 122, 1611 (1996).
[5] P. L. Ryan et al., Proc. Natl. Acad. Sci. USA 98, 4327
(2001).
[6] R. A. Foty and M. S. Steinberg, Int. J. Dev. Biol. 48, 397
(2004).
Physics, 3rd ed., Vol. 7: Theory of Elasticity (Elsevier
Science, 1984).
[21] Y. S. Chu et al., J. Cell Biol. 167, 1183 (2004).
[22] P. G. de Gennes et al., Capillarity and Wetting Phenom-
ena: Drops, Bubbles, Pearls, Waves (Springer, 2004).
[23] Z. Dagan et al., J. Fluid Mech. 115, 505 (1982).
[24] G. Forgacs et al., Biophys. J. 74, 2227 (1998).
[25] P. Marmottant et al., Proc. Natl. Acad. Sci. USA 106,
17271 (2009).
[7] K. Jakab et al., Tissue Eng. Part A 14, 413 (2008).
[8] C. Norotte et al., Europhys. Lett. 81, 46003 (2008).
[9] A. Mgharbel et al., HFSP Journal 3, 213 (2009).
[26] K. Jakab et al., Dev. Dyn. 237, 2438 (2008).
[27] Y. Caia et al., Curr. Opin. Cell Biol. 21, 47 (2009).
[28] P. A. Janmey and D. A. Weitz, Trends Biochem. Sci. 29,
[10] J. Mombach et al., Physica A: Statistical Mechanics and
364 (2004).
its Applications 352, 525 (2005).
[29] N. Borghi and W. J. Nelson, Curr. Top. Dev. Biol. 89, 1
[11] H. M. Phillips and M. S. Steinberg, Biophys. J. 64, 121
(2009).
(1969).
[30] F. Sbrana et al., Am. J. Physiol., Cell Physiol. 295, C160
[12] A. Kalantarian et al., Biophys. J. 96, 1606 (2009).
[13] M. Sato et al., J. Biomech. Eng. 112, 263 (1990).
[14] E. Evans et al., Biophys. J. 56, 151 (1989).
[15] T. Aoki et al., Ann. Biomed. Eng. 25, 581 (1997).
[16] J. T. Butcher et al., Circ. Res. 100, 1503 (2007).
[17] T. Ohashi et al., J. Biomech 38, 2248 (2005).
[18] E. Washburn, Phys. Rev. 17, 273 (1921).
[19] Y. S. Chu et al., Phys. Rev. Lett. 94, 028102 (2005).
[20] L. D. Landau and E. M. Lifshitz, Course of Theoretical
(2008).
[31] O. Chaudhuri et al., Nat Meth. 6, 383 (2009).
[32] N. Desprat et al., Biophys. J. 88, 2224 (2005).
[33] H. Delanoe-Ayari et al., Proc. Natl. Acad. Sci. USA 101,
2229 (2004).
[34] P. A. Pouille et al., Sci. Signal. 2, ra16 (2009).
[35] M. Basan et al., HFSP Journal 3, 265 (2009).
|
1911.00421 | 1 | 1911 | 2019-11-01T15:25:39 | Cockroaches use diverse strategies to self-right on the ground | [
"physics.bio-ph",
"q-bio.QM"
] | Terrestrial animals often must self-right from an upside-down orientation on the ground to survive. Here, we compared self-righting strategies of the Madagascar hissing, American, and discoid cockroaches on a challenging flat, rigid, low-friction surface to quantify the mechanical principles. All three species almost always self-righted (97% probability) when given time (30 seconds), frequently self-righted (63%) on the first attempt, and on that attempt did so in one second or less. When successful, two of the three species gained and used pitch and/or roll rotational kinetic energy to overcome potential energy barriers (American 63% of all attempts and discoid 78%). By contrast, the largest, heaviest, wingless cockroach (Madagascar hissing) relied far less on the energy of motion and was the slowest to self-right. Two of the three species used rolling strategies to overcome low potential energy barriers. Successful righting attempts had greater rolling rotation than failed attempts as the center of mass rose to the highest position. Madagascar hissing cockroaches rolled using body deformation (98% of all trials) and the American cockroach relied on leg forces (93%). By contrast, the discoid cockroach overcame higher and a wider range of potential energy barriers with simultaneous pitching and rolling using wings (46% of all trials) and legs (49%) equally to self-right. Our quantification revealed the performance advantages of using rotational kinetic energy to overcome the potential energy barrier and rolling more to lower it, while maintaining diverse strategies for ground-based self-righting. | physics.bio-ph | physics | Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Cockroaches use diverse strategies to self-right on the ground
Chen Li1,2*, Toni Wöhrl3, Han K. Lam2, Robert J. Full2
1Department of Mechanical Engineering, Johns Hopkins University
2Department of Integrative Biology, University of California, Berkeley
3Institute of Sports Science, Friedrich-Schiller-Universität Jena
*[email protected]
KEY WORDS
Locomotion, potential energy barrier, insects, Periplaneta americana, Blaberus discoidalis,
Gromphadorhina portentosa
SUMMARY STATEMENT
Comparative study of cockroach self-righting reveals performance advantages of using
rotational kinetic energy to overcome potential energy barrier and rolling more to lower it, while
maintaining diverse strategies.
ABSTRACT
Terrestrial animals often must self-right from an upside-down orientation on the ground to
survive. Here, we compared self-righting strategies of the Madagascar hissing, American, and
discoid cockroaches on a challenging flat, rigid, low-friction surface to quantify the mechanical
principles. All three species almost always self-righted (97% probability) when given time (30
seconds), frequently self-righted (63%) on the first attempt, and on that attempt did so in one second
or less. When successful, two of the three species gained and used pitch and/or roll rotational kinetic
energy to overcome potential energy barriers (American 63% of all attempts and discoid 78%). By
contrast, the largest, heaviest, wingless cockroach (Madagascar hissing) relied far less on the
energy of motion and was the slowest to self-right. Two of the three species used rolling strategies
to overcome low potential energy barriers. Successful righting attempts had greater rolling rotation
than failed attempts as the center of mass rose to the highest position. Madagascar hissing
cockroaches rolled using body deformation (98% of all trials) and the American cockroach relied
on leg forces (93%). By contrast, the discoid cockroach overcame higher and a wider range of
potential energy barriers with simultaneous pitching and rolling using wings (46% of all trials) and
legs (49%) equally to self-right. Our quantification revealed the performance advantages of using
1
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
rotational kinetic energy to overcome the potential energy barrier and rolling more to lower it, while
maintaining diverse strategies for ground-based self-righting.
INTRODUCTION
Righting oneself from upside down on the ground is a prevalent locomotor transition that
many animals must perform to survive. Even on flat, level ground with high friction, legged
locomotion can induce large pitch and roll moments (Ting et al., 1994) that can result in overturning.
During locomotion in complex terrain with inclinations (Minetti et al., 2002), uneven topology
(Chiari et al., 2017; Daley and Biewener, 2006; Sponberg and Full, 2008), low friction (Clark and
Higham, 2011), uncertain contact (Spagna et al., 2007), flowable ground (Li et al., 2012), and
cluttered obstacles (Li et al., 2015; Li et al., 2017), overturning is even more likely. Other forms of
terrestrial locomotion like jumping (Faisal and Matheson, 2001; Libby et al., 2012) and climbing
(Jusufi et al., 2008), as well as flying (Faisal and Matheson, 2001) and swimming (Vosatka, 1970),
can suffer instability and loss of body control resulting in overturning. Non-locomotor behaviors
such as fighting and courtship can also produce overturning (Mann et al., 2006; Willemsen and
Hailey, 2003). Under these circumstances, animals must be able to self-right promptly to avoid
predation, starvation, and dehydration, as well as to sense, locomote, and reproduce.
Small animals like insects are particularly susceptible to overturning, because they are
more sensitive to perturbations resulting from small body inertia (Walter and Carrier, 2002) and
terrain irregularities negligible to larger animals (Kaspari and Weiser, 1999). Ground-based self-
righting has been studied in many insect species, including beetles (Evans, 1973; Frantsevich, 2004;
Frantsevich and Mokrushov, 1980), cockroaches (Camhi, 1977; Delcomyn, 1987; Full et al., 1995;
Reingold and Camhi, 1977; Sherman et al., 1977; Zill, 1986), stick insects (Graham, 1979), locusts
(Faisal and Matheson, 2001), and springtails (Brackenbury, 1990). Many self-righting strategies
have been described (Brackenbury, 1990; Camhi, 1977; Evans, 1973; Faisal and Matheson, 2001;
Frantsevich, 2004; Full et al., 1995; Zill, 1986), including: (1) using appendages (legs, wings, tail,
and antennae) and head to grasp, pivot, push, or pull, (2) deforming the body, and (3) jumping with
elastic energy storage and release. Some insects use multiple strategies and transition among them
to self-right (Frantsevich, 2004). In addition, insects can use diverse body rotation including
pitching, diagonal rotations (simultaneous pitching and rolling), and rolling (Brackenbury, 1990;
Camhi, 1977; Delcomyn, 1987; Evans, 1973; Frantsevich, 2004; Frantsevich and Mokrushov, 1980;
Full et al., 1995; Reingold and Camhi, 1977; Sherman et al., 1977; Zill, 1986). Furthermore, neural
control and motor patterns of self-righting have been investigated in a variety of insect species
(Camhi, 1977; Delcomyn, 1987; Faisal and Matheson, 2001; Frantsevich and Mokrushov, 1980;
2
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Graham, 1979; Reingold and Camhi, 1977; Sherman et al., 1977; Zill, 1986). Although these
strategies have been well described, the mechanical principles of ground-based self-righting of
small animals remain less understood. Here, we quantify the performance and body rotation of self-
righting cockroaches and model the mechanical challenges to gain insight into what governs a small
animal's use of various strategies and body rotation.
Previous observations and modeling in turtles have provided insight into the mechanics of
how body and appendage morphology affects ground-based self-righting of larger animals (Ashe,
1970; Domokos and Várkonyi, 2008). Ground-based self-righting is the change of body orientation
during which the body overcomes gravitational potential energy barriers (Domokos and Várkonyi,
2008). Based on this concept, a planar geometric model explained how shell shape and appendage
length together determine whether turtles use active or passive strategies to self-right in the
transverse plane (Domokos and Várkonyi, 2008). Turtles primarily rely on passive rotations of
unstable shells and/or active, quasi-static pushing of necks and legs to overcome large, primary
potential energy barriers. To assist self-righting, turtles also use head and leg bobbing to gain
modest amounts of rotational kinetic energy to overcome small, secondary potential energy barriers
(Domokos and Várkonyi, 2008). In addition, the dependence of potential energy barriers on body
rotation explained why many turtles almost always self-right via body rolling in the transverse
plane on level, flat surfaces (Domokos and Várkonyi, 2008; Malashichev, 2016; Rubin et al., 2018;
Stancher et al., 2006). Turtles have shells longer in the fore-aft than in the lateral direction, so body
pitching overcomes higher potential energy barriers than body rolling does. Because turtles cannot
gain sufficient body rotational kinetic energy to overcome the large potential energy barriers
required for self-righting using pitching, they roll to self-right.
Here, inspired by these insights, we take the next step in understanding the mechanical
principles of ground-based self-righting of small animals. First, we hypothesized that small insects'
self-righting strategies can be dynamic, being able to gain and use pitch and/or roll rotational kinetic
energy to overcome primary potential energy barriers. Dynamic behavior is plausible because many
insects like cockroaches and beetles are capable of rapid locomotion and generating large impulses
relative to body weight (Koditschek et al., 2004; Sponberg and Full, 2008; Ting et al., 1994; Zurek
and Gilbert, 2014). Second, we hypothesized that, given the diverse three-dimensional body
rotations possible (Brackenbury, 1990; Camhi, 1977; Delcomyn, 1987; Evans, 1973; Frantsevich,
2004; Frantsevich and Mokrushov, 1980; Full et al., 1995; Reingold and Camhi, 1977; Sherman et
al., 1977; Zill, 1986), insects roll more when they succeed in self-righting than when they fail
because increased rolling lowers potential energy barriers.
3
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
To test our hypotheses, we studied self-righting on a flat, rigid, low friction surface of three
species of cockroaches, the Madagascar hissing cockroach (Gromphadorhina portentosa), the
American cockroach (Periplaneta americana), and the discoid cockroach (Blaberus discoidalis),
which differ in body size, body shape, leg length, and availability of wings (Fig. 1). The selection
of multiple species (Chiari et al., 2017; Domokos and Várkonyi, 2008) from a common super order
(Dictyoptera) (Bell et al., 2007) provided access to observing a greater number of strategies and
body rotations, but with phylogenetic control that allows comparison. We used high-speed imaging
to measure the animals' self-righting performance and body rotation. We used a locomotor
transition ethogram analysis to quantify probability distribution of and transitions between self-
righting strategies. We developed a simple geometric model to examine how the animal body
moved to overcome barriers on a potential energy landscape. We compared successful and failed
attempts to reveal what factors among body deformation and body and appendage behaviors
contributed to successful self-righting (Rubin et al., 2018).
Fig. 1. Interspecies comparison of body and appendage morphology and schematic defining
4
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
digitized markers and variables. (A) Madagascar hissing cockroach. (B) American cockroach.
(C) Discoid cockroach. The animals are scaled to the same width in top and front views to illustrate
differences in body elongation and flatness (Table 1). The body shape of each species is well
approximated by an ellipsoid, with length, width, and thickness of 2a, 2b, and 2c, respectively.
Four colored points in the top views are the four digitized markers. (D, E) Schematics of a self-
righting animal, showing the four digitized markers, head (H), abdomen (A), left (L), right (R), and
definitions of body pitch , body roll , body flexion , head twisting H, and abdomen twisting A.
A' and L' are downward projections of A and L to the same height levels of H and R, respectively.
M is a point midway between L and R. 𝑛⃑ is the plane normal of the estimated sagittal plane. H"
and A" are projections of H and A into the sagittal plane. In the example shown (discoid cockroach
using wings), body is flexing, head is twisting to the right, and abdomen is twisting to the left.
Table 1. Sample size and morphological measurements (mean ± 1 s.d.).
Species
Madagascar American
Discoid
Number of individuals
Number of trials
Number of successful trials
within 30 seconds
Number of failed trials
within 30 seconds
Number of successful trials
on first attempt
Number of successful trials
needing more than one attempt
Number of attempts
Number of successful attempts
Number of failed attempts
Mass (g)
Body length 2a (cm)
Body width 2b (cm)
Body thickness 2c (cm)
Front leg length (cm)
Mid leg length (cm)
Hind leg length (cm)
Body elongation
(body length / body width)
7
59
56
3
40
7
61
58
3
29
16
95
56
39
0.66 ± 0.05
3.34 ± 0.14
1.19 ± 0.07
0.70 ± 0.01
1.62 ± 0.03
2.20 ± 0.09
3.12 ± 0.03
2.81 ± 0.20
29
205
58
147
2.14 ± 0.15
4.98 ± 0.17
2.38 ± 0.11
0.96 ± 0.02
1.91 ± 0.10
2.67 ± 0.06
3.60 ± 0.00
2.09 ± 0.12
6
55
54
1
41
13
78
54
24
7.44 ± 1.17
6.03 ± 0.42
2.24 ± 0.10
1.32 ± 0.10
2.08 ± 0.08
2.93 ± 0.03
3.65 ± 0.10
2.69 ± 0.22
5
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
MATERIALS AND METHODS
Animals
We used six Madagascar hissing cockroach, seven American cockroaches, and seven
discoid cockroaches. We used adult males because females were often gravid and under different
load-bearing conditions. Prior to experiments, we kept the cockroaches in individual plastic
containers at room temperature (28 °C) on a 12 h: 12 h light: dark cycle and provided water and
food (fruit and dog chow) ad libitum. See Table 1 for animal body mass and body and leg
dimensions.
The Madagascar hissing and American cockroaches are both relatively elongate and have
similar body aspect ratios (body length vs. body width vs. body thickness) (Table 1, Fig. 1A, B).
By contrast, the discoid cockroach is less elongate (ANOVA, P < 0.05) and flatter (ANOVA, P <
0.05) (Table 1, Fig. 1C). The American and discoid cockroach have wings, whereas the Madagascar
hissing cockroach are wingless.
Experimental setup and protocol
We used a low friction, level, flat, rigid surface as the righting arena. The surface was
covered with low-friction cardstock, with static friction coefficient = 0.10 ± 0.01 (mean ± 1 s.d.)
between the ground and dorsal surface of the animal body (measured by the inclined plane method).
Sidewalls around the arena prevented animals from escaping. Four 500W work lights above and
three fluorescent lights around the righting arena provided lighting for the high-speed cameras. The
temperature during experiments was 36.5 °C. Two webcams (Logitech C920) recorded the entire
experiments from top and side views at 30 frame/s. Four synchronized high-speed cameras (AOS
and Fastec) recorded up to 30 seconds of each trial from four sides of the arena at 250 frames/s and
800 × 600 resolution.
For every trial, we held the animal in an upside-down orientation by grasping the edges of
its pronotum and gently released it from a small height (< 0.5 cm) above the center of the area. The
small drop was to ensure that the animal did not begin leg searching, a common strategy used for
self-righting, before it was set to be upside down on the ground. From high-speed videos, we
verified that kinetic energy from the small drop dissipated so that the animal was stationary before
it initiated the self-righting response. If the animal did not right within 30 seconds, it was picked
up and placed back into its container for rest. We tested all individuals of all three species by
alternating individuals and species to ensure sufficient time (> 10 minutes) for each individual to
rest between trials to minimize the effect of fatigue (Camhi, 1977).
6
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Sample size
Excluding trials in which the animals touched the sidewalls when attempting to self-right,
we collected a total of 176 trials from a total of 20 individuals from the three species of cockroaches,
with approximately 9 trials from each individual. Because the animal often needed more than one
attempt to self-right, from the 176 trials, we identified a total of 378 attempts (see definition below),
including 168 successful attempts and 210 failed attempts. See Table 1 for details of sample size.
Definition of attempts
Because the animal was allowed up to 30 seconds during each trial, much longer than the
time of a typical self-righting attempt (Fig. S1A), the animal may make more than one attempt in
a trial. Thus, for each trial, we observed the videos to record how many attempts the animal made,
whether each attempt was successful or not, and measured the duration of each attempt.
We defined an attempt as the entire process during which the animal moved its body and
appendages to eventually generate a pitching and/or rolling motion, because change in body yaw
did not contribute to self-righting. We separated two consecutive attempts by when the animal
returned to an upside-down orientation in between the two pitching and/or rolling motions. By this
definition, each failed attempt not only included the duration of the body pitching and/or rolling
motion, but also the duration prior to it during which the body and appendages moved to generate
the attempt. We note that attempts by this definition may and often do include multiple movement
cycles of wing opening/closing or leg pushing or flailing, which often occur at higher frequencies
than body pitching and/or rolling motion. We did not use wing or leg motion to define attempts
because they do not necessarily generate body pitching or rolling, which are defining features
towards self-righting.
We then separated attempts into successful and failed ones depending on whether it
resulted in self-righting. Each trial can have up to one successful attempt preceded by zero to
several failed attempts.
Performance analysis
For each trial, we recorded whether the animal succeeded in self-righting within 30 seconds.
We also recorded whether the animal succeeded in self-righting on the first attempt of each trial.
For each successful trial, we recorded the total number of attempts it took the animal to self-right.
We measured total self-righting time, defined as the duration from the instant the animal's dorsal
surface touched the surface in an upside-down orientation to the instant when all its six legs touched
7
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
the ground after the body became upright. We also measured successful attempt time, defined as
the duration of the final successful attempt of each successful trial. We calculated the probabilities
of self-righting within 30 seconds and on the first attempt, as the ratio of their occurrences to the
total number of trials for each species.
Strategy transition analysis
To quantify the transitions between strategies during self-righting, we created a locomotor
ethogram analysis using each trial (Blaesing, 2004; Li et al., 2015). For each species, we first
recorded the sequence of locomotor strategies and the outcome (either successful self-righting or
failure). We then calculated the animal's probabilities of entering various self-righting strategies,
transitioning between them, and attaining a final outcome. The probability of each transition
between nodes was defined as the ratio of the number of occurrences of that transition to the total
number of trials of each species. To quantify the often-repeated failed attempts before the final
successful attempt, we also counted the number of times the animal continued to use the same
strategy consecutively for each trial, and we averaged this number across all the trials of each
species to obtain the probability of self-transitions.
Body rotation and deformation analysis
To quantify body rotation and deformation during self-righting for each attempt, we
digitized four markers on the animal's body (Fig. 1D, E) at the start and end of the attempt and
when the body was highest. The instance when the body CoM was highest was determined from
high-speed videos by observing when the body stopped pitching and/or rolling upward and began
pitching and/or falling downward.
The four markers included: a head marker at the tip of the head (H), an abdomen marker at
the tip of the abdomen (A), a left marker on the left side of the abdomen (L), and a right marker on
the right side of the abdomen (R). Both the left and right markers were located at about 60% body
length from the head, close to the fore-aft position of the center of mass (Kram et al., 1997). Each
marker was digitized in at least two high-speed videos from different views using DLTdv5
(Hedrick, 2008), which were used to reconstruct 3-D positions using DLTcal5 (Hedrick, 2008) and
a custom 27-point calibration object. The position midway (M) between the left and right markers
was calculated.
We approximated the CoM position using the average position of all four markers. Using
positions of the tips of the head (H) and abdomen (A), we calculated body pitch and body yaw
relative to the ground. Using positions of the left (L) and right (R) points on the sides of the
8
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
abdomen, we calculated body roll relative to the ground. In addition, we calculated body flexion
as the angle within the sagittal plane formed between the in-plane components (𝑟 H and 𝑟 A) of two
vectors 𝑟 H and 𝑟 A, which started from the midway point (M) and pointed to the head (H) and
abdomen (A) markers, respectively. 𝑟 H┴ and 𝑟 A┴ are components of 𝑟 H and 𝑟 A perpendicular to the
sagittal plane. A negative body flexion meant body hyperextension. Further, we calculated head
and abdomen twisting, and , as the angles between the sagittal plane and the vectors 𝑟 H and
𝑟 A, respectively. Sagittal plane was approximated by a plane whose normal vector 𝑛⃑ was the vector
from the left (L) to the right (R) marker. See Fig. 1D, E for details. Equations are summarized
below.
CoM position:
xCoM 1/4(xHxAxLxR)
yCoM 1/4(yHyAyLyR)
zCoM 1/4(zHzAzLzR)
body orientation:
pitch tan1[(zA zH)/sqrt((xA xH)2 + (yA yH)2]
roll tan1[(zL zR)/sqrt((xL xR)2 + (yL yR)2]
yaw = tan1[(yA yH)/(xA xH)]
body flexion:
head twisting:
cos1[(𝑟 H • 𝑟 A)/(𝑟 H 𝑟 A)]
tan1(𝑟 H┴/𝑟 H)
abdomen twisting:
tan1(𝑟 A┴/𝑟 A)
where:
𝑛⃑ (xL, yL, zL) (xR, yR, zR)
𝑟 H┴𝑟 H • 𝑛⃑
𝑟 H𝑟 H𝑟 H • 𝑛⃑
𝑟 A┴𝑟 A • 𝑛⃑
𝑟 A𝑟 A𝑟 A • 𝑛⃑
To study how the animal moved in an attempt to self-right, we calculated the changes in
each of these variables from the start of each attempt to when the body CoM was highest. When
doing this, we used absolute values of pitch, roll, and yaw considering symmetry of the animal's
9
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
ellipsoid-shaped body. We also set head and abdomen twisting at the start of the attempt to be
always non-negative, considering lateral symmetry of the animal.
Body and appendage behavior analysis
To further identify what contributed to successful self-righting, for each attempt, we
recorded the following events (or lack thereof) to quantify the animal's body and appendage
behaviors:
(1) Dynamic: whether the animal's body rotation was dynamic, being able to gain and use
pitch and/or roll rotational kinetic energy to overcome potential energy barriers. Dynamic behavior
was determined by observing whether the animal's body was still moving upward when its
appendage used for self-righting (wings or legs) had stopped pushing against the ground. A wing
stopped pushing against the ground when its distal section lifted off the ground as the body pitched
and/or rolled. A leg stopped pushing against the ground when its distal segments, which engaged
the surface for self-righting, slipped, reducing vertical force production (Full et al., 1995). When
either of these occurred, the body could only continue to move upward if it still had rotational
kinetic energy.
(2) Body lift-off: whether the body lifted off from the surface.
(3) Body hold: whether the body was held in the air after pitching up so that the abdomen
remained raised, when using wings to self-right.
(4) Body sliding: whether the animal's body slid on the ground as it pitched/rolled toward
self-righting.
(5) Leg assist: whether legs assisted by pushing against the surface to generate body
pitching and/or rolling towards self-righting, when using wings to self-right.
(6) Leg slip: whether the leg engaging the surface to self-right (both as the primary and
assisting mechanisms) slipped on the surface.
(7) Accelerate: whether the assisting leg accelerated body pitching and/or rolling motion
towards self-righting.
(8) Overshoot: whether there was any overshooting in body pitching and/or rolling motion
beyond the upright orientation that must be corrected by legs.
We calculated the probabilities of these body and appendage behaviors as the ratios of the
occurrences of each to the total number of attempts for each strategy, separated by whether the
attempt was successful or not.
All data analyses were performed using Microsoft Excel and MATLAB.
10
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Statistics
Before pooling trials, for each species, we performed a mixed-design ANOVA (for
continuous variables) or a chi-square test (for nominal variables), both with trial number as a fixed
factor and individual as a random factor to account for individual variability. We found no effect
of trial for any measurements relevant to a trial (P > 0.05, ANOVA or P > 0.05, chi-square test),
including number of attempts to self-right, self-righting probabilities, righting times, and transition
probabilities. Thus, we pooled all trials from each individual to calculate their means and
confidence intervals (for nominal variables) or standard deviations (for continuous variables).
Before pooling attempts, for each species, we performed a mixed-design ANOVA (for
continuous variables) or a chi-square test (for nominal variables), both with attempt number as a
fixed factor and individual as a random factor to account for individual variability. We found no
effect of attempt for most (72 out of 84) measurements relevant to an attempt (P > 0.05, ANOVA
or P > 0.05, chi-square test), including attempt time, changes in body pitch, roll, yaw, CoM height,
body flexion, head and abdomen twisting, and body and leg behavior probabilities. Thus, we
pooled all attempts for each of the self-righting strategies, separated by whether the attempt was
successful or not, to calculate their means and confidence intervals (for nominal variables) or
standard deviations (for continuous variables).
To test whether measurements relevant to an attempt differed between successful and failed
attempts, for each species using each strategy, we used a mixed-design ANOVA (for continuous
variables) or a chi-square test (for nominal variables), with the successful/failure record as a fixed
factor and individual as a random factor to account for individual variability.
To test whether measurements relevant to the strategy used (winged or legged) differed
between winged and legged attempts, for each species separated by whether the attempt was
successful or not, we used a mixed-design ANOVA or a chi-square test (for nominal variables),
with the strategy used as a fixed factor and individual as a random factor to account for individual
variability.
To test whether measurements relevant to successful trials differed between species, we
used a mixed-design ANOVA (for continuous variables) or a chi-square test (for probabilities),
with species as a fixed factor and individual as a nested, random factor to account for individual
variability.
Wherever possible, we used Tukey's honestly significant difference test (HSD) to perform
post-hoc analysis. All the statistical tests followed (McDonald, 2009) and were performed using
JMP.
11
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Potential energy landscape model using simple body geometry
To visualize how the animal rotated during self-righting attempts and how this differed
between strategies and species, we used a simple geometric model to calculate the potential energy
landscape of the body (Fig. 2). Because the animal rarely lifted off the ground during self-righting
for all three species (7 out of 378 attempts, Fig. S2B), as a first-order approximation, we considered
the animal body as an ellipsoid with its lowest point in contact with a horizontal, flat surface (Fig.
2A-C). Ellipsoid length 2a, width 2b, and thickness 2c were body length, width, and thickness from
morphological measurements (Table 1). We approximated the CoM position with the ellipsoid's
geometric center (Kram et al., 1997).
The simple geometric model allowed us to visualize the state of an ellipsoidal body on a
potential energy landscape (Fig. 2D). For an elongate ellipsoid body, self-righting by pitching
overcomes the highest potential energy barrier (Fig. 2A), whereas self-righting by rolling
overcomes the lowest barrier (Fig. 2B). Self-righting by a diagonal body rotation (Frantsevich,
2004), with simultaneous pitching and rolling, overcomes an intermediate barrier (e.g., Fig. 2C, an
ideal diagonal rotation about a fixed axis in the horizontal plane between pitch and roll axes). Body
yawing did not affect CoM position or barrier height (because we used the yaw-pitch-roll
convention of Euler angles).
Fig. 2. Potential energy landscape from the simple geometric model. (A-C) An ellipsoid
approximating the animal body in contact with the ground, either pitching (A), rolling (B), or
rotating diagonally (simultaneous pitching and rolling) (C). Dashed line shows rotation axis.
12
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Diagonal rotation shown is about a fixed axis within the horizontal ground plane for simplicity;
actual diagonal rotation of the animal may be about a time-varying axis. Red, blue, and yellow
arrows on each ellipsoidal body show its three major axes to illustrate body rotation. Vector g
shows the direction of gravity. (D) Potential energy landscape, shown as CoM height as a function
of body pitch and body roll (using Euler angles with yaw-pitch-roll convention), calculated from
the geometric model. We use absolute values of body pitch and roll considering symmetry of the
ellipsoid. Downward and upward arrows indicate an upside-down and upright body orientation,
respectively. Cyan, green, yellow, and magenta curves with arrows are representative trajectories
for pure pitching, two different diagonal rotations, and pure rolling, all about a fixed axis in the
horizontal plane, to illustrate the fact that more body rolling decreases the potential energy barrier.
White curves on the landscape are iso-height contours. Small yellow arrows on the landscape are
gradients. Model results shown are using the discoid cockroach's body dimensions as an example.
RESULTS
Self-righting attempts
For all three species, self-righting on a flat, rigid, low friction surface was a challenging
task and often required more than one attempt to succeed (Table 1, Fig. 3A; 24%, 29%, and 48%
of all trials had multiple attempts for the Madagascar hissing, American, and discoid cockroaches,
respectively). Repeated attempts were consistent with previous observations in the discoid
cockroach (Full et al., 1995). The Madagascar hissing, American, and discoid cockroaches needed
an average of 1.3, 1.8, and 3.2 attempts to self-right. The difference was significant only between
the Madagascar and discoid cockroaches (P < 0.05, ANOVA, Tukey HSD).
For all three species, we found no dependence on trial number or only a few cases of
dependence on attempt number (see Statistics). The lack of dependence on trial number showed
that there was only a minor effect of history dependence on self-righting and that the animal's
motion and use of strategies was stochastic and unpredictable (Full et al., 1995) over consecutive
attempts.
Self-righting probability
All three species self-righted with high probability when given time (30 seconds in our
experiments; Fig. 3B, white bar; averaging 97% for all three species) and self-righted on the first
attempt in over half of all trials (Fig. 3B, gray; averaging 63% for all species) with no significant
difference across species (P > 0.05, chi-square test).
13
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Fig. 3. Self-righting performance. (A) Number of attempts to achieve self-righting. n = 55, 59,
61 for (B) Self-righting probability within 30 seconds (white) and on the first attempt (gray). Error
bars represent 95% confidence intervals. (C) Total time to achieve self-righting. (D) Time to
achieve self-righting on the first attempt. From left to right are Madagascar hissing, American, and
discoid cockroaches. In A, C, and D, data are shown using violin plots. Black and red lines show
the mean and median, respectfully. Width of graph shows the frequency of the data along the y-
axis. Brackets show whether there is a significant difference (asterisk, P < 0.05, ANOVA) or not
(n.s.). See Table 1 for sample size.
Self-righting time
All three species were capable of self-righting rapidly. The fastest self-righting took only
0.14 s for the American cockroach, 0.31 s for the discoid cockroach, and 0.46 s for the Madagascar
hissing cockroach. The median total time to achieve self-righting including failed attempts was 1.1
s, 0.6 s, and 1.6 s for the Madagascar hissing, American, and discoid cockroaches, respectively (Fig.
3C). The maximal time was 19.9 s, 3.9 s, and 17.7 s for the Madagascar hissing, American and
discoid cockroaches, respectively. The difference was only significant between the American and
discoid cockroaches (P < 0.05, ANOVA, Tukey HSD). The mean self-righting time on the first
attempt (Fig. 3D) was 1.0 s for the Madagascar hissing cockroach, longer than the American
14
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
cockroach's 0.6 s (P < 0.05, ANOVA, Tukey HSD), although neither differed from the discoid
cockroach's 0.9 s.
Fig. 4. Representative snapshots of self-righting strategies. (A) Madagascar hissing cockroach
using body arching. (B) American cockroach using wings. (C) Discoid cockroach using wings. (D)
American cockroach using legs. (E) Discoid cockroach using legs. i-v are five snapshots forward
in time.
Self-righting strategies
Body arching. The Madagascar hissing cockroach's self-righting relied primarily on
changing body shape assisted by legs (Fig. 4A, Movie 1). When lying upside down (Fig. 4A, i), the
animal hyperextended its body into an arch to raise the CoM (Fig. 4A, ii) (Camhi, 1977), similar
15
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
to some beetles (Frantsevich, 2004). The narrow static stability region between the head and tip of
the abdomen in contact with the ground and lateral perturbations from flailing legs induced the
body to roll (Fig. 4A, iii). As the body fell onto one side, rolling stopped due to resistance from the
legs and the metastable body shape in the transverse plane (Camhi, 1977), resembling that of
medium-height turtle shells (Domokos and Várkonyi, 2008). Then, the legs on the lowered side
kept pushing, resulting in skidding and yawing on the surface, while the body continued
hyperextending (Fig. 4A, iv). When a body arching attempt failed, the animal sometimes quickly
flexed its body straight (occurring at a 25% probability per attempt) to reverse the direction of body
rolling using rotational kinetic energy gained due to falling of the CoM to start another body arching
attempt. When one of the pushing legs eventually managed to wedge under the body, its thrust
rolled the body further over protruding legs to overcome their secondary potential energy barriers
to achieve self-righting (Fig. 4A, v).
Wing use. Both the American and discoid cockroaches can self-right primarily using wings
(Fig. 4B, C, Movie 2). When lying upside down (Fig. 4B, C, i), the animal separated its wings
laterally and pronated them so that their outer edges pushed against the surface while the head
remained in contact as a pivot, which pitched the abdomen upward (Fig. 4B, C, ii) and often resulted
in additional body rolling. When a winged attempt failed, the animal closed its wings to pitch back
downward and sometimes started the same process again in another attempt (occurring at a 3%
probability per attempt for the American cockroach and an average of 1.1 times per attempt for the
discoid cockroaches). When a winged attempt succeeded, the animal fell with additional body
pitching and/or rolling to become upright (Fig. 4B, C, iii, iv, v). Legs flailed in this process,
resulting in small lateral perturbations. Flailing legs frequently hit and pushed against the ground
(91% of attempts) providing impulses to change body rotation.
Leg use. The American and discoid cockroaches can also self-right primarily using legs
(Fig. 4D, E, Movie 3). When lying upside down, these insects always continuously kicked their
legs outward in an attempt to push against the ground (Reingold and Camhi, 1977; Zill, 1986).
Frequent legs slipping (55% of attempts) due to the low friction of the surface resulted in continuous
body sliding (41% of attempts). In failed attempts, body rolling and pitching induced by kicking
legs were too small to achieve self-righting, and the animal started the same process in another
attempt (occurring at a 51% and 21% probability per attempt for the American and discoid
cockroaches, respectively). When a legged attempt succeeded (Fig. 4D, E, i), two legs engaged the
surface simultaneously (Fig. 4D, E, ii), typically a hind leg and a contralateral middle leg (76% and
93% of attempts for the American and all attempts for the discoid cockroaches, respectively). The
16
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
two legs pushed to thrust the body forward, pitched it head up, and rolled it such that the abdomen
cleared the surface to self-right (Fig. 4D, E, iii, iv, v).
Probability of dynamic self-righting
For both the American and discoid cockroaches using both wings and legs, self-righting
attempts were often dynamic (Fig. 5; American: 67% of winged attempts, 55% of legged attempts,
56% of all attempts; discoid: 37% of winged attempts, 80% of legged attempts, 51% of all attempts),
being able to gain and use pitch and/or roll rotational kinetic energy in an attempt to overcome
potential energy barriers. By contrast, the Madagascar hissing cockroach's self-righting using body
arching was never dynamic (0%, Fig. 5).
Fig. 5. Probability of self-righting dynamically. A dynamic attempt is one in which the animal
is able to gain and use pitch and/or roll rotational kinetic energy in an attempt to overcome potential
energy barriers, whether the attempt is successful or not. See Table 1 for sample size.
Self-righting transitions
All three species attempted more than one strategy and often transitioned between them to
self-right, even though not all of them led to successful righting on the flat, rigid, low-friction
surface (Fig. 6). Both the Madagascar hissing and American cockroaches' self-righting was more
stereotypical and primarily used one successful strategy, in contrast to the discoid cockroach that
used two successful strategies nearly equally.
The Madagascar hissing cockroach (Fig. 6A) most frequently used body arching to self-
right (85% of all trials). When not successful, this cockroach always continued to use body arching,
17
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
leading to a high probability of self-righting (98%). It occasionally used body twisting (13%)
(Camhi, 1977) which never succeeded and after which it always transitioned to body arching (13%).
Fig. 6. Self-righting locomotor transition ethograms. (A) Madagascar hissing cockroach. (B)
American cockroach. (C) Discoid cockroach. Arrow widths are proportional to transition
probabilities between nodes, with probability values shown by numbers. Transition probabilities
are defined as the ratio of the number of occurrences of each transition to the total number of trials
for each species. Red arrows and numbers are for self-transition and represent the average number
of times of continuing the same strategy during each trial. A self-transition probability greater than
one means that on average it occurred more than once for each trial. The sum of transition
probabilities out of each node equals that into the node, except for start with a total probability of
18
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
1 going out, and righting and failure with a total probability of 1 into both together. See Table 1 for
sample size.
The American cockroach (Fig. 6B) most frequently used legs (93%) and occasionally used wings
(2%), despite being capable of self-righting using both strategies. When not successful, it often
continued to use the same legged or winged strategy, but also occasionally transitioned between
them. It also infrequently used flapping (2%) which never succeeded.
By contrast, the discoid cockroach (Fig. 6C) initially used either wings (49%) or legs (34%)
to self-right. When unsuccessful, it continued to use the same legged or winged strategy, but also
frequently transitioned between them, resulting in high probabilities of self-righting (46% or 49%
eventually using wings or legs to self-right, respectively).
All three species occasionally entered a temporary quiescence mode (Camhi, 1977) without
apparent body or appendage movement (2%, 3%, and 16% for Madagascar hissing, American, and
discoid cockroaches, respectively). The Madagascar cockroach occasionally used body twisting
(13%) and the American cockroach showed wing flapping (2%) in an attempt to self-right, but
these two strategies never succeeded.
Body state on potential energy landscape
For all three species, because the body rarely lifted off the ground for all three species (7
out of 378 attempts, Fig. S2B), the measured state of the animal (body pitch, body roll, and CoM
height) lied on the surface of the potential energy landscape using the simple geometric model (Fig.
7). Being on the surface of the energy landscape allowed us to examine how the animal's body
moved through three stages (start, highest CoM height, and end) of an attempt to overcome
potential energy barriers (or lack thereof).
For the Madagascar hissing cockroach using body arching and the American cockroach
using legs, body rotation was mainly rolling during both successful and failed attempts (Fig. 7A, i,
ii; Fig. 7C, i, ii), which overcame the lowest potential energy barrier if successful (Fig. 7A, i). For
the American cockroach using wings, body rotation was mainly pitching during both successful
and failed attempts (Fig. 7B, i, ii), which overcame the highest potential energy barrier if successful
(Fig. 7B, i). For the discoid cockroach using both wings and legs, body rotation involved
simultaneous pitching and rolling during successful attempts (Fig. 7D, i; Fig. 7E, i), which
overcame intermediate potential energy barriers, and body rotation was mainly pitching during
failed attempts (Fig. 7D, ii; Fig. 7E, ii). In failed attempts, the animal was unable to overcome the
potential energy barriers (Fig. 7A-E, ii).
19
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Fig. 7. State of the body on the potential energy landscape at the start (1), highest CoM
position (2), and end (3) of the attempt during successful (i) vs. failed (ii) attempts. (A)
Madagascar hissing cockroach using body arching. (B) American cockroach using wings. (C)
20
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
American cockroach using legs. (D) Discoid cockroach using wings. (E) Discoid cockroach using
legs. Landscape is defined in Fig. 2D. On each landscape, the ellipsoids show means (center of
ellipsoid) ± 1 s.d. (principal semi-axis lengths of ellipsoid) of body pitch, body roll, and CoM height
at each stage of the attempt. For failed attempts (ii), the end state (3) is not shown because it nearly
overlaps with the start state (1). Sample size of each case is shown. Note that the sample size here
combined for each species is slightly smaller than its total number of attempts, because in some
attempts the animal markers are out of the field of view and cannot be digitized.
In addition, both the Madagascar hissing and American cockroaches had a large number of
successful attempts (50 and 43, respectively) using strategies (body arching and legged,
respectively) that overcame low potential energy barriers (Fig. 7A, C, i). The American cockroach
had only one successful attempt using wings which overcame high potential energy barriers (Fig.
7B, i). By contrast, the discoid cockroach had similar numbers of successful attempts to overcome
potential energy barriers using the two strategies, winged (28%) and legged (23%) self-righting
(Fig. 7D, E, i).
Body rotation and center of mass height increase
Madagascar hissing cockroach. Using body arching to self-right, the Madagascar hissing
cockroach pitched little towards 90º (Fig. 8A, i) but rolled substantially towards 90º (Fig. 8B, i) as
the body attained its highest CoM position. Rolling resulted in a small CoM height increase relative
to the highest potential barrier height possible (a c) (Fig. 8C, i). The body rolled more in
successful than in failed attempts (roll = 69º vs. 50º; P < 0.05, ANOVA).
American cockroach. Using wings to self-right, the American cockroach pitched
substantially towards 90º (Fig. 8A, ii) and rolled little towards 90º (Fig. 8B, ii) as the body attained
its highest CoM position. This resulted in a large CoM height increase relative to the highest
potential barrier height possible (a c) (Fig. 8C, ii).
Using legs to self-right, the American cockroach pitched little towards 90º (Fig. 8A, iii)
and rolled substantially towards 90º (Fig. 8B, iii) as the body attained its highest CoM position.
This resulted in a small CoM height increase relative to the highest potential barrier height possible
(a c) (Fig. 8C, iii). The body rolled more in successful than in failed attempts (roll = 61º vs.
20º; P < 0.05, ANOVA).
For successful attempts, the American cockroach pitched more (pitch = 78º vs. 5º; P <
0.05, ANOVA), rolled less (roll = 6º vs. 61º; P < 0.05, ANOVA), and its CoM height increased
21
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
more (zCoM = 1.8 cm vs. 0.7 cm; P < 0.05, ANOVA) when using wings than when using legs (Fig.
8A-C, ii vs. iii).
Fig. 8. Body pitch increase (A), body roll increase (B), and CoM height increase (C) when the
body was highest for successful (S, red) vs. failed (F, blue) self-righting attempts. (i)
Madagascar hissing cockroach using body arching. (ii) American cockroach using wings. (iii)
American cockroach using legs. (iv) Discoid cockroach using wings. (v) Discoid cockroach using
legs. We used absolute values of body pitch and roll considering symmetry of the ellipsoid
representing the body. Data are shown using violin plots. Black and red lines show the mean and
median. Width of graph shows the frequency of the data along the y-axis. Black asterisks and
brackets indicate a significant difference between successful and failed attempts (P < 0.05,
ANOVA). Red asterisks and brackets indicate a significant difference between winged and legged
attempts for the same species (P < 0.05, ANOVA). In (C), two horizontal dashed lines show the
lowest and highest barriers from the ellipsoid model, b c and a c, for pure rolling and pure
pitching, respectively. For successful attempts, CoM height increase is the measured barrier height.
For failed attempts, barrier height is not measured because the animal did not overcome it. See
Table 1 for sample size.
22
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Discoid cockroach. Using wings to self-right, the discoid cockroach pitched substantially
towards 90º (Fig. 8A, iv) and rolled less than it pitched (Fig. 8B, iv) towards 90º as the body attained
its highest CoM position. This resulted in a large CoM height increase relative to the highest
potential barrier height possible (a c) (Fig. 8C, iv). The body rolled more in successful than in
failed attempts (roll = 17º vs. 2º; P < 0.05, ANOVA).
Using legs to self-right, the discoid cockroach both pitched (Fig. 8A, v) and rolled (Fig.
8B, v) a little towards 90º as the body attained its highest CoM position. This resulted in a small
CoM height increase relative to the highest potential barrier height possible (a c) (Fig. 8C, v).
The body rolled more in successful than in failed attempts (roll = 34º vs. 5º; P < 0.05, ANOVA).
For successful attempts, the discoid cockroach pitched more (pitch = 51º vs. 21º, P <
0.05, ANOVA), rolled less (roll = 17º vs. 34º, P < 0.05, ANOVA), and its CoM height increased
more (zCoM = 1.3 cm vs. 0.8 cm, P < 0.05, ANOVA) when using wings than when using legs (Fig.
8A-C, iv vs. v).
All three species. For all three species, body rolling was the best predictor of whether an
attempt succeeded or failed. Roll increase when the CoM was highest was greater in successful
than in failed attempts for all the cases (P < 0.05, ANOVA; Fig. 8B, i, iii-v), except for the
American cockroach using wings (Fig. 8B, ii) which had a small sample size (1 successful and 4
failed attempts).
Because CoM height increase was the measured potential energy barrier height for
successful attempts, both the American and discoid cockroaches overcame higher barriers when
using wings than when using legs, and this difference was greater for the American cockroach.
Other factors contributing to successful self-righting
Besides body rolling, three factors were important in differentiating successful from failed
attempts (Fig. S2). First, except for the American cockroach using wings, leg slip was less frequent
in successful attempts for all three species (Fig. S2A, i, iii-v; Madagascar arching: successful: 0%,
failed: 100%; American legged: successful: 0%, failed: 90%; discoid winged: successful: 45%,
failed: 97%; discoid legged: successful: 0%, failed: 100%; P < 0.05, chi-square test). Second, for
the American cockroach using legs and the discoid cockroach using both wings and legs, legs more
frequently hit the ground in successful attempts to accelerate body rotation, after wings or legs
generated the initial body pitching and/or rolling (Fig. S2B, iii-v; American legged: successful:
51%, failed: 10%; discoid winged: successful: 62%, failed: 5%; discoid legged: successful: 34%,
failed: 0%; P < 0.05, chi-square test). Third, for both the American and discoid cockroaches using
wings, the body was held in the air with the abdomen pitched upward less frequently in successful
23
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
attempts (Fig. S2C, ii, iv; American winged: successful: 0%, failed: 80%; discoid winged:
successful: 52%, failed: 98%; P < 0.05, chi-square test). Body holding was not observed in the
legged and arching strategies.
We did not observe significant differences between successful and failed attempts that were
consistent across species and strategies for all other measurements (Figs. S1, S3), including attempt
time, body yaw change, body flexion change, head and abdomen twisting changes, dynamic
probability, body lift-off probability, body sliding probability, leg assist probability, and overshoot
probability. We did find significant differences between successful and failed attempts in attempt
time for the discoid cockroach using both wings and legs (Fig. S1A, iv, v), in body yaw change for
the American cockroach using wings (Fig. S1B, ii), in both head and abdomen twisting change for
the Madagascar hissing cockroach using body arching (Fig. S1D, E, i), in the probability of
dynamic self-righting for the discoid cockroach using legs (Fig. S3A, v), and in body sliding
probability for the American cockroach using wings (Fig. S3C, ii).
DISCUSSION
Our study quantified self-righting attempts (Fig. 3A), performance (Fig. 3B-D), probability
of using kinetic energy (Fig. 5), use of and transitions among strategies (Figs. 4, 6), body rotation
(Figs. 7, 8, S1B) and deformation (Fig. S1C-E), and body and appendage behaviors (Figs. S2, S3)
in the context of a potential energy landscape (Figs. 2, 7).
Advantages of dynamic self-righting using rotational kinetic energy
As we hypothesized, self-righting strategies in insects like cockroaches can be dynamic.
The ability to self-right dynamically (Fig. 5) by gaining and using pitch and/or roll rotational kinetic
energy to overcome potential energy barriers offered the American and discoid cockroaches several
performance advantages. First, with all else being equal and confirmed using a physical model (Li
et al., 2017), the larger its pitch and/or roll rotational kinetic energy, the faster the body pitched
and/or rolled, and the shorter the time to self-right. In addition, although each dynamic attempt
costs more energy, as our physical modeling demonstrated (Li et al., 2017), greater body rotational
kinetic energy increased the chance of self-righting for each attempt and could save energy overall
by reducing the number of failed attempts. Further, pitch and/or roll rotational kinetic energy
allowed the animal to reach a broad range of body rotation states of higher potential energy on the
landscape (Fig. 7B-E). This gives them the opportunity to overcome energy barriers using a greater
number of self-righting strategies. Finally, on slippery surfaces or sand where self-righting using
24
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
quasi-static leg grasping may be difficult (see Fig. 16B of (Frantsevich, 2004)), pushing appendages
rapidly to gain body rotational kinetic energy to self-right can be more effective.
Successful attempts revealed three body and appendage behaviors favoring dynamic self-
righting performance (Fig. S2). First, the animal's legs slipped less frequently in successful
attempts (Fig. S2A). This was beneficial because leg slipping leads to body yawing, sliding, and
premature falling of the CoM, which either dissipates pitch and/or roll rotational kinetic energy or
converts it into yaw rotational kinetic energy or horizontal translational kinetic energy that does
not contribute to self-righting. Second, the animal's assisting leg(s) more frequently accelerated
body rolling and/or pitching in successful attempts (Fig. S2B), adding pitch and/or roll rotational
kinetic energy. Third, when using wings to self-right, the animal's body was held during pitching
less frequently in successful attempts (Fig. S2C), and therefore did not lose the pitch and/or roll
rotational kinetic energy generated by prior wing pushing.
Body rolling facilitates self-righting by lowering potential energy barrier
As we hypothesized, for all but one strategy (Fig. 8B), cockroaches rolled their body more
during successful than failed attempts as the center of mass rose, because increased rolling lowers
the potential energy barrier (Figs. 2D, 7). This is important because ground-based self-righting is a
strenuous task. For example, a single hind leg of the discoid cockroach may need to generate ground
reaction forces during self-righting as large as eight times that during high speed running (at 8 body
length/s) (Full et al., 1995). Using the potential energy landscape model (Fig. 2), if the discoid
cockroach self-righted using wings with pure pitching, the mechanical work needed to overcome
the highest potential energy barrier (420 J) would be seven times that needed per stride during
medium speed running (at 5 body length/s) (Kram et al., 1997). Using the observed body rotation
during winged self-righting (Fig. 7D, Fig. 8, iv), this mechanical work is reduced by 40% (to 260
J). Consistent with this finding, winged self-righting of a cockroach-inspired physical model/robot
(Li et al., 2016; Li et al., 2017) demonstrated that body rolling increased the chances of successful
self-righting by lowering the potential energy barrier.
Both the American and discoid cockroaches are capable of self-righting using both wings
and legs. For both species, using legs with greater body rolling and less pitching is more favorable
because it overcomes a lower potential energy barrier than using wings with greater body pitching
and less rolling (Fig. 8A-C, ii vs. iii, iv vs. v, red). Given this, the American cockroach's successful
self-righting is more stereotyped than the discoid's (Figs. 6, B vs. C; Fig. 7, B, C vs. D, E) partly
because its potential energy barrier difference between the strategies is larger. For the American
cockroach, the potential energy barrier height difference is 1.7 cm for pitching vs. 0.7 cm for rolling
25
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
(Fig. 8C, ii vs. iii, red). By contrast for the discoid cockroach, the energy barrier difference was
only 1.3 cm for pitching vs. 0.8 cm for rolling (Fig. 8C, iv vs. v, red).
Advantages of diverse self-righting strategies
The ability of cockroaches and other insects (Frantsevich, 2004) to use and transition
among more than one strategy to self-right offers several possible performance advantages. First,
if damaged or lost appendages (Fleming et al., 2007; Jayaram et al., 2011) preclude the use of one
strategy, the animal still has an opportunity to self-right using an alternative strategy. Second, the
observed unsuccessful strategies such as body twisting and wing flapping (Fig. 4), as well as body
yawing and deformation and various body and appendage behaviors (Fig. S1, S3), which seemed
not beneficial here, may allow the animal to self-right in novel ways in natural environments by
interacting with slopes, uneven and deformable surfaces, or nearby objects (Golubovic et al., 2013;
Peng et al., 2015; Sasaki and Nonaka, 2016). Third, even the seemingly stochastic and
unpredictable motion over consecutive attempts may be an adaptation to heterogeneous, stochastic
natural environments (Kaspari and Weiser, 1999).
More broadly, the use of and transitions among diverse self-righting strategies may be an
adaptation for many animals. Studies of ground-based self-righting of beetles (Frantsevich, 2004)
and turtles (Ashe, 1970; Domokos and Várkonyi, 2008), and aquatic self-righting of marine
invertebrates on underwater substrates (Vosatka, 1970; Young et al., 2006), also observed diverse
strategies, including leg pivoting, head bobbing, tail pushing, body dorsiflexion, leg pushing, body
flexion, and tail bending.
Future work
Our quantification of motion on the potential energy landscape using a simple rigid body
only offers initial insights into the mechanical principles of self-righting of small insects. Future
work should expand the potential energy landscape by adding degrees of freedoms to better
understand how appendage motion and body deformation change energy barriers and stability to
result in self-righting (Othayoth et al., 2017). Our quantification of self-righting on a flat, rigid,
low-friction surface represents a very challenging scenario. Future experiments should test and
model how animals interact with slopes, uneven and deformable surfaces, or nearby objects
(Golubovic et al., 2013; Peng et al., 2015; Sasaki and Nonaka, 2016) using potential energy
landscapes to reveal principles of self-righting in nature. In addition, given our finding that rolling
facilitates self-righting by lowering the potential energy barrier, we speculate that searching to
grasp the ground or nearby objects (Frantsevich, 2004; Sasaki and Nonaka, 2016), leg flailing
26
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
(Othayoth et al., 2017), and body twisting, and during self-righting may induce lateral perturbations
to increase rolling. Further, experiments (Rubin et al., 2018) and multi-body dynamics simulations
(Xuan et al., 2019) to obtain three-dimensional ground reaction forces of the body and appendages
in contact with the substrate will help elucidate the dynamics of self-righting. Finally,
electromyography measurements will shed light on how animals control or coordinate (Xuan et al.,
2019) their wings, legs, and body deformation to self-right.
Acknowledgements
We thank Ratan Othayoth for technical assistance; Ratan Othayoth, Qihan Xuan, Sean Gart,
Kaushik Jayaram, Nate Hunt, Tom Libby, Chad Kessens, Peter Várkonyi, and two anonymous
reviewers for helpful discussion and suggestions, and Will Roderick, Mel Roderick, Armita
Manafzadeh, Jeehyun Kim, and Kristine Cueva for help with preliminary data collection and animal
care.
Competing interests
The authors declare no competing or financial interests.
Author contributions
Conceptualization: C.L., R.J.F.; Methodology: C.L., T.W., H.K.L., R.J.F.; Software: C.L.,
T.W.; Validation: C.L., R.J.F.; Formal analysis: C.L., T.W.; Investigation: T.W., H.K.L.;
Resources: C.L., R.J.F.; Data Curation: C.L., T.W.; Writing - original draft: C.L.; Writing - review
& editing: C.L., R.J.F.; Visualization: C.L.; Supervision: C.L., R.J.F.; Project administration: C.L.;
Funding acquisition: C.L., T.W., R.J.F.
Funding
This work is supported by a Miller Research Fellowship from the Miller Institute for Basic
Research in Science, University of California, Berkeley, a Burroughs Wellcome Fund Career
Award at the Scientific Interface, and an Army Research Office Young Investigator Award under
grant number W911NF-17-1-0346 to C.L., an FSU Jena Academic Exchange Program to T.W.,
and Army Research Laboratory Micro Autonomous Science and Technology Collaborative
Technology Alliances (MAST CTA) to R.J.F.
Supplementary information
Supplementary information is available.
27
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
References
Ashe, V. M. (1970). The righting reflex in turtles : A description and comparison. 20, 1 -- 3.
Bell, W. J., Roth, L. M. and Nalepa, C. A. (2007). Cockroaches: Ecology, Behavior, and
Natural History. Johns Hopkins University Press.
Blaesing, B. (2004). Stick insect locomotion in a complex environment: climbing over large
gaps. J. Exp. Biol. 207, 1273 -- 1286.
Brackenbury, J. (1990). A novel method of self-righting in the springtail Sminthurus viridis
(Insecta: Collembola). J. Zool. 222, 117 -- 119.
Camhi, J. M. (1977). Behavioral switching in cockroaches: transformations of tactile reflexes
during righting behavior. J. Comp. Physiol. A 113, 283 -- 301.
Chiari, Y., Van Der Meijden, A., Caccone, A., Claude, J. and Gilles, B. (2017). Self-righting
potential and the evolution of shell shape in Galápagos tortoises. Sci. Rep. 7, 1 -- 8.
Clark, A. J. and Higham, T. E. (2011). Slipping, sliding and stability: locomotor strategies for
overcoming low-friction surfaces. J. Exp. Biol. 214, 1369 -- 78.
Daley, M. A. and Biewener, A. A. (2006). Running over rough terrain reveals limb control for
intrinsic stability. Proc. Natl. Acad. Sci. 103, 15681 -- 15686.
Delcomyn, F. (1987). Motor activity during searching and walking movements of cockroach
legs. J. Exp. Biol. 133, 111 -- 120.
Domokos, G. and Várkonyi, P. L. (2008). Geometry and self-righting of turtles. Proc. R. Soc.
London B Biol. Sci. 275, 11 -- 7.
Evans, M. E. G. (1973). The jump of the click beetle (Coleoptera : E1ateridae)-energetics and
mechanics. J Zool 181 -- 194.
Faisal, A. A. and Matheson, T. (2001). Coordinated righting behaviour in locusts. J. Exp. Biol.
204, 637 -- 48.
Fleming, P. A., Muller, D. and Bateman, P. W. (2007). Leave it all behind: A taxonomic
perspective of autotomy in invertebrates. Biol. Rev. 82, 481 -- 510.
Frantsevich, L. (2004). Righting kinematics in beetles (Insecta: Coleoptera). Arthropod Struct.
Dev. 33, 221 -- 35.
Frantsevich, L. and Mokrushov, P. (1980). Turning and righting in Geotrupes (Coleoptera,
Scarabaeidae). J. Comp. Physiol. 289, 279 -- 289.
Full, R., Yamauchi, A. and Jindrich, D. (1995). Maximum single leg force production:
cockroaches righting on photoelastic gelatin. J. Exp. Biol. 198, 2441 -- 2452.
Golubovic, A., Bonnet, X., Djordjević, S., Djurakic, M., Tomović, L., Golubović, A., Bonnet,
28
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
X., Djordjević, S., Djurakic, M., Tomović, L., et al. (2013). Variations in righting
behaviour across Hermann's tortoise populations. J. Zool. 291, 69 -- 75.
Graham, D. (1979). Effects of circum-oesophageal lesion on the behaviour of the stick insect
Carausius morosus. Biol. Cybern. 145, 139 -- 145.
Hedrick, T. L. (2008). Software techniques for two- and three-dimensional kinematic
measurements of biological and biomimetic systems. Bioinspir. Biomim. 3, 034001.
Jayaram, K., Merritt, C., Cherian, A. and Full, R. (2011). Running without feet: the role of
tarsi during high-speed horizontal locomotion in cockroaches. Integr. Comp. Biol. 51, E64.
Jusufi, A., Goldman, D. I., Revzen, S. and Full, R. J. (2008). Active tails enhance arboreal
acrobatics in geckos. Proc. Natl. Acad. Sci. 105, 4215 -- 9.
Kaspari, M. and Weiser, M. (1999). The size-grain hypothesis and interspecific scaling in ants.
Funct. Ecol. 13, 530 -- 538.
Koditschek, D. E., Full, R. J. and Buehler, M. (2004). Mechanical aspects of legged
locomotion control. Arthropod Struct. Dev. 33, 251 -- 72.
Kram, R., Wong, B. and Full, R. J. (1997). Three-dimensional kinematics and limb kinetic
energy of running cockroaches. J. Exp. Biol. 200, 1919 -- 29.
Li, C., Hsieh, S. T. and Goldman, D. I. (2012). Multi-functional foot use during running in the
zebra-tailed lizard (Callisaurus draconoides). J. Exp. Biol. 215, 3293 -- 3308.
Li, C., Pullin, A. O., Haldane, D. W., Lam, H. K., Fearing, R. S. and Full, R. J. (2015).
Terradynamically streamlined shapes in animals and robots enhance traversability through
densely cluttered terrain. Bioinspir. Biomim. 10, 046003.
Li, C., Kessens, C. C., Young, A., Fearing, R. S. and Full, R. J. (2016). Cockroach-inspired
winged robot reveals principles of ground-based dynamic self-righting. In IEEE
International Conference on Intelligent Robots and Systems, pp. 2128 -- 2134.
Li, C., Kessens, C. C., Fearing, R. S. and Full, R. J. (2017). Mechanical principles of dynamic
terrestrial self-righting using wings. Adv. Robot. 31, 881 -- 900.
Libby, T., Moore, T. Y., Chang-Siu, E., Li, D., Cohen, D. J., Jusufi, A. and Full, R. J. (2012).
Tail-assisted pitch control in lizards, robots and dinosaurs. Nature 481, 181 -- 184.
Malashichev, Y. (2016). Asymmetry of righting reflexes in sea turtles and its behavioral
correlates. Physiol. Behav. 157, 1 -- 8.
Mann, G. K. H., O'Riain, M. J. and Hofmeyr, M. D. (2006). Shaping up to fight: Sexual
selection influences body shape and size in the fighting tortoise (Chersina angulata). J. Zool.
269, 373 -- 379.
McDonald, J. H. (2009). Handbook of Biological Statistics. Baltimore: Sparky House
29
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Publishing.
Minetti, A. E., Moia, C., Roi, G. S., Susta, D. and Ferretti, G. (2002). Energy cost of walking
and running at extreme uphill and downhill slopes. J. Appl. Physiol. 93, 1039 -- 46.
Othayoth, R., Xuan, Q. and Li, C. (2017). Leg vibrations help cockroaches self-right using
wings. In Integrative and Comparative Biology, p. 57.
Peng, S., Ding, X., Yang, F. and Xu, K. (2015). Motion planning and implementation for the
self-recovery of an overturned multi-legged robot. Robotica 5, 1 -- 14.
Reingold, S. C. and Camhi, J. M. (1977). A quantitative analysis of rhythmic leg movements
during three different behaviors in the cockroach, Periplaneta americana. J. Insect Physiol.
23, 1407 -- 1420.
Rubin, A. M., Blob, R. W. and Mayerl, C. J. (2018). Biomechanical factors influencing
successful self-righting in the pleurodire turtle, Emydura subglobosa. J. Exp. Biol.
jeb.182642.
Sasaki, M. and Nonaka, T. (2016). The Reciprocity of Environment and Action in Self-Righting
Beetles: The Textures of the Ground and an Object, and the Claws. Ecol. Psychol. 28, 78 --
107.
Sherman, E., Novotny, M. and Camhi, J. (1977). A modified walking rhythm employed during
righting behavior in the cockroach Gromphadorhina portentosa. J. Comp. Physiol. 316,
303 -- 316.
Spagna, J. C., Goldman, D. I., Lin, P.-C., Koditschek, D. E. and Full, R. J. (2007).
Distributed mechanical feedback in arthropods and robots simplifies control of rapid
running on challenging terrain. Bioinspir. Biomim. 2, 9 -- 18.
Sponberg, S. and Full, R. J. (2008). Neuromechanical response of musculo-skeletal structures in
cockroaches during rapid running on rough terrain. J. Exp. Biol. 211, 433 -- 446.
Stancher, G., Clara, E., Regolin, L. and Vallortigara, G. (2006). Lateralized righting behavior
in the tortoise (Testudo hermanni). Behav. Brain Res. 173, 315 -- 9.
Ting, L. H., Blickhan, R. and Full, R. J. (1994). Dynamic and static stability in hexapedal
runners. J. Exp. Biol. 197, 251 -- 69.
Vosatka, E. D. (1970). Observations on the Swimming , Righting , and Burrowing Movements
of Young Horseshoe Crabs , Limulus Polyphemus. Ohio J. Sci. 70, 276 -- 283.
Walter, R. M. and Carrier, D. R. (2002). Scaling of rotational inertia in murine rodents and two
species of lizard. 2141, 2135 -- 2141.
Willemsen, R. E. and Hailey, A. (2003). Sexual dimorphism of body size and shell shape in
European tortoises. J. Zool. 260, 353 -- 365.
30
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Xuan, Q., Othayoth, R. and Li, C. (2019). In silico experiments reveal the importance of
randomness of motions in cockroach's winged self-righting. Integr. Comp. Biol. 59,.
Young, J. S., Peck, L. S. and Matheson, T. (2006). The effects of temperature on walking and
righting in temperate and Antarctic crustaceans. Polar Biol. 29, 978 -- 987.
Zill, S. N. (1986). A model of pattern generation of cockroach walking reconsidered. J.
Neurobiol. 17, 317 -- 28.
Zurek, D. B. and Gilbert, C. (2014). Static antennae act as locomotory guides that compensate
for visual motion blur in a diurnal , keen-eyed predator. Proc. R. Soc. B 281, 20133072.
31
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
32
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Supplementary Information
Fig. S1. Attempt time and body yaw and deformation. Attempt time (A), body yaw change (B),
body flexion change (C), head twisting change (D), and abdomen twisting change (E) when the
body was highest for successful (S, red) vs. failed (F, blue) self-righting attempts. (i) Madagascar
hissing cockroach using body arching. (ii) American cockroach using the wings. (iii) American
cockroach using the legs. (iv) Discoid cockroach using the wings. (v) Discoid cockroach using the
legs. We used absolute values of body yaw considering rotational symmetry on the level, flat
surface. Data are shown using violin plots. Black and red lines indicate the mean and median. Width
of graph indicates the frequency of the data along the y-axis. Black asterisks and braces indicate a
33
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
significant difference between successful and failed attempts (P < 0.05, ANOVA). In (C), negative
flexion changes in (i) and positive flexion changes in (ii-v) mean increase in hyperextension and
flexion, respectively. In (D,E), positive and negative changes in twisting mean increase and
reduction in twisting, respectively. See Table 1 for sample size.
34
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Fig. S2. Body and appendage behaviors that show a consistent difference between successful
(S, red) and failed (F, blue) attempts. (A) Leg slip probability. (B) Accelerate probability. (C)
Body hold probability. (i) Madagascar hissing cockroach using body arching. (ii) American
cockroach using the wings. (iii) American cockroach using the legs. (iv) Discoid cockroach using
the wings. (v) Discoid cockroach using the legs. Asterisks and braces indicate a significant
difference between successful and failed attempts (P < 0.05, chi-square test). The large differences
between successful and failed attempts in (A, ii-iv) are due to individual variation. See Table 1 for
sample size.
35
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Fig. S3. Body and appendage behaviors that do not show a consistent difference between
successful (S, red) and failed (F, blue) attempts. (A) Dynamic probability. (B) Body lift-off
probability. (C) Body sliding probability. (D) Leg assist probability. (E) Overshoot probability. (i)
Madagascar hissing cockroach using body arching. (ii) American cockroach using the wings. (iii)
American cockroach using the legs. (iv) Discoid cockroach using the wings. (v) Discoid cockroach
using the legs. Asterisk and braces indicate a significant difference between successful and failed
attempts (P < 0.05, chi-square test). See Table 1 for sample size.
36
Journal of Experimental Biology (2019), 222, jeb186080; https://li.me.jhu.edu
Movie 1: Madagascar hissing cockroach self-righting using body arching.
https://www.youtube.com/watch?v=DNJL3ATHbtQ
Movie 2: American and discoid cockroaches self-righting using wings.
https://www.youtube.com/watch?v=W3pIJdcv2nw
Movie 3: American and discoid cockroaches self-righting using legs.
https://www.youtube.com/watch?v=CoP1-n89DpI
37
|
1606.08507 | 2 | 1606 | 2016-09-29T10:07:58 | Structural Protein-based Flexible Whispering Gallery Mode Resonators | [
"physics.bio-ph",
"cond-mat.soft"
] | Nature provides a set of solutions for photonic structures that are finely tuned, organically diverse and optically efficient. Exquisite knowledge of structure-property relationships in proteins aids in the design of materials with desired properties for building devices with novel functionalities, which are difficult to achieve or previously unattainable. Recent bio-inspired photonic platforms made from proteinaceous materials lay the groundwork for many functional device applications, such as electroluminescence in peptide nucleic acids, multiphoton absorption in amyloid fibers and silk waveguides and inverse opals. Here we report whispering-gallery-mode (WGM) microresonators fabricated entirely from semi-crystalline structural proteins (i.e., squid ring teeth, SRT, from Loligo vulgaris and its recombinant) with unconventional thermo-optic response. We demonstrated waveguides, add-drop filters and flexible resonators as first examples of energy-efficient, highly flexible, biocompatible and biodegradable protein-based photonic devices. Optical switching efficiency in these devices is over thousand times greater than the values reported for Silica WGM resonators. This work opens the way for designing energy efficient functional photonic devices using structure- property relationships of proteins. | physics.bio-ph | physics | STRUCTURAL PROTEIN-BASED FLEXIBLE WHISPERING GALLERY MODE
RESONATORS
Huzeyfe Yılmaz1,*, Abdon Pena-Francesch2,*, Robert Shreiner2, Huihun Jung2, Şahin Kaya
Özdemir1, Lan Yang1, Melik C. Demirel2,3
1. Electrical and Systems Engineering Department, Washington University, St. Louis, MO 63130
2. Department of Engineering Science and Mechanics, Pennsylvania State University, University
Park, PA, 16802
3. Materials Research Institute and Huck Institutes of Life Sciences, Pennsylvania State
University, University Park, PA, 16802
*These authors contributed equally
Corresponding Authors E-mails: LY [email protected], ŞKÖ [email protected], MCD
[email protected]
Abstract
Nature provides a set of solutions for photonic structures that are finely tuned, organically
diverse and optically efficient. Exquisite knowledge of structure-property relationships in
proteins aids in the design of materials with desired properties for building devices with novel
functionalities, which are difficult to achieve or previously unattainable. Recent bio-inspired
photonic platforms made from proteinaceous materials lay the groundwork for many functional
device applications, such as electroluminescence in peptide nucleic acids[1], multiphoton
absorption in amyloid fibers[2] and silk waveguides and inverse opals[3]. Here we report
whispering-gallery-mode (WGM) microresonators fabricated entirely from semi-crystalline
structural proteins (i.e., squid ring teeth, SRT, from Loligo vulgaris and its recombinant) with
unconventional thermo-optic response. We demonstrated waveguides, add-drop filters and
flexible resonators as first examples of energy-efficient, highly flexible, biocompatible and
biodegradable protein-based photonic devices. Optical switching efficiency in these devices is
over thousand times greater than the values reported for Silica WGM resonators. This work
opens the way for designing energy efficient functional photonic devices using structure-
property relationships of proteins.
The ability to manipulate and control light flow and light–matter interactions using whispering-
gallery-mode (WGM) resonators has created significant interest in various fields of science,
including but not limited to biosensing[4] and detection[5], cavity-QED[6], optomechanics[7], and
parity-time symmetric photonics[8]. WGM resonators are currently manufactured using
microelectronics technologies with conventional materials such as silica[9], silicon[10], and silicon
nitride[11]. However, recent developments in optical technologies have revealed the strong need
for developing soft, biocompatible, and biodegradable photonic devices and photonic structures
with novel functionalities that cannot be attained with current optical materials[12]. Towards this
aim, all-polymer[13] and polymer-coated silica WGM resonators[14], as well as silica WGM
resonators encapsulated in low-index polymers[15] have been fabricated using conventional and
commercially available polymers, such as polydimethylsiloxane (PDMS) and polystyrene (PS) to
address the need for flexible structures. Nature, on the other hand, has optimized proteins
through millions of years of evolution to provide diverse materials with complex structures. Thus
studying structural proteins at the molecular level will help us in our endeavor for efficiently
designing diverse materials, which are finely tunable, flexible, biodegradable, and have physical
properties and functionalities different or superior to those currently present in conventional
materials. This will also enable the production of functional devices with less chemical
processing and energy consumption, addressing growing environmental concerns.
In this Letter, we studied energy efficiency of whispering-gallery-mode microresonators
fabricated solely from semi-crystalline structural proteins. We fabricated microresonators using
native SRT (Fig 1a), and recombinant SRT (Fig. 1b) to demonstrate the possibility of building
flexible all-protein functional photonic devices. All-protein microtoroid WGM resonators were
fabricated using a molding and solution casting technique (Fig. S1). From initial silica templates
fabricated as previously described[16], we prepared polydimethylsiloxane (PDMS) negative
molds for solution casting of the proteins. The microtoroid structures were then manufactured by
filling the respective protein solutions into the corresponding PDMS molds. Light from a tunable
laser was evanescently coupled in and out of the fabricated resonators through a tapered optical
fiber to measure their resonance properties.
Structural proteins are characterized by long-range ordered molecular secondary structures (e.g.,
β-sheets, coiled coils, or triple helices) that arise due to highly repetitive primary amino acid
sequences within the proteins. These features promote the formation of structural hierarchy via
self-assembly. SRT is a semi-crystalline structural proteins, which are flexible, biodegradable,
and thermally and structurally stable materials. Semi-crystalline morphology of these proteins,
which emanates from their β-sheet secondary structures, is ideal for tunability of their physical
properties. SRT protein complex is composed of several proteins varying in molecular weight
(e.g., 10-55 kDa) [17]. Although optical properties do not vary with polydispersity in molecular
weight due to linear dielectric response or statistically driven size, they may change in special
cases due to cooperative non-linear response
or thermal effects. Therefore, a recombinant SRT
(Rec) protein with unique molecular weight (i.e., 18 kDa) is also selected to create a
monodisperse material[18]. A combination of RNA-sequencing, protein mass spectroscopy, and
bioinformatics tools (i.e., transcriptome assembly) was performed to produce 18 kDa
recombinant SRT protein[18] (Fig. 1c). Rec, and SRT proteins were analyzed using X-ray
diffraction (XRD), which showed that SRT protein had lower crystallinity compared to Rec
protein at room temperature (Fig. S2a-c). Rec, and SRT proteins were also analyzed using
infrared spectroscopy (FTIR). Amide regions of FTIR data also confirm that the crystalline
regions are composed of β-sheets and α-turns (Fig. S2e-g). Figures 1d present an example
protein-based WGM resonator and related transmission spectra of in the 977 nm band. The
quality factors of these proteinaceous WGM resonators ranged from 105 to 106 (Fig. S3).
a"
e""
b"
c"
d"
SRT"
f"
60
50
40
30
20
15
10
3.5
Rec"
70
50
40
35
25
15
10
Figure 1. All-protein whispering-gallery-mode (WGM) microresonators. (a)-(b) Optical images of SRT
rings and recombinant SRT (rec) powder proteins. (c) Scanning electron microscope (SEM) image of a
WGM resonator fabricated from SRT protein (Scale bar 20 µm). Insets present typical transmission
spectra of WGM resonances in the 977 nm. (d) SDS-Page shows protein molecular weight varying
between 15-55 kDA for native SRT, and 18 kDa for recombinant SRT. (e) Green fluorescence emission
from an all-SRT chip, which is doped with 1 mM of thioflavin-T (ThT) dye confirming the large red-shift
due its binding to β-sheets in the protein. Absorbance and emission spectra of ThT dye. The absorption
peak of ThT in solution is at 412 nm. When it is free in a solution, it emits light at 438 nm, and when it is
bound to β-sheets in proteins it emits at 482 nm. (f) The temperature dependence of the refractive indices
of the respective protein-based WGM resonators made of SRT, recombinant SRT (Rec), and methanol
treated SRT (m-SRT).
We probed the nano-crystalline regions in these structural proteins, delineated by anti-parallel β-
sheets, using fluorescent microscopy. We selected a fluorescent dye, thioflavin-T (ThT), which
binds specifically to the β-sheets, but does not bind to amorphous polypeptides[19]. When ThT
dye binds to β-sheets, it undergoes a characteristic red shift of its excitation/emission spectrum
(Fig. 1e) that may be selectively excited with blue light at 450 nm, resulting in a green
fluorescence signal with maximum at 482 and extending to 570 nm. We prepared an all-SRT
chip with microresonators, an SRT-coated silica microresonator, a silica resonator partially
coated with SRT, and a PDMS chip with microresonators (see Supplementary Information, Fig.
S4) to test selective binding of ThT dye. The green fluorescence emitted by the all-SRT chip
when exposed to blue light is indicative of a large red-shift in emission due to the binding of the
ThT dye to β-sheets (Fig. 1e).
We experimentally determined the thermo-optic coefficient of SRT proteins by monitoring the
resonant wavelength of a WGM resonator while the temperature of the all-protein chip substrate
was varied (Fig. 1f). We also chemically treated SRT resonator (i.e., sample kept in methanol for
4 hours) and measured the resonant wavelength shift as a function of temperature. For all
resonators, we observed that the resonant wavelengths, λ, experience a blue shift (i.e., decrease
in wavelength) as the temperature was increased (Fig. 1f), i.e., 𝑑𝜆(𝑛,𝑟)/𝑑𝑇<0. Subsequently,
using the measured parameters in the relation 1/𝜆 𝑑𝜆(𝑛,𝑟)/𝑑𝑇 =1/𝑛 𝑑𝑛/𝑑𝑇+1/𝑟 𝑑𝑟/𝑑𝑇,
where T is temperature, n is the refractive index, 1/𝑟 𝑑𝑟/𝑑𝑇 is the linear thermal expansion
normalized thermo-optic coefficients 1/𝑛 𝑑𝑛/𝑑𝑇 of SRT and Rec proteins at 1420 nm
coefficient, and r is the radius of the resonator. We measured the thermal expansion coefficient
for all protein samples (i.e., -95x10-6 ± 7x10-6 K-1 for SRT proteins), and hence we calculated the
wavelength.
By coupling proteinaceous WGM resonators to two separate fiber-taper waveguides, we
fabricated add-drop filters (Fig. 2a) that are frequently integrated in optical communication
architectures (such as optical filters, multiplexers and routers), as well as used as schemes for
high performance optical sensing devices. As seen in Fig. 2b, the add-drop filter routed the light
from the input waveguide to the drop port in the second waveguide when the wavelength of the
light coincided with the resonance wavelength of the SRT resonator. Non-resonant light passed
through the input waveguide to the transmission port. The dip in the transmission port and the
peak in the drop port seen in Fig. 2b clearly confirm that the SRT add-drop filter performed its
function with an add-drop efficiency of 51%, which may be further increased by more careful
setting of the waveguide-resonator coupling and by decreasing scattering losses. We also
produced proteinaceous waveguides (Fig. 2c) using electro-spinning (e-spin) technique (Fig. S5).
Engaging a single protein fiber to a silica fiber taper via van der Waals attractions, we achieved
evanescent coupling of light from the silica fiber taper to the protein fiber, clearly demonstrating
the waveguiding capability of proteinaceous fibers. The flexible nature of these protein-based
WGM resonators is shown in Fig. 2d, where the soft base of the WGM microresonators is bent
in a bowed shape. This characteristic enables significant bending on a macroscopic scale (see
inset). We performed three-point bending dynamic mechanical analysis (DMA) for the
recombinant and native SRT films to quantitatively describe the flexibility of the fabricated
photonic microstructures (Fig. 2d-inset). The load-deflection curve shows similar bending
modulus (~1 GPa) for all films. Figure 2e shows bending of the protein substrate as a function of
the wavelength of the recombinant SRT resonator. The resonance frequency did not change as
the strain in the substrate increased (Fig. 2e-inset), which is an important attribute of flexible
protein resonators for opto-mechanics applications.
a"
Drop"
port"
d"
Input"
port"
b"
c"
Input:"silica"
fiber"
"
T
R
S
"
r
e
b
fi
e"
Figure 2. Device examples of semi-crystalline proteins. (a-b) An all-SRT microtoroid resonator with two
coupling tapered-fiber waveguides forming an add-drop filter. Transmission spectrum at 667 nm band
shows a resonant dip in the transmission port and a resonance peak in the drop port (scale bars 20 µm).
(c) Light transmission through a fiber waveguide prepared from recombinant SRT by electrospinning.
Coupling of laser light into the protein fiber was done using a silica tapered fiber (scale bar 50 µm). (d)
Flexibility of the substrate (scale bar 1 cm) and of the WGM resonators fabricated from SRT (scale bar 1
mm). Inset presents the results of dynamic mechanical analysis (DMA) measurements used to quantify the
bending of native and recombinant SRT films. (e) Wavelength measurements of flexible substrate as a
function of strain (inset) show stable resonance in the protein resonator.
Finally, in order to utilize the strong thermo-optic response of structural proteins, we constructed
an all-optical on/off switch using protein-based WGM structure (Fig. 3a) where a pump laser
(1450nm band) controlled the transmission of a probe light (980 nm band) via the thermal
response of the protein. First, two resonance modes were identified (i.e., 1451.7 nm and 974.8
nm). We fixed the wavelength of the probe laser to the probe resonance wavelength where the
normalized transmission of the probe along the fiber was set to zero. Then we fixed the
wavelength of the pump laser slightly detuned from the pump resonance where the normalized
transmission of the pump through the fiber taper was set to unity. Using a square signal, we then
tuned the pump laser wavelength to the pump mode wavelength where the normalized
transmission of the pump became zero, which in turn heated the cavity and shifted the probe
resonance wavelength, effectively moving the probe laser out of resonance with the protein
microresonator and the normalized probe transmission became unity, which completed the all-
optical switching with an SRT protein microtoroid (Fig. 3b). Repeating this experiment for
various input powers, we found that the minimum power to perform switching with 12 dB
isolation is only 1.6 µW where the circulating power was 143 µW. With silica microtoroid
resonators, we obtained 8 dB isolation at 22.75 µW of input power and 223 mW of circulating
power. Comparing circulating powers, the protein resonator performs the switching with three
orders of magnitude less pump power (Fig. 3c). This striking performance is due to the strong
thermo-optic coefficients of recombinant SRT protein. Our results show that protein-based
photonic devices can be used for low power consumption applications. Moreover, we showed
that functional and efficient protein-based photonic devices could be fabricated by tuning their
crystallinity at the molecular level, which in turn tunes their thermo-optic response.
a"
b"
Switch"Off"
Switch"On"
c"
Figure 3. All-optical switch based on thermal shift of WGMs in an all-protein microtoroid resonator. (a)
Signal in 980 nm band is initially coupled to the resonator where no light transmits and control light in
1450 nm band is just off resonance (left). When the control light is detuned to the microtoroid resonance,
cavity temperature increases and the WGM in the 980 nm band becomes off-resonance (right). (b)
Normalized transmission of the signal and the control light at 1.6 µW. (c) Ratio of normalized
transmission for various input powers. For a silica microtoroid 22.75 µW of control light is required to
perform switching.
In summary, proteinaceous WGM microcavities demonstrated here provide a platform to
manipulate photons within microscale volumes at high power efficiency. As illustrated here,
protein-based materials could provide the next generation of adaptive photonic structures whose
optical, mechanical and thermal properties can be engineered at the molecular level for energy
efficient applications in photonics. This approach will also expedite the design, fabrication and
synthesis of eco-friendly, recyclable, flexible materials and devices.
Methods
Native squid ring teeth (SRT): European common squid (Loligo vulgaris) were caught from the
coast of Tarragona (Spain). The squid ring teeth (SRT) were removed from the tentacles,
immediately soaked in deionized water and ethanol mixture (70:30 ratio v/v) and vacuum dried
in a desiccator.
Expression and characterization of recombinant squid ring teeth (SRT) proteins: The SRT
protein family is comprised of SRTs of different size that exhibit different physical properties.
Heterologous expression of the smallest (~18 kDa) SRT protein extracted from Loligo vulgaris
(LvSRT) was performed using the protocols described earlier[18]. Briefly, the full length
sequence was cloned into Novagen's pET14b vector system and transformed into E. coli strain
BL21(DE3). Recombinant SRT expression was achieved with a purity of ~90% and an estimated
yield of ~50 mg/L. The yield increases approximately ten fold (i.e., ~0.5 g/L) in auto induction
media. The size of the protein was confirmed via an SDS-page gel (Fig. 2d).
WGM resonator fabrication: Details on the Si master preparation and etching for micro-WGM
fabrication were reported earlier[20]. Briefly, the microtoroids were fabricated from a 2 µm thick
oxide layer on a silicon wafer. First, series of circular pads of 80 µm in diameter were created
through a combination of standard photolithography technique and buffered HF etching. These
circular pads serve as etch mask for isotropic etching of silicon in XeF2 gas chamber, leaving
under-cut silica disks supported by silicon pillar. The microdisks were then selectively reflowed
using a 30W carbon dioxide (CO2) laser to form a toroidal shape. A negative mold was made
from polydimethylsiloxane (PDMS, Dow Corning Sylgard® 184 silicone elastomer kit). PDMS
and curing agent were mixed in a 10:1 ratio, stirred and degased under vacuum for 30 minutes.
The mix was poured on the silicon master and it was cured at room temperature for 24 hours.
Due to mechanical processing requirements (e.g., viscosity in solvent depends on protein
molecular size and solubility), the shape and size of WGM resonators varied. SRT
microresonators had a major diameter of 54 µm and a minor diameter of 9 µm, which were used
in quality factors and dn/dT measurements. For ThT dye experiments, we used larger SRT
microresonators (i.e., 100 µm).
Protein casting: SRT solution (either recombinant or native protein) was prepared by dissolving
50 mg/mL of protein in hexafluoro-2-propanol (HFIP). The solution was sonicated for 1 hour
and vacuum-filtered in a 4-8 µm mesh size filter. 80 µL of SRT solution were poured into a
PDMS toroid mold in successive 20 µL additions 1 minute apart. After the last addition, the
HFIP was evaporated at room temperature in the fume hood for 5 minutes and the mold-SRT
system was immersed in butadiene (plasticizer) at 80°C (above Tg) for 30 minutes. The
thermoplastic SRT film was peeled off with tweezers and excess butanediol was removed by
rinsing with ethanol.
WGM / resonance: Taper-fibers fabricated by heating and pulling single mode fibers were used
to couple light from a tunable laser into and out of the microtoroid resonators. The coupling
distance between the taper and the resonators was controlled by a nanopositioning system. Four
tunable lasers with emission in the wavelength bands of 670 nm, 780 nm, 980 nm and 1450 nm
were used to characterize the resonators in different bands of the spectrum. Their wavelengths
were linearly scanned around resonances of the resonators. The real-time transmission spectra
were obtained by a photodetector connected to an oscilloscope. The scanning speeds and powers
of the lasers were optimized in order to eliminate thermally-induced (due to heat build-up in
high-Q resonators) line-width broadening and narrowing effects. Typical operating conditions
for scanning speed and laser power were 10-20 nm/s and 15 µW, respectively.
Mechanical characterization: Protein films were prepared by casting 50 mg/mL SRT/HFIP
solution on PDMS molds, resulting in 40x8x0.03 mm rectangular films (length, width, thickness).
The samples were analyzed in a dynamic mechanical analysis instrument (TA 800Q DMA) with
a three-point bending clamp of 20 mm length between supports. Strain ramp experiments were
performed at room temperature at a rate of 250 µm/min with a preload of 0.02 N. The flexural
modulus was calculated as described in ASTM D790 – 10.
Fluorescence imaging: Thioflavin-T (Sigma) aqueous solution is prepared at 0.125 µM
concentration. The absorbance and emission spectra are measured using a fluorescence
microscope (Zeiss LSM 5 PASCAL system coupled to a Zeiss Axiovert 200M microscope) at
458 nm wavelength.
SRT fibers: Fibers were obtained through an electrospinning process that utilized a conventional
horizontal setup from Dr. Seong Kim's laboratory at the Pennsylvania State University. The
schematic design consisted of three components (Fig. S6): a syringe for supplying the SRT
solution, a collecting platform for accumulating the array of fibers, and a voltage source for
generating the necessary electric field. A standard procedure began with the preparation of an
approximately 50 mg/mL solution of SRT in hexafluoro-2-propanol (HFIP). Once fully
dissolved, 0.5 mL of the solution was drawn into the 1 cm diameter plastic syringe, whose 22
gauge bevel tip needle was previously cut perpendicularly to its cylindrical axis, straightening
the eventual orientation of the fiber jet. Placing the syringe securely in the pumping device
(PHD 2000 Programmable, Harvard Apparatus, Holliston, MA) contained in the ventilation hood,
the collecting platform was then configured. Supported by a ring stand and clamps, the
collecting platform, constructed from cut pieces of aluminum held 0.8 cm apart by glass slides
and epoxy, was positioned such that its center was aligned with the syringe needle, 6.5 cm away
from the tip. With the setup complete, the syringe needle was attached to the 10 kV source (HV
Power Supply, Gamma High Voltage Research, Ormond Beach, FL), and the aluminum plates of
the collecting platform were grounded. After establishing the voltage difference, the pumping
device was activated with an infusion rate of 250 µL/min. Following the discharge of
approximately 0.25 mL of the solution, the pumping process was terminated and the voltage
source detached. To ensure that the HFIP evaporated completely, five minutes were allowed for
drying, and the collecting platform was then removed and inspected for fibers between its plates.
Acknowledgments
MCD, AP, HJ and RS were supported by the Office of Naval Research under grant No.
N000141310595, Army Research Office under grant No. W911NF-16-1-0019, and Materials
Research Institute of the Pennsylvania State University. LY, SKO, and HY were supported by
the Army Research Office under grant No. W911NF-12-1-0026 and the National Science
Foundation under grant No 1264997. We thank Dr. Seong Kim for providing electro-spinning set
up.
Author contributions
MCD and ŞKÖ conceived the idea and planned the research. MCD derived structure-property
equations. MCD, ŞKÖ, and LY supervised the research. HY fabricated the resonators and
performed the photonics measurements and data analysis. AP performed the mechanical and
spectroscopic measurements and data analysis. RS prepared the fibers and HJ worked on the
cloning, recombinant expression and purification of structural proteins. All authors contributed
to writing and revising the manuscript, and agreed on the final content of the manuscript.
Additional information
The authors declare no competing financial interests. Supplementary information accompanies
this paper. Correspondence and requests for materials should be addressed to MCD, ŞKÖ and LY.
O. Berger, L. Adler-Abramovich, M. Levy-Sakin, A. Grunwald, Y. Liebes-Peer, M.
T. Lu, H. Lee, T. Chen, S. Herchak, J. H. Kim, S. E. Fraser, R. C. Flagan, K. Vahala,
S. Spillane, T. Kippenberg, O. Painter, K. Vahala, Physical Review Letters 2003, 91,
B. Peng, Ş. K. Özdemir, F. Lei, F. Monifi, M. Gianfreda, G. L. Long, S. Fan, F. Nori, C.
D. Armani, T. Kippenberg, S. Spillane, K. Vahala, Nature 2003, 421, 925.
P. Hanczyc, M. Samoc, B. Norden, Nature Photonics 2013, 7, 969.
H. Guerboukha, G. Yan, O. Skorobogata, M. Skorobogatiy, Advanced Optical Materials
References
[1]
Bachar, L. Buzhansky, E. Mossou, V. T. Forsyth, T. Schwartz, Nature Nanotechnology 2015, 10,
353.
[2]
[3]
2014, 2, 1181; N. Huby, V. Vié, A. Renault, S. Beaufils, T. Lefevre, F. Paquet-Mercier, M.
Pézolet, B. Bêche, Applied Physics Letters 2013, 102, 123702; S. Kim, A. N. Mitropoulos, J. D.
Spitzberg, H. Tao, D. L. Kaplan, F. G. Omenetto, Nature Photonics 2012, 6, 818.
[4] M. A. Santiago-Cordoba, S. V. Boriskina, F. Vollmer, M. C. Demirel, Applied Physics
Letters 2011, 99, 073701; M. A. Santiago-Cordoba, M. Cetinkaya, S. V. Boriskina, F. Vollmer,
M. C. Demirel, Journal of Biophotonics 2012, 5, 629.
[5]
Proc. Natl. Acad. Sci. U.S.A. 2011, 108, 5976; F. Vollmer, S. Arnold, D. Keng, Proc. Natl. Acad.
Sci. U.S.A. 2008, 105, 20701.
[6]
043902.
[7] M. Aspelmeyer, T. J. Kippenberg, F. Marquardt, Reviews of Modern Physics 2014, 86,
1391.
[8]
M. Bender, L. Yang, Nature Physics 2014, 10, 394.
[9]
[10] M. Borselli, T. Johnson, O. Painter, Optics Express 2005, 13, 1515.
[11] P. E. Barclay, K. Srinivasan, O. Painter, B. Lev, H. Mabuchi, Applied Physics Letters
2006, 89, 131108.
[12]
[13]
2009, 17, 2573.
[14] B.-B. Li, Q.-Y. Wang, Y.-F. Xiao, X.-F. Jiang, Y. Li, L. Xiao, Q. Gong, Applied Physics
Letters 2010, 96, 251109.
[15] F. Monifi, S. K. Odemir, J. Friedlein, L. Yang, Photonics Technology Letters, IEEE 2013,
25, 1458.
[16] Ş. K. Özdemir, J. Zhu, X. Yang, B. Peng, H. Yilmaz, L. He, F. Monifi, S. H. Huang, G. L.
Long, L. Yang, Proceedings of the National Academy of Sciences 2014, 111, E3836.
[17] M. C. Demirel, M. Cetinkaya, A. Pena‐Francesch, H. Jung, Macromolecular Bioscience
2015, 15, 300.
[18] A. Pena-Francesch, S. Florez, H. Jung, A. Sebastian, I. Albert, W. Curtis, M. C. Demirel,
Advanced Functional Materials 2014, 24, 7393.
[19] D. Rogers, American Journal of Clinical Pathology 1965, 44, 59.
[20]
Photonics 2010, 4, 46.
J. Hu, L. Li, H. Lin, P. Zhang, W. Zhou, Z. Ma, Optical Materials Express 2013, 3, 1313.
J. R. Schwesyg, T. Beckmann, A. S. Zimmermann, K. Buse, D. Haertle, Optics Express
J. G. Zhu, S. K. Ozdemir, Y. F. Xiao, L. Li, L. N. He, D. R. Chen, L. Yang, Nature
|
1806.06768 | 1 | 1806 | 2018-06-18T15:24:36 | Epithelial Wound Healing Coordinates Distinct Actin Network Architectures to Conserve Mechanical Work and Balance Power | [
"physics.bio-ph"
] | How cells with diverse morphologies and cytoskeletal architectures modulate their mechanical behaviors to drive robust collective motion within tissues is poorly understood. During wound repair within epithelial monolayers in vitro, cells coordinate the assembly of branched and bundled actin networks to regulate the total mechanical work produced by collective cell motion. Using traction force microscopy, we show that the balance of actin network architectures optimizes the wound closure rate and the magnitude of the mechanical work. These values are constrained by the effective power exerted by the monolayer, which is conserved and independent of actin architectures. Using a cell-based physical model, we show that the rate at which mechanical work is done by the monolayer is limited by the transformation between actin network architectures and differential regulation of cell-substrate friction. These results and our proposed molecular mechanisms provide a robust quantitative model for how cells collectively coordinate their non-equilibrium behaviors to dynamically regulate tissue-scale mechanical output. | physics.bio-ph | physics | Epithelial Wound Healing Coordinates Distinct Actin
Network Architectures to Conserve Mechanical Work and
Balance Power
Visar Ajeti#,2,3, A. Pasha Tabatabai#,2,3, Andrew J. Fleszar+4, Michael F. Staddon+5, Daniel S.
Seara1,3, Cristian Suarez6, M. Sulaiman Yousafzai2,3, Dapeng Bi7, David R. Kovar6, Shiladitya
Banerjee5, Michael P. Murrell*1,2,3
1Department of Physics, Yale University, 217 Prospect Street, New Haven, Connecticut 06511,
USA
2Department of Biomedical Engineering, Yale University, 55 Prospect Street, New Haven,
Connecticut 06511, USA
3Systems Biology Institute, Yale University, 850 West Campus Drive, West Haven, Connecticut
06516, USA
4Department of Biomedical Engineering, University of Wisconsin-Madison, 1550 Engineering
Drive, Madison, WI, 53706, USA
5Department of Physics and Astronomy, Institute for the Physics of Living Systems, University
College London, Gower Street, London WC1E 6BT, UK
6Department of Molecular Genetics and Cell Biology, University of Chicago, 920 E. 58th St,
Chicago, IL, 60637, USA
7Department of Physics, Northeastern University, 111 Dana Research Center, Boston, MA 02115,
USA
*=corresponding author
#These authors contributed equally
+These authors contributed equally
1
Abstract
How cells with diverse morphologies and cytoskeletal architectures modulate their mechanical
behaviors to drive robust collective motion within tissues is poorly understood. During wound
repair within epithelial monolayers in vitro, cells coordinate the assembly of branched and bundled
actin networks to regulate the total mechanical work produced by collective cell motion. Using
traction force microscopy, we show that the balance of actin network architectures optimizes the
wound closure rate and the magnitude of the mechanical work. These values are constrained by
the effective power exerted by the monolayer, which is conserved and independent of actin
architectures. Using a cell-based physical model, we show that the rate at which mechanical work
is done by the monolayer is limited by the transformation between actin network architectures and
differential regulation of cell-substrate friction. These results and our proposed molecular
mechanisms provide a robust quantitative model for how cells collectively coordinate their non-
equilibrium behaviors to dynamically regulate tissue-scale mechanical output.
2
Introduction
The collective motion of cells drives early embryonic events such as egg shell rotation1 and midgut
invagination2 in Drosophila and epidermal morphogenesis in C. elegans3. Poor coordination is
associated with pathological states, as cell sheets4, clusters5 and strands6 extravasate from tumors
during cancer metastasis7. The coordination of cell movement is driven by a balance of mechanical
stresses occurring at cell interfaces and between the cell and the extracellular matrix (ECM) 8, 9, 10.
The regulated maintenance of this balance determines the homeostatic movement of the epithelium
along mucosal surfaces11 and the repair of epithelial wounds12, 13, 14.
Epithelial wound repair is driven by both the cooperation of distinct modes of migration
and the coordination of mechanical force production which involves the assembly and contractility
of disparate actin architectures across diverse timescales. At the early stages of wound repair, cells
both proximal and distal to the wound migrate to close the space 15, 16, 17. Migration through
"crawling" is driven by forward lamellipodial protrusions, which couple to focal adhesions and
induce contraction and rearward motion of the substrate 8. By contrast, at later stages of wound
repair, lamellipodial protrusions coexist with multi-cellular actomyosin bundles, called "purse
strings", which assemble at the wound periphery ("leading edge")18 and contract laterally to pull
cells forward19. Purse string formation and the dynamics of closure depend upon the curvature of
the wound 20, 21 and can occur in the presence8 or absence of underlying adhesion22. Consequently,
the relationship between contractility and adhesion generated via purse strings is unclear as is the
extent to which it cooperates with lamellipodial protrusion to drive efficient wound closure.
In this study, we investigate the cooperation between Arp2/3-driven lamellipodial
protrusion and actomyosin purse string contraction in controlling the mechanical output within
epithelial monolayers after inducing single cell wounds by laser ablation. Monolayers adhere to
substrates of varying rigidity and constant adhesivity. After ablation, we correlate the dynamics of
closure with the applied mechanical work subject to perturbation in substrate stiffness and
pharmacology. Specifically, we aim to identify the extent to which cells regulate their modes of
migration to optimize the collective dynamic and mechanical outputs of the monolayer.
3
Results
Substrate Stiffness and Actin Filament Disassembly Regulate Monolayer Viscoelasticity
Confluent monolayers, either of MDCK or Caco-2 cells, are adhered to collagen-coated
polyacrylamide (PA) gels (Fig 1a). A 337 nm laser ablates a single cell and creates a small hole
in the gel (SFig 1). Ablation induces quick outward retraction in the surrounding cells for ~120 s,
as the wound reaches its maximum size (Fig 1b, Movie 1). By applying particle image velocimetry
(PIV) to differential interference contrast images (DIC), we quantify the velocity of the retracting
monolayer over time (Fig 1c). We find that the velocity of the retraction decays exponentially
with time, with a characteristic time scale, τ0. Applying a Kelvin-Voigt model, τ0 provides the
ratio of the monolayer viscosity ηm to monolayer modulus Em (Fig 1d). We find that τ0 varies
between 10 and 30 s, roughly consistent with F-actin (filamentous actin) disassembly during this
period (SFig 2, Movie 2). We find that τ0 decreases for increasing substrate rigidity, suggestive of
an increase in monolayer elasticity and/or a decrease in monolayer viscosity (Fig 1e). Thus, cell-
substrate interactions contribute to the viscoelasticity of the epithelial monolayer. We show that
these results are not limited by cell-ECM adhesion levels (SFig 3).
Substrate Stiffness Regulates the Proportion of Distinct Cellular Actin Architectures
Between 120 s and 20 min, F-actin assembles into both lamellipodial protrusions and
actomyosin bundles within cells at the leading edge of the wound concomitant with the
establishment of focal adhesions (Fig 2a, Movies 3, 4, 5). The proportion of cells that exhibit
lamellipodia or purse string architectures at the leading edge vary during closure of the wound (Fig
2a, b). While it is understood that lamellipodial protrusion and cell spreading are associated with
enhanced rigidity of the substrate23, 24, it has thus far remained unclear the extent to which
coordination between lamellipodia and purse string assembly depends on substrate rigidity during
wound closure.
We measure the extent of lamellipodial protrusion and actin purse string formation as a
function of substrate rigidity E at the leading edge of the wound over time. To this end, we trace
the boundary of the wound through F-actin fluorescence intensity in MDCK cells stably
transfected by F-tractin (Fig 2b). We trace the leading edge boundary of length L (SFig 4), which
yields the area of the wound, A. No assumption is made regarding the shape of the wound. We also
4
trace the lamellar border across the wound, as defined by the peak in actin fluorescence intensity
to define the area enclosed by the lamella, AL. Therefore, AL represents the sum of the wound area
and the area of the lamellipodia. The quotient ψ = (AL – A)/ AL, where AL≥A, quantifies the extent
of total lamellipodial protrusion and is analogous to the total number of cells at the leading edge
that exhibit lamellipodia (SFig 4). Therefore, larger ψ corresponds to a predominantly
lamellipodial leading edge, while smaller ψ corresponds to a predominantly purse string edge. We
further distinguish between lamellipodia and purse string by calculating the local nematic order of
actin filaments (SFig 5). We find that lamellipodial edges have lower F-actin alignment than purse
string edges. In addition, we complement these calculations at the leading edge with transient
transfections, where individual cells within the monolayer express fluorescent F-actin, providing
single cell statistics (Fig 2c). The relative abundance of lamellipodial protrusion versus purse string
evolves as a function of A (Fig 2d). While the wound is large (A >1200 µm2) there is significant
F-actin assembly without protrusion, and ψ is low. As A decreases from its initial value,
lamellipodial protrusion reaches a maximum, ψmax, before decreasing for the remainder of closure.
Within a single experiment for E=12.2 kPa, ψ varies slightly (δψsingle = ψmax-ψmin = 0.29 +/- 0.09),
whereas the variation in ψ across control experiments is large (δψall = 0.7).
The abundance of lamellipodial protrusion versus purse string also varies with substrate
rigidity (Fig 2d-inset). For low rigidity (E < 4.3 kPa), ψmax is low, indicating the predominance of
purse strings. By contrast, for high rigidity (E > 4.3 kPa), ψmax is high, indicative of principally
lamellipodial morphology. While ψ is a measure of the proportion of area covered by each
architecture, we also measure the proportion of cells at the leading edge expressing each
architecture individually. In the case of single cell transfections, we qualitatively categorize each
cell as initially choosing either a lamellipodial or purse string morphology, as we image only single
cells and not the entire wound boundary. Indeed, we confirm through single cell transfection that
the dominant F-actin structure depends on substrate stiffness (Fig 2e). The proportion of
lamellipodia and purse string varies for substrate stiffnesses ranging between 1.3 kPa and 55 kPa
and also for glass coverslips (Supplementary Note 3).
Cell morphology and F-actin architecture can be modulated with pharmacological
treatments to bias the formation of lamellipodia or purse strings. Blebbistatin, a myosin ATPase
5
inhibitor, limits the formation of the purse string, and motion is driven by Arp2/3 lamellipodial
protrusion (Fig 2f)25. Similarly SMIFH2 inhibits formin-based F-actin nucleation, and will restrict
purse string formation26. In contrast, CK666 inhibits Arp2/3 related polymerization, minimizing
lamellipodial protrusions and promotes the formation of purse strings (Fig 2g)27. Characterization
of the F-actin architecture through ψmax captures the differences amongst drug and untreated
wound closures (Fig 2h). Thus, these two primary architectures can be controlled through both
substrate rigidity and pharmacological perturbation.
The Balance of Actin Architectures Maintains a Constant Wound Closure Rate
Independent of Substrate Rigidity
We next assess the impact of the balance of F-actin architecture within cells at the leading
edge on wound closure speed. Wound closure rate is measured in two ways. First, as indicated
previously, the perimeter of the leading edge of length L is traced to determine the wound area A
the monolayer distal to the leading edge ("bulk") are calculated using PIV on DIC images (Fig
and tracked over time (Fig 3a). Second, the cumulative displacement field 𝑥𝑥⃗, and the velocity 𝑣𝑣⃗, of
3b). We calculate the strain 𝜀𝜀 of the bulk cells from the divergence of the cumulative displacement
field, 𝜀𝜀=<∇�⃗∙𝑥𝑥�⃗> and the subsequent strain rate 𝜀𝜀 of the monolayer is the time rate of change of
the strain in the linear regime. The rate of change in A predominantly reflects the motion of the
leading edge, and the rate of change in ε reflects the motion of the bulk cells across a fixed field
of view. A and ε are highly correlated, indicating that motion at the leading edge is correlated to
the movement of bulk cells distal to the wound (Fig 3c-inset). A decreases exponentially with time
scale, τ1, and transitions to a second, shorter time scale, τ2 in 45% of wounds (Fig 3c). The latter
correlates with very small wounds (L<100 µm), at which point the leading edge is close to the
ablated hole in the gel. We therefore restrict our analysis to wounds with L > 100µm and
approximate the characteristic time scale for closure as the larger timescale, τ1.
We find that despite differences in monolayer viscoelasticity (Fig 1e), the wound closure
rate is independent of substrate rigidity, as we find no statistically significant correlation (N=36)
between τ1 and E (Fig 3d). However, as there is a difference in the rates of motion between
lamellipodia and purse string in single cells (Fig 2c), we sought to understand how different
architectures regulate closure speed for the wound. Given the presence of both lamellipodia and
6
purse string for E>4.3 kPa, the experiments are partitioned into cases in which the maximum
lamellipodial area at the leading edge is more than 50% ("LP") or less than 50% ("PS") of the total
wound area as found through visual inspection of DIC images. LP wounds have shorter τ1 than PS
wounds, indicating that epithelial wounds with a predominantly lamellipodial leading edge heal
faster, consistent with Fig. 2c. For E<4.3 kPa, all closures are purse string mediated and faster than
purse string closure on more rigid substrates (Fig 3e). Thus, we find that there is a critical rigidity
E*, for which there is a shift in the dynamics of F-actin architecture-specific closure rates. A
similar architecture shift is observed at a different E* for Caco-2 monolayers (SFig 6).
In addition to stochastic variations in the balance of F-actin architectures within cells at the
leading edge, pharmacological perturbations to the balance of F-actin architecture can influence
the rate of closure. On soft substrates (E =1.8 kPa) for which the purse string dominates,
blebbistatin treatment reduces the strain rate (Fig 3f). Conversely, the strain rate for closures on
stiffer substrates (E=12.2 kPa) are less affected by blebbistatin treatment consistent with
lamellipodial closures being the standard closure mode. Thus, motion is most sensitive to myosin
inhibition for E<E* due to the presence of actomyosin purse string and less sensitive for E>E* due
to the majority presence of lamellipodia.
Wounds Balance Closure Speed and Mechanical Work to Maintain a Constant Effective
Power
Using traction force microscopy (TFM) for E>E* and monolayer cell densities
1500< ρ<2500 cells mm-2 (Fig 4a,b), we define the local force 𝐹𝐹⃗ (Fig 4c) and strain energy, ω (Fig
4d) . The total strain energy for a wound is calculated as W=Σω for a wound of length L with a
thickness 2δ+/- = 8 µm at the leading edge (Fig 4c,d, SFig 7, Methods). When wounds are
pharmacologically inhibited from closing, W is not strongly dependent on L (Fig 4e-g, Movie 6).
However, the total energy of the leading edge decreases linearly with L during successful wound
closure (Fig 4h, Methods). Thus, wound closure occurs with a constant energy density W/L
(tension), while maintaining a constant average velocity of the wound within an experiment (Fig
4h).
We next investigate whether the energy density of a wound is set by the balance of F-actin
architecture. We first correlate the variations in ψmax with W for closures on E=12.2 kPa substrates
7
and separate F-actin architectures into PS (ψmax<0.3) and LP (ψmax>0.3) phenotypes (SFig 5).
Indeed, the energy density is lower for LP than for PS (Fig 4i). Consistent with single cell and PIV
measurements (Fig 2c, 3e), closure velocities are higher with LP than with PS (Fig 4j).
Surprisingly, the average change in energy density precisely counter-balances the velocity
difference between the two closure types. Therefore, the product of the energy density and the
wound closure velocity, what we term the effective power P, of the wound is conserved across
phenotypes (Fig 4k). The effective power represents the rate of all mechanical work done at the
leading edge (Methods).
To test the robustness of the relationship between F-actin architecture and mechanical
work, we induce LP and PS phenotypes through blebbistatin and CK666 drug treatments
respectively (Movie 7). Overall, any perturbation to actin assembly decreases the magnitude of the
mechanical work (Fig 4l), suggesting that the work is maximized for mixed architectures. Similar
effects are seen when inhibiting Rho-associated protein kinases and microtubule assembly (Movie
8). Consistent with the LP and PS phenotypes in control (untreated) samples, the energy densities
of blebbistatin-induced lamellipodial wounds are less than CK666-induced purse strings (Fig 4l),
with no significant differences in the velocities of these two groups (Fig 4m). As observed
previously, the effective power remains constant across LP and PS phenotypes (Fig 4n).
Total F-actin Remains Constant as Lamellipodia Transitions to Purse String
The decrease in ψ during wound closure is due to the transformation of F-actin architecture
within a cell from a lamellipodium to a purse string. Prior to this conversion, the lamellar actin
moves at a constant speed (Fig 5a). The lamellipodia then ceases to protrude and the lamellar actin
surpasses the lamellipodial edge to become the actin purse string (Fig 5a,b). Concomitantly, the
purse string increases in F-actin fluorescence intensity locally (Fig 5c, SFig 8). This mechanism is
similar but opposite in direction to what has been observed for lamellipodia to lamellar F-actin
flow in single cells28. The rate at which F-actin decreases in the lamellipodia, mLP is strongly
correlated with the rate of increase in purse string F-actin intensity, mPS. Comparing the integrated
fluorescence intensities, we find that there is sufficient F-actin in the lamellipodia to account for
the increase in F-actin in the purse string and within the cell body (Fig 5d), suggesting that de novo
actin polymerization may not be necessary to form the purse string. By contrast, there is no
significant increase in purse string/lamellar band intensity within pharmacologically perturbed
8
wounds (SFig 9). These results suggest that lamellipodial F-actin reinforces the lamellar band to
form the purse string through the rearrangement of existing F-actin and/or (de)polymerization to
keep the total F-actin constant (Fig 5e).
The total intensity of F-actin along the wound boundary, Iγ decreases during closure (Fig
5f). However, as the boundary length, L, also decreases during closure, the density of F-actin,
ρ=Iγ/L, increases (Fig 5g). The amount of myosin in the lamella/purse string is proportional to the
amount of F-actin (Fig 5h-j, SFig 8) and the line tension of the lamella/purse string is directly
correlated with the total F-actin fluorescence intensity (Fig 5k). Therefore, myosin and F-actin
maintain an approximately constant ratio throughout the transformation from lamella to purse
string during closure.
Differential Friction Between Lamellipodia and Purse String Establishes a Constant
Effective Power
We introduce an active adherent vertex model to simulate collective cell dynamics at the
leading edge of the wound (Fig 6a, Movie 9). Each cell within a two-dimensional confluent tissue
is modeled by a polygon, whose mechanical energy arises from cell-cell adhesion, cell elasticity,
and actomyosin contractility29, 30, 31 (Supplementary Text; Methods). The elastic substrate is
modeled by a triangular mesh of springs which is anchored to the polygonal cells via stiff springs
(focal adhesion) that undergo stochastic turnover. Forces are measured analogously to
experimental traction force microscopy (Fig 6b). As observed in experiments, the wound area
decays exponentially with a characteristic timescale τ1 (Fig 6c, 3c). Additionally, simulations of
wound closure also exhibit a constant strain energy density and velocity (Fig 6d, 4h).
Cells at the leading edge begin motion by protrusive crawling, which stochastically
to actomyosin contractility. We simulate the closure dynamics of an initially circularly shaped
transitions to a purse-string at a rate 𝑘𝑘𝑝𝑝𝑝𝑝, resulting in an increased tension at the leading edge due
wound for a range of purse-string assembly rates, 𝑘𝑘𝑝𝑝𝑝𝑝 (0.50-2.00 hr-1) and different values of
substrate stiffness. Consistent with ψmax used experimentally, we classify wound repair as purse
string dominant if the mean proportion of the wound perimeter covered by purse string is greater
than 50% (PS), and lamellipodia dominant (LP) if the purse-string coverage is less than 50% (SFig
9
PS*2.5=koff
5). When cell-substrate adhesion lifetimes vary between lamellipodial and purse string cells such
LP (Fig 6e), mechanical measurements are in quantitative agreement with
that koff
experimental data. We find that the closure rate increases with substrate stiffness for purse string
closure, while lamellipodial closure shows no dependence on stiffness (Fig 6f). Prolonged cell-
substrate adhesion lifetime on purse-string edges is necessary for sensitivity to substrate stiffness
(SFig 10).
Our model allows us to investigate the role of actin architectures on wound closure
dynamics and traction forces generated on the substrate. We show that the energy density of purse
strings is higher than lamellipodial protrusion (Fig 6g) and that the velocity of protrusion is faster
than the purse string (Fig 6h) as found experimentally (Fig 4i,j). Additionally, the invariance in
effective power between lamellipodia and purse string is observed (Fig 6i). Interestingly, these
results only hold if cell-substrate adhesions have a longer attachment time for purse strings than
for lamellipodia cells (Fig 6j, SFig 10). Seen in both experiments and simulations, focal adhesion
orientations differ between the two phenotypes consistent with previous results (Fig 6k-m)32. The
necessary asymmetry in adhesion lifetimes between purse string and lamellipodia required to
maintain a constant effective power is consistent with the experimentally measured asymmetry in
focal adhesion size (Fig 6n).
Since differential rates of attachment and detachment of focal adhesions will lead to
differences in cell-substrate friction33, we calculate the effective friction between the monolayer
and the substrate, given by the ratio of the forces along the leading edge to the velocity of closure
<�𝐹𝐹⃗�> 𝑣𝑣� . Since the forces can be measured directly, our estimation is model-independent.
6o), due to an increased adhesion lifetime on purse string edges. Furthermore, by varying 𝑘𝑘𝑜𝑜𝑜𝑜𝑜𝑜𝑝𝑝𝑝𝑝 we
Consistent with the difference in focal adhesion size, we find that lamellipodial wounds exhibit
less friction than purse string wounds on 12.2 kPa substrates (Fig 6o). In agreement with
experimental data, we find a significantly higher effective friction for purse string simulations (Fig
see that an increase in effective friction corresponds to an increase in wound closure time 34 (SFig
10).
Discussion
10
Using the F-actin cytoskeleton, tissues generate tensile forces, and the balance of this
tension against external loading and resistance is referred to as "tensional" or "energy"
homeostasis35, 36. Previous studies have shown that to maintain mechanical homeostasis in cell
colonies, traction stress maxima localize to colony boundaries37, and that the total elastic strain
energy increases linearly with the size of the colony38, 39. In addition, traction stress maxima
localize to the leading edge of migrating cell monolayers 8 32. Similarly, in wound healing, traction
stress maxima localize to the boundary (Movie 10) and drive the elastic strain energy to decrease
linearly with the perimeter of the wound maintaining tensional homeostasis during the dynamics
of closure. At the maximum wound perimeter, lamellipodial protrusions initiate closure through
retrograde traction stresses32. As the perimeter decreases, lamellipodia cease to protrude and the
purse string generates larger, anterograde traction stresses (SFig 6). Thus, the total energy
decreases with perimeter, keeping the average energy density constant in single wounds.
The average wound closure rate is independent of substrate rigidity. However, the
proportion of cells within the leading edge that exhibit lamellipodia increases with rigid substrates.
When mixed architectures are observed (E=12.2 kPa), lamellipodial protrusions generate low
energy densities and move at high velocity. By contrast, the purse string produces high energies
and velocity is low. Surprisingly, the product of velocity and energy density, quantifying the
effective power, remains, on average, conserved and invariant of F-actin architecture regulation.
The same trend is established through spontaneous variation in ψmax across control experiments
and with pharmacological perturbation.
Lamellipodial protrusion rates have previously been correlated with substrate rigidity,
suggesting that the stabilization of lamellipodial protrusions is a sufficient condition for the
mechano-sensitivity of the wound40 41. Surprisingly however, when individual experiments are
partitioned into predominantly purse string experiments (PS), the wound closes quickly on highly
compliant substrates and slowly on rigid substrates suggesting mechano-sensitivity.
Finally, we show through a cell-based computational model, that the effective power is
conserved during wound closure only when focal adhesions are more stable in time for purse
strings than for lamellipodia, yielding a difference in the friction with the surface. With equal
stability, the strain energy density is proportional to the velocity of closure, resulting in greater
effective power for the faster closing wounds. Longer adhesion lifetimes result in increased strain
11
transmitted to the substrate and slower movement, balancing the effective power. We confirm by
experiment, that purse string-associated adhesions are approximately 2.5 times the size as those
associated with lamellipodia (Fig 6), suggesting their increased stability based on previous studies
that indicate an association between size and stability42, 43. Furthermore, we show that experimental
estimates of the friction are consistent with both the measurements of focal adhesion size as well
as the analytical estimates. Thus, the limitation to the rate at which work is applied (i.e. the
effective power) is dependent upon the stability of adhesions and the timescales over which
mechanical forces are transmitted to the ECM.
Taken together, the results presented here relate the non-equilibrium self-assembly of the
cytoskeletal machinery to the flow of mechanical energy at tissue-scales. We have identified
multiple conservation relationships from the molecular machinery (F-actin, focal adhesions) to
cellular-scale mechanical outputs (work, effective power) that regulate the dynamics of collective
motion. These fundamental identities have broader implications for the relationship between the
mechanical outputs of tissues and their constituent cells.
Acknowledgements
We acknowledge funding ARO MURI W911NF-14-1-0403 to MM, DK, CS, VA & APT. DSS
acknowledges support from NSF Fellowship grant # DGE1122492. We acknowledge funding
CMMI-1525316 and NIH RO1 GM126256 to MM and NIH U54 CA209992 to MM & MSY. We
also acknowledge fellowship support from the Yale Endowed Fund to VA. SB and MFS
acknowledge support from Institute for the Physics of Living Systems at the University College
London, and EPSRC funded PhD studentship for MFS. Any opinion, findings, and conclusions or
recommendations expressed in this material are those of the authors(s) and do not necessarily
reflect the views of the National Science Foundation, NIH, or EPSRC.
Author Contributions
MPM designed and conceived the experimental work. SB designed and conceived the
computational model. VA, APT, AF, MSY acquired experimental data. MPM, APT, DSS, CS
and VA analyzed experimental data. MFS implemented the model and performed simulations. DB
provided computational tools and design. MPM and SB contributed analytical tools. MPM, APT,
SB, MFS, & DK wrote the paper.
12
Competing Financial Interests
The authors declare no competing financial interests.
13
Methods
Polyacrylamide Gel Formation
Polyacrylamide gels are polymerized onto coverslips of 25mm diameter (#1.5, Dow Corning).
Briefly, the coverslips are treated with a combination of aminopropylsilane (Sigma Aldrich) and
glutaraldehyde (Electron Microscopy Sciences) to make the surface reactive to the acrylamide.
Varying concentrations of bis-acrylamide are mixed with 0.05% w/v ammonium persulfate (Fisher
BioReagents) to yield a gel with an elastic modulus E of 1.5 kPa to 55 kPa. 40nM beads (Molecular
Probes) are embedded in the gel mixture prior to polymerization. The ratios of polyacrylamide to
bis-acrylamide for the gels used in this study are 5%:0.1% (E=1.3 kPa) 5%:0.175% (E=1.8 kPa),
7.5%:0.153% (E=4.3 kPa), 7.5%:0.3% (E=8.6 kPa), 12%:0.086% (E=12.2 kPa), 12%:0.145%
(E=16 kPa), 12%:0.19% (E=24 kPa), and 12%:0.6% (E=55 kPa) 40, 44.
Before the gels are fully polymerized, 9 µl of the gel is added to the coverslip and covered with
another coverslip, which has been made hydrophobic through treatment with Rain-X®. The gels
are polymerized on the coverslips for 30 minutes at room temperature. The gels are then reacted
with the standard 2mg/mL Sulfo-SANPAH (Thermo Fisher Scientific) protocol44. The surface of
the gels is then coated with rat tail collagen Type I (Corning®; high concentration). Concentration
of collagen used range between (0.01 - 1 mg/mL) to establish the role of adhesivity on traction
force measurement (SFig 3). In the main text, we limit the scope to 1 mg/mL collagen to not limit
lamellipodial activity. The reaction proceeds for 2 hours in the dark, and the coverslips are then
rinsed with 1X PBS.
Cell Culture
Madin-Darby Canine Kidney (MDCK.2) cells (CRL-2936™; ATCC, Manassas, VA) were
maintained in Eagle's Minimum Essential Medium (ATCC), supplemented with 10% fetal bovine
serum (GIBCO Life Technologies) and 1% penicillin/streptomycin at 37°C and 5% CO2 in a
humidified incubator. Early passage cells (< 20 passages) were used for experiments. Similarly,
Caco-2 (HTB-37™; ATCC, Manassas, VA) were maintained in the same culture conditions with
the exception of 20% fetal bovine serum. Caco-2 cells were used primarily for their large size
which enables superior visualization of their cell cytoskeleton during wound repair.
14
F-TRActin Stable Line Transfection
MDCK.2 cells were stably transfected with plasmid construct encoding for FTRActinEGFP (kind
gift from Sergey Plotnikov, University of Toronto). Briefly, cells were transfected using
FuGENE®HD where 2 µg of the DNA plasmid was added to the transfection reagent and added
to a cell dish. Cells were incubated for over 24 hrs with the plasmid to complete the transfection
process. Following 1 week of incubation with a selection media containing G418 (Mirus Bio LLC)
at 0.5mg/mL, population of cells were selected based on fluorescently expressing the construct.
The isolated population were cultured and expanded on the selective media and used for
experimental purposes.
LifeAct Transient Transfection of Caco-2
Transfection was performed on Caco-2 cells using Lipofectamine 3000 (Life Technologies).
Briefly, 3.75 µl of Lipofectamine 3000 was added to 125 µl of serum free media. 2.5 µg of GFP-
LifeAct DNA (Bement Lab, University of Wisconsin-Madison) and 5 µl of P3000 reagent were
added to the mixture. The solution was mixed and incubated at room temperature for 5 minutes
before being added to 70% confluent cells in a drop-wise manner. Cells were incubated overnight
at 37°C before imaging.
Laser Ablation and Time Lapse Imaging
Laser ablation was performed using a 435nm laser (Andor Technology, Belfast, Northern Ireland).
A 60X oil-immersion objective (Leica Microsystems) was used for ablation and the laser power
was held between 60% and 65%. Images were acquired at 5 second intervals for the first 20
minutes following ablation with a confocal microscope (Andor Technologies, Belfast, Northern
Ireland). Images were then collected at 5 minute intervals until wound closure.
Traction Force Microscopy
Traction force microscopy is used to measure the forces exerted by cells on the substrate44. 'Force-
loaded' images (with cells) of the beads embedded in the polyacrylamide gels were obtained using
a 60X oil-immersion objective (Leica Microsystems). The 'Null-force' image was obtained at the
end of each experiment by adding trypsin to the cells for 1 hour. If the wound was exceedingly
large (i.e. of the size of the image), the ablated image was used for the null-force image. Images
15
were aligned to correct drift (StackReg for ImageJ) 45 and compared to the reference image using
particle imaging velocimetry (PIV) software (http://www.oceanwave.jp/softwares/mpiv/) in
MATLAB to produce a grid spacing of 7.0 µm. Forces can be reliably measured between 4.3 and
24 kPa.
The traction forces are used to calculate the energies of deformation in the substrate using the
spatial distribution of forces 𝐹𝐹⃗ and displacements 𝑥𝑥⃗. Since lamellipodia and purse strings have
been shown to apply forces in different directions32, we choose a directionally-independent metric
(the strain energy ω) as opposed to summing up individual force contributions. Consequently, the
effective power is the rate at which all mechanical work is done instead of isolating the amount of
work done in the direction of motion. The spatial distribution of ω is calculated, for each coarse-
grained grid size, using the equation 𝜔𝜔=12𝐹𝐹⃗∙𝑥𝑥⃗. The total strain energy W for a given geometry is
calculated by summing up the local values of ω within a range of 8 µm (~size of the hot-spots in
strain energy maps) at the leading edge. Forces along the leading edge are reliably calculated when
wounds are far from the ablation-induced hole in the traction force substrate. Therefore, we apply
a wound perimeter cutoff and only report forces for wounds greater than 100 µm in perimeter. For
12.2 kPa substrates, traction force damage does not affect force measurements with this perimeter
cutoff in 83% (N=29 of 36) of wounds.
Immunofluorescence Microscopy
Cells were fixed with a solution of 4% paraformaldehyde in PBS and permeabilized with 0.2%
Triton X-100. The cells were then blocked with 1% BSA in PBS at room temperature for 1 hour,
incubated with primary antibodies for Paxillin (Rabbit monocolonal [Y113] to Paxillin, Abcam
ab32084; 1:250 dilution) and E-Cadherin (Rat monoclonal [DECMA-1] to E Cadherin, Abcam ab
11512; 1:200 dilution) in PBS for 1 hour at room temperature. The samples were incubated with
complimentary secondary antibodies in PBS for 1 hour at room temperature. The secondary
antibodies used were Alexa Fluor 647 (Donkey anti- Rabbit, Abcam ab 150075, 1:500 dilution)
and Alexa Fluor 405 (Goat anti-Rat, Abcam ab 175671, 1:500 dilution). Phalloidin stainining was
performed with Alexa Fluor 488 Phalloidin (Life Technologies; 1:40) diluted in PBS with 1%
BSA at room temperature for 20 minutes. Images were taken with a 60X oil immersion objective.
Analysis was performed with ImageJ.
16
Super-resolution Microscopy
Ablated wounds were fixed in 4% paraformaldehyde and stained for F-actin using SiR-Actin
(Cytoskeleton Inc. CY-SC001, 1:1000 dilution) and p-myosin light chain2 (Rabbit, S15 - Cell
Signaling 3617s #9284, 1:250 dilution) . To visualize spatial distribution of myosin, we used Atto
594 (Anti-Rabbit Atto 594, Sigma Aldrich 77671, 1:500 dilution) as a secondary antibody tailored
for super resolution imaging. Super resolution imaging was performed on the Abberior STED
system (Abberior Instrument GmbH – Pulsed STED laser @775nm) using a water immersion lens
1.3NA. Images were taken with a scanning resolution of 20nm and 2.5D scan line."
Pharmacological Treatments
Formin FH2 domain inhibitor (SMIFH2) was purchased from MilliporeSigma (34409) and used
at a 10µM concentrations throughout all the experiments. Myosin II inhibitor ( (-)Blebbistatin)
was purchased from Sigma Aldrich (B0560) and used at a concentration of 50 µM unless otherwise
noted. Arp2/3 inhibitor (CK666), purchased from Tocris (3950) and used at 200 µM concentration.
All inhibitors were reconstituted in DMSO at high stock concentration and diluted in media to
their appropriate concentration without any further purification.
Focal Adhesion Analysis
Focal adhesion size and orientation were calculated using the "Analyze Particle" plugin in Fiji
(ImageJ software) 46. A threshold was applied to confocal images stained for paxillin, and the focal
adhesion is fit to an ellipse with a given area. The angle of orientation is determined as the angle
between the major axis of the fitted ellipse and either the vector normal to the purse string (for
purse strings) or the vector normal to the nearest lamellipodial protrusion (for lamellipodia).
Myosin Intensity Calculations
The fluorescence intensities of myosin filaments were determined on super resolution images
through spot tracking in Imaris software (Bitplane AG, Zurich, Switzerland). Myosin were found
as "spots" with 200nm diameter.
Cell-based computational model
17
The model is implemented using Surface Evolver47. A wound is generated by removing cells
within a circle of radius of 22.5 µm from a colony of 250 cells. The vertices surrounding the wound
are then moved onto the edge of the circle, and the system relaxed so that all wound vertices lie on
the perimeter of the circle and the system is at an energy minimum. To initiate gap closure, cells
around the wound are set to the crawling mode. We then run the simulation until closure. At each
time step we update adhesions binding, cell modes, perform neighbor exchanges and apply
mechanical and active forces on vertices.
Dimensionless parameters for the cell energy were taken from a previous study using a vertex
model for an MDCK monolayer 48. Using length and force scales of the cells and substrate we
recover the dimensional values. Other parameters, including purse string tension, protrusion force
and focal adhesion binding rates were chosen to qualitatively match wound closure dynamics and
traction force maps.
Using displacements in the substrate mesh we can interpolate a displacement field on a square
grid, and use finite difference discretization to calculate elastic stress and strain. We measure the
total work during wound closure as the amount of strain energy within a thickness of 8 µm of the
wound border. For a full description of the model see the Supplementary Information.
Statistical Tests
All statistical comparisons between two distributions were done with a two-sided t-test. When
distributions are presented as a single value with error bars, the value is the mean of the
distribution, and the error bars are the standard deviations. We use the symbols *, **, and *** for
p< 0.05, 0.01, and 0.001 respectively. When fitting lines to data, we quote the p-value as
significance values to rejections of the null hypothesis.
Data Availability
Data that support plots and other findings within this manuscript are available from the
corresponding authors upon reasonable request.
Code Availability
18
Custom codes that were used to analyze experimental data within this manuscript are available
from the corresponding authors upon reasonable request.
19
Figure 1: Monolayer Viscoelasticity Depends on Substrate Rigidity (a) Schematic of experimental setup
showing laser ablation of an epithelial sheet and subsequent sheet retraction velocity, v. Star marks location
of ablation. Cell monolayers adhere to collagen covalently bound to polyacrylamide substrates. Fluorescent
particles are embedded within the polyacrylamide substrate allowing the calculation of traction forces.
Kelvin-Voigt model used to quantify retraction. (b) DIC images immediately after ablation on a 1.3 kPa
substrate. Scale bar is 25 µm. (c) PIV of initial deformation of the monolayer 0.5s after ablation, compared
to immediately prior to ablation. Red star indicates the position of ablation. (d) Average monolayer
retraction velocity over time, t, and viscoelastic timescale, τ0, representative of the ratio between monolayer
viscosity ηm and elasticity Em within a Kelvin-Voigt model. (e) τ0 is a function of substrate elasticity, E
(ntotal=38).
20
21
Figure 2. F-actin Architecture Varies with Wound Size and with Substrate Stiffness. (a) F-actin (red),
cadherin (blue), and paxillin (green) stains for a wound closing on a 4.3 kPa substrate. (b) F-Tractin stably
transfected in an MDCK monolayer during healing. Ablation occurs at t=0 min. Outlined is the wound area
A, and the area of the wound and lamellipodial protrusions, AL. This yields the fraction of wound area
covered by lamellipodia, ψ. (c) Single cells within monolayers expressing LifeAct during healing form
either lamellipodia (LP- top) or a purse string (PS- bottom). Dotted lines show regions used for kymograph
analysis. Kymographs (right) indicate the rate of protrusion by both lamellipodium and purse string. (d)
Example of lamellipodial fraction ψ as a function of wound area on a 12.2 kPa substrate with maximimal
lamellipodial area fraction ψmax. (d-inset) The purse string fraction,1- ψmax, as a function of substrate
stiffness, E (N=32). Dashed line is line of best fit with p<0.001. (e) Percent of cells at wound boundary
closing via lamellipodia (LP) versus purse string (PS) as determined by eye. (f) F-actin image of wound
treated with blebbistatin has pronounced lamellipodial protrusions. (g) F-actin image of wound treated with
CK666 has an enhanced purse string. (h) Average purse string fraction 1-< ψmax> and standard deviation
across drug treatments of wounds closing on 12.2 kPa substrates. (Ncontrol= 9, Nbleb= 3, NSMIFH2=6, and
NCK666=5). Scale bars are 25 µm. Stars indicate positions of ablations.
22
Figure 3. Purse String Coordinates with Lamellipodia to Maintain Closure Time. (a) DIC images of
epithelial wound closure over time on an 8.5 kPa substrate. Red arrows point to boundaries between purse
string and lamellipodia. (b) PIV showing accumulated strain of DIC images in (a). (c) Wound area over
time, indicating the initial wound size A0, and exponential decay constants τ1 and τ2, where 𝐴𝐴=𝐴𝐴0𝑒𝑒−𝑡𝑡/𝜏𝜏.
(d) τ1 as a function of substrate stiffness (N=36). (e) τ1 as a function of substrate stiffness separating the
samples from (d) into predominantly purse string samples (PS) or lamellipodial crawling (LP). (f) Strain
rates during closure on very soft (E=1.8 kPa) and stiff (E=12.2 kPa) gels show difference in sensitivity to
myosin-inhibition by blebbistatin (E=1.8kPa: control N=3, 10µM Bleb N=3, 40µM Bleb N=2; E=12.2kPa:
control N=4, 40µM Bleb N=4).
23
Figure 4. Mechanical Work but not Effective Power is Architecture Dependent. (a) Definition of
strain energy, W, and effective power, P, in wound healing schematic. (b) DIC image sequence of the
closing of a wound on E=4.3 kPa and closing principally by purse string (PS). Associated traction force
vectors (c) and strain energy maps (d) for the wound in (b) with definition of the wound perimeter, L, and
wound thickness, 2δ+/-. Total strain energy W versus L for pharmacologically inhibited closures (e-f) and
timelapse (g). (h) Successful closure on a substrate with stiffness E=12.2 kPa exhibiting a linear W-L
relation (closed symbols) and constant velocity (open symbols). Dashed line is linear best fit of W vs L. (i)
Wound energy densities W/L for E=12.2 kPa control is split into LP (N=12) (ψmax>0.3) or PS (N=8)
24
(ψmax<0.3) subsets. (j) Closure velocity and effective power (k) for data in (i). Solid lines in (i-k) represent
the average value for all control samples (N=20) before creating LP and PS subsets. (l-n) Wound energy
densities, velocities, and effective powers for CK666 (N=8) and Blebbistatin treated monolayers (N=11).
Dashed lines are average values for control samples, indicating that both drug treatments decrease
mechanical work for a wound.
25
Figure 5. Total F-actin is Constant During Lamellipodial to Purse String Transition. (a) Montage of
fluorescent F-actin within a single cell in a MDCK monolayer on a 12.2 kPa substrate showing a transition
from lamellipodial crawling to purse string. (b) Kymograph of F-tractin intensities, measuring the quantity
of F-actin along the dashed line in (a). Regions outlined are for the cell body (CB), the purse string (PS)
and the lamellipodium (LP). (c) The spatial sum of F-actin fluorescence per unit time for the regions
outlined in (b). The ratio of mass flux into the purse string mPS and out of the lamellipodium mLP shows
26
correlation in F-actin transfer. (d) F-actin fluorescence intensity integrated over the time series in (c). The
decrease in F-actin from the lamellipodium is consistent with the sum of F-actin increases in the purse string
and cell body. (e) Cartoon schematic of lamellipodial actin being incorporated into the lamellar band to
form a purse string. From t=0 to t=tf (when the wound is closed) the total F-actin intensity along the wound
boundary (f) decreases, but the actin density (ρ=Iγ/L) increases (g). Lines are average and grey regions are
standard deviation of (N=4). Super-resolution images of F-actin (h) and myosin (i) of a cell at the leading
edge. Myosin localizes to lamellar band. Scale bar is 5 µm. (j) In the lamella, the total myosin intensity Imyo
scales with the total amount of actin. (k) Line tension T=W/L depends on actin content of purse string in
CK666 treated wounds (N=7). Red dots are from one sample.
27
Figure 6. Differential friction is necessary to balance effective power. (a) Schematic of vertex model
and inset where cells at the leading edge can either exhibit purse string or lamellipodial crawling. (b) Spatial
distribution of strain energy measured through the simulation during closure. (c) Vertex model wound area
dependence on time. (d) Strain energy-perimeter relation and wound closure velocity. (e) Differences in
unbinding probabilities koff
LP=0.208 min-1(green) lead to differences in focal
adhesion lifetimes. (e-inset) Equal focal adhesion off-rates koff
LP=0.208 min-1 lead to equal focal
adhesion lifetimes. (f) Decay time τ1 as a function of substrate stiffness (E=1 kPa nLP=26, nPS=16; E=2 kPa
nLP=29, nPS=13; E=4 kPa nLP=24, nPS=18; E=8 kPa nLP=22, nPS=20; E=16 kPa nLP=24, nPS=18). (g-i) Energy
density, velocity, and effective power for wounds exhibiting either purse string or lamellipodial crawling.
koff
PS for E=4kPa (nLP=24, nPS=18). (j) The effective power of wounds on E=4 kPa substrate is not
balanced between LP and PS if koff
LP (nLP=14, nPS=6). Vertex model and confocal image show focal
adhesions parallel to leading edge for purse strings (k) and perpendicular to leading edge for lamellipodia
PS=0.083 min-1 (blue) and koff
PS= koff
LP>koff
PS= koff
28
(l). Scale bars are 10µm. (m) Histograms of focal adhesion angular distributions for cells exhibiting PS
(N=257) and LP (N=342). (n) Mean focal adhesion size differs between PS (N=412) and LP (N=1075). (o)
Experimentally and analytically calculated friction for LP and PS.
29
References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Haigo SL, Bilder D. Global tissue revolutions in a morphogenetic movement controlling elongation.
Science 2011, 331(6020): 1071-1074.
Edwards KA, Demsky M, Montague RA, Weymouth N, Kiehart DP. GFP-moesin illuminates actin
cytoskeleton dynamics
living tissue and demonstrates cell shape changes during
morphogenesis in Drosophila. Dev Biol 1997, 191(1): 103-117.
in
Simske JS, Hardin J. Getting into shape: epidermal morphogenesis in Caenorhabditis elegans
embryos. Bioessays 2001, 23(1): 12-23.
Nabeshima K, Inoue T, Shimao Y, Kataoka H, Koono M. Cohort migration of carcinoma cells:
differentiated colorectal carcinoma cells move as coherent cell clusters or sheets. Histol
Histopathol 1999, 14(4): 1183-1197.
Friedl P, Noble PB, Walton PA, Laird DW, Chauvin PJ, Tabah RJ, et al. Migration of coordinated cell
clusters in mesenchymal and epithelial cancer explants in vitro. Cancer Res 1995, 55(20): 4557-
4560.
Gaggioli C, Hooper S, Hidalgo-Carcedo C, Grosse R, Marshall JF, Harrington K, et al. Fibroblast-led
collective invasion of carcinoma cells with differing roles for RhoGTPases in leading and following
cells. Nat Cell Biol 2007, 9(12): 1392-1400.
Friedl P, Gilmour D. Collective cell migration in morphogenesis, regeneration and cancer. Nat Rev
Mol Cell Biol 2009, 10(7): 445-457.
Trepat X, Wasserman MR, Angelini TE, Millet E, Weitz DA, Butler JP, et al. Physical forces during
collective cell migration. Nat Phys 2009, 5(6): 426-430.
Tambe DT, Hardin CC, Angelini TE, Rajendran K, Park CY, Serra-Picamal X, et al. Collective cell
guidance by cooperative intercellular forces. Nat Mater 2011, 10(6): 469-475.
Maruthamuthu V, Sabass B, Schwarz US, Gardel ML. Cell-ECM traction force modulates
endogenous tension at cell-cell contacts. Proc Natl Acad Sci U S A 2011, 108(12): 4708-4713.
Schmidt GH, Winton DJ, Ponder BA. Development of the pattern of cell renewal in the crypt-villus
unit of chimaeric mouse small intestine. Development 1988, 103(4): 785-790.
30
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
Brugues A, Anon E, Conte V, Veldhuis J, Colombelli J, Munoz J, et al. Physical forces driving wound
healing. Molecular Biology of the Cell 2013, 24.
Zhao M, Song B, Pu J, Forrester JV, McCaig CD. Direct visualization of a stratified epithelium reveals
that wounds heal by unified sliding of cell sheets. FASEB J 2003, 17(3): 397-406.
Bement WM, Mandato CA, Kirsch MN. Wound-induced assembly and closure of an actomyosin
purse string in Xenopus oocytes. Curr Biol 1999, 9(11): 579-587.
Farooqui R, Fenteany G. Multiple rows of cells behind an epithelial wound edge extend cryptic
lamellipodia to collectively drive cell-sheet movement. J Cell Sci 2005, 118(Pt 1): 51-63.
Nusrat A, Delp C, Madara JL. Intestinal Epithelial Restitution - Characterization of a Cell-Culture
Model and Mapping of Cytoskeletal Elements in Migrating Cells. Journal of Clinical Investigation
1992, 89(5): 1501-1511.
Tamada M, Perez TD, Nelson WJ, Sheetz MP. Two distinct modes of myosin assembly and
dynamics during epithelial wound closure. J Cell Biol 2007, 176(1): 27-33.
Bement WM, Forscher P, Mooseker MS. A Novel Cytoskeletal Structure Involved in Purse String
Wound Closure and Cell Polarity Maintenance. J Cell Biol 1993, 121(3): 565-578.
Murrell M, Kamm R, Matsudaira P. Substrate viscosity enhances correlation in epithelial sheet
movement. Biophys J 2011, 101(2): 297-306.
Anon E, Serra-Picamal X, Hersen P, Gauthier NC, Sheetz MP, Trepat X, et al. Cell crawling mediates
collective cell migration to close undamaged epithelial gaps. Proc Natl Acad Sci U S A 2012,
109(27): 10891-10896.
Ravasio A, Cheddadi I, Chen T, Pereira T, Ong HT, Bertocchi C, et al. Gap geometry dictates
epithelial closure efficiency. Nat Commun 2015, 6: 7683.
Vedula SR, Peyret G, Cheddadi I, Chen T, Brugues A, Hirata H, et al. Mechanics of epithelial closure
over non-adherent environments. Nat Commun 2015, 6: 6111.
Califano JP, Reinhart-King CA. Substrate Stiffness and Cell Area Predict Cellular Traction Stresses
in Single Cells and Cells in Contact. Cell Mol Bioeng 2010, 3(1): 68-75.
31
Han SJ, Bielawski KS, Ting LH, Rodriguez ML, Sniadecki NJ. Decoupling substrate stiffness, spread
area, and micropost density: a close spatial relationship between traction forces and focal
adhesions. Biophys J 2012, 103(4): 640-648.
Limouze J, Straight AF, Mitchison T, Sellers JR. Specificity of blebbistatin, an inhibitor of myosin II.
J Muscle Res Cell Motil 2004, 25(4-5): 337-341.
Rizvi SA, Neidt EM, Cui J, Feiger Z, Skau CT, Gardel ML, et al. Identification and characterization of
a small molecule inhibitor of formin-mediated actin assembly. Chem Biol 2009, 16(11): 1158-1168.
Nolen BJ, Tomasevic N, Russell A, Pierce DW, Jia Z, McCormick CD, et al. Characterization of two
classes of small molecule inhibitors of Arp2/3 complex. Nature 2009, 4640: 1031-1034.
Burnette DT, Manley S, Sengupta P, Sougrat R, Davidson MW, Kachar B, et al. A role for actin arcs
in the leading-edge advance of migrating cells. Nat Cell Biol 2011, 13(4): 371-U388.
Honda H, Eguchi G. How Much Does the Cell Boundary Contract in a Monolayered Cell Sheet. J
Theor Biol 1980, 84(3): 575-588.
Farhadifar R, Roper JC, Algouy B, Eaton S, Julicher F. The influence of cell mechanics, cell-cell
interactions, and proliferation on epithelial packing. Curr Biol 2007, 17(24): 2095-2104.
Fletcher AG, Osterfield M, Baker RE, Shvartsman SY. Vertex Models of Epithelial Morphogenesis.
Biophys J 2014, 106(11): 2291-2304.
Brugues A, Anon E, Conte V, Veldhuis JH, Gupta M, Colombelli J, et al. Forces driving epithelial
wound healing. Nat Phys 2014, 10(9): 684-691.
Krzyszczyk P, Wolgemuth CW. Mechanosensing can result from adhesion molecule dynamics.
Biophys J 2011, 101(10): L53-55.
Cochet-Escartin O, Ranft J, Silberzan P, Marcq P. Border forces and friction control epithelial
closure dynamics. Biophys J 2014, 106(1): 65-73.
Brown RA, Prajapati R, McGrouther DA, Yannas IV, Eastwood M. Tensional homeostasis in dermal
fibroblasts: mechanical responses to mechanical loading in three-dimensional substrates. J Cell
Physiol 1998, 175(3): 323-332.
32
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
Chien S. Mechanotransduction and endothelial cell homeostasis: the wisdom of the cell. Am J
Physiol Heart Circ Physiol 2007, 292(3): H1209-1224.
du Roure O, Saez A, Buguin A, Austin RH, Chavrier P, Silberzan P, et al. Force mapping in epithelial
cell migration. Proc Natl Acad Sci U S A 2005, 102(7): 2390-2395.
Mertz AF, Che Y, Banerjee S, Goldstein JM, Rosowski KA, Revilla SF, et al. Cadherin-based
intercellular adhesions organize epithelial cell-matrix traction forces. Proc Natl Acad Sci U S A
2013, 110(3): 842-847.
Mertz AF, Banerjee S, Che Y, German GK, Xu Y, Hyland C, et al. Scaling of traction forces with the
size of cohesive cell colonies. Phys Rev Lett 2012, 108(19): 198101.
Yeung T, Georges PC, Flanagan LA, Marg B, Ortiz M, Funaki M, et al. Effects of substrate stiffness
on cell morphology, cytoskeletal structure, and adhesion. Cell Motil Cytoskeleton 2005, 60(1): 24-
34.
Pelham RJ, Jr., Wang Y. Cell locomotion and focal adhesions are regulated by substrate flexibility.
Proc Natl Acad Sci U S A 1997, 94(25): 13661-13665.
Stricker J, Aratyn-Schaus Y, Oakes PW, Gardel ML. Spatiotemporal constraints on the force-
dependent growth of focal adhesions. Biophys J 2011, 100(12): 2883-2893.
Goffin JM, Pittet P, Csucs G, Lussi JW, Meister JJ, Hinz B. Focal adhesion size controls tension-
dependent recruitment of alpha-smooth muscle actin to stress fibers. J Cell Biol 2006, 172(2): 259-
268.
Sabass B, Gardel ML, Waterman CM, Schwarz US. High resolution traction force microscopy based
on experimental and computational advances. Biophys J 2008, 94(1): 207-220.
Thevenaz P, Ruttimann UE, Unser M. A pyramid approach to subpixel registration based on
intensity. Ieee T Image Process 1998, 7(1): 27-41.
Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, et al. Fiji: an open-source
platform for biological-image analysis. Nat Methods 2012, 9(7): 676-682.
Brakke KA. The surface evolver. Experiment Math 1992, 1(2): 141-165.
33
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
Kuipers D, Mehonic A, Kajita M, Peter L, Fujita Y, Duke T, et al. Epithelial repair is a two-stage
process driven first by dying cells and then by their neighbours. J Cell Sci 2014, 127(6): 1229-1241.
34
|
1808.03192 | 1 | 1808 | 2018-08-06T09:16:53 | Multi-purpose SLM-light-sheet microscope | [
"physics.bio-ph",
"physics.optics"
] | By integrating a phase-only Spatial Light Modulator (SLM) into the illumination arm of a cylindrical-lens-based Selective Plane Illumination Microscope (SPIM), we have created a versatile system able to deliver high quality images by operating in a wide variety of different imaging modalities. When placed in a Fourier plane, the SLM permits modulation of the microscope's light-sheet to implement imaging techniques such as structured illumination, tiling, pivoting, autofocusing and pencil beam scanning. Previous publications on dedicated microscope setups have shown how these techniques can deliver improved image quality by rejecting out-offocus light (structured illumination and pencil beam scanning), reducing shadowing (light-sheet pivoting), and obtaining a more uniform illumination by moving the highest-resolution region of the light-sheet across the imaging Field of View (tiling). Our SLM-SPIM configuration is easy to build and use, and has been designed to allow all of these techniques to be employed on one optical setup compatible with the OpenSPIM design. It also offers the possibility to choose between three different light-sheets, in thickness and height, which can be selected according to the characteristics of the sample and the imaging technique to be applied. We demonstrate the flexibility and performance of the system with results obtained by applying a variety of different imaging techniques on samples of fluorescent beads, Zebrafish embryos, and optically cleared whole mouse brain samples. Thus our approach allows easy implementation of advanced imaging techniques while retaining the simplicity of a cylindrical-lens-based light-sheet microscope. | physics.bio-ph | physics |
Multi-purpose SLM-light-sheet microscope
CHIARA GARBELLOTTO, JONATHAN M. TAYLOR∗
University of Glasgow, University Avenue, Glasgow, G12 8QQ, UK
*[email protected]
Abstract: By integrating a phase-only Spatial Light Modulator (SLM) into the illumination
arm of a cylindrical-lens-based Selective Plane Illumination Microscope (SPIM), we have
created a versatile system able to deliver high quality images by operating in a wide variety of
different imaging modalities. When placed in a Fourier plane, the SLM permits modulation of the
microscope's light-sheet to implement imaging techniques such as structured illumination, tiling,
pivoting, autofocusing and pencil beam scanning. Previous publications on dedicated microscope
setups have shown how these techniques can deliver improved image quality by rejecting out-of-
focus light (structured illumination and pencil beam scanning), reducing shadowing (light-sheet
pivoting), and obtaining a more uniform illumination by moving the highest-resolution region
of the light-sheet across the imaging Field of View (tiling). Our SLM-SPIM configuration is
easy to build and use, and has been designed to allow all of these techniques to be employed on
one optical setup compatible with the OpenSPIM design. It also offers the possibility to choose
between three different light-sheets, in thickness and height, which can be selected according to
the characteristics of the sample and the imaging technique to be applied. We demonstrate the
flexibility and performance of the system with results obtained by applying a variety of different
imaging techniques on samples of fluorescent beads, Zebrafish embryos, and optically cleared
whole mouse brain samples. Thus our approach allows easy implementation of advanced imaging
techniques while retaining the simplicity of a cylindrical-lens-based light-sheet microscope.
© 2018 Optical Society of America
OCIS codes: (180.2520) Fluorescence microscopy; (180.6900) Three-dimensional microscopy; (170.3880) Medical and
biological imaging; (230.6120) Spatial light modulators.
References and links
1.
J. Huisken, J. Swoger, F. Del Bene, J. Wittbrodt, E. H. K. Stelzer, "Optical sectioning deep inside live embryos by
selective plane illumination microscopy," Science 305(5686), 1007 -- 1009 (2004) .
2. S. Daetwyler, J. Huisken, "Fast Fluorescence Microscopy with Light Sheets," Biol. Bull. 231(1), 14 -- 25 (2016).
3. P. J. Keller, A. D. Schmidt, J. Wittbrodt, E. H. K. Stelzer. "Reconstruction of zebrafish early embryonic development
by scanned light sheet microscopy," Science 322(5904), 1065 -- 1069 (2008).
J. Huisken, D. Y. R. Stainier, "Even fluorescence excitation by multidirectional selective plane illumination microscopy
(mSPIM)," Opt. Lett. 32(17), 2608 -- 2610 (2007).
5. R. Itoh, J. R. Landry, S. S. Hamann, O. Solgaard, "Light sheet fluorescence microscopy using high-speed structured
4.
and pivoting illumination," Opt. Lett. 41(21), 5015 -- 5018 (2016).
6. O. E. Olarte, J. Licea-Rodriguez, J. A. Palero, E. J. Gualda, D. Artigas, J. Mayer, J. Swoger, J. Sharpe, I. Rocha-
Mendoza, R. Rangel-Rojo, and P. Loza-Alvarez, "Image formation by linear and nonlinear digital scanned light- sheet
fluorescence microscopy with Gaussian and Bessel beam profiles," Biomed. Opt. Express 3(7), 1492 -- 1505 (2012).
7. F. O. Fahrbach, A. Rohrbach, "Propagation stability of self-reconstructing Bessel beams enables contrast-enhanced
imaging in thick media," Nat. Commun. 3, 632 (2012).
8. T. A. Planchon, L. Gao, D. E. Milkie, M. W. Davidson, J. A. Galbraith, C. G. Galbraith, E. Betzig, "Rapid
three-dimensional isotropic imaging of living cells using Bessel beam plane illumination," Nat. Methods 8(5),
417 -- 423 (2011).
9. T. Vettenburg, H. I. C. Dalgarno, J. Nylk, C. Coll -- Lladó, D. E. K. Ferrier, T. Cižmár, F. J. Gunn-Moore, K. Dholakia,
"Light-sheet microscopy using an Airy beam," Nat. Methods 11(5), 541 -- 544 (2014).
10. E. Baumgart, U. Kubitscheck, "Scanned light sheet microscopy with confocal slit detection," Opt. Express 20(19),
21805 -- 21814 (2012).
Biomed. Opt. 15(1), 016027 (2010).
11. J. Mertz, J. Kim, "Scanning light-sheet microscopy in the whole mouse brain with HiLo background rejection," J.
12. T. Breuninger, K. Greger, E. H. K. Stelzer, "Lateral modulation boosts image quality in single plane illumination
fluorescence microscopy," Opt. Lett. 32(13), 1938 -- 1940 (2007).
13. L. Gao, "Extend the field of view of selective plan illumination microscopy by tiling the excitation light sheet," Opt.
Express 23(5), 6102 -- 6111 (2015).
14. D. Wilding, P. Pozzi, O. Soloviev, G. Vdovin, M. Verhaegen, "Adaptive illumination based on direct wavefront
sensing in a light-sheet fluorescence microscope," Opt. Express 24(22), 24896 -- 24906 (2016).
15. T. J. Mitchell, "Adaptive beam control and analysis in fluorescence microscopy," PhD thesis, University of Durham
16. R. Li, X. Zhou, D. Wu, T. Peng, "Selective plane illumination microscopy with structured illumination based on
spatial light modulators," Proc. SPIE 8949, 89491S (2014).
17. C. Garbellotto and J. M. Taylor, "SLM-SPIM data http://dx.doi.org/10.5525/gla.researchdata.648," (2018).
18. C. Maurer, A. Jesacher, S. Bernet, M. Ritsch-Marte, "What spatial light modulators can do for optical microscopy,"
Laser Photon. Rev. 5(1), 81 -- 101 (2011).
19. S. Quirin, D. S. Peterka, R. Yuste,"Instantaneous three-dimensional sensing using spatial light modulator illumination
with extended depth of field imaging," Opt. Express 21(13), 16007 -- 16021 (2013).
20. M. P. Lee, G. M. Gibson, R. Bowman, S. Bernet, D. B. Phillips, M. J. Padgett, "A multi-modal stereo microscope
based on a spatial light modulator," Opt. Express 21(14), 16541-16551 (2013).
21. P. Zammit, A. Harvey, G. Carles, "Extended depth-of-field imaging and ranging in a snapshot," Optica 1(4), 209 -- 2016
(2015).
(2014).
22. M. Aakhte, E. A. Akhlaghi, H. A. J. Müller, "SSPIM: a beam shaping toolbox for structured selective plane
illumination microscopy," Sci. Rep. 8(1), 10067 (2018).
23. P. G. Pitrone, J. Schindelin, L. Stuyvenberg, S. Preibisch, M. Weber, K. W. Eliceiri, J. Huisken, P. Toman-
cak,"OpenSPIM: an open-access light-sheet microscopy platform," Nat. Methods 10(7), 598 (2013).
24. K. Chung, J. Wallace, S. Y. Kim, S. Kalyanasundaram, A. S. Andalman, T. J. Davidson, J. J. Mirzabekov, K. A.
Zalocusky, J. Mattis, A. K. Denisin, S. Pak, H. Bernstein, C. Ramakrishnan, L. Grosenick, V. Gradinaru, K. Deisseroth,
" Structural and molecular interrogation of intact biological systems," Nature 497(7449), 332 (2013).
25. T. J. Schröter, S. B. Johnson, K. John, P. A. Santi, "Scanning thin-sheet laser imaging microscopy (sTSLIM) with
structured illumination and HiLo background rejection," Biomed. Opt. Express 3(1), 170 -- 177 (2012).
26. M. A. Neil, R. Juškaitis, T. Wilson, "Method of obtaining optical sectioning by using structured light in a conventional
microscope," Opt. Lett. 22(240), 1905 -- 1907 (1997).
27. M. A. Neil, R. Juškaitis, T. Wilson, "Real time 3D fluorescence microscopy by two beam interference illumination,"
Opt. Commun. 153(1 -- 3), 1 -- 4 (1998).
28. P. J. Keller, A. D. Schmidt, A. Santella, K. Khairy, Z. Bao, J. Wittbrodt, E. H. K. Stelzer, "Fast, high-contrast imaging
of animal development with scanned light sheet-based structured-illumination microscopy," Nat. Methods 7(8),
637 -- 642 (2010).
29. B. Judkewitz, C. Yang, "Axial standing-wave illumination frequency-domain imaging (SWIF)," Opt. Express 24(9),
11001 -- 11010 (2014).
30. Q. Fu, B. L. Martin, D. Q. Matus, L. Gao, "Imaging multicellular specimens with real-time optimized tiling light-sheet
selective plane illumination microscopy," Nat. Commun. 7, 11088 (2016).
31. K. N. Walker, R. K. Tyson, "Wavefront correction using a Fourier-based image sharpness metric," Proc. SPIE 7468,
32. C. Bourgenot, C. D. Saunter, J. M. Taylor, J. M. Girkin, G. D. Love, "3D adaptive optics in a light sheet microscope,"
74680O (2009).
Opt. Express 20(12), 13252 (2012).
Introduction
1.
Selective Plane Illumination Microscopy (SPIM) [1] is becoming increasingly popular for live
fluorescence imaging in developmental biology. Compared to confocal microscopy, which also
performs optical sectioning, the use of a static light-sheet to illuminate only the imaging focal
plane means that SPIM offers important advantages including rapid snapshot acquisition and
reduced photobleaching of the sample [2]. Snapshot acquisition is particularly important for
capturing the 3D structure of rapidly changing organisms or scenes; photobleaching is minimized
by the confinement of the illumination light to a single plane, thus ensuring that the only
fluorophores excited are those that lie in the plane currently being imaged. Despite these attributes
making SPIM particularly well-suited for in vivo imaging, a classical SPIM implementation
suffers from a number of issues including:
• Shadow artefacts: parts of the sample will absorb or scatter the side-launched light-sheet,
generating dark stripes behind them, elongated parallel to the illumination direction.
• Scattering: when illuminating a plane inside the sample, light emitted by the excited
fluorophores has to travel through sample tissue in order to be collected by the imaging
objective, which means it inevitably undergoes some scattering on the imaging path. This
results in undesirable out-of-focus background in the images, leading to reduced image
contrast. Tissue scattering also affects the propagation of the light-sheet itself, resulting in
even more out-of-focus light, in this case coming from out-of-focus fluorophores excited
by the scattered light-sheet.
• Limited field of view: even in the absence of scattering, the illumination delivered by a
Gaussian light-sheet is not uniform in thickness across the image Field of View (FoV). A
Gaussian beam generates a light-sheet with a certain waist size (thickness of the light-sheet
at its focus) and extent (Rayleigh length). This shape of the light-sheet results in an uneven
illumination across the image FoV, with better optical sectioning around the beam waist,
where the light-sheet is at its thinnest, and poorer optical sectioning (more out-of-focus
excitation) at the sides of the image, generated by the thicker parts of the sheet.
A variety of modifications to SPIM light-sheet microscopy have been proposed to tackle
some of these issues, and similarly for the closely-related technique of DSLM light-sheet
microscopy (Digital Scanned Laser Light-sheet Fluorescence Microscopy [3]), where a light-
sheet is synthesized by rapid scanning of a focused 2D Gaussian beam. Shadows can be reduced
by illuminating the sample from multiple directions [4, 5], and using Bessel [6 -- 8] or Airy [9]
beams instead of Gaussian beams permits a more uniform illumination across a larger FoV. One
way to reduce the effects of scattering from tissue surrounding the imaged plane is to use a
DSLM configuration (where a synthetic light-sheet is formed from a scanned Gaussian beam)
in conjunction with a rolling confocal slit on the detection camera to reject scattered light [10].
Methods based on structured illumination, such as HiLo [11] and the method proposed in [12]
which we refer to as the 3-phase method, can also help enhance image contrast by reducing the
out-of-focus contribution. The technique known as tiling can be used to extend the limited FoV
over which high-quality depth sectioning can be achieved using a simple Gaussian beam [13].
In this case each plane in the sample is imaged multiple times, each time with the light-sheet
focused at a different lateral position in the FoV. The final image of the plane is then created by
stitching together adjacent vertical stripes taken from the different images, each stripe containing
only the part of the image generated by the thinnest part of the light-sheet.
The aim of the present paper is to present a single SPIM system that uses a spatial light
modulator (SLM) to create a simple and flexible platform for performing the aforementioned
advanced light-sheet microscopy techniques, as well as permitting automatic adaptive light-sheet
positioning for best focus (autofocusing). Our setup consists of a basic SPIM microscope
(illumination arm with cylindrical lens, imaging chamber with sample holder and translation
stage, imaging arm) with the addition of a phase-modulating SLM in the illumination arm.
With the SLM conjugated to a Fourier plane in the optical path of the illumination beam (SLM
illuminated by a collimated light beam), we demonstrate that it is possible to modulate and move
the resultant light-sheet to implement a wide range of imaging techniques without having to
move any mechanical part in the microscope. While SLMs have previously been used for certain
specific purposes in light-sheet illumination [9,14 -- 16], we will show that a single microscope
design is well-suited for applying a wide variety of different optical techniques on the same
imaging platform.
This work is organized as follows. In the next section we present the optical design of our
SLM-SPIM microscope, discuss the motivation behind our design choices and give details about
the different samples used in our experiments and the way they were mounted for imaging. In
Section 3 we introduce some of the imaging techniques that can be performed with our system, and
include experimental results obtained imaging fluorescent beads in agarose, Zebrafish embryos,
and cleared mouse brain samples. All the raw data and the MATLAB codes used to produce
figures, plots and calculations presented in Section 3 can be found in the data repository [17].
2. SLM-SPIM microscope
SLMs are versatile pixellated phase (or amplitude) modulation devices that have already found
many applications in the field of optical microscopy [18], where they have been used to control the
sample illumination [19] or as a Fourier mask in the imaging path [20,21]. SLMs have previously
been described by a few authors in SPIM systems, both in the illumination and in the imaging
path (for example to deliver structured illumination [16] or to correct for aberrations [14]).
Recent work in parallel with our own has presented the SSPIM (Structured SPIM [22]), a
DSLM microscope with an incorporated SLM which offers the option to shape the beam profile
(Gaussian, Airy, Bessel, Lattice) as well as performing tiling and structured illumination. In
contrast, our work showcases the use of a programmable SLM device in a more classical scan-free,
cylindrical-lens-based SPIM such as that used for the OpenSPIM design [23].
In our SPIM we use a reflective phase-only liquid crystal SLM, placed in a Fourier plane
in the optical path of the illumination beam. The optical setup is illustrated in Figure 1, while
details of the components and devices used are summarized in Table 1. The laser beam coming
out of the optical fiber is collimated, expanded (to fill the size of the SLM active surface), and
directed onto the SLM, which modulates the phase of the beam's wavefronts. Two lenses and a
cylindrical lens are then used to create the light-sheet and conjugate the plane of the SLM to the
back focal plane of the illumination objective. This conjugation means that what the objective
focuses (in one direction) on the sample is the far field diffraction pattern corresponding to the
phase-shift pattern displayed on the SLM, i.e. its Fourier transform. A second system optimized
for imaging cleared mouse organs (glycerol/CLARITY) shares common laser launch optics and
SLM with our water-immersion system, but has a separate final part of the illumination arm
(last spherical lens, cylindrical lens and launching objective), glycerol chamber, and a vertically
mounted imaging arm (components included in Table 1). The patterns displayed on the SLM
can easily be rotated and customized for either one of the systems, permitting us to perform the
different imaging techniques on both water- and glycerol-immersed samples.
As illustrated in Figure 1, our water-immersion system also allows easy switching between
three different optical configurations on the laser launch path. Each of these interchangeable
modules has a different positioning of the cylindrical lens and final spherical lens. This allows us
to choose between three light-sheets of different thickness and height, as well as offering different
conjugations between the SLM and the objective lens (discussed in detail below). Experimental
profiling of the light-sheet (using a small fluorescent bead translated through the light-sheet) has
confirmed a FWHM (at sheet waist) of around 2 µm for setup 1, 3 µm for setup 2 and 5 µm for
setup 3. The sheet height is ∼4 mm for setup 1, ∼2 mm for setup 2, and ∼0.6 mm for setup 3.
2.1. Design considerations
In the light-sheet launch path, it is necessary to place the cylindrical lens after the SLM. Otherwise,
if the SLM is appropriately conjugated to a pupil or image plane, a line focus will be formed
on the SLM and no meaningful phase control would be available along one axis. However, it
follows from this choice that it is not possible for the SLM to be simultaneously conjugate to the
pupil plane in both the horizontal and vertical axes simultaneously (and similarly, neither can it
be simultaneously conjugate to the object plane in both axes). These considerations mean that
care is required in designing the light-sheet launch optics to ensure that the SLM provides the
necessary degrees of freedom to manipulate the light-sheet as required for an experiment.
The three interchangeable lens modules mentioned above offer different conjugations between
the plane of the SLM and the focal plane of the illumination objective, each of which is optimal
for different families of applied beam shaping techniques. In a top-down view of our system
design (Figure 1a), it can be seen that setups 1 conjugates the SLM with the back focal plane
of the illumination objective: the two lenses L3 and L4 are separated by the sum of their focal
lengths, the SLM is at f3 from L3 and the back focal plane of the objective is at f4 from L4.
Fig. 1: Optical scheme of our SLM-SPIM, with a top view of the system in (a) and a side view of
its launching arm in (b) (see Table 1 for details of the individual components). Changing the
position of the last two lenses before the illumination objective allows us to switch between three
different setups, yielding different sheet heights and thicknesses, and changing the conjugation
of the SLM with the sample plane. (a) View of the SLM-SPIM from above. The cylindrical
lens has no optical power in this plane. A mirror placed before the SLM (bottom right corner)
permits adjustment of the vertical position of the light-sheet in the sample plane. A second mirror
can easily be inserted after L3 and used to redirect the laser beam to the side (upwards in this
figure), onto a second illumination arm, (not included in this scheme) ending in the glycerol
chamber. The glycerol illumination arm consists of (see Table 1 for details): L3 (shared with the
water-imaging system) ←160mm→ CL2 ←40mm→ L6 ←100mm→ obj3 → glycerol chamber.
The glycerol imaging arm is mounted vertically above the glycerol chamber and is composed of a
glycerol dipping objective (obj4), a tube lens (L7) and the same camera used for the images in
water. (b) Side view of the final part of the SLM-SPIM illumination arm, with three different
possible configurations.
BS L3 Setup 2: Setup 3: X Y Z Setup 1: obj1 BS polarizer L2 mirror L1 L3 L4 CL obj1 obj2 filter L5 L4 CL Setup 2: Setup 3: sample SLM L0 fibre L4 CL 100mm 50mm 50mm 50mm 50mm 20mm 100mm X Y Z Setup 1: water camera mirror chamber (a) (b) L4 CL L4 CL L4 CL Lenses L0: achromatic doublet (Thorlabs, AC064-013-A-ML), f = 13 mm;
L1: plano-convex, f = 35 mm; L2: plano-convex, f = 100 mm;
L3: plano-convex, f = 100 mm; L4: plano-convex, f = 100 mm
for setup1, f = 50 mm for setup 2 and f = 25.4 mm for setup 3;
CL: cylindrical lens, f = 50 mm for setup 1 and 2, f = 80 mm for
setup 3; L5: plano-convex, f = 100 mm;
Other optical elements BS: beam splitter; polarizer: linear polarizer;filter: Green Fluo-
rescent Protein (GFP) filter (central wavelength 525 nm); obj1:
10× Nikon Plan Fluorite Imaging Objective, 0.3 NA, 16 mm WD
(Working Distance); obj2: 40× Nikon CFI APO NIR Objective,
0.80 NA, 3.5mm WD;
SLM head
SLM controller
SLM Hamamatsu LCOS-SLM (Liquid Crystal on Silicon Spatial Light
Modulator) serie X13138;
pixels: 1272 × 1024; pixel size: 12.5 µm; effective area size: 15.9
mm × 12.8 mm; fill factor: 96 %;
input signal: DVI-D; DVI signal format (pixels): 1280 × 1024;
input signal levels: 256; DVI used frame rate: 60Hz;
Laser OBIS coherent laser; wavelength: 488 nm;
Camera XIMEA MD028xU-SY; sensor active area: 8.8 mm × 6.6 mm;
resolution: 1934 × 1456, 2.8 Mp; pixel size: 4.54 µm; frame rate:
56.9 fps; dynamic range: 71.1 dB;
Glycerol system Illumination arm, after the SLM, L3 and the pop-in mirror: CL2:
cylindrical lens, f = 60 mm; L6: plano-convex, f = 100 mm; obj3:
5× ZEISS EC Plan-Neofluar Objective, 0.16 NA, 18.5mm WD.
Imaging arm: L7: plano-convex, f = 150 mm; obj4: 20× ZEISS
Clr Plan-Neofluar Objective, 1.0 NA, 5.6mm WD;
Table 1: List of components used (with reference to Figure 1).
However, the optical power of the cylindrical lens means that, when viewing the system from
the side (Figure 1b), the SLM is not conjugate to either a pupil or image plane. In contrast,
setup 2 gives the opposite situation and, in its side view, it conjugates the plane of the SLM
with the focal plane of the launching objective: L4 and the cylindrical lens are separated by the
sum of their focal lengths and the back focal plane of the objective is at fcl from the cylindrical
lens. Setup 3 gives a conjugation similar to the one of setup 2, but was designed to give a good
compromise between a perfect conjugation (achievable with the use of a cylindrical lens with
fcl = 75 mm, instead of the fcl = 80 mm we decided to use for this setup) and the combination of
high flexibility in tilting the light-sheet and a high demagnification of the incoming beam (both
of which are achieved using a longer fcl), resulting in a light-sheet with thicker waist and longer
Rayleigh length (see below for a more detailed discussion about these choices).
Most of the imaging techniques we performed on our system can be achieved using any of the
three setups, but because of the different sheet height, sheet thickness and SLM conjugation they
provide, different setups would be the preferred choice for different imaging techniques. Setup 1
gives the thinnest light-sheet at beam waist, and is therefore the best one to use for experiments
such as the light-sheet tiling ones (Section 3.2), where the only part of the light-sheet which
is actually used to create the final image is its central, thin waist. The conjugation of the SLM
with the focal plane of the launching objective generally makes setup 2 the most appropriate
for shadow suppression experiments (Section 3.3): the light-sheet is in this case tilted in the
sample plane, and the perfect conjugation of the SLM with the center of the FoV assures that the
light-sheet does not shift vertically while being tilted (which would result in a change in image
brightness dependent on the light-sheet tilt). Setup 3 gives a thicker light-sheet, which on the
other hand also means a more even illumination across the FoV (longer Rayleigh length). This
makes it a good choice for experiments such as the pencil beam scanning in Section 3.4, where
the light-sheet illumination is substituted with a vertically scanned focused beam. The use of a
thicker but more uniform beam also helps reduce the time needed to generate a homogeneously
illuminated image of the entire FoV.
The choice of what setup to use for a particular experiment should also depend on the
characteristic of the sample to be imaged. For the experiments which we could perform with
more than one of the three setups, we chose to use Setup 1, which is the one that we find gives the
best light-sheet for imaging fluorescent beads and Zebrafish embryos, with a vertically uniform
illumination across the used FoV and a 2 µm thick light-sheet waist. In Section 3.3 (shadow
suppression experiments) we give an example of a situation in which the type of sample strongly
influences the setup choice, further demonstrating the advantages of working with an easily
reconfigurable system.
In our experiments, we use the SLM in a refractive mode (applying blazed gratings). This
means that most of the light modulated by the SLM is concentrated in the first diffracted order,
but some power is always lost in higher orders and in the so-called 0th order (containing the
specularly-reflected light that is not modulated by the SLM). In order to eliminate the 0th order
we apply a further, constant phase ramp to the SLM, to displace the desired first order from the
0th order. The SLM is tilted such that the first order lies on the optical axis, and the 0th (and
higher) orders can then be blocked by a mask placed in the first focal plane after the SLM. When
we apply different phase patterns to the SLM, the trajectory and structure of the first order beam
is altered slightly, but it is still allowed to pass by the mask.
2.2. Samples
For our experiments with the water objective, the samples where mounted in a length of FEP
tubing (Fluorinated Ethylene Propylene, 1.3 mm ID × 1.6mm OD, Adtech Polymer Engineering
Ltd) attached to the end of a syringe. For calibration and alignment purposes, and for the tiling
experiments, the FEP tube was filled with beads (polystyrene beads, 0.2 µm Dragon Green, Bangs
Laboratories Inc) in a 1.5 % low melting point agarose solution (Agarose, High-EEO/Protein
Electrophoresis Grade, Fisher Scientific). When imaging ex-vivo Zebrafish embryos, the fish
was placed in a FEP tube filled with system water. For our experiments we used the Zebrafish
nacre mutant (reduced pigment), with the cardiac muscle labelled with green fluorescent protein
(myl7:eGFP). The fish specimen was preserved in formalin (10% Formalin solution, neutral
buffered, Sigma-Aldrich). For the experiments on cleared mouse samples, we used a mouse brain
prepared with the CLARITY method [24]. The sample was placed in a quartz cuvette (UV fused
quartz glass, Thorlabs part CV10Q3500F) filled with 85% glycerol, which was sealed while
excluding any air bubble from inside. The cuvette was then immersed in the chamber filled with
85% glycerol, and held horizontally underneath the dipping imaging objective.
3. Results and analysis
3.1. Structured Illumination
Two mutually-coherent light-sheets that propagate in the same plane (x y plane), but at a different
angle with respect to the optical axis of the illumination arm, will interfere in the sample
(cid:104)(I1 − I2)2 + (I1 − I3)2 + (I2 − I3)2(cid:105)1/2
√
2
3
plane and generate a light-sheet with sinusoidal modulation along y (Figure 2b). This type of
intensity-modulated light-sheet can be used to perform two different structured illumination
techniques: HiLo [11,25], and the 3-phase method [26,27]. In the 3-phase method, three images
are taken by translating the same sinusoidal light-sheet by precise spatial phase shifts of: 0, 2/3π
and 4/3π. The three images I1,2,3 are then combined using the following formula:
,
I =
(1)
to yield a final image I with reduced out-of-focus background and hence enhanced contrast.
This 3-phase technique has been implemented on a digitally-scanned light-sheet microscope
by fast time-modulation of the scanned beam [28], and in 2007 on a SPIM microscope using
a grid-projection approach [12]. HiLo imaging has also been implemented previously on a
digitally-scanned light-sheet microscope; it requires only two raw images, one acquired with a
normal, uniform light-sheet and one with a modulated light-sheet [11].
To generate the two interfering sheets, we display two opposite sawtooth patterns simultaneously
on the top and bottom halves of the SLM, as in Figure 2a. As the beam diffracts off the SLM it is
thereby split into two half-beams, propagating in the image plane (x y plane) but at two opposite
angles α and −α from the optical axis of the launching arm. When the two half-beams meet
again and interfere in the sample plane, they generate the desired sinusoidal pattern (Figure 2b).
The period of the final illumination pattern is defined by the propagation angle, which is in turn
determined by the sawtooth period on the SLM. In order to shift the illumination pattern by a
desired phase (to perform the 3-phase method), the equivalent optical phase shift can simply
be added to one of the two halves of the SLM. Figure 3 shows results from the application of
this structured illumination technique to image the heart of a formalin-fixed Zebrafish embryo
specimen (4 days post-fertilization) expressing GFP fluorescence.
To quantify the improvement in image contrast obtained as a result of the out-of-focus back-
ground reduction, we calculated the standard deviation of the energy-normalized histograms [28]
of the structured illumination (Figure 3c) and the normal light-sheet images (Figure 3b). This
standard deviation can be calculated using the following formula:
(cid:118)(cid:117)(cid:117)(cid:117)(cid:117)(cid:116)
(cid:19)2
(cid:18) Ii − ¯I Ii
i
,
σN =
C − 1
(2)
where C is the total number of pixels in the image, i ranges from 1 to C, Ii is the intensity value
of the i-th pixel, Ii is the sum of all the pixels values in the image, and ¯I = Ii/C is the mean
intensity value of the image. As more extensively explained in [28], the ratio between the σN
values of two images can be used to quantify the change in image contrast, with an higher σN
value corresponding to a better image contrast. For the images shown in Figure 3b and 3c we
calculated this ratio to be σN(c)/σN(b) ∼ 2.6.
It should be noted that our approach to generate the two interfering light-sheets cannot be used
with setup 2 and 3. In fact, in the case of perfect conjugation between the SLM and the waist of
the light-sheet (or almost perfect conjugation with setup 3) the desired interference appears only
on one side of the imaging FoV. Instead, the mis-conjugation offered by the side view of setup 1
moves the edge of the interference region to the side, allowing the two half-beams to interfere
across the whole FoV. A possible future alternative could be to develop an approach analogous to
the one used in [29]. The SLM would be divided into vertical stripes instead of into two halves,
and the two blazed gratings displayed on alternate stripes. This method would allow structured
illumination experiments to be performed with all our three setups, but it would give rise to extra
diffraction orders which would reduce efficiency, increase out-of-focus excitation and bleaching,
and might therefore require masking out.
Fig. 2: (a) Pattern displayed on the active area of the SLM to create a sinusoidally modulated
light-sheet; the blue circle indicates the footprint of the collimated input beam. (b) Experimental
image of a modulated light-sheet obtained with our method, imaged in aqueous Fluorescein
dye diluted in water, revealing an interference pattern with a period of ∼10µm (white to white);
scale bar: 50 µm. (c) Schematic showing the optical path (distances and sizes not to scale) of the
light reflected off the SLM. The two beams follow the same optical path when viewed in the xz
plane, forming two light-sheets on the same plane. (d) In the xy plane, the two sheets propagate
at different angles, generating an interference pattern in the sample plane.
Fig. 3: 3-phase structured illumination performed using setup 1. (a) Cropped views of the three
images, I1,2,3, taken with a modulated light-sheet (period = 20 µm on the sample plane). The
sinusoidal pattern for the second and third images is shifted by a phase of 2
3 π with
respect to the first image. (b) Image acquired with a normal, non-modulated light-sheet. (c) Image
obtained by combining I1, I2 and I3 using Equation 1. (d) Intensity profile along the same row in
images (b) and (c), to visualize the achieved background reduction and improved image contrast
(values normalized to the global maximum of the two plotted lines). Scale bars: 50 µm.
3 π and 4
SLMView from above:SLML3 L4 CL Obj. 1 ZXBSYXBSSLMYX(a)(c)View from the side:Structured sheet(b)(d)abcI 1I 2I 3IMicronsdNormalized intensity 3.2. Tiling
Tiling is a technique proposed to work around the trade-off between light-sheet thickness and
length [13,30]. The ideal light-sheet would stay thin across the whole imaging FoV. Light-sheets
created by a focused Gaussian beam only remain thin over the beam's Rayleigh length, and thus
thinner sheets diffract and spread more rapidly as they propagate. To mimic the illumination
delivered by an ideal thin and long light-sheet, one can use the waist of a short, thin sheet to
image only one part of the FoV, then move the sheet's waist laterally and build up, step by step,
an image of the entire FoV. In our case, the light-sheet can be focused at different distances from
the launching objective by displaying a Fresnel lens pattern on the SLM. Images acquired by
illuminating the same plane inside the sample but changing the focus of the light-sheet (along its
propagation direction, i.e. in the x direction in the imaging coordinate system) are then combined
to obtain a final image. Of each of the initial images, only a restricted vertical stripe generated by
the thin part of the light-sheet is allowed to contribute to the final image. Results obtained using
this technique imaging fluorescent beads can be seen in Figure 4. For the tiling experiments we
chose to use setup 1, which is the one that gives the thinnest light-sheet (best z resolution, but
smallest effective FoV in x).
Fig. 4: Tiling technique demonstrated with 0.2 µm fluorescent beads. We acquired eight images of
the same plane of beads, each image with the light-sheet focused at a different position (laterally)
in the FoV. Left: (a,b,c) Same horizontal stripe taken from the second, fifth and eighth of the
eight images taken (only three images shown for sake of clarity): in each image the position of
the sheet waist (indicated by the yellow arrowheads) can be recognized by the brightness of the
beads and the reduced number of out-of-focus beads. (d) Image obtained tiling the eight images,
i.e. retaining only the sharpest vertical stripe taken from each of them. Right: zooms on beads
taken from three different lateral positions (1,2,3) in the images, with each row showing how
the same beads appear in the corresponding image to the left. Each of these zoomed images has
been normalized to its own maximum value for clarity. Notice how, looking at one column at a
time (i.e. the same sets of beads), as the position of the sheet's waist gets further away from the
position of the beads, the relative amount of light illuminating out-of-focus features increases.
The composite tiled image (d) gives the best optical sectioning (most of the light concentrated on
in-focus features) throughout the entire FoV.
1231 2 3abcd 3.3. Multi-angle illumination for shadow suppression
To be imaged using SPIM, a sample first of all needs to be transparent enough for the light-sheet to
propagate through it. Parts of the sample that strongly scatter or absorb the excitation light create
a visible shadow behind them in the "downstream" direction of beam propagation (in our case to
their right, since the light-sheet illuminates the sample from the left). This shadow effect can be
reduced by combining illumination generated by light-sheets propagating at different angles [4,5].
In both these previous works the illumination angle was modulated at high speed (> 1 kHz) to
provide a range of illumination angles within a single image exposure. In our system (using
a slower SLM) light-sheets with different propagation directions can be created by displaying
different sawtooth patterns on the SLM. A shadow-free image can be obtained by switching
between the different light-sheets within the exposure time of a single image, or by recording
one image for each light-sheet inclination and combining the different images afterwards. This
second approach, despite being less efficient in terms of image acquisition/computing time, offers
the flexibility to permit what we propose as an alternative and improved algorithm for combining
the acquired images to obtain shadow suppression: Maximum Intensity Projection (MIP). When
the light-sheet is tilted through a range of different angles within a single image exposure, the
resulting shadow-free image is generated from the sum of all the fluorescence excited by each
light-sheet inclination. This technique can be replicated by computing the sum of a set of images
acquired each with the light-sheet propagating at a different inclination. An alternative way of
combining these images is to compute the Maximum Intensity Projection (MIP) of the whole
stack: for each pixel compare the values assigned to that pixel on each image and only keep
the maximum one. This results in a final image where each pixel takes on the value from the
raw image where that region was experiencing minimal shadowing. As can be seen in Figure
5, the image obtained using MIP not only preserves a better image contrast when compared to
the one obtained by averaging, a but it also assures a more accurate representation of the true
intensity profile across the image. In fact, computing the average of a set of images acquired with
the light-sheet propagating at different angles results in an alteration of the true image intensity
profile: parts of the sample which are well-illuminated in all the images (i.e. are not affected
by shadows) are inherently seen as brighter than those that are only illuminated in some of the
images. Using MIP on the other hand ensures that the final intensity of each part of the sample
only depends on the intensity observed when it is illuminated without obstruction, and not on the
number of images which agree with that value. In Figure 5 we compare a normal light-sheet
image and the two alternative shadow suppression algorithms, averaging and MIP. A zoom-in
on a region strongly affected by shadows shows the shadow suppression results obtained by
combining seventeen images acquired with the light-sheet propagating at different angles, equally
spaced within ±8 degrees. Before being combined either with MIP or averaging, the seventeen
images were properly rescaled to account for the diffraction efficiency of the SLM, which changes
with the angle of propagation of the diffracted first order, with a higher tilt corresponding to
a dimmer light-sheet. We selected an horizontal line across a region of the sample which is
not affected by any shadows in the normal light-sheet image (green line in Figure 5(d)), and
compared its intensity profile with the intensity profile of this same line in the two images
computed for shadow suppression. This plot helps viualising the reduced image contrast offered
by the averaging technique, and also its unpredictable distortion of the original intensity profile.
Because of the conjugation of the SLM plane with the center of the FoV, setup 2 is the most
appropriate setup to be used for the shadow suppression experiments, provided it can deliver
a high enough tilt angle with high diffraction efficiency. As the SLM is used to send the first
diffraction order to different directions, it generates light-sheets that propagate at different angles
but overlap entirely again on each plane conjugate to the plane of the SLM, as they do on the
SLM itself. In the case of setup 2, as the light-sheet is tilted using the SLM, its rotation in the
sample plane happens around the center of the FoV, which is in fact conjugated to the plane
Fig. 5: Shadow suppression using the light-sheet pivoting technique. (a) Image of a formalin-
preserved Zebrafish embryo heart (4 dpf) acquired with a normal light-sheet, using setup 3. (b)
Image obtained by computing the Maximum Intensity Projection (MIP) of a stack of seventeen
images, acquired with the light-sheet propagating at different angles, equally spaced within ±8
degrees. (c) Image obtained by averaging the same seventeen images used for (b). (d-f) Zoomed-in
views of the dashed line boxes in images (a-c). Each of these images has been normalized to its
own maximum value. (g) Intensity profile of the same horizontal line in images (d-f). This plot
shows how the MIP allows to preserve the original image contrast and, with respect to averaging,
a more accurate representation of the true intensity profile: notice how averaging (blue dashed
line) distorts the relative intensity of the two peaks indicated by the black arrowheads, making
the left peak appear as brighter than the one on the right. Intensity values in these plots are
normalized to the global maximum of the three plotted lines. Scale bars: 50 µm.
of SLM. In setup 1 the situation is different: the rotation of the light-sheet happens around a
position that is to the side of the FoV, such that a tilt also corresponds to an undesired vertical
shift of the light-sheet in the images. As the tilt angle increases, the bright central part of the
Gaussian sheet shifts away (vertically) from the FoV, which is illuminated by less and less light.
One other thing to keep in mind is that for shadow suppression the best results are obtained by
using a large range of angles for the incoming light-sheets. In our system, a limit to the maximum
achievable tilt angle is set by the SLM pixel spacing. By considering the magnification of the relay
optics within our microscope, we can find the relation between the tilt angle of the light-sheet in
ab+ 8-8Maximum Intensity ProjectionAverageNormal imageabcMicronsNormalized intensitydefgdef the sample plane and the angle at which it originally propagates as it is diffracted off the SLM:
sin(θ2) =
sin(θ),
f1 f3
f2 f4
(3)
where θ is the propagation angle after the SLM, θ2 is the propagation angle in the sample plane,
and f1, f2, f3 and f4 are the focal lengths of the four lenses through which the beam passes before
reaching the sample plane (see Figure 1b), with f4 being the focal length associated with the
launch objective. Finally, we can approximate sin(θ) with λ/p and obtain:
sin(θ2) =
f1 f3
f2 f4
λ
p,
(4)
where λ is the wavelength (in our case 488 nm) and p is the pixel period of the sawtooth pattern
displayed on the SLM: the smaller the period of the sawtooth pattern on the SLM, the larger the
angle at which the light-sheet generated by the first diffracted order propagates. On the other hand,
decreasing the sawtooth pattern period size also reduces the number of SLM pixels used for each
period, and this coarser approximation of the ideal pattern results in a less efficient concentration
of the diffracted light in the first diffracted order, i.e. the resultant light-sheet is dimmer. Even
for larger pixel periods, this varying diffraction efficiency means that larger-angle sheets will be
somewhat dimmer, and prior to further processing we rescale each image to compensate for the
different brightnesses of the light-sheets they have been generated with.
In order to maintain a sufficiently bright first order, we required a minimum period of four SLM
pixels. Using Equation 4 we can verify that a sawtooth period of four SLM pixels corresponds,
with setup 2, to a tilt angle of ∼ 3◦. For our experiments on the embryonic Zebrafish heart, we
found that this maximum tilt angle was insufficient for good shadow suppression. We therefore
decided to perform these experiments with setup 3, which gives adequate conjugation between
the SLM plane and the center of the imaging FoV but offers a much bigger range of possible tilt
angles, with a 50 µm period (four SLM pixels) corresponding to a tilt angle of ∼ 8.8◦. Figure
5 illustrates the results obtained combining seventeen images acquired with setup 3, with the
light-sheet propagating at different angles, equally spaced within ±8◦, imaging the heart of an
ex-vivo 4 dpf Zebrafish embryo.
3.4. Pencil beam scanning (synthetic DSLM)
Part of the fluorescence excited in the illuminated plane undergoes scattering by the intervening
tissue before reaching the detector, resulting in an increased diffused background signal and a loss
of contrast. In DSLM light-sheet systems, where the light-sheet is formed by a rapidly-scanned
2D Gaussian beam (a "pencil" beam), the beam scan can be combined with a confocal rolling
shutter on the camera to suppress the background signal [10]. e can replicate this same approach
by recording a sequence of full-frame images as the pencil beam is scanned across the FoV, and
then create the final image by applying a synthetic confocal slit to each raw image (i.e. masking
out all rows except those where the pencil beam should have appeared). This post-acquisition
confocal slit method is implemented for example in [7], where each image is multiplied by a
smooth Gaussian mask centered and aligned with the illumination beam. In order for the mask to
correctly select the desired part of each image, the position and inclination of the illumination
beam needs to be known, either from previous calibration or from the acquired images. In fact,
both implementations of the confocal slit technique (rolling shutter, and post-acquisition masking)
require precise alignment and size-matching between the illumination beam and the detection
line used.
Our system enables us to implement the post-acquisition confocal slit method, and we here
propose an alternative way to process the raw images and achieve background rejection, obtaining
a technique that combines the ease of a standard, full-frame acquisition with a simple and
calibration-free postprocessing procedure. We scan the illumination beam across the entire FoV,
acquiring a single full image for each position of the beam. We then combine these images into a
3D stack of size n × m × i, where n × m is the image size in pixels and i is the number of images
taken. The final contrast-enhanced image can be obtained by simply computing the Maximum
Intensity Projection of the stack of images along its third dimension (dimension of size i).
To explain and justify this procedure, let us concentrate on how the intensity value of a single
pixel changes as we scan through the stack of images. Let (x,y) be the position of the pixel in
the image, and (xs,ys) its corresponding location in the sample plane, and assume we expect to
detect some non-scattered signal from (xs,ys). The value of pixel (x,y) will be very low in the
images taken with the pencil beam positioned far away from (xs,ys), somewhat higher as the
beam gets closer to it and more scattered light reaches pixel (x,y), and it will be at its highest
when (xs,ys) is directly in the path of the incoming beam. The Maximum Intensity Projection
therefore gives, for each pixel (x,y), the intensity received by it in the image recorded with the
illumination beam giving best overlap with (xs,ys), the position in the sample that maps onto that
pixel. And this is exactly what we wish to retain in our reconstructed image.
Acquiring entire images but then only retaining some pixel rows from each image is of
course a relatively slow and inefficient way of implementing the pencil beam scanning technique.
However, this shows how our flexible imaging system can easily be used to explore the feasibility
of new imaging modalities, obtaining good pilot results without investing the time and effort
Fig. 6: Pencil beam scanning technique applied on a cleared whole mouse brain sample, imaged
in glycerol. (a) Image acquired with a normal light-sheet. (b) Image generated with the scanning
pencil beam technique, using our reconstruction procedure (based on a Maximum Intensity
Projection) on a set of 200 raw images, each taken with the horizontal pencil beam focused at a
different height in the sample plane. Scale bars: 50 µm. (c,d) Zoomed-in views of the dashed
line rectangles in images (a,b), to highlight some of the faint features (left arrowhead) and fine
structures (right arrowhead) revealed by the pencil beam scanning technique. Scale bars are here
of 20 µm. (e) Cross section of the normalized intensity along the same column in images (a) and
(b) (values normalized to the global maximum of the two plotted lines).
abacdMicronseNormalized intensity needed to build and calibrate a high-speed dedicated rolling-confocal-slit DSLM system. Our
post-processing approach based on a simple Maximum Intensity Projection also has the benefit
of not requiring any calibration of the pencil beam position and orientation on each image plane.
In our implementation the sample is illuminated by a regular 2D-focused beam, which we
generate by displaying an opposing cylindrical lens phase function on the SLM (note that the
same could be achieved by physical removal of the cylindrical lens from the system). The beam
is scanned using the SLM in order to sequentially illuminate, line by line, the entire in-focus
plane. The phase function applied to our SLM is the sum of the cylindrical lens pattern and a
linear phase ramp. In practice the combination of these two functions yields a cylindrical phase
function that translates vertically on the SLM as the pencil beam is scanned (or laterally on the
SLM, in case of our experiments with the glycerol setup).
The improved contrast achievable using this technique is particularly valuable when imaging
deep in highly scattering samples. To demonstrate this we performed the pencil beam scanning
technique on cleared whole mouse brain samples, with results shown in Figure 6. For this
experiments we programmed the SLM to make the pencil beam translate with steps of 1.4 µm
in the sample plane (corresponding to ∼6 pixels in the image, with the pencil beam having a
FWHM of ∼ 25 pixels). A total of 200 images were taken to cover the region of interest shown in
Figure 6 (1146×1556 pixels).
3.5. Autofocusing
By displaying a horizontal phase ramp on the SLM, the light-sheet can be moved in z, i.e. towards
and away from the imaging objective. This offers a natural method for optimizing the position of
the light-sheet to coincide with the focal plane of the camera, without having to move the imaging
objective or the tube lens. Particularly for high-numerical-aperture imaging in thick samples,
this optimization must often be performed on a per-sample basis, and even when moving to a
different location in the sample.
We developed a MATLAB script to automatically optimize the light-sheet position to the
plane of best focus. First, the light-sheet is scanned over a certain range in z (chosen by the user)
around its rest position (flat SLM), recording one image for each position of the light-sheet. The
images are analyzed and the light-sheet is moved to the position that yielded the image with best
focus. To evaluate the quality of the focus of each image, we use the sharpness metric proposed
in [31] and used for adaptive optics on a SPIM in [32]. This metric quantifies the image focus
through a measure of the ratio between the high and low spatial frequency content of the image,
and is defined as follows:
,
(5)
Np
F[I(x, y)] masked
F[I(x, y)] unmasked
S =
Np
where I(x, y) is the intensity of pixel (x,y), Np is the number of pixels in the image, and F denotes
the Fourier transform. A rectangular mask is applied to the 2D power spectral density (PSD) of
the original image (F[I(x, y)]), in order to mask out its central values, representing the lowest
spatial frequencies contained in the image. The sharpness value S is then given by the sum of the
absolute values of F[I]masked divided by sum of the absolute values of F[I]unmasked (0 < S < 1).
As the images become more blurred (moving away from the plane of best focus), their low spatial
frequency content increases with respect to the high frequency content, which means that the
mask that subtracts the lowest frequencies has a stronger effect on F[I], resulting in a lower value
for S. The maximum S value identifies the image with best focus (see sharpness plot in Figure 7).
In the experiment presented in Figure 7 the sharpness metric was calculated using a band-pass
mask with a cut-on spatial frequency (in the sample plane) of 0.009 µm−1(11×9 pixels) and a
cut-off spatial frequency of 0.2275 µm−1 (201×151 pixels).
Fig. 7: Autofocusing experiment using sharpness metric of Equation 5 on an ex-vivo 4 dpf
Zebrafish embryo's heart (using setup 1). In this illustration, eighty-one images were taken using
the SLM to move the sheet to different positions with respect to the imaging objective (with steps
of 0.5 µm, for a total range of 40 µm). (a) and (c) show images taken with the light-sheet in an
out-of-focus plane, while (b) is the image that the sharpness measurements identified as the one
with best focus (i.e. light-sheet at the correct distance from the imaging objective). (d) Sharpness
values, one for each image, with highest value indicating the plane of best focus. Note that in
practice a smaller number of images would be taken, and the sharpness interpolated using an
appropriate function, to quickly find the optimum focus. Scale bars: 50 µm.
4. Conclusion
We have shown how a phase-only liquid crystal SLM can be introduced as a simple modification
to the illumination arm of a SPIM, to give a flexible, versatile system able to deliver high quality
images by applying a range of advanced light-sheet imaging techniques. Imaging fluorescent
beads, Zebrafish embryos and optically cleared whole mouse brain samples, we have demonstrated
how the SLM-SPIM can be used to apply: structured illumination and pencil beam scanning
techniques to reduce the out-of-focus content of the images; light-sheet pivoting to reduce the
effect of shadows; light-sheet tiling to obtain a more uniform illumination across the image FoV
and improve optical sectioning; automated focus optimization. Our modular system also gives the
option to choose between three different light-sheets, allowing to select the sheet's thickness and
height according to the characteristic of the sample and the imaging technique to be performed.
We have also proposed new, computationally-undemanding image reconstruction methods based
on the maximum intensity projection operation.
With its simple, functional design and the use of a computer-reconfigurable SLM, we believe
our system represents an ideal platform for manipulating the illuminating light-sheet to apply
a range of advanced imaging techniques on a single SPIM microscope, and also to explore
combinations of multiple techniques and potentially trial new ones.
abcImage numberSharpnessabcd Acknowledgments
Chiara Garbellotto is supported by a studentship from the EPSRC CDT in Intelligent Sensing
and Measurement, Grant Number EP/L016753/1. We thank Andrew Tobin, Sophie Bradley and
team (Glasgow University) for the CLARITY-treated mouse brain samples, and Martin Denvir,
Carl Tucker and team (Edinburgh University) for the zebrafish samples.
Declaration
The Authors declare no conflict of interest.
|
1607.05304 | 1 | 1607 | 2016-07-18T20:13:39 | A nonequilibrium diffusion and capture mechanism ensures tip-localization of regulating proteins on dynamic filaments | [
"physics.bio-ph",
"cond-mat.stat-mech",
"q-bio.SC"
] | Diffusive motion of regulatory enzymes on biopolymers with eventual capture at a reaction site is a common feature in cell biology. Using a lattice gas model we study the impact of diffusion and capture for a microtubule polymerase and a depolymerase. Our results show that the capture mechanism localizes the proteins and creates large-scale spatial correlations. We develop an analytic approximation that globally accounts for relevant correlations and yields results that are in excellent agreement with experimental data. Our results show that diffusion and capture operates most efficiently at cellular enzyme concentrations which points to in vivo relevance. | physics.bio-ph | physics | A nonequilibrium diffusion and capture mechanism ensures tip-localization of
regulating proteins on dynamic filaments
Arnold Sommerfeld Center for Theoretical Physics (ASC) and Center for NanoScience (CeNS),
Emanuel Reithmann, Louis Reese,∗ and Erwin Frey†
Department of Physics, Ludwig-Maximilians-Universitat Munchen,
Theresienstrasse 37, 80333 Munchen, Germany
(Dated: October 2, 2018)
Diffusive motion of regulatory enzymes on biopolymers with eventual capture at a reaction site
is a common feature in cell biology. Using a lattice gas model we study the impact of diffusion
and capture for a microtubule polymerase and a depolymerase. Our results show that the capture
mechanism localizes the proteins and creates large-scale spatial correlations. We develop an analytic
approximation that globally accounts for relevant correlations and yields results that are in excellent
agreement with experimental data. Our results show that diffusion and capture operates most
efficiently at cellular enzyme concentrations which points to in vivo relevance.
The diffusive motion of proteins on filamentous struc-
tures in the cell is vital for several cellular functions such
as gene regulation [1] and cytoskeletal dynamics [2, 3]:
To find their target sites, transcription factors are likely
to employ one-dimensional diffusion on the DNA and the
dynamics of this process largely determine the kinetics of
gene regulation [4, 5]. Similarly, actin and microtubule
(MT) binding proteins diffuse on the respective filaments
and fulfill regulatory functions primarily at the filament
ends. Adam and Delbruck [6] suggested that a reduc-
tion in dimensionality of the diffusive motion enhances
the effective rate of association of particles with binding
sites on the membrane or on DNA and filaments, and
this concept has been widely applied and extended [7, 8],
see also Refs. [9] for recent reviews on the topic.
With regard to cytoskeletal architectures, efficient as-
sociation and localization of enzymes to specific sites is
relevant for a variety of cellular processes throughout the
cell cycle and for cell motility and dynamics [10]. It was
recently shown experimentally that one-dimensional dif-
fusion is utilized [2, 11] by two proteins with important
roles in the regulation of MT dynamics [12 -- 15], MCAK
and XMAP215. These proteins strongly localize at their
respective reaction sites and show association rates for
these sites that are significantly higher than expected for
binding via three-dimensional diffusion [11, 16]. Both
proteins carry out vital tasks, with MCAK acting as de-
polymerase of tubulin protofilaments [17] and XMAP215
as a poylmerase [16] when bound to ends of MTs. Note
that similar mechanisms are also assumed to be rele-
vant for actin associated proteins [18]. However, diffu-
sive motion on filaments does not lead to a localization
and efficient association of proteins per se: As we have
shown previously [19], it is crucial to include a captur-
ing mechanism at the reaction site, which suppresses
the one-dimensional diffusive motion of a protein that
reaches this site; without such a capturing mechanism
no increase in the effective association rate for the tip
occurs. For MCAK and XMAP215 protein capturing is
observed in experiments: Diffusive motion stops once the
proteins reach the MT tip [11, 16]. Yet, the underlying
interactions with the MT tip are still elusive and being
studied [20].
Here we present a theoretical description of enzyme
diffusion and capture at MT tips where the enzymes cat-
alyze filament polymerization or depolymerization. Pre-
vious studies of similar systems have lacked either a cap-
turing mechanism [21, 22] or a dynamic filament [19], al-
though both features are critical. To overcome both lim-
itations, we employ a one-dimensional lattice gas [23, 24]
with particle capturing in a dynamic system, in which
growth or shrinkage of the filament is triggered by the
interactions of particles with the lattice end. Our moti-
vation is twofold: Firstly, we seek for a detailed mathe-
matical understanding of the capturing mechanism. Sec-
ondly, based on a fully quantitative model, we wish
to elucidate the specific biomolecular mechanisms em-
ployed by XMAP215 and MCAK. Our results show that
the capturing process induces large-scale spatial correla-
tions in the protein distribution along the filament. We
develop a mathematical framework that systematically
includes relevant correlations on a global scale. This
conceptual advancement allows us to quantitatively ex-
plain the results of in vitro experiments with XMAP215
and MCAK [16, 25]. We demonstrate that the diffusion
and capture mechanism strongly localizes XMAP215 and
MCAK at the MT tip and that the process operates op-
timally under physiological conditions for both proteins,
which suggests that it is relevant in vivo.
Model definition. We consider a one-dimensional lat-
tice with lattice spacing a and a semi-infinite geometry
which corresponds to one protofilament, as depicted in
Fig. 1. In the case of MTs, a is the length of a tubulin
heterodimer, 8.4 nm. The configuration of enzymes on
the lattice is described by occupation numbers ni, taking
values ni=0 for empty, and ni=1 for occupied sites. The
particles symmetrically hop to neighboring sites at rate
, and interact via hard-core repulsion. We implement
Langmuir kinetics to model a surrounding reservoir of
particles with a constant concentration c. Particles at-
6
1
0
2
l
u
J
8
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
4
0
3
5
0
.
7
0
6
1
:
v
i
X
r
a
tach to and detach from the lattice at rates ωac and ωd,
respectively [27, 28]. Sites i≥3 are considered as bulk
sites. There the dynamics differs from that in the bulk
as we implement a protein capturing mechanism: Hop-
ping from site i=1 to site i=2 is disallowed, as suggested
experimentally for MCAK and XMAP215 [11, 16].
In
this way, detailed balance is broken which leads to strong
tip-localization due to a particle flux along the filament;
in equilibrium models such a significant localization is
absent, see Fig. S1 in the Supporting Material [26]. Par-
ticles detach from the first lattice site at a distinct off-
rate, ωd(cid:54)=ωd. We refer to site i=1 as a reaction site at
which new lattice sites may be added or removed. For
the moment, we specify our discussion to polymerases
such as XMAP215 [16]. However, our considerations are
largely independent of whether polymerization or depoly-
merization occurs -- an equivalent formulation can also be
found for the depolymerase MCAK [26]. For XMAP215,
we specify that lattice growth is triggered at rate δ if
the protein is bound to the first lattice site. Hence, the
average speed of lattice growth v for the MT is propor-
tional to the average particle occupation (cid:104)n1(cid:105) and the
XMAP215 polymerization rate: v=δa(cid:104)n1(cid:105). Here we as-
sume one catalyzing protein per protofilament end at
saturating conditions [26]. The actual maximum num-
ber of catalytically active proteins is unknown; in exper-
imental literature approximately 10 XMAP215 proteins
at the MT tip are estimated at 50 nM XMAP215 [16].
As shown in recent experiments, XMAP215 acts proces-
sively, i.e. one molecule adds multiple tubulin dimers
to the growing MT end [16]. To implement such behav-
ior in our model the particle at the tip is transferred
to newly incorporated lattice sites.
In our analysis we
neglect uncatalyzed tubulin addition or removal as typi-
cal corresponding experiments [11, 16, 25, 29] were per-
formed under conditions where these processes did not
occur with a significant rate. An extension is, however,
possible in a straightforward fashion and does not affect
tip-localization significantly; see Fig. S5 and S6 in the
FIG. 1.
Illustration of the model for XMAP215. Particles
bind to empty lattice sites with rate ωac, where c is the par-
ticle concentration in solution, and detach with rate ωd. The
proteins hop symmetrically to neighboring sites at rate but
exclude each other. We assume a distinct off-rate ωd at the
first site. Particles bound there cease hopping but add new
lattice sites at rate δ. The particle which stimulates poly-
merization moves with the tip. An analogous model can be
defined for MCAK, where depolymerization occurs if the lat-
tice end is occupied, see Supporting Material for details [26].
2
Supporting Material [26]. Therefore we expect validity of
our further considerations also with intrinsic MT dynam-
ics, for example as a consequence of hydrolysis of tubulin
bound GTP which was studied extensively in previous
models [30].
Mathematical analysis. We set up the equations of
motion for the average occupation numbers of the
stochastic process defined above. All equations will be
formulated in the frame of reference comoving with the
dynamic lattice end. In the bulk of the lattice, i≥3, we
obtain
dt(cid:104)ni(cid:105) = (cid:0)(cid:104)ni+1(cid:105)−2(cid:104)ni(cid:105)+(cid:104)ni−1(cid:105)(cid:1) + δ(cid:0)(cid:104)n1ni−1(cid:105)−(cid:104)n1ni(cid:105)(cid:1)
+ ωac(cid:0)1−(cid:104)ni(cid:105)(cid:1) − ωd(cid:104)ni(cid:105) .
(1)
d
This equation comprises contributions from hopping
while obeying the exclusion principle [31] (terms propor-
tional to ) and a displacement current due to polymer-
ization (terms proportional to δ) as well as particle at-
tachment and detachment (terms proportional to ωa and
ωd, respectively). The tip occupations complement these
bulk dynamics in the following manner:
d
d
d
(2)
dt ρ1 = (ρ2−g2) + ωac(1−ρ1) − ωdρ1 ,
dt ρ2 = (ρ3−2ρ2+g2) − δg2 + ωac(1−ρ2) − ωdρ2 ,
dt g2 = (g3−g2) + δg2 + ωac(ρ1+ρ2−2g2) − (ωd+ωd)g2 .
Here we have defined the average density, ρi:=(cid:104)ni(cid:105), and
the correlation function, gi:=(cid:104)n1ni(cid:105). Moreover, since the
polymerization process facilitated by XMAP215 is pro-
cessive, an empty lattice site at i=2 is created and site
i=1 remains occupied each time a new site is added.
We fully quantify our model with the experimental data
available for XMAP215 [16, 29]; see Supporting Material
for parameter values [26].
In the first step we test the quality of standard ap-
proximation techniques for driven lattice gases against
stochastic simulation data obtained from Gillespie's al-
gorithm [32]. The set of equations which determines the
lattice occupations (Eq. 1 and Eqs. 2) is not closed; the
dynamics of the density ρi and the correlation functions
gi=(cid:104)n1ni(cid:105) are coupled.
In fact, there is a hierarchy of
equations, which,
in general, precludes the derivation
of an exact solution for many driven lattice gas sys-
tems. A common and often quite successful approxi-
mation scheme for exclusion processes is to assume that
there are no correlations and that one may factorize all
correlation functions, (cid:104)n1ni(cid:105)≈(cid:104)n1(cid:105)(cid:104)ni(cid:105). In this mean-field
(MF) approximation one obtains a closed set of differen-
tial equations for the particle density ρi which may be
solved subject to proper boundary conditions; see Sup-
porting Material for details. Fig. 2 shows the average
occupation number of the first site, (cid:104)n1(cid:105), as a function
of the protein concentration in solution c. A comparison
with our stochastic simulation data shows that the MF
solution strongly overestimates (cid:104)n1(cid:105) and thus the average
polymerization speed v.
3
ρi, Eq. 1, and the correlation function gi; for details
see the Supporting Material [26]. Due to the capturing
mechanism a continuous description is not valid at sites
i=1, 2, and we retain the local dynamics there, Eqs. 2.
These equations constrain the boundary conditions of
ρ(x) and g(x) at x=a. We further impose that the density
equilibrates asymptotically at the Langmuir isotherm,
limx→∞ ρ(x)=ρLa=ωac/(ωac+ωd), and that correlations
vanish, limx→∞ g(x)=(cid:104)n1(cid:105)ρLa. Solving the equations of
this correlated MF (CMF) theory for the steady state
tip density we obtain the results shown in Fig. 2, which
are in excellent agreement with the stochastic simulation
data. We therefore conclude that there are long-ranged
correlations along the MT and that they are essential
in explaining the observed average tip density and the
ensuing polymerization speed.
Fig. 3(a) shows the density profile along the lattice ob-
tained by stochastic simulations and the CMF approach.
The particle occupation is obtained with high precision
within the CMF framework along the whole lattice. The
density profiles also agree with recent data from time
and ensemble averaged high resolution fluorescence in-
tensity profiles for XMAP215 [36]. Notably, there is a
discontinuity at sites i=1, 2, which is due to particle cap-
ture and which demonstrates the strong tip-localization
of the proteins.
FIG. 3. Comparison of density and tip-bulk correlation pro-
files obtained by the CMF approximation (lines) and stochas-
tic simulations (symbols) for XMAP215 concentrations of 10
and 100 nM (see Supporting Material [26] for parameter val-
ues [16, 29]). (a) XMAP215 strongly localizes to the MT tip
and the density profile drops abruptly at sites i=1, 2.
(b)
The correlation coefficient corr(n1, ni) (see Eq. (4)) along the
lattice shows the significance of tip-bulk correlations over hun-
dreds of lattice sites.
(3a)
In Figure 3(b), the Pearson product-momentum corre-
lation coefficient
(3b)
where we defined ρ(x, t)=(cid:104)ni+1(cid:105) and g(x, t)=(cid:104)n1ni+1(cid:105)
with x=a(i−1) for i≥3. We have further introduced
the macroscopic diffusion constant D=a2 and the max-
imum polymerization speed v0=δa. Equations 3 can
be derived from the discrete equations for the density
corr(n1, ni) =
cov(n1, ni)
σ(n1)σ(ni)
,
(4)
which quantifies the correlations between the tip site i=1
and sites i≥2 in the bulk, is plotted against lattice posi-
tion. Here cov(·,·), and σ(·) signify the covariance and
the standard deviation, respectively. The correlation co-
Average occupation of the first lattice site (cid:104)n1(cid:105).
FIG. 2.
The MF approach as well as the FSMF approximation for
segment sizes of N =2, 5 deviate strongly from stochastic sim-
ulation data (open circles), in complete contrast to the CMF
approximation. Parameter values are detailed in the Support-
ing Material [26].
One possible reason for the failure of the MF calcula-
tion lies in correlations that arise close to the reaction
site. Local correlations can efficiently be accounted for
by employing a finite segment mean-field (FSMF) the-
ory [33, 34]. Here, the idea is to retain all correlations
close to the catalytic site by solving the full master equa-
tion for the first N sites and to use the MF assump-
tion only outside of this segment. The density profile is
then obtained by matching the tip solution and the MF
solution [21, 35]; see Supporting Material [26]. While
the results show the right trend towards the numerical
data, the improvement over the MF results is insignifi-
cant. These observations suggest that correlations extend
far beyond the immediate vicinity of the reaction site.
To account for such correlations we extend the MF
theory by retaining both the density and the correlation
function as dynamic variables. In order to close the set of
equations we employ the following factorization scheme:
(cid:104)n1n2ni(cid:105)≈(cid:104)n1n2(cid:105)(cid:104)ni(cid:105), and (cid:104)n2ni(cid:105)≈(cid:104)n2(cid:105)(cid:104)ni(cid:105) for i≥3, i.e.
we retain correlations with respect to the reaction site
but neglect them within the bulk of the lattice. We con-
firmed this approximation scheme for typical biological
parameter values by stochastic simulations; see Fig. S2
in the Supporting Material [26]. With the above closure
relations one obtains for the bulk dynamics in a contin-
uous description
∂tρ(x, t) = D∂2
∂tg(x, t) = D∂2
xρ − v0∂xg + ωac(1−ρ) − ωdρ ,
xg − v0∂xg + ρ(cid:0)ρ2−g2
(cid:1)
+ ωac(cid:0)ρ+ρ1−2g(cid:1) −(cid:0)ωd+ωd
(cid:1)g ,
4
ticle interactions become important for lattice diffusion.
This explains the deviations in the computed correlation
profile, Fig 3(b). As the CMF method is based on a
non-perturbative ansatz there is no analytic expression
that exactly quantifies its error. However, we observe
very good agreement with our Gillespie algorithm based
simulations over a very broad parameter range and, im-
portantly, for typical biological parameters, see Fig. S4
in the Supporting Material [26].
Comparison with experimental data. We now turn to
a comparison with experimental data for the polymeriza-
tion velocity [16, 29] and, to supplement the results for
XMAP215, we apply our methods to an analogous model
for MCAK particles which depolymerize MTs [11, 25]. In
essence, we adapt the above model to account for lattice
shrinkage triggered by an occupied reaction site, see Sup-
porting Material for details [26]. Similar to the processive
polymerization of XMAP215 also MCAK is assumed to
depolymerize processively [11, 25]. The parameters em-
ployed in the model are again drawn from available exper-
imental data [25]. For both MCAK and XMAP215, we
find excellent quantitative agreement between our the-
oretical approach and experimentally determined poly-
merization and depolymerization velocities; see Fig. 4(a).
This quantitative agreement is achieved without an ad-
justable parameter; see Supporting Material [26]. We
then used the quantified models to investigate the im-
pact of the diffusion and capture process for XMAP215
and MCAK. Fig. 4(b) shows the increase of protein lo-
calization at the reaction site due to diffusion and cap-
ture on the filament: We plot the difference between tip
densities in the presence (ρCMF
) and absence of diffu-
1
sion on filaments (ρno diffusion
=ωac/(ωd+ωac)). For both
enzymes, diffusive motion and subsequent capturing at
the MT lattice strongly increases the occupation den-
sity at the tip and therefore constitutes a highly efficient
means of increasing the effective attachment rate to the
reaction site. Moreover, the ensuing curve shows a pro-
nounced maximum, indicating an optimal concentration
range at which the enhancement of tip occupancy due
to diffusion on the MT reaches its peak. Strikingly, this
maximum coincides with the physiological concentration
range for each protein: 100−1000 nM for XMAP215 [13]
and 10−100 nM [37] for MCAK. This strongly supports
the importance of diffusion and capture for MCAK and
XMAP215 in vivo.
1
It is interesting to speculate about possible biomolecu-
lar mechanisms that could generate particle capturing at
the MT tip as such a mechanism would probably require
an energy source to drive the system out of equilibrium.
Concerning MCAK, it was recently hypothesized, that
an ATP is required to stop its diffusive motion at the
MT tip [20] which is consistent with our proposed non-
equilibrium model. Since XMAP215 does not bind nu-
cleotides such as ATP or GTP itself [16], one might spec-
ulate that a non-equilibrium capturing mechanism relies
FIG. 4. Panel (a) demonstrates excellent agreement of poly-
merization and depolymerization velocities obtained from our
theoretical analysis (CMF approximation) with existing ex-
perimental data for XMAP215 [16, 29] and MCAK [25],
respectively. Panel (b) depicts the difference between the
occupation density at the tip ρ1 with and without diffu-
sion on the MT, where ρno diffusion
= ωac/(ωd + ωac), and
shows the impact of diffusion and capture on tip localization
of MCAK (blue) and XMAP (orange). The concentration
range for maximum efficiency coincides with the physiologi-
cal concentration range for each protein: 100 − 1000 nM for
XMAP215 [13] and 10 − 100 nM for MCAK [37] (shaded ar-
eas). In (c) the reaction site density with lattice diffusion (ρ1,
solid lines) and without lattice diffusion (ρno diffusion
, dashed
lines) is depicted. Kinetic parameters are given in the Sup-
porting Material [26].
1
1
efficient decays very slowly over a broad region at the
tip. The capturing mechanism and the resulting particle
flux towards the filament tip ensue strong positive cor-
relations with respect to the first lattice site and sites
in its vicinity. This effect is antagonized by weak nega-
tive correlations caused by the creation of empty lattice
sites due to polymerization. With diffusion taking place
on a faster time scale than polymerization, the positive
correlations dominate. This is confirmed by stochastic
simulations where either capturing or growth is switched
off: We find anti-correlations if capturing is turned off,
and positive correlations if there is no growth of the lat-
tice; see Fig. S3 in the Supporting Material [26]. Note
that for higher growth rates the correlation profile can
also become negative. We conclude that the spatial cor-
relations which emerge over several hundred lattice sites
are a direct consequence of protein capture and proces-
sive growth. Further, it becomes evident why the MF
and the FSMF approaches do not lead to the correct tip
density: Correlations extend into the system on a length-
scale which exceeds the scope of these and other previous
approaches [19, 21, 22].
In contrast the CMF approx-
imation captures and quantifies significant correlations
and successfully reproduces simulation data. Note that
also higher order correlations of the form (cid:104)n1njnk(cid:105) and
(cid:104)njnk(cid:105) with j, k≥2 and k>j, which are neglected in the
CMF approximation, might be of relevance when par-
on tubulin polymerization or depolymerization. Possi-
bly, a conformational change of XMAP215 coupled to
processes involved in MT depolymerization or polymer-
ization could lead to protein capture.
Summary and Conclusion.
In this work, we studied
the regulatory influence of an explicit capture process on
the distribution of MT polymerases and depolymerases
that are subject to one-dimensional diffusion on MTs.
To model these biologically relevant situations we em-
ployed a model based on a symmetric simple exclusion
process [24] extended by a detailed balance breaking cap-
turing process at the lattice end, which acts as a biasing
mechanism. Our results show that the occupation of the
MT tip with a protein spatially correlates with the oc-
cupation of the MT lattice. This is a direct consequence
of protein capturing which in turn strongly localizes the
proteins at the MT tip. Correlations decay slowly along
the lattice and have a large impact on the occupation of
the MT tip. This is of relevance as the latter quantity de-
termines the velocity of enzyme-dependent MT growth or
shrinking. We derive a generalized set of hydrodynamic
equations which couple the evolution of the particle den-
sity with the evolution of relevant correlations. In that
way it is possible to account for those correlations on
a global scale. Similar correlations have been identified
in two-dimensional diffusive systems [38] or in diffusive
systems with a small, local drive [39].
Our findings are not limited to MTs and their asso-
ciated enzymes, but might also be applicable to other
enzymatic processes with spatial degrees of freedom and,
quite generally, non-equilibrium physics.
We thank Linda Wordeman, Gary Brouhard and
their respective co-workers for sharing data points of
XMAP215 and MCAK induced MT growing and shrink-
ing velocities, respectively, as shown in Fig. 4(a). This
research was supported by the German Excellence Initia-
tive via the program "NanoSystems Initiative Munich"
(NIM) and the Deutsche Forschungsgemeinschaft (DFG)
via project B02 within the Collaborative Research Center
(SFB 863) "Forces in Biomolecular Systems".
University of Technology, Delft, Netherlands.
∗ Present address: Department of Bionanoscience, Delft
† [email protected]
[1] P. H. von Hippel and O. G. Berg, J. Biol. Chem. 264, 675
(1989).
5
molecular biology (Freeman, 1968) pp. 198 -- 215.
[7] P. H. Richter and M. Eigen, Biophys. Chem. 2, 255 (1974);
O. G. Berg, R. B. Winter, and P. H. Von Hippel, Biochem-
istry 20, 6929 (1981).
[8] L. Bintu, N. E. Buchler, H. G. Garcia, U. Gerland, T. Hwa,
J. Kondev, and R. Phillips, Curr. Opin. Genet. Dev. 15,
116 (2005).
[9] L. Mirny, M. Slutsky, Z. Wunderlich, A. Tafvizi, J. Leith,
and A. Kosmrlj, J. Phys. A: Math. Theor. 42, 434013
(2009); A. B. Kolomeisky, Phys. Chem. Chem. Phys. 13,
2088 (2011); O. B´enichou, C. Loverdo, M. Moreau, and
R. Voituriez, Rev. Mod. Phys. 83, 81 (2011); M. Shein-
man, O. B´enichou, Y. Kafri, and R. Voituriez, Rep. Prog.
Phys. 75, 026601 (2012).
[10] C. G. Dos Remedios, D. Chhabra, M. Kekic, I. V. De-
dova, M. Tsubakihara, D. A. Berry, and N. J. Nosworthy,
Physiol. Rev. 83, 433 (2003); A. Akhmanova and M. O.
Steinmetz, Nat. Rev. Mol. Cell Biol. 9, 309 (2008).
[11] J. Helenius, G. J. Brouhard, Y. Kalaidzidis, S. Diez, and
J. Howard, Nature 441, 115 (2006).
[12] R. Tournebize, A. Popov, K. Kinoshita, A. J. Ashford,
S. Rybina, A. Pozniakovsky, T. U. Mayer, C. E. Walczak,
E. Karsenti,
and A. A. Hyman, Nat. Cell Biol. 2, 13
(2000).
[13] K. Kinoshita, I. Arnal, A. Desai, D. N. Drechsel, and
A. A. Hyman, Science 294, 1340 (2001).
[14] J. D. Wilbur and R. Heald, eLife 2, e00290 (2013).
[15] S. B. Reber, J. Baumgart, P. O. Widlund, A. Pozni-
akovsky, J. Howard, A. A. Hyman, and F. Julicher, Nat.
Cell Biol. 15, 1116 (2013).
[16] G. J. Brouhard, J. H. Stear, T. L. Noetzel, J. Al-Bassam,
K. Kinoshita, S. C. Harrison, J. Howard, and A. A. Hy-
man, Cell 132, 79 (2008).
[17] A. Desai, S. Verma, T. J. Mitchison, and C. E. Walczak,
Cell 96, 69 (1999).
[18] S. Romero, C. Le Clainche, D. Didry, C. Egile, D. Pan-
taloni, and M.-F. Carlier, Cell 119, 419 (2004); D. Vavy-
lonis, D. R. Kovar, B. O'Shaughnessy, and T. D. Pollard,
Mol. Cell 21, 455 (2006); S. D. Hansen and R. D. Mullins,
J. Cell Biol. 191, 571 (2010); H. Mizuno, C. Higashida,
Y. Yuan, T. Ishizaki, S. Narumiya, and N. Watanabe,
Science 331, 80 (2011).
[19] E. Reithmann, L. Reese, and E. Frey, Biophys. J. 108,
787 (2015).
[20] C. T. Friel and J. Howard, EMBO J. 30, 3928 (2011);
and
A. Ritter, N. Kreis, F. Louwen, L. Wordeman,
J. Yuan, Crit. Rev. Biochem. Mol. Bio. 51, 228 (2016).
[21] G. Klein, K. Kruse, G. Cuniberti, and F. Julicher, Phys.
Rev. Lett. 94, 108102 (2005).
[22] M. Schmitt and H. Stark, Europhys. Lett. 96, 28001
(2011).
[23] T. Chou, K. Mallick, and R. K. P. Zia, Rep. Prog. Phys.
74, 116601 (2011).
[24] B. Derrida, Phys. Rep. 301, 65 (1998).
[25] J. R. Cooper, M. Wagenbach, C. L. Asbury,
and
[2] J. R. Cooper and L. Wordeman, Curr. Opin. Cell Biol. 21,
L. Wordeman, Nat. Struct. Mol. Biol. 17, 77 (2010).
68 (2009).
[3] J. Howard and A. A. Hyman, Curr. Opin. Cell Biol. 19,
31 (2007).
[26] See Supplemental Material.
[27] R. Lipowsky, S. Klumpp, and T. Nieuwenhuizen, Phys.
Rev. Lett. 87, 108101 (2001).
[4] A. D. Riggs, H. Suzuki, and S. Bourgeois, J. Mol. Biol.
[28] A. Parmeggiani, T. Franosch, and E. Frey, Phys. Rev.
48, 67 (1970).
[5] P. Hammar, P. Leroy, A. Mahmutovic, E. G. Marklund,
O. G. Berg, and J. Elf, Science 336, 1595 (2012).
[6] G. Adam and M. Delbruck, in Structural chemistry and
Lett. 90, 86601 (2003); Phys. Rev. E 70, 46101 (2004).
[29] P. O. Widlund, J. H. Stear, A. Pozniakovsky, M. Zanic,
S. Reber, G. J. Brouhard, A. A. Hyman, and J. Howard,
Proc. Nat. Acad. Sci. USA 108, 2741 (2011).
6
[30] T. Antal, P. L. Krapivsky, S. Redner, M. Mailman, and
B. Chakraborty, Phys. Rev. E 76, 041907 (2007); R. Pad-
inhateeri, A. B. Kolomeisky, and D. Lacoste, Biophys. J.
102, 1274 (2012); T. Niedermayer and R. Lipowsky, Phys.
Rev. E 92, 052137 (2015).
[31] B. Derrida, J. Stat. Mech. Theor. Exp. 2007, P07023
(2007).
[32] D. T. Gillespie, Annu. Rev. Phys. Chem. 58, 35 (2007).
[33] T. Chou and G. Lakatos, Phys. Rev. Lett. 93, 198101
(2004).
[34] G. Lakatos, T. Chou, and A. Kolomeisky, Phys. Rev. E
71, 011103 (2005).
[35] S. Nowak, P.-W. Fok, and T. Chou, Phys. Rev. E 76,
31135 (2007).
[36] S. P. Maurer, N. I. Cade, G. Bohner, N. Gustafsson,
E. Boutant, and T. Surrey, Curr. Biol. 24, 372 (2014).
[37] A. W. Hunter, M. Caplow, D. L. Coy, W. O. Hancock,
S. Diez, L. Wordeman, and J. Howard, Mol. Cell 11, 445
(2003).
[38] D. C. Markham, M. J. Simpson, and R. E. Baker, Phys.
Rev. E 87, 062702 (2013); D. C. Markham, M. J. Simpson,
P. K. Maini, E. A. Gaffney, and R. E. Baker, Phys. Rev.
E 88, 052713 (2013).
[39] T. Sadhu, S. N. Majumdar, and D. Mukamel, Phys. Rev.
E 90, 012109 (2014).
Supplemental Material: A nonequilibrium diffusion and capture mechanism ensures
tip-localization of regulating proteins on dynamic filaments
Arnold Sommerfeld Center for Theoretical Physics (ASC) and Center for NanoScience (CeNS), Department of
Physics, Ludwig-Maximilians-Universitat Munchen, Theresienstrasse 37, 80333 Munchen, Germany
Emanuel Reithmann, Louis Reese, and Erwin Frey
1
TIP LOCALIZATION DUE TO PARTICLE CAPTURING
In this work, we investigate a model where the diffusive motion of particles on a filament ceases as soon as they
arrive at a reaction site. This feature, which we refer to as particle capturing, is a key element of our model, as it
drives the system out of thermal equilibrium. In order to investigate the impact of particle capture on tip localization
of particles, we also investigated a model where particles are not captured at the tip, but where a hopping from the
tip into the bulk occurs such that detailed balance is not broken. In detail, we introduce a release rate , at which
particles hop from site i = 1 to site i = 2. Then, to implement equilibrium conditions for particle hopping (i.e. with
respect to a system without lattice growth or shrinkage), we impose / = ωd/ωd. This condition ensures detailed
balance in a static system and for a constant on-rate along the lattice. Since we also implement lattice growth,
detailed balance is still broken, which manifests itself in a net particle drift away from the tip in the comoving frame
of reference. In Fig. S1 we compare density profiles of the hopping-equilibrium model and the one with strict (i.e.
irreversible) particle capturing as defined in the main text with parameters as for XMAP215.
In the equilibrium
model, the density profile is almost constant whereas in the model with capturing a strong tip-localization occurs (1-2
orders of magnitude increase in the tip-density). Although an irreversible capturing is, of course, a simplification, we
expect similar effects to occur for release rates much smaller than the equilibrium release rate, (cid:28) eq := ( ωd)/ωd.
In this case, capturing generates a particle current towards the MT tip which conversely leads to spatial correlations
subject of this work.
FIG. S1. Diffusion and capture ensures tip-localization. Density profiles from MC simulations (open symbols) of (a) a model
where particle hopping obeys detailed balance with respect to a static lattice and (b) the model from the main text. In an
"equilibrium" model no localization occurs and the density profile is almost constant due to fast diffusion. With strict (i.e.
irreversible capturing), the tip is highly occupied as compared to its equilibrium occupation (dotted lines). Note that in both
models we implement off-rates which differ at the tip and lattice growth which results in non-constant density profiles also when
particle hopping obeys detailed balance. Further, also the "equilibrium" model is out of equilibrium due to lattice growth. In
(a) = 3.0 × 103 s−1, in (b) = 0. Other parameters as for the XMAP215 model, see Table I.
MEAN-FIELD (MF) APPROXIMATION
In the mean-field approximation all correlations are neglected; we set (cid:104)ninj(cid:105) = (cid:104)ni(cid:105)(cid:104)nj(cid:105). This closes the hierarchy
of equations stated in the main text:
2
d
d
dt(cid:104)ni(cid:105) = ((cid:104)ni+1(cid:105) − 2(cid:104)ni(cid:105) + (cid:104)ni−1(cid:105)) + δ((cid:104)n1(cid:105)(cid:104)ni−1(cid:105) − (cid:104)n1(cid:105)(cid:104)ni(cid:105)) + ωac(1 − (cid:104)ni(cid:105)) − ωd(cid:104)ni(cid:105) for i ≥ 3
dt(cid:104)n1(cid:105) = ((cid:104)n2(cid:105) − (cid:104)n1(cid:105)(cid:104)n2(cid:105)) + ωac(1 − (cid:104)n1(cid:105)) − ωd(cid:104)n1(cid:105)
dt(cid:104)n2(cid:105) = ((cid:104)n3(cid:105) − 2(cid:104)n2(cid:105) + (cid:104)n1(cid:105)(cid:104)n2(cid:105)) − δ(cid:104)n1(cid:105)(cid:104)n2(cid:105) + ωac(1 − (cid:104)n2(cid:105)) − ωd(cid:104)n2(cid:105) .
d
(S1)
(S2)
(S3)
Instead of solving the recurrence relation, we use a continuous description for Eq. S1. At sites i = 1, 2 such an
approximation is not valid due to a discontinuity in the density profile. Performing a Taylor expansion for small
lattice spacings a up to second order we obtain
∂tρ(x, t) = a2∂2
xρ(x, t) − δa∂xρ(x, t)(cid:104)n1(cid:105) + ωac(1 − ρ(x, t)) − ωdρ(x, t) .
(S4)
In the above equation the continuous labeling x = a(i − 1) is used for ρ(x, t) = (cid:104)ni+1(cid:105). Further, we use that for
typical biological systems (cid:29) δ holds true and neglect the second order term due to the particle drift in the comoving
frame, 1
xρ(x). Since we are only interested in the steady state solution we set the time derivative to zero. As
boundary condition, we impose that the density equilibrates at the Langmuir density for large distances to the tip,
limx→∞ ρ(x) = ρLa = ωac/(ωac + ωd). The boundary condition at x = a has to be consistent with the solution of
Eqs. S2 and S3, ρ(a) = (cid:104)n2(cid:105). We can use the continuous solution to express (cid:104)n3(cid:105) = ρ(2a) and solve Eqs. S2 and S3.
This self-consistent solution can be obtained numerically and determines the MF density profile along the whole
lattice.
2 δa2∂2
THE FINITE SEGMENT MEAN-FIELD (FSMF) APPROXIMATION
The finite segment mean-field approach is based on the idea to account for correlations locally within a small
segment.
In detail, all correlations within this segment are retained whereas outside the segment correlations are
neglected. An efficient implementation is achieved by using the transition matrix corresponding to the master equation
for occupations of the segment. Since in our model correlations are strongest close to the tip, we choose to keep
correlations with respect to the first N sites. For example, for N = 2 the corresponding transition matrix Mij with
i, j ∈ {0, . . . , 3} reads
−2ωac − (cid:104)n3(cid:105)
ωac
0
M =
ωac + (cid:104)n3(cid:105) −ωd − ωac − (1 + (cid:104)n3(cid:105))
ωd + (cid:104)n3(cid:105)
ωd
0
−ωd − ωac − (cid:104)n3(cid:105)
ωac + (cid:104)n3(cid:105)
0
ωd
δ + (cid:104)n3(cid:105) + ωd
−ωd − ωd − (cid:104)n3(cid:105) − δ
ωac
.
Here we introduced (cid:104)n3(cid:105) = (1 − (cid:104)n3(cid:105)). Further, the enumeration of states is chosen such that it corresponds to the
respective binary number, e.g. M01 describes transitions from state (n1 = 0, n2 = 1) to state (n1 = 0, n2 = 0).
Note that correlations with respect to nN +1 are already neglected. The eigenvector of M with eigenvalue 0 is then
computed, which yields steady state occupations within the segment in dependence of (cid:104)nN +1(cid:105). A self-consistent
solution of these occupations and those for sites i > N is obtained in analogous fashion to the MF procedure: We
use the continuous MF solution for densities with i > N and the discrete solutions for sites in the segment to express
all densities in terms of (cid:104)nN +1(cid:105). The master equation for (cid:104)nN +1(cid:105) (given by Eq. S1) is then solved numerically in the
steady state to compute the complete density profile. This procedure is, however, strongly limited by the size of the
finite segment as the corresponding transition matrix is of size 2N × 2N .
THE CORRELATED MEAN-FIELD (CMF) APPROXIMATION
In the following we will show how to perform the CMF approximation for the model presented in the main text.
This approach systematically includes the relevant correlations arising due to the capturing mechanism.
The CMF calculations can be separated in three steps: a) Computation of the continuous solution for the density
ρ(x) and correlation profile g(x) in the bulk, i ≥ 2. b) Computation of the discrete solution for i = 1. c) Matching of
the continuous solution and the discrete solution.
d
We start with deriving the continuous bulk solutions. The density profile is governed by Eq. 1 of the main text:
dt(cid:104)ni(cid:105) = (cid:0)(cid:104)ni+1(1 − ni)(cid:105)−(cid:104)ni(1 − ni+1)(cid:105)+(cid:104)ni−1(1 − ni)(cid:105) − (cid:104)ni(1 − ni−1)(cid:105)(cid:1) + δ(cid:0)(cid:104)n1ni−1(cid:105)−(cid:104)n1ni(cid:105)(cid:1)
+ ωac(cid:0)1−(cid:104)ni(cid:105)(cid:1) − ωd(cid:104)ni(cid:105)
= (cid:0)(cid:104)ni+1(cid:105)−2(cid:104)ni(cid:105)+(cid:104)ni−1(cid:105)(cid:1) + δ(cid:0)(cid:104)n1ni−1(cid:105)−(cid:104)n1ni(cid:105)(cid:1) + ωac(cid:0)1−(cid:104)ni(cid:105)(cid:1) − ωd(cid:104)ni(cid:105) .
(S5)
Here, we account for particle hopping with exclusion (terms ∝ ), lattice growth (terms ∝ δ), particle attachment
(terms ∝ ωa), and particle detachment (terms ∝ ωd). In the main text we show that it is essential to account for
tip-bulk correlations on a large scale. In the CMF approach this is achieved globally by coupling the evolution of
the density with the one for tip-bulk correlations. The discrete equation governing the evolution of correlations with
respect to the reaction site reads
3
d
dt(cid:104)n1ni(cid:105) = ((cid:104)n1ni−1(cid:105) − 2(cid:104)n1ni(cid:105) + (cid:104)n1ni+1(cid:105) + (cid:104)n2ni(cid:105) − (cid:104)n1n2ni(cid:105)) + δ((cid:104)n1ni−1(cid:105) − (cid:104)n1ni(cid:105))
+ωac((cid:104)n1(cid:105) + (cid:104)ni(cid:105) − (cid:104)n1ni(cid:105)) − (ωd + ωd)(cid:104)n1ni(cid:105).
(S6)
The above equation, which follows from the master equation, describes changes of the joint probability for a simulta-
neous occupation of the first and the i-th site: All probabilities for processes that lead to a simultaneous occupation
of both lattice sites multiplied with the respective rate are added and all probabilities for processes where one of
the two sites is emptied multiplied with the respective rate are subtracted. Again, contributions arise from particle
hopping with exclusion (terms ∝ ), lattice growth (terms ∝ δ), particle attachment (terms ∝ ωa), and particle
detachment (terms ∝ ωd), respectively. For example, for particle hopping we have contributions from hopping pro-
cesses with respect to the i-th site ((cid:104)n1ni−1(cid:105) − 2(cid:104)n1ni(cid:105) + (cid:104)n1ni+1(cid:105)) as well as the capturing of a particle at the
first site ((cid:104)n2ni(cid:105) − (cid:104)n1n2ni(cid:105)). Note that higher order correlators can be obtained in complete analogy. In order to
close the hierarchy of moments, we use the factorization scheme stated in the main text: (cid:104)n1n2ni(cid:105) ≈ (cid:104)n1n2(cid:105)(cid:104)ni(cid:105) and
(cid:104)n2ni(cid:105) ≈ (cid:104)n2(cid:105)(cid:104)ni(cid:105) for i ≥ 3. Fig. S2 shows that this is justified, as the corresponding correlation coefficients are one
to two orders of magnitude lower than corr(n1, ni). In the continuous limit a → 0 the recurrence relations given by
the dynamic equations for (cid:104)ni(cid:105) and (cid:104)n1ni(cid:105) translate into a set of coupled differential equations. Up to a second order
Taylor expansion we obtain
∂tρ(x, t) = a2∂2
∂tg(x, t) = (a2∂2
xρ(x, t) − δa∂xg(x, t) + ωac(1 − ρ(x, t)) − ωdρ(x, t)
xg(x, t) + (cid:104)n2(cid:105)(t)ρ(x, t) − (cid:104)n1n2(cid:105)(t)ρ(x, t)) − δa∂xg(x, t) + ωac((cid:104)n1(cid:105)(t) + ρ(x, t) − 2g(x, t))
(S7)
2 δa2∂2
xg(x)) since (cid:29) δ for typical biological situations.
(S8)
Here, we used again a continuous labeling x = a(i − 1) and neglected second order terms due to lattice growth
(∝ 1
In this work, we are interested in the steady state
properties of the system, ∂tρ(x, t) = 0 and ∂tg(x, t) = 0. Under this condition, Eqs. S7 and S8 are solved for the
continuous solutions ρ(x) and g(x). Further, we impose the following boundary conditions to obtain a meaningful
solution: limx→∞ ρ(x) = ρLa = ωac/(ωac + ωd), limx→∞ g(x) = (cid:104)n1(cid:105)ρLa, ρ(a) = (cid:104)n2(cid:105) and g(a) = (cid:104)n1n2(cid:105). Note that
the solutions depend on the yet unknown variables (cid:104)n1(cid:105), (cid:104)n2(cid:105) and (cid:104)n1n2(cid:105).
−(ωd + ωd)g(x, t) .
In the second step, we solve the equation for the occupancy of the reaction sites, i = 1,
d
dt(cid:104)n1(cid:105) = 0 = ((cid:104)n2(cid:105) − (cid:104)n1n2(cid:105)) + ωac(1 − (cid:104)n1(cid:105)) − ωd(cid:104)n1(cid:105) ,
to express (cid:104)n1(cid:105) in terms of (cid:104)n2(cid:105) and (cid:104)n1n2(cid:105).
(cid:104)n1n2(cid:105). To this end we employ the "master equations" for the latter variables.
Lastly, we self-consistently match the discrete and continuous solutions in that we determine the values of (cid:104)n2(cid:105) and
d
dt(cid:104)n2(cid:105) = 0 = ((cid:104)n3(cid:105) − 2(cid:104)n2(cid:105) + (cid:104)n1n2(cid:105)) − δ(cid:104)n1n2(cid:105) + ωac(1 − (cid:104)n2(cid:105)) − ωd(cid:104)n2(cid:105)
d
dt(cid:104)n1n2(cid:105) = 0 = ((cid:104)n1n3(cid:105) − (cid:104)n1n2(cid:105)) + δ(cid:104)n1n2(cid:105) + ωac((cid:104)n1(cid:105) + (cid:104)n2(cid:105) − 2(cid:104)n1n2(cid:105)) − (ωd + ωd)(cid:104)n1n2(cid:105).
(S11)
We insert the continuous bulk solutions derived in the first step for (cid:104)n3(cid:105) = ρ(2a) and (cid:104)n1n3(cid:105) = g(2a). Finally, the
discrete solution for (cid:104)n1(cid:105) is used to express all variables in terms of (cid:104)n2(cid:105) and (cid:104)n1n2(cid:105). This allows us to solve Eqs. S10
and S11 numerically which, as a consequence, fixes the entire density and correlation profile.
The behavior of correlations is also demonstrated in Fig S3: Without a capturing mechanism, correlations are
purely negative due to the creation of empty sites resulting from the processive polymerization scheme. Opposed to
that, purely positive correlations arise in a static lattice with capturing.
The CMF approach neglects correlations within the diffusive compartment (i.e. we assume (cid:104)ninj(cid:105) = (cid:104)ni(cid:105)(cid:104)nj(cid:105) and
(cid:104)n1ninj(cid:105) = (cid:104)n1ni(cid:105)(cid:104)nj(cid:105) for i, j ≥ 3 and i < j). As this approximation is a non-perturbative ansatz, it is in general not
(S9)
(S10)
4
FIG. S2. In panel (a) and (b) we show that correlations corr(n2, ni) and corr(n1n2, ni) for i ≥ 3 are negligible since they are
one to two orders of magnitude smaller than the tip bulk correlations, corr(n1, ni). Parameters as for the XMAP215 model.
FIG. S3. Tip-bulk correlation profile obtained from stochastic simulations. Without particle capturing (orange data points)
correlations are negative due to the processive growth of the lattice and the resulting creation of empty lattice sites. Correlations
are positive in a static system with a capturing mechanism (blue points). Parameter values are equal to the ones used for the
XMAP215 model; concentrations are c = 10 nM for the case without polymerization and c = 5000 nM for the case without
capturing.
possible to quantify its error. In order to ensure the validity over a broad and biologically relevant parameter range,
we performed extensive MC simulations and compared the result with CMF computations. In detail, we performed
parameter sweeps for (from 300 − 10000 s−1), ωd (from 0.1 − 10 s−1), δ (from 5 − 95 s−1) and c (for each parameter
point at five equidistant values between c1 and c5, such that ρCMF
(c5) = 0.9). The results are
shown in Fig. S4. The CMF approximation delivers good results over this very broad parameter range; the maximum
relative deviation for ρ1 over the 1000 different tested parameter sets is 6.5%.
(c1) = 0.1 and ρCMF
1
1
In a previous publication, we derived an effective theory that allows for the calculation of reaction site occupations
that are subject to a diffusion and capture mechanism in a static lattice (i.e. without lattice growth or shrinkage) [R1].
While both approaches consider protein diffusion and capture on filaments, they differ significantly on a conceptual
level and with respect to the scope of their predictions: Whereas the previous approach is based an on a heuristic
theory and a priori only valid in the absence of polymerization and depolymerization, respectively, the CMF ap-
proach specifically accounts for lattice growth and shrinkage. Further, the CMF approximation is derived from more
conceptual considerations: It assumes that diffusion and capture creates correlations which primarily affect the tip
100101102103Sitei−0.010.000.010.020.03corr(n1ni)NocapturingNopolymerization 5
1
FIG. S4. Error of CMF approximation. We compared results for the tip density obtained from the CMF approximation (ρCMF
)
) for 1000 different parameter sets. For each set {, δ, ωa, ωd, ωd} we determined five equidistant
and MC simulations (ρMC
concentrations between c1 and c5, such that ρCMF
(c5) = 0.9. For these concentrations, we computed the
average relative deviation between simulation results and analytic approximation to get an estimate ∆CMF of the error along a
ρ1 − c curve (right side). Note that we expect the error to vanish for very low and very high occupations. We performed sweeps
with respect to and δ (a), and and ωd (b). Deviations are small, with the maximal c-averaged deviation being 5% and the
maximal relative deviation being 6.5%. Color encodes the c-averaged deviations ∆CMF with white denoting 0% deviation and
dark blue denoting more significant deviations. As expected, we observe a small trend of increasing errors whenever interactions
in the lattice bulk become more frequent, i.e. for high , small δ and small ωd. Opposed to Eq. S8 we include the second order
term that arises due to lattice polymerziation, 1
xg(x), as (cid:29)δ does not necessarily hold true any more.
1
1
(c1) = 0.1 and ρCMF
1
2 δa2∂2
occupation while the diffusive motion of proteins on the MT depends less significantly on mutual correlations [R2].
As a consequence, the CMF approach yields density and tip-bulk correlation profiles for protein occupations along
the MT, which are beyond the scope of our previous approach. As shown in the main text, the latter quantities are
key to a quantitative understanding of tip-localization due to diffusion and capture and related processes.
UNCATALYZED GROWTH AND SHRINKAGE OF MTS
The model described in the main text does not account for MT growth or shrinkage in the absence of depoly-
merziation or polymerization factors like MCAK or XMAP215. The reason for this assumption is twofold: a) In
the experiments with XMAP215 [R3] and MCAK [R4] low concentrations of free tubulin were used such that no
spontaneous MT growth was observed. Also, the measurements in Widlund et al. [R3] suggest that the rate of tubulin
detachment in the corresponding experiments is negligible. b) Concerning MT depolymerization, we aim for a descrip-
tion of protein induced tubulin removal from stabilized MTs in analogy to in vitro experiments with MCAK [R4, R5].
In this way, our model neglects the dynamic instability seen for unstabilized MTs [R6, R7], but provides a description
how a stabilizing structure at the MT tip (e.g. GTP-tubulin) can be removed by regulatory enzymes.
6
That being said, let us emphasize that an extension towards uncatalyzed tubulin attachment and detachment is
feasible based on the model described in the main text. To this end we include further processes in the model: If the
terminal lattice site is unoccupied, a new site can be added at rate δpoly
spont . For completeness,
we also include catalyzed (processive) growth and shrinkage with corresponding rates δpoly
, respectively.
The resulting equations for the CMF framework then read
spont or removed at rate δdepoly
cat and δdepoly
cat
∂tρ(x, t) = 0 = (δdepoly
spont − δpoly
spont)a∂xρ(x, t) + ( +
δpoly
spont)a2∂2
xρ(x, t)
− δpoly
cat
+(δdepoly
+ωac(1 − ρ(x, t)) − ωdρ(x, t) ,
cat + δpoly
spont − δdepoly
1
2
1
δdepoly
spont +
2
1
(δpoly
2
spont )a ∂xg(x, t) +
cat + δdepoly
cat
− δpoly
spont − δdepoly
spont )a2∂2
xg(x, t)
∂tg(x, t) = 0 = (δdepoly
spont ((cid:104)n2(cid:105)(t) − (cid:104)n1n2(cid:105)(t))(a∂xρ(x, t) +
spont + )((cid:104)n2(cid:105)(t) − (cid:104)n1n2(cid:105)(t))ρ(x, t) + δdepoly
δdepoly
cat
xρ(x, t))
xg(x, t) + ωac((cid:104)n1(cid:105)(t) − ρ(x, t) − 2g(x, t))
cat )a∂xg(x, t) + ( +
δpoly
cat +
)a2∂2
a2∂2
1
2
1
2
1
2
− δpoly
+(δdepoly
−(ωd + ωd)g(x, t) ,
cat
d
dt(cid:104)n1(cid:105)(t) = 0 = ((cid:104)n2(cid:105)(t) − (cid:104)n1n2(cid:105)(t)) + δdepoly
dt(cid:104)n2(cid:105)(t) = 0 = ((cid:104)n3(cid:105)(t) − 2(cid:104)n2(cid:105)(t) + (cid:104)n1n2(cid:105)(t)) − δpoly
spont ((cid:104)n2(cid:105)(t) − (cid:104)n1n2(cid:105)(t)) + ωac(1 − (cid:104)n1(cid:105)(t)) − ωd(cid:104)n1(cid:105)(t) ,
((cid:104)n1n3(cid:105)(t) − (cid:104)n1n2(cid:105)(t))
cat (cid:104)n1n2(cid:105)(t) + δdepoly
cat
d
−δpoly
spont((cid:104)n2(cid:105)(t) − (cid:104)n1n2(cid:105)(t)) + δdepoly
spont ((cid:104)n1n2(cid:105)(t) − (cid:104)n1n3(cid:105)(t) + (cid:104)n3(cid:105)(t) − (cid:104)n2(cid:105)(t)) + ωac(1 − (cid:104)n2(cid:105)(t)) − ωd(cid:104)n2(cid:105)(t) ,
(S15)
d
dt(cid:104)n1n2(cid:105)(t) = 0 = ((cid:104)n1n3(cid:105)(t) − (cid:104)n1n2(cid:105)(t)) − δpoly
cat (cid:104)n1n2(cid:105)(t) + δdepoly
cat
((cid:104)n1n3(cid:105)(t) − (cid:104)n1n2(cid:105)(t))
+δdepoly
spont ((cid:104)n2(cid:105)(t)(cid:104)n3(cid:105)(t) − (cid:104)n1n2(cid:105)(t)(cid:104)n3(cid:105)(t)) + ωac((cid:104)n1(cid:105)(t) + (cid:104)n2(cid:105)(t) − 2(cid:104)n1n2(cid:105)(t)) − (ωd + ωd)(cid:104)n1n2(cid:105)(t).
(S16)
(S12)
(S13)
(S14)
The equations are solved in analogy to the case without spontaneous lattice dynamics.
As mentioned above, our models neglect intrinsic MT dynamics such as dynamic instability. However, we expect
validity of our results for tip-localization also under such circumstances. We studied the extended model with spon-
taneous growth and shrinkage rates over a variety of parameter values (up to spontaneous growth and shrinkage rates
spont − δdepoly
of 24 µm/min). For a comparison, we estimated the rate of spontaneous MT growth (vspont = a(δpoly
spont ))
at tubulin concentrations slightly above 5 µM from the experiments performed by Widlund et al. [R3]. At such
tubulin concentrations, MTs were observed to start growing also without the presence of XMAP215 at a speed of
approximately vspont = 0.5 µm/min. Given this resulting spontaneous MT growth rate, we compared a model with
and without fast intrinsic MT dynamics (δpoly
spont = 51 s−1 and
δdepoly
spont = 50 s−1 for a dynamic lattice). The results are shown in Figs. S5 and S6. They show the robustness of the
protein distribution ρ(x) and, in particular, the tip occupation against changes in the lattice growth or shrinkage rates.
Moreover, the CMF approximation is also applicable for rapidly fluctuating MT lengths. Note that XMAP215 also
catalyzes tubulin removal under certain conditions [R8] which could readily be accounted for in the above approach.
spont = 0 for a stable lattice; δpoly
spont = 1 s−1 and δdepoly
MCAK MODEL
spont = δpoly
spont = δdepoly
Similar to the model for XMAP215 stated in the main text we can set up a model for the depolymerase activity of
MCAK, see Fig. S7. The ensuing set of equations corresponding to the CMF approach in the bulk are a special case of
Eqs. S12-S16 with δpoly
cat = 0. We implement a processive depolymerization scheme [R4, R5, R9]. In
detail, MCAK particles stay at the terminal site during depolymerization (i.e. move along with the tip) whenever the
neighboring site is empty. Otherwise, they dissociate from the tip during depolymerization. This means that MCAK
particles fall off the MT tip whenever they hit another particle during the depolymerization process. The results
of the CMF approach for the MCAK model agree excellently with simulation data, as shown in Fig. S8. Further,
also for the MCAK model the MF approximation and FSMFT produce results that deviate from simulation data at
intermediate concentrations.
7
FIG. S5. Extended model that accounts for uncatalyzed growth and shrinkage of MTs. We compare diffusion and capture
cat = 9.5 s−1, blue) with diffusion and capture on a
on a slowly growing lattice (δpoly
= 50 s−1,
lattice with fast intrinsic dynamics but the same average growth speed (δpoly
cat = 59.5 s−1, orange). The average MT growing velocity, and therefore also the tip density, deviate little which implies the
δpoly
validity of our results also on dynamic lattices. MC simulations (symbols) agree well with solutions of the CMF approximation
(lines). Other parameter values are as for the XMAP215 model, see Table I.
spont = 50 s−1, δdepoly
spont = 51 s−1 , δdepoly
spont = 1 s−1, δdepoly
spont = 0, δdepoly
cat
cat
= 0, δpoly
FIG. S6. Density profiles of an adapted model with an intrinsically dynamic lattice (orange) in comparison to the model
presented in the main text (blue) for c = 10 nM and c = 100 nM. Tip-localization occurs also on a lattice with fast spontaneous
growth and shrinkage. The tip-density is almost unaffected by rapid fluctuations of the MT length, suggesting the validity of
our results also for dynamic MTs. The results of our simulations (symbols) agree well with the CMF results (lines). Model
cat = 59.5 s−1 for the dynamic lattice. Other parameters
parameters are δpoly
and parameters for the stable lattice as for the XMAP215 model.
spont = 50 s−1, δdepoly
spont = 51 s−1 , δdepoly
= 50 s−1, δpoly
cat
100200300400XMAP215Concentration[nM]12345v[µm/min]CMFsolutiondynamiclatticeMCsimulationdynamiclatticeCMFsolutionstablelatticeMCsimulationstablelattice 8
FIG. S7. Illustration of the MCAK model. Particle movement is identical to the XMPAP215 model. Depolymerization occurs
whenever the first lattice site is occupied. Particles depolymerize processively in that they move along with the shrinking tip.
When the second site is occupied, a particle on the tip that stimulates shrinkage falls off together with the first lattice site.
FIG. S8. Comparison of different analytic approaches (lines) with simulations of the MCAK model (circles). Whereas the MF
and FSMFT approaches (dashed lines) predict the depolymerization velocity insufficiently, the CMF approximation (solid line)
delivers results which are in excellent agreement with simulation data. Model parameters are given in Table I.
PARAMETER VALUES
The parameter values used for the XMAP215 and MCAK model were extracted from experimental data [R3, R4, R8].
Model parameters were computed based on measured diffusion coefficients (for ), particle dwell times on the MT
tip (for ωd) and bulk (for ωd), attachment rates (for ωa), and maximal (de)polymerization velocities at saturated
(de)polymerase concentrations (for δ). A conversion factor ntubulins from µm into tubulin subunits was adapted to the
assumed protofilament numbers nprotofilaments of the MTs used in the respective experiments: 1625 tubulin dimers/µm
for XMAP215 [R3] and 1750 tubulin dimers/µm for MCAK [R4]. Note that the polymerization velocity refers to one
MT tip [R3, R8], wheres the depolymerization rate refers to the average shrinkage rate of both ends [R4]. Opposed
to the measurements for MCAK, where the maximal depolymerization velocity was determined [R4], Widlund et al.
do not directly state the maximal MT polymerization velocity due to XMAP215 induced growth [R3]. To get a good
estimate for the maximal growing velocity vmax of MTs at saturating polymerase (XMAP215) concentrations, we fitted
a Michaelis-Menten curve to the experimental data. The rate of tubulin attachment and detachment per regulating
10−1100101102103MCAKconcentration[nM]0.00.10.20.30.40.5v[µm/min]MFFSMFT;N=5CMFMCSimulation 9
Experiment
MCAK-FL
XMAP215
Theory
MCAK-FL
XMAP215
D
(µm)2 s−1
7.6 ×10−2
D
(µm)2 s−1
3.0 ×10−1
kon
events /(s µm nM)
4.56 ×10−1
kon
events /(s µm nM)
1×10−1
koff
events/s
1.70
koff
events/s
4.1×10−1
vmax
µm/min
5.0 ×10−1
vmax
µm/min
4.6
KM
µm/(min nM)
4.3
Koff
s−1
2.6 ×10−1
s−1
1.2 ×103
4.7 ×103
ωa
−1
(nM s)
2.61 ×10−4
6 ×10−5
ωd
s−1
1.70
4.1 ×10−1
δ
s−1
5.2 ×10−1
9.5
ωd
s−1
3.0 ×10−2
2.6 ×10−1
TABLE I. Rate constants for MCAK-FL [R4] and XMAP215 [R3, R8]. The diffusion constant D and the on- and off-rates of
enzymes to the MT lattice, kon and koff , were measured directly. The measured depolymerization and polymerization profiles
yield the maximal depolymerization and polymerization velocities vmax and the effective Michaelis constant KM . Conversion to
the theoretical values was achieved by translating kon, koff , and vmax, into appropriate lattice units. The hopping rate is related
to the diffusion coefficient by = D/a2. The off-rate at the first site for MCAK was, in contrast to the one for XMAP215, not
measured directly. It can, however, be estimated from KM by using the depolymerization behavior at low concentrations and
a MF argument which exploits the fact that the system is uncorrelated at asymptotically low occupations [R1].
protein δ depends on the maximal number of catalytically active proteins at the MT tip ntip: vmax = δ ntip n−1
tubulins.
Since the specific number for ntip is elusive (there are estimates for approximately 10 XMAP215s at the MT tip at
50 nM XMAP215. [R8]), we have to make an assumption. Here, we choose one protein per protofilament, ntip =
tubulins = (cid:104)n1(cid:105) δ a, where a is
nprotofilaments. In doing so the MT tip velocity then reduces to v = (cid:104)n1(cid:105) δ nprotofilaments n−1
the length of a tubulin dimer.
As the dwell time of proteins on the tip (i.e. 1/ωd) was not measured for MCAK particles, we used the measured
Michaelis constant KM to estimate this value: Since the Michaelis constant determines the linear increase in the
depolymerization velocity for asymptotically low MCAK concentrations, vlow c = 1/KM × c + O(c2), we can use it to
estimate the tip-dwell time for MCAK particles. In detail, we analytically computed the depolymeriztion velocity for
asymptotically low concentrations using a MF and low-density approximation of our model up to first order in c [R1].
As correlations vanish under these conditions, we expect the result to be exact which allows us to infer the MCAK
off-rate at the tip ωd. The list of ensuing parameters is given in Table I.
∗ Present address: Department of Bionanoscience, Delft University of Technology, Delft, Netherlands.
† [email protected]
[R1] E. Reithmann, L. Reese, and E. Frey, Biophys. J. 108, 787 (2015).
[R2] B. Derrida, J. Stat. Mech. Theor. Exp. 2007, P07023 (2007).
[R3] P. O. Widlund, J. H. Stear, A. Pozniakovsky, M. Zanic, S. Reber, G. J. Brouhard, A. A. Hyman, and J. Howard, Proc.
Nat. Acad. Sci. USA 108, 2741 (2011).
[R4] J. R. Cooper, M. Wagenbach, C. L. Asbury, and L. Wordeman, Nat. Struct. Mol. Biol. 17, 77 (2010).
[R5] J. Helenius, G. J. Brouhard, Y. Kalaidzidis, S. Diez, and J. Howard, Nature 441, 115 (2006).
[R6] T. Antal, P. L. Krapivsky, S. Redner, M. Mailman, and B. Chakraborty, Phys. Rev. E 76, 041907 (2007); R. Padinhateeri,
A. B. Kolomeisky, and D. Lacoste, Biophys. J. 102, 1274 (2012); T. Niedermayer and R. Lipowsky, Phys. Rev. E 92,
052137 (2015).
[R7] P. Zakharov, N. Gudimchuk, V. Voevodin, A. Tikhonravov, F. I. Ataullakhanov, and E. L. Grishchuk, Biophys. J. 109,
2574 (2015).
[R8] G. J. Brouhard, J. H. Stear, T. L. Noetzel, J. Al-Bassam, K. Kinoshita, S. C. Harrison, J. Howard, and A. A. Hyman,
Cell 132, 79 (2008).
[R9] G. Klein, K. Kruse, G. Cuniberti, and F. Julicher, Phys. Rev. Lett. 94, 108102 (2005).
|
1512.03247 | 1 | 1512 | 2015-12-10T13:28:52 | Amyloid Fibril Solubility | [
"physics.bio-ph",
"q-bio.BM"
] | It is well established that amyloid fibril solubility is protein specific, but how solubility depends on the interactions between the fibril building blocks is not clear. Here we use a simple protein model and perform Monte Carlo simulations to directly measure the solubility of amyloid fibrils as a function of the interaction between the fibril building blocks. Our simulations confirms that the fibril solubility depends on the fibril thickness and that the relationship between the interactions and the solubility can be described by a simple analytical formula. The results presented in this study reveal general rules how side-chain side-chain interactions, backbone hydrogen bonding and temperature affect amyloid fibril solubility, which might prove a powerful tool to design protein fibrils with desired solubility and aggregation properties in general. | physics.bio-ph | physics | Amyloid Fibril Solubility
L. G. Rizzi1 and S. Auer1, ∗
1School of Chemistry, University of Leeds, Leeds LS2 9JT, United Kingdom
(Dated: August 14, 2018)
It is well established that amyloid fibril solubility is protein specific, but how solubility depends
on the interactions between the fibril building blocks is not clear. Here we use a simple protein
model and perform Monte Carlo simulations to directly measure the solubility of amyloid fibrils as
a function of the interaction between the fibril building blocks. Our simulations confirms that the
fibril solubility depends on the fibril thickness and that the relationship between the interactions
and the solubility can be described by a simple analytical formula. The results presented in this
study reveal general rules how side-chain side-chain interactions, backbone hydrogen bonding and
temperature affect amyloid fibril solubility, which might prove a powerful tool to design protein
fibrils with desired solubility and aggregation properties in general.
5
1
0
2
c
e
D
0
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
7
4
2
3
0
.
2
1
5
1
:
v
i
X
r
a
INTRODUCTION
The maintenance of proteins in their soluble state is
crucial to protein homeostasis in living organisms because
small changes in their concentration can lead to amyloid-
related diseases1,2. At a given temperature, the solubility
Ce (also known as the "critical concentration"3 -- 6) is the
concentration of monomers C at which the solution is
saturated, for then the fibrils in solution cannot lengthen
or shorten. Its importance for example in protein home-
ostasis arises, because the fibril solubility determines the
concentration at which the fibrils can form, as well as
the fibrilallion kinetics6,7 as it is a central parameter in
existing nucleation theories (e.g. refs.4,6,8 -- 11).
Experimental measurements of fibrils solubilities are
intrinsically difficult, because they can be very low12 -- 18
(typically 0.1-100 µM), and are beyond the detection
limit of commonly used techniques for small oligomers
and fibrils19. Experiments are usually made at high
concentrations where fibrils form spontaneously, or in
the presence of preformed fibrils, and fibril solubilities
can be determined by the ratio of the on and off rates of
fibril elongation13,19. These experiments show that fibril
solubilities can vary strongly with point mutations of the
amino acid sequence20 -- 22, and the fibril structure23,24.
Theoretically, a statistical mechanical description of
the fibril solubility can be obtained in the case of linear
aggregation in terms of molecular partition functions
(e.g. refs.3,5,25,26). Also, aggregation propensities might
be predicted from protein physicochemical properties by
using knowledge-based computational approaches27 -- 30
but they are not able to distinguish between different
fibril structures. The thermodynamic stability of amyloid
fibrils has been studied by computer simulations using
full-atomistic description of proteins31,32 but they are
restricted to fibrils composed of small peptide fragments
and are computationally too costly to be repeated under
different conditions to obtain a full phase diagram. Sim-
ulations using coarse-grained models to explore the self-
assembly of large peptide systems have been performed,
but they often determine a kinetic rather than a thermo-
dynamic phase diagram (e.g. refs.33,34), which have been
shown to be fundamentally different35. A direct calcula-
tion of the fibril solubility has been performed by Auer et
al.36,37 that revealed the existence of various metastable
fibrillar aggregates and that the fibril solubility depends
of fibril thickness. However, such simulations are also too
costly and have only been performed for one set of inter-
action parameters. Computer simulations using a lattice
model for proteins have been performed to determine the
phase diagram from the peak in the heat capacity, but
the interpretation of this in terms of fibril solubility is
not clear38. Although lattice peptide models have been
employed to investigate fibril formation and growth39 -- 43,
they did not provide a general understanding of how
amyloid fibril solubility is determined by protein-protein
interactions.
A theoretical relationship between the interactions of
the building blocks of an aggregate and its solubility can
be provided by a combination of the well known van't
Hoff equation and the Haas-Drenth (HD) model44,45 but
its applicability has never been verified by computer
simulations.
In its integrated form, the van't Hoff
equation is given by
Ce = Cre−λ ,
(1)
where Cr is a reference concentration and λ = L/kBT
is the dimensionless latent heat of peptide aggregation
into fibrillar aggregates The corresponding latent heat
L can be estimated theoretically by the HD model46 for
protein crystals, and λ is given by half the binding energy
of peptides in the aggregates.
PEPTIDE MODEL
To test this, we consider a peptide model that was
recently used to explore the formation of amyloid fibril
networks47.
In this model, all peptides are considered
in their aggregation-prone state and are represented
by hexagons on a two-dimensional (2D) lattice with
2
FIG. 1:
Illustration of an amyloid fibril composed of three
β-sheets. The fibril building blocks (i.e. the peptides) are
represented by hexagons on a triangular lattice. The four
red sides of the hexagons model the hydrophobicity-mediated
side-chain side-chain interactions between peptides, while
the black sides model directional hydrogen bonds between
peptides.
triangular symmetry, Figure 1. Each hexagon has two
opposing strong bonding sides (shown in black) and
four weaker bonding sides (shown in red), so that they
can self-assemble into a highly ordered cross-β structure
characteristic of amyloid fibrils48,49. Here the strong
bonding sides correspond to hydrogen bonding which
drives the formation of β-sheets and the weaker bonding
sides correspond to hydrophobicity-mediated side-chain
side-chain interactions that drive the fibril thickening.
The corresponding dimensionless surface energies for
the strong hydrogen and weak hydrophobicity-mediated
bonds can be written as
ψ = aσ/kBT = E/2kBT
and
ψh = ahσh/kBT = Eh/2kBT ,
(2)
(3)
respectively. Here σ and σh are the corresponding specific
surface energies, and a and ah are the corresponding
surface areas of the peptide faces. The second equality
results from an approximation between the surface ener-
gies and the binding energies E and Eh between peptides.
The energy E corresponds to the formation of directional
backbone hydrogen bonds, whereas Eh is the energy due
to the formation of weaker bonds modeling side-chain
hydrophobic interactions between peptides. The fibril
solubility in the combined van't Hoff and HD model is
then given by
Ce = Cre−2ψ−4ψh ,
(4)
where we used that the dimensionless latent heat is given
by λ = 2ψ + 4ψh.
FIG. 2: Concentration ranges showing the fibril solubility Ce,
the metanucleation border C1β, and intermediary concentra-
tions Ciβ for i ≥ 2.
An important feature characteristic of amyloid fib-
rils is that their solubility depends on the fibril thick-
ness11,36,37,40.
It has been shown that the solubility
of fibrils composed of iβ-sheet layers is given by10(i =
1, 2, 3, . . . )
Ciβ = Cee4ψh/i .
(5)
Substitution of the fibril solubility Ce given by Eq. 4 into
Eq. 5 yields (i = 1, 2, 3, . . . )
Ciβ = Cre−2ψ−4ψh(1−1/i)
.
(6)
The analytical expressions given by Eqs. 4 and 6 define
the concentration range between the fibril solubility
Ce and the threshold concentration C1β at the nucle-
ation/metanucleation border (termed also "supercritical
concentration"4,6) in which fibrils nucleate10,11. As
illustrated in Figure 2, no fibrils can nucleate when the
peptide concentration C < Ce, whereas for C > C1β
the fibril nucleus is a single peptide and fibrils form
in the absence of any nucleation barrier4,10,11,45,50,51.
Importantly, for intermediate concentrations, Ce < C <
C1β, fibril nucleation and elongation depends on their
thickness41. Despite the fundamental
importance of
those expressions, they have not yet been verified by
neither simulations nor experiments.
SIMULATION DETAILS
To do this we perform Monte Carlo (MC) simulations
in which we determine the equilibrium concentration
at which a fibril of a fixed thickness neither grows
nor shrinks. The simulations are performed on a 2D
triangular lattice with 5122 sites. At the beginning of
the simulation one single fibril of fixed thickness i is put
in the center of the simulation box and peptide monomers
3
of Ciβ are given in units of peptides per lattice site.
Even though our results have been obtained in 2D, an
estimate for the fibril solubilities in 3D can be obtained
by scaling them as C 3/2
Such scaling implicitly
iβ .
assumes that the interactions of the peptide in the third
dimension, and thereby to the latent heat, are negligible.
Assuming that the area of the lattice sites is 1 nm2, the
number concentrations in Figure
4a yields threshold
concentrations C1β in 3D around 1 mM and solubilities
Ce in 3D equal to 238 µM for ψh = 0.25 and 0.1 µM for
ψh = 1.5. Figure 4b shows the same data in a ln(Ciβ)-
vs-ψh plot. A fit of these data points to Eq. 6 with
fixed i illustrates the predicted linear dependence with
slope −4(1 − 1/i). Furthermore, a plot of ln(Ciβ)-vs-1/i
(Figure 4c) can be used to fit Eq. 6, which provides
estimates for the solubility Ce in the limit of i → ∞ and
the reference concentration Cr. There is good agreement
between the estimates for Ce obtained by this scaling
analysis and the direct measurements (Figure 4b), and
Cr is independent of ψh (Figure 4d). We note that the
value Cr ≈ 3 (in units of peptides per lattice site) is
related to the number of states that peptides can assume
in our lattice model (Figure 1). The reason for this is
that only one of the three possible orientations can lead
to a successful monomer attachment to the fibril, so that
there will be 3 times more monomers in solution than
would be if all of them were in the same alignment of
the peptides in the fibril. Furthermore, the fact that Cr
is independent of ψh is important for theoretical studies
to predict changes in the solubility as it can be assumed
ref.45). Thus, the main effect of
constant (see e.g.
decreasing the hydrophobicity-mediated interactions ψh
is that it increases the solubilities Ce and Ciβ for i > 1,
but does not alter the metanucleation border C1β. This
rationalizes the experimental observations that amino
acid substitutions that decrease the hydrophobicity-
mediated interaction between the fibril building blocks
decrease its aggregation propensity20,21,52 -- 54.
Hydrogen Bonding Effect. The lengthening of amyloid
fibrils is driven by the formation of backbone hydrogen
bonds in the direction of the fibril
lengthening axis.
The strength of the hydrogen bonding network between
neigboring peptides in the fibril varies depending on
the amino acid sequence, because it determines the
number of hydrogen bonds that can be formed55. To
determine the effect of changes in the hydrogen bonding
interactions we vary ψ while keeping the strength of
the hydrophobicity-mediated interactions ψh fixed. Fig-
ure 5a shows that Ciβ decreases with increasing ψ. A
plot of these data as ln(Ciβ)-vs-ψ coordinates, illustrates
a linear dependence and that the slope is independent
of the fibril thickness. Indeed, a fit of Eq. 6 with fixed
i to these data points shows that the slope is equal to
−2 for all i (Figure 5b). As before, a scaling analysis
to obtain the bulk solubility Ce in the limit of i → ∞
and the reference concentration Cr can be obtained by
FIG. 3: Solubility measurements for a 1β-sheet fibril showing
that, through random attachment and detachment events, the
number concentration of peptides in solution approaches the
equilibrium concentration C1β instead of the solubility Ce.
The blue circle and red squares data were obtained from fibril
elogantion and fibril schrinkage simulations, respectively. The
horizontal dashed and dash-dotted lines are theoretical values
predicted, respectively, by Eqs. 6 and 4 considering ψ = 3 and
ψh = 0.25.
are randomly placed on the remaining lattice sites (see
Figure 1 for a snapshot). In the MC simulations we only
perform diffusion-like and rotational moves as described
in previous simulations39,47. Depending on the peptide
concentration the fibril will either elongate or shrink until
the equilibrium concentration is reached, as illustrated in
Figure 3. We emphasize that during the simulation the
fibril can only lengthen and is not allowed to thicken.
Typically in each simulation we perform 107 MC steps
and each solubility measurement is an average over four
independent simulations. In the following we present our
simulation results for the fibril solubility as a function
of the hydrophobicity parameter ψh, hydrogen bonding
parameter ψ, and temperature to verify the applicability
of Eqs. 4 and 6.
RESULTS AND DISCUSSIONS
presence
Weak Hydrophobic Effect. The
of
hydrophobicity-mediated bonds has been shown to
be an important factor in the formation of fibrillar
aggregates. Fibril structure and point mutations change
the interactions between side-chains of the peptides in
the fibril and can alter their aggregation propensity52.
In Figure 4a we present results of the ψh dependence
of Ciβ at fixed ψ. As can be seen from the figure,
Ciβ decreases with increasing ψh for all fibrils with
a thickness larger than one, and this effect becomes
stronger with increasing fibril thickness. The values
4
there is again good agreement of the results obtained
from our scaling analysis and direct measurements of Ce.
Furthermore, also in this case the so obtained reference
concentration Cr ≈ 3 is independent of ψ (Figure 5d).
The main effect of decreasing the hydrogen bond energy
interaction ψ is that it increases the solubility Ce and
the threshold concentration C1β while keeping the ratio
C1β/Ce constant. This effect might explain experiments
using hydrogen-deuterium exchange mass spectrometry
to demonstrate that the solubilities of different Aβ1−40
polymorphs increases as hydrogen bond formation be-
tween proteins decreases23.
Temperature Dependence. Amyloid fibrils are known to
have high thermodynamic stability, and the temperatures
required for the dissociation of the fibril structure are far
greater than needed for e.g. the denaturation of native
proteins56. To test the effect of temperature on the
fibril solubility we performed simulations in which the we
fixed the binding energies and varied the temperature.
Figure 6a illustrates that Ciβ increases with increasing
kBT , and that this effect is less pronounced with the
increasing number of fibril layers i.
In Figure 6b we
present the same data but in a ln(Ciβ)-vs-1/kBT plot. A
fit of Eq. 6 with fixed i to these data points using a ref-
erence concentration Cr = 3 shows a linear dependence
with a slope equal to −(E + 2Eh(1 − 1/i)) as predicted
by Eq. 6. Furthermore, presenting the same data in a
ln(Ciβ)-vs-1/i plot (Figure 6c), a fit of Eq. 5 to these
data points also provides estimates for the solubility Ce
from the intercept in the limit of i → ∞. As shown in
Figure 6b, such estimates are in good agreement with the
direct measurements of Ce for very thick fibrils (i.e. for
i = 100). In the limiting case when i → ∞, the slope
obtained for the bulk solubility line Ce corresponds to
the latent heat L = λkBT (Eq. 1), which remarkably
is equal to half of the binding energies E + 2Eh, as
predicted by the HD lattice model (see Figure 6d). The
fact that L is half the binding energy cross validates our
assumption that Cr = 3 in the linear fit of Eq. 5 to the
data presented in Figure 6c. Finally, the slopes obtained
from the fit of Eq. 5 to data in Figure 6c, in combination
with Eq. 3, provide values for the surface energy ahσh at
each temperature. As shown in Figure 6e, such values are
equal to half the values of the weak bond energy Eh/2.
Furthermore, combined with the estimate for the latent
heat λ, the values of ahσh can be used to estimate the
strong bond surface energies aσ. Figure 6f confirms that
the values of aσ are independent of kBT and equal to half
the binding energy of the strong bond E/2. Note that the
dashed lines in Figs. 6e and f are estimates obtained from
the fit of Eq. 6 to data presented in Figure 6 considering
i = 1. The main effect of increasing the temperature is
that it increases both C1β and Ce, which is in agreement
with experimental measurements of solubilities for a
short Aβ-model peptide57. In the example presented in
Figure 6, increasing the temperature from 10 oC to 60 oC
FIG. 4:
(a) Dependence of solubilities Ciβ on ψh: diamonds,
up triangles, right triangles, circles, left triangles - simulation
data for 1β, 2β, 3β, 4β, 5β-sheet, respectively; squares - data
obtained for a bulk fibril (i.e. i = 100); dotted lines - guide to
the eyes only. (b) monolog plot of the same data presented in
(a); dashed lines - best fit of Eq. 6; asterisks - data obtained
for i → ∞ from scaling analysis.
(c) ln(Ciβ)-vs-1/i plot;
dashed lines - best fit of Eq. 6. (d) reference concentration Cr
obtained from the scaling analysis described in text; dashed
line is the Cr = 3 line. All results here were obtained for
ψ = 3 at room temperature, i.e. kBT = 2.5 kJ.mol−1.
FIG. 5:
(a) Dependence of solubilities Ciβ on ψ. Symbols
are the same as in Figure 4. Dotted lines are guides to the
eyes only. (b) monolog plot of the same data presented in (a);
dashed lines - best fit of Eq. 6; asterisks - data obtained for
i → ∞ from scaling analysis as described in main text. (c)
ln(Ciβ)-vs-1/i plot for the scaling analysis; dashed lines - best
fit of Eq. 6. (d) reference concentration Cr obtained from the
scaling analysis; dashed line - Cr = 3 line. All results here
were obtained for ψh = 0.5 and kBT = 2.5 kJ.mol−1.
a fit of Eq. 6 to our data presented in a ln(Ciβ)-vs-
1/i plot (Figure 5c). As can be seen in Figure 5b,
5
fibrils of various thickness and the dimensionless specific
surface energies that are given by half the dimensionless
binding energies between the peptides in the fibril (Eqs. 2
and 3). This is important, as these equations can be
used to e.g.
estimate changes in the fibril solubility
upon point mutations or fibril structure. In combination
with existing nucleation theories they can also be used to
predict how such changes affect their fibrillation kinetics
(e.g. refs.44,45). To design peptides that assemble into
fibrils with desired solubility our analysis reveals some
general rules: (i) decreasing the hydrophobicity-mediated
interaction ψh will increase the solubility Ce, corrobo-
rating what was previously suggested by experimental
studies20,21,54; (ii) changes in ψh do not alter the value
of C1β, but the solubilities Ce and Ciβ for i > 1; (iii)
decreasing the hydrogen bond energy interaction ψ will
increase both the solubility line Ce and the metanucle-
ation border C1β while keeping ratio C1β/Ce constant;
(iv) higher temperatures increases both C1β and Ce.
Finally, we note that our model is limited to peptides
in an aggregate-prone state immersed in an implicity
athermal solvent; thus neither conformational changes
nor solvent effects57 are considered.
L. G. Rizzi acknowledge support by the Brazilian
agency CNPq (Grant No 245412/2012-3).
FIG. 6:
(a) Dependence of solubilities Ciβ on kBT . Symbols
are the same as in Figure 4. Dotted lines are guides to
the eyes only. (b) monolog plot of the same data presented
in (a); dashed lines - best fit of Eq. 1 assuming Cr = 3;
asterisks - data obtained for i → ∞ from the scaling analysis
described in main text. (c) ln(Ciβ)-vs-1/i plot for the scaling
analysis; dashed lines - best fit of Eq. 5 with Cr = 3. (d)
Specific heat L extracted at each temperature from data in
(c) by considering Eqs. 1 and 5 assuming Cr = 3; dashed
line - estimate of L from the slope of ln(Ce)-vs-1/kBT in
(b) considering Eq. 1. Panels (e) and (f) show, respectively,
estimates for the surface energies ahσh and aσ evaluated from
the scaling presented in (a) by considering Eq. 1 and Eq. 5 at
each temperature assuming Cr = 3; dashed lines in (e) and (f)
are estimates extracted from the scaling in (b) for comparison
(see text). All results here were obtained for binding energies
E = 15 kJ.mol−1 and Eh = 2.5 kJ.mol−1.
changes the solubilities Ce in 3D from 25 µM to 171 µM,
and the threshold concentration C1β in 3D from 0.61 mM
to 2.56 mM, respectively.
CONCLUSIONS
In summary, the main results of the analysis presented
is that we numerically verified Eqs. 4 and 6 which
provide simple relationships between the solubilities of
∗ Electronic address: [email protected]
[1] Vendruscolo, M.; Knowles, T. P. J.; Dobson, C. M.
Protein Solubility and Protein Homeostasis: A Generic
View of Protein Misfolding Disorders. Cold Spring Harb.
Perspect. Biol. 2011, 3, a010454.
[2] Ciryam, P.; Tartaglia, G. G.; Morimoto, R. I.; Dob-
son, C. M.; Vendruscolo, M. Widespread Aggregation
and Neurodegenerative Diseases Are Associated with
Supersaturated Proteins. Cell Reports 2013, 5, 781.
[3] Hill, T. L. Linear Aggregation Theory in Cell Biology;
Springer-Verlag New York Inc., 1987.
[4] Powers, E. T.; Powers, D. L. The Kinetics of Nucleated
Polymerizations at High Concentrations: Amyloid Fibril
Formation Near and Above the "Supercritical Concen-
tration". Biophys. J. 2006, 91, 122.
[5] Lee, C. F. Self-Assembly of Protein Amyloids: A Com-
petition Between Amorphous and Ordered Aggregation.
Phys. Rev. E 2009, 80, 031922.
[6] Gillam, J. E.; MacPhee, C. E. Modelling Amyloid Fib-
ril Formation Kinetics: Mechanisms of Nucleation and
Growth. J. Phys.: Condens. Matter 2013, 25, 373101.
[7] Schmittschmitt, J. P.; Scholtz, J. M. The Role of Protein
Stability, Solubility, and Net Charge in Amyloid Fibril
Formation. Prot. Sci. 2003, 12, 2374.
[8] Jarrett, J. T.; Lansbury, P. T. Seeding "One-Dimensional
Crystallization" of Amyloid: A Pathogenic Mechanism in
Alzheimer's Disease and Scrapie? Cell 1993, 73, 1055.
[9] Lomakin, A. L.; Teplow, D. B.; Kirschner, D. A.;
Benedek, G. B. Kinetic Theory of Fibrillogenesis of
Amyloid β-Protein. Proc. Natl. Acad. Sci. USA 1997,
94, 7942.
6
[10] Kashchiev, D.; Auer, S. Nucleation of Amyloid Fibrils.
J. Mol. Biol. 2004, 341, 1317.
J. Chem. Phys. 2010, 132, 215101.
[11] Kashchiev, D.; Cabriolu, R.; Auer, S. Confounding the
Paradigm: Peculiarities of Amyloid Fibril Nucleation. J.
Am. Chem. Soc. 2013, 135, 1531.
[12] Jarrett, J. T.; Costa, P. R.; Griffin, R. G.; Lansbury, P. T.
Models of the β Protein C-terminus: Differences in
Amyloid Structure May Lead to Segregation of "Long"
and "Short" Fibrils. J. Am. Chem. Soc. 1994, 116, 9741.
[13] O'Nuallain, B.; Shivaprasad, S.; Kheterpal, I.; Wetzel, R.
Thermodynamics of Aβ(1-40) Amyloid Fibril Elongation.
Biochemistry 2005, 44, 12709.
[14] Garai, K.; Sahoo, B.; Sengupta, P.; Maiti, S. Quasiho-
mogeneous Nucleation of Amyloid Beta Yields Numerical
Bounds for the Critical Radius, the Surface Tension, and
the Free Energy Barrier for Nucleus Formation. J. Chem.
Phys. 2008, 128, 045102.
[15] Hellstrand, E.; Boland, B.; Walsh, D. M.; Linse, S.
Amyloid β-Protein Aggregation Produces Highly Re-
producible Kinetic Data and Occurs by a Two-Phase
Process. ACS Chem. Neurosci. 2010, 1, 13.
[16] Baldwin, A. J.; Knowles, T. P. J.; Tartaglia, G. G.;
Fitzpatrick, A. W.; Devlin, G. L.; Shammas, S. L.;
Waudby, C. A.; Mossuto, M. F.; Meehan, S.; Gras, S. L.
et al. Metastability of Native Proteins and the Phe-
nomenon of Amyloid Formation. J. Am. Chem. Soc.
2011, 133, 14160.
[17] Davies, R. P. W.; Aggeli, A. Self-Assembly of Am-
phiphilic β-Sheet Peptide Tapes Based on Aliphatic Side
Chains. J. Pept. Sci. 2011, 17, 107 -- 114.
[18] Yagi, H.; Hasegawa, K.; Yoshimura, Y.; Goto, Y. Ac-
celeration of the Depolymerization of Amyloid β-Fibrils
by Ultrasonication. Biochimica et Biophysica Acta 2013,
1834, 2480.
[19] Harper, J. D.; Lansbury, P. T. Models of Amyloid
Seeding in Alzheimer's Disease and Scrapie: Mechanistic
Truths and Physiological Consequences of the Time-
Dependent Solubility of Amyloid Proteins. Annu. Rev.
Biochem. 1997, 66, 385.
[20] Wood, S. J.; Wetzel, R.; Martin, J. D.; Hurle, M. R.
Prolines and Amyloidogenicity in Fragments of the
Alzheimer's Peptide βA4. Biochemistry 1995, 34, 724.
[21] Wurth, C.; Guimard, N. K.; Hecht, M. H. Mutations
That Reduce Aggregation of the Alzheimer's Aβ42 Pep-
tide: An Unbiased Search for the Sequence Determinants
of Aβ Amyloidogenesis. J. Mol. Biol. 2002, 319, 1279.
[22] Williams, A. D.; Shivaprasad, S.; Wetzel, R. Alanine
Scanning Mutagenesis of Aβ(1-40) Amyloid Fibril Sta-
bility. J. Mol. Biol. 2006, 357, 1283.
[23] Kodali, R.; Williams, A. D.; Chemuru, S.; Wetzel, R.
Aβ(1-40) Forms Five Distinct Amyloid Structures Whose
β-sheet Contents and Fibril Stabilities Are Correlated. J.
Mol. Biol. 2010, 401, 503.
[24] Qiang, W.; Kelley, K.; Tycko, R. Polymorph-Specific Ki-
netics and Thermodynamics of β-Amyloid Fibril Growth.
J. Am. Chem. Soc. 2013, 135, 6860.
[25] Ferrone, F. A. Analysis of Protein Aggregation Kinetics.
Methods Enzymol. 1999, 309, 256.
[26] Schreck, J. S.; Yuan, J.-M. Statistical Mechanical Treat-
ments of Protein Amyloid Formation. Int. J. Mol. Sci.
2013, 14, 17420.
[27] DuBay, K. F.; Pawar, A. P.; Chiti, F.; Zurdo, J.; Dob-
son, C. M.; Vendruscolo, M. Prediction of the Absolute
Aggregation Rates of Amyloidogenic Polypeptide Chains.
[28] Fernandez-Escamilla, A.; Rousseau, F.; Schymkowitz, J.;
Serrano, L. Prediction of Sequence-Dependent and Mu-
tational Effects on the Aggregation of Peptides and
Proteins. Nature Biotech. 2004, 22, 1302.
[29] Conchillo-Sol´e, O.; de Groot, N. S.; Avil´es, F. X.;
Vendrell, J.; Daura, X.; Ventura, S. AGGRESCAN: A
Server for the Prediction and Evaluation of "Hot Spots"
of Aggregation in Polypeptides. BMC Bioinfo. 2007, 8,
65.
[30] Tartaglia, G. G.; Pawar, A. P.; Campioni, S.; Dob-
son, C. M.; Chiti, F.; Vendruscolo, M. Prediction of
Aggregation-Prone Regions in Structured Proteins. J.
Mol. Biol. 2008, 380, 425.
[31] Wu, C.; Shea, J.-E. Coarse-Grained Models for Protein
Aggregation. Curr. Opin. Struct. Biol. 2011, 21, 209.
[32] Morriss-Andrews, A.; Shea, J.-E. Simulations of Pro-
tein Aggregation: Insights From Atomistic and Coarse-
Grained Models. J. Phys. Chem. Lett. 2014, 5, 1899.
[33] Nguyen, H. D.; Hall, C. K. Phase Diagrams Describing
Fibrillization by Polyalanine Peptides. Biophys. J. 2004,
87, 4122.
[34] Nguyen, H. D.; Hall, C. K. Spontaneous Fibril Forma-
tion by Polyalanines; Discontinuous Molecular Dynamics
Simulations. J. Am. Chem. Soc. 2006, 128, 1890.
[35] Ricchiuto, P.; Brukhno, A. V.; Auer, S. Protein Aggrega-
tion: Kinetics Versus Thermodynamics. J. Phys. Chem.
B 2012, 116, 8703.
[36] Auer, S.; Kashchiev, D. Phase Diagram of α-Helical and
β-Sheet Forming Peptides. Phys. Rev. Lett. 2010, 104,
168105.
[37] Auer, S. Phase Diagram of Polypeptide Chains. J. Chem.
Phys. 2011, 135, 175103.
[38] Ni, R.; Abeln, S.; Schor, M.; Stuart, M. A. C.; Bol-
huis, P. G. Interplay Between Folding and Assembly of
Fibril-Forming Polypeptides. Phys. Rev. Lett. 2013, 111,
058101.
[39] Zhang, J.; Muthukumar, M. Simulations of Nucleation
and Elongation of Amyloid Fibrils. J. Chem. Phys. 2009,
130, 035102.
[40] Cabriolu, R.; Kashchiev, D.; Auer, S. Breakdown of Nu-
cleation Theory for Crystals with Strongly Anisotropic
Interactions Between Molecules. J. Chem. Phys. 2012,
137, 204903.
[41] Bingham, R. J.; Rizzi, L. G.; Cabriolu, R.; Auer, S.
Non-Monotonic Supersaturation Dependence of the Nu-
cleus Size of Crystals with Anisotropically Interacting
Molecules. J. Chem. Phys. 2013, 139, 241101.
[42] Irback, A.; Jonsson, S.; Linnemann, N.; Linse, B.;
Wallin, S. Aggregate Geometry in Amyloid Fibril Nu-
cleation. Phys. Rev. Lett. 2013, 110, 058101.
[43] Irback, A.; Wess´en, J. Thermodynamics of Amyloid
Formation and the Role of Intersheet Interactions. J.
Chem. Phys. 2015, 143, 105104.
[44] Cabriolu, R.; Auer, S. Amyloid Fibrillation Kinetics:
Insight From Atomistic Nucleation Theory. J. Mol. Biol.
2011, 411, 275.
[45] Auer, S. Nucleation of Polymorphic Amyloid Fibrils.
Biophys. J. 2015, 108, 1176.
[46] Haas, C.; Drenth, J. The Interaction Energy Between
Two Protein Molecules Related to Physical Properties of
Their Solution and Their Crystals and Implications for
Crystal Growth. J. Cryst. Growth 1995, 154, 126.
[47] Rizzi, L. G.; Head, D. A.; Auer, S. Universality in the
Morphology and Mechanics of Coarsening Amyloid Fibril
Networks. Phys. Rev. Lett. 2015, 114, 078102.
[48] Sawaya, M. R.;
I.;
Sievers, S. A.; Apostol, M.
Sambashivan, S.; Nelson, R.;
Ivanova, M.
I.;
Thompson, M. J.; Balbirnie, M.; Wiltzius, J. J. W.;
McFarlane, H. T. et al. Atomic Structures of Amyloid
Cross-β Spines Reveal Varied Steric Zippers. Nature
2007, 447, 453.
[49] Fitzpatrick, A. W. P.; Debelouchina, G. T.; Bayro, M. J.;
Claire, D. K.; Caporini, M. A.; Bajaj, W. S.; Ja-
roniec, C. P.; Wang, L.; Ladizhansky, V.; Muller, S. A.
et al. Atomic Structure and Hierarchical Assembly of a
Cross-β Amyloid Fibril. P. Natl. Acad. Sci. USA 2013,
110, 5468.
[50] Auer, S. Amyloid Fibril Nucleation: Effect of Amino Acid
Hydrophobicity. J. Phys. Chem. B 2014, 118, 5289.
[51] Ferrone, F. A. Assembly of Aβ Proceeds via Monomeric
Nuclei. J. Mol. Biol. 2015, 427, 287.
[52] Chiti, F.; Stefani, M.; Taddei, N.; Ramponi, G.; Dob-
son, C. M. Rationalization of the Effects of Mutations on
Peptide and Protein Aggregation Rates. Nature 2003,
424, 805.
7
[53] Chiti, F.; Taddei, N.; Baroni, F.; Capanni, C.; Ste-
fani, M.; Ramponi, G.; Dobson, C. M. Kinetic Parti-
tioning of Protein Folding and Aggregation. Nat. Struct.
Biol. 2002, 9, 137.
[54] de Groot, N. S.; Aviles, F. X.; Vendrell, J.; Ventur, S.
Mutagenesis of the Central Hydrophobic Cluster in
Aβ42 Alzheimer's Peptide Side-Chain Properties Corre-
late with Aggregation Propensities. FEBS Journal 2006,
273, 658.
[55] Cheng, P. N.; Pham, J. D.; Nowick, J. S. The
Supramolecular Chemistry of β-Sheets. J. Am. Chem.
Soc. 2011, 135, 5477.
[56] Meersman, F.; Dobson, C. M. Probing the Pressure-
Temperature Stability of Amyloid Fibrils Provides New
Insights into Their Molecular Properties. Biochimica et
Biophysica Acta 2006, 1764, 452.
[57] Hamley, I. W.; Nutt, D. R.; Brown, G. D.; Miravet, J. F.;
Escuder, B.; Rodr´ıguez-Llansola, F. Influence of the
Solvent on the Self-Assembly of a Modified Amyloid Beta
Peptide Fragment. II. NMR and Computer Simulation
Investigation. J. Phys. Chem. B 2010, 114, 940.
|
1210.4332 | 1 | 1210 | 2012-10-16T09:50:15 | Collective Chemotactic Dynamics in the Presence of Self-Generated Fluid Flows | [
"physics.bio-ph",
"physics.flu-dyn"
] | In micro-swimmer suspensions locomotion necessarily generates fluid motion, and it is known that such flows can lead to collective behavior from unbiased swimming. We examine the complementary problem of how chemotaxis is affected by self-generated flows. A kinetic theory coupling run-and-tumble chemotaxis to the flows of collective swimming shows separate branches of chemotactic and hydrodynamic instabilities for isotropic suspensions, the first driving aggregation, the second producing increased orientational order in suspensions of "pushers" and maximal disorder in suspensions of "pullers". Nonlinear simulations show that hydrodynamic interactions can limit and modify chemotactically-driven aggregation dynamics. In puller suspensions the dynamics form aggregates that are mutually-repelling due to the non-trivial flows. In pusher suspensions chemotactic aggregation can lead to destabilizing flows that fragment the regions of aggregation. | physics.bio-ph | physics | Collective Chemotactic Dynamics in the Presence of Self-Generated Fluid Flows
Enkeleida Lushi1,2, Raymond E. Goldstein3, Michael J. Shelley1
1Courant Institute of Mathematical Sciences, New York University, NY, USA
2Department of Mathematics, Imperial College London, London, UK
3Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, UK
In micro-swimmer suspensions locomotion necessarily generates fluid motion, and it is known
that such flows can lead to collective behavior from unbiased swimming. We examine the comple-
mentary problem of how chemotaxis is affected by self-generated flows. A kinetic theory coupling
run-and-tumble chemotaxis to the flows of collective swimming shows separate branches of chemo-
tactic and hydrodynamic instabilities for isotropic suspensions, the first driving aggregation, the
second producing increased orientational order in suspensions of "pushers" and maximal disorder in
suspensions of "pullers". Nonlinear simulations show that hydrodynamic interactions can limit and
modify chemotactically-driven aggregation dynamics. In puller suspensions the dynamics form ag-
gregates that are mutually-repelling due to the non-trivial flows. In pusher suspensions chemotactic
aggregation can lead to destabilizing flows that fragment the regions of aggregation.
PACS numbers: 87.17.Jj, 87.18.Hf, 47.63.Gd, 05.20.Dd
A growing body of experimental work has established
that suspensions of motile microorganisms can develop
complex large-scale patterns of collective swimming at
sufficiently high concentration [1, 2]. This behavior gen-
erally occurs in the absence of directional cues for swim-
ming, purely as a consequence of steric and hydrody-
namic interactions between the cells. Yet, there are many
circumstances in which cells exhibit chemotaxis, directed
motion in response to chemical gradients, and this pro-
cess by itself can lead to complex spatio-temporal pattern
formation [3]. As swimmer-generated flows may also ad-
vect any chemoattractant field, it is natural then to ask
how self-generated fluid flows in suspensions of microor-
ganisms affect modes of communication [4] and aggrega-
tion. Here we present an analysis of this issue and suggest
potential realizations of this pattern-forming system.
Chemotactic focusing of cell concentration has been
studied using the classical Keller-Segel (KS) model [5]
and in theories incorporating the run-and-tumble (RT)
[6] dynamics of bacterial motion [7, 8]. We extend a
well-known kinetic model for modulated RT dynamics
to include flows produced by the active stresses due to
swimming. A simpler version of our model is consid-
ered in [7] to study swimmer transport and rotation in
a given background shear flow. Without RT dynam-
ics, our model reduces to one for active suspensions [9]
which captures the large-scale flows seen in experiments
[1, 10] and illuminates the effect of propulsion mechanism
(pusher vs. puller) on large-scale dynamics and stability.
When swimmers produce a chemoattractant leading to
aggregation, the self-generated flows can have a large ef-
fect; pushers create complex flows that can bound growth
in organism density, while pullers show limited pattern
coarsening and isolated aggreggates repelling due to non-
trivial flows. Merging the RT and active suspension mod-
els is seamless as both are kinetic theories with conforma-
tion variables the particle position and orientation [11].
Consider a suspension of swimmers at local concentra-
tion Φ(x, t), each of which moves at a constant speed
U in a run-and-tumble dynamics. They move in a fluid
of local velocity u(x, t) and produce a chemoattractant
of concentration C(x, t) and molecular diffusivity DC.
We choose rescalings based on the swimmer contribution
to the fluid stress tensor (below), with a characteristic
length (cid:96) = l/φ, where l is the swimmer size and φ ≡ l3(cid:104)Φ(cid:105)
is the effective mean volume fraction in suspension. Scal-
ing time by (cid:96)/U , C evolves as
Ct + u · ∇C = P e−1∇2C − β1C + β2Φ ,
(1)
where β1 and β2 are rate constants for chemoattractant
self-degradation and production, and the P´eclet number
P e = U (cid:96)/DC measures the strength of diffusion to ad-
vection on the intrinsic scale (cid:96). Collective swimming may
generate coherence on larger scales with higher speeds,
increasing the importance of advection. Without advec-
tion this is the KS model [5]. In the case of E. coli [12]
gives U ∼ 25 µm/s, l ∼ 5 µm, φ ∼ 0.1 at a cell concen-
tration of 109 cm−3, so (cid:96) ∼ 50 µm. With DC ∼ 5× 10−6
cm2/s we obtain P e (cid:39) 2.5, β1 (cid:39) 0.008 and β2 (cid:39) 0.004.
For faster-swimming organisms, such as marine bacteria
[13], this intrinsic P´eclet number can reach O(10 − 20).
The configuration of swimmers is given by a distribu-
tion function Ψ(x, p, t) of the center of mass position x
and orientation p satisfying the Fokker-Planck equation
(cid:20)
(cid:90)
(cid:21)
Ψt = −
λ(DtC)Ψ − 1
4π
dp(cid:48)λ(DtC)Ψ(x, p(cid:48), t)
− ∇x · (Ψ x) − ∇p · (Ψ p) ,
(cid:82) dpΨ(x, p, t). The bracketed term in (2) describes the
where the local swimmer concentration is Φ(x, t) =
(2)
effect of RT chemotaxis based on a swimming dynamics
of straight runs and modulated reorientations (tumbles)
where λ(DtC) is a tumbling frequency, the probability of
2
1
0
2
t
c
O
6
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
2
3
3
4
.
0
1
2
1
:
v
i
X
r
a
2
FIG. 1: (color online). Chemotactic suspensions at large times, t = 3000. Swimmer concentration Φ (top) and mean direction
n (bottom) for (i) chemotactic neutral swimmers (u = 0), and (ii) Φ (top), fluid streamlines and velocity field u (bottom) for
chemotactic pullers. (iii,iv) Φ (top), streamlines and u (bottom) for chemotactic pushers. Parameters β1 = β2 = 0.25, P e = 20,
γ = 1, DT = DR = 0.025. Parameters λ0 and χ are indicated in Figs. 2c,d for cases (i-iii) and are λ0 = 5, χ = 0.6 for case (iv).
See [16] for movies of swimmer concentration dynamics.
a bacterial tumble event as a function of the chemoat-
tractant temporal gradient DtC = Ct + (p + u) · ∇C
along a swimmer's path. Experiments [14] show that
when DtC > 0 the tumbling frequency is reduced, and
is otherwise constant, as captured by the biphasic form
λ(DtC) = λ0 max(min(1 − χDtC, 1), 0), a linearized ver-
sion of an earlier model [8, 12]. The fluxes in (2) are
x = p + u,
p = (I − pp) (γE + W) p.
(3)
The particle velocity x includes swimming at constant
speed (non-dimensionalized to unity) in the axis direc-
tion p (p = 1), translation by the fluid velocity u. (In
the KS model [5], swimmer speed is linear in the chem-
ical gradient.) The angular velocity p follows Jeffrey's
equation [15] where E = (∇xu + ∇xuT )/2 is the rate-
of-strain tensor, W = (∇xu − ∇xuT )/2 is the vorticity
tensor. For rod-like swimmers, the shape factor is γ ∼ 1.
The fluid velocity u(x, t) produced by the suspension
satisfies the Stokes equations driven by an "active" stress
Σa arising from particle locomotion:
− ∇2
xu + ∇xq = ∇x · Σa, ∇x · u = 0
Σa(x, t) = α(cid:82) dpΨ(x, p, t)pp.
(4)
(5)
The active stress is an orientational average of the force
dipoles αpp the cells exert on the fluid [9], where α is an
O(1) constant by our rescaling. A cell that self-propels
by front-actuation (a puller) has stresslet strength α > 0,
and a rear-actuated cell (pusher) has α < 0. The case
of "neutral" cells (α = u = 0) is the closest this model
approaches the KS [5] and RT models [8, 12].
We first illustrate the effect hydrodynamics has on
aggregation. From nonlinear simulations of Eqs. (1-6),
Fig. 1 shows the swimmer concentration Φ(x, t) and mean
orientation n = (cid:82) dpΨ/Φ at late times, having started
near uniform isotropy, for neutral, puller, and pusher
suspensions. All share a dominant self-aggregation in-
stability, but differing (or no) hydrodynamic interactions.
Neutral swimmers show aggregation and pattern coarsen-
ing. Pullers show limited aggregation into circular spots
kept apart by non-trivial fluid flows. Pushers create com-
plex fluid flows and fragmented aggregation regions.
These behaviors can be understood through a stability
analysis of uniform isotropic suspensions. For simplicity,
consider rod-like (γ = 1) swimmers and a quasi-static
chemoattractant field P e−1∇2C − β1C = −β2Φ, which
slaves C to Φ. The tumbling frequency is simplified to
λ(p) = λ0 (1 − χp · ∇C). A steady state is Ψ0 = 1/4π
(Φ = 1), u = 0, and C0 = β1/β2. Perturbations of the
form ( Ψ(p, k), C(k)) exp(ik · x + σt), yield
ikχβ2(k · p)
β1 + k2P e−1 + 1
(σ + λ0 + ik · p) Ψ =
(cid:33)
(cid:32)
Φ
where Φ =(cid:82) dp(cid:48) Ψ(cid:48) and k = kk. Since Σp =(cid:82) dp(cid:48) Ψ(cid:48)p(cid:48)p(cid:48),
(k · p)p · (I − kk) Σpk ,
(6)
this is a linear eigenvalue problem for Ψ and σ. The first
term on the RHS is chemotactic (C) and has unstable
dynamics restricted to the zeroeth azimuthal mode on
p = 1. The second is hydrodynamic (H), with unstable
dynamics restricted to the first azimuthal mode. This
λ0
4π
− 3α
4π
3
FIG. 2: (color online). Linear stability analysis. (a) Branches of the hydrodynamic instability, with and without tumbling,
for pushers. (b) Chemotactic branch for χ = 35, λ0 = 0.25, β1 = β2 = 1/4 and P e = 20. Regimes diagram for (c) neutral
swimmers and pullers, and (d) pushers. Solid curves give linear stability boundaries for long waves. Dashed lines show shifted
boundaries in nonlinear simulations at finite box size. Encircled labels (i-iii) denote parameters used in simulations (Fig. 1a-c).
yields uncoupled relations for growth rates σC,H ,
(cid:20)
(cid:18) aC − 1
(cid:19)(cid:21)
(cid:19)
(cid:18) aC − 1
(cid:19)
aC + 1
,
(7)
− 1
ik
log
(cid:18) aH − 1
aH + 1
2
λ0
4k
3iα
= R
2 + aC log
aC + 1
H − a2
aH + (a4
H ) log
= 2a3
H − 4
3
where aC,H = (σC,H +λ0)/ik and R = χβ2/(β1 +k2/P e).
We refer to these as the chemotactic and hydrodynamic
relations, respectively. The first induces growth in con-
centration fluctuations, while the second increases orien-
tational order. The two are coupled only through the
basal tumbling rate λ0 which in the hydrodynamic re-
lation only shifts the growth rate. Further, the chemo-
tactic instability gives rise to normal stresses of the form
Σp = kk − k⊥ k⊥, while the hydrodynamic instability
gives shear stresses of the form Σp = kk⊥ + k⊥ k.
For pushers (α < 0) the hydrodynamic instability
has a finite bandwidth (Fig. 2a), though with maxi-
mal growth rates at k = 0. Tumbling shifts down
the Re(σH (k)) branch by λ0 for all k, further stabi-
lizing the system. Long-wave asymptotics of (7) give
two solution branches: σH1 (cid:39) −α/5 − λ0 + 15/7αk2
and σH2 (cid:39) −λ0 + O(−αk2). There is no hydrody-
namic instability for pullers [9]. Fig. 2b shows the
chemotactic growth rate. Small k asymptotics yields
σC ≈ k2/(3λ0)[(χβ2/β1)λ0 − 1]:
for (χβ2/β1) > 1/λ0
there are wavenumbers with Re(σC(k)) > 0, shown in
one case as a finite band of unstable modes whose width
is controlled by chemo-attractant diffusion.
From Fig. 2a, we can obtain a range for λ0 for which
there is a hydrodynamic instability in pusher suspen-
sions. Heuristically, λ0 sets an effective rotational dif-
fusivity, and λ0 ≥ 0.2 turns off the hydrodynamic insta-
bility for any system size. For L = 50 and the diffusion
constants used in simulations, λ0 ≥ 0.09 suffices. This
information is assembled in Fig. 2 c,d as phase diagrams
that relate the parameters to various dynamical regimes.
Numerical studies of the full nonlinear dynamics (1–5)
were done in 2D, with a box size L = 50 large enough to
include several unstable linear modes. Swimmer transla-
tional and rotational diffusions are added in the model to
control the growth of steep gradients over long-times. An
initial random perturbation of the uniform isotropic state
is used: Ψ(x, p, 0) = 1/2π+Σii cos(ki·x+ξi)Qi(pi) with
random coefficients i < 0.01, ξi an arbitrary phase and
Qi a low-order polynomial. The initial chemo-attractant
concentration is uniform with C(x, 0) = β1/β2 = 1. Fig-
ure 1 shows long-time swimmer concentration Φ for four
illustrative cases. In each case, concentration C closely
tracks Φ. Cases (i-iii) share the same chemotactic insta-
bility, but differ in swimming actuation: α = 0, 1,−1.
The expected regimes of these three cases are shown in
Figs. 2c,d. For neutral swimmers, aggregation dominates
and the dynamics is typified by the formation of a few
regions of steadily increasing concentration that slowly
coarsen (Fig. 1a). The maximum swimmer concentra-
tion (Fig. 3) shows little sign of the rapid self-focussing
associated with finite-time chemotactic collapse [17] of
the KS model, which here may be due to the fixed swim-
ming speed [18]. While the initial aggregation for pullers
(Fig. 1b) is similar to that for neutral swimmers (Fig.
3) its long-time behavior is very different. Concentration
growth and coarsening cease as the dynamics enters a
near steady-state with circular regions of high concentra-
tion. Active-stress driven flows suppress further coarsen-
ing by pushing nearby peaks apart and apparently main-
tain the few remaining high concentration regions.
For pushers (Fig. 1c), linear theory gives only a chemo-
tactic instability, and the dynamics is indeed initially
dominated by aggregation as is evidenced by the early
rapid growth of normal stresses relative to shear stresses
(not shown). However, aggregation into a regions with
high swimmer concentration creates a destabilizing ac-
tive stress, giving rise to unsteady fluid flows. These
flows fragment the peaks while pushing them around the
4
Nonetheless, we expect similar results when large-scale
coherence is driven by steric effects [10, 19]. On that
note, steric effects with no hydrodynamics may also limit
aggregation of chemotactic random walkers [20].
Finally, these auto-chemotactic effects can be seen as
complementary to the enhanced mixing by swimmers
[11] that has also been explored for microfluidic appli-
cations [21]. Systematic studies of the interplay between
chemotaxis and locomotion-generated fluid flow should
be possible through controlled introduction of exoge-
neous chemoattractants to trigger aggregation, through
the interplay of quorum sensing and chemotaxis [22], and
perhaps by specific genetic engineering of the dynamics
of locomotion and chemosensing [23].
This work was supported in part by NSF grants DMS-
0652775 and DMS-0652795, and DOE grant DEFG02-
00ER25053 (E.L., M.J.S.) and an ERC Advanced Inves-
tigator Grant 247333 (R.E.G.).
[1] C. Dombrowski, et al., Phys. Rev. Lett. 93, 098103
(2004); L.H. Cisneros, et al., Exp. Fluids 43, 737 (2007).
[2] S. Ramaswamy, Ann. Rev. Condens. Matter Phys. 1, 323
(2010).
[3] E.O. Budrene and H.C. Berg, Nature 349, 630 (1991).
[4] B.L. Bassler, Cell 109, 421 (2002).
[5] E.F. Keller, L.A. Segel, J. Theor. Biol. 30, 225 (1971).
[6] H.C. Berg, Random Walks in Biology Princeton Univer-
sity Press, 1993
[7] R.N. Bearon, T.J. Pedley, Bull. Math. Bio. 62, 775
(2000).
[8] K.C. Chen, et al., J. Math. Bio. 47, 518 (2003).
[9] D. Saintillan and M.J. Shelley, Phys. Rev. Lett. 100,
178103 (2008); also Phys. Fluids 20, 123304 (2008).
[10] L.H. Cisneros, et al., Phys. Rev. E 83, 061907 (2011).
[11] E. Lushi and M.J. Shelley, preprint (2012).
[12] J. Saragosti, et al., Proc. Natl. Acad. Sci. USA 108,
16235 (2011).
[13] R. Stocker, et al., Proc. Natl. Acad. Sci. USA 105, 4209
(2008).
[14] R.M. Macnab and D.E. Koshland, Proc. Natl. Acad. Sci.
USA 69, 2509 (1972); N. Mittal, et al., Proc. Natl. Acad.
Sci. USA 100, 13259 (2003).
[15] G.B. Jeffery, Proc. R. Soc. London, Ser. A 102, 161
(1922).
[16] See Supplementary Material at [URL to be inserted by
publisher] for movies of the swimmer concentration.
[17] S. Childress and J.K. Percus, Math. BioSci. 56, 217
(1981).
[18] M.J. Schnitzer, et al., Symp. Soc. Gen. Microbiol 46, 15
(1990).
[19] K. Drescher, et al., Proc. Natl. Acad. Sci. USA 108,
10940 (2011).
[20] J. Taktikos, et al., Phys. Rev. E 85, 051901 (2012).
[21] M.J. Kim and K.S. Breuer, Anal. Chem. 79, 955 (2007).
[22] S. Park, et al.. Science 301, 188 (2003).
[23] C. Liu, et al.. Science 334, 238 (2011).
FIG. 3: (color online) Measures of growth: maximum swim-
mer concentration for cases (i-iii) in Fig. 1.
domain. The dynamics is one of constant aggregation
and flow instability, which apparently suppresses further
growth in swimmer concentration (Fig. 3).
Lastly, we examine in Fig. 1(iv) the dynamics that
arises with parameters close to those measured by
Saragosti et al [12] (before our rescaling) in their experi-
ments of E. coli chemotaxis. These parameters lie far to
the right of the aggregation regime of Fig. 2d as λ0 is 20
times higher than at the predicted threshold for suppress-
ing hydrodynamic instabilities. Not surprisingly, the
simulations show chemotactic aggregation into very high
peaks. Once the swimmer concentration in those peaks
is large enough, the active stresses give rise to small-scale
and localized fluid flows (cf. Fig. 1(iii)). These local flows
do appear to be implicated in the slow "wriggling" we ob-
serve of the saturated aggregates (see Supplementary Ma-
terial [16]). The experiments of Saragosti et al [12], which
are performed in confined micro-channels and capillaries,
show instead the development of traveling concentration
waves of chemotactic bacteria. These traveling waves
were initiated in the experiments through an initial con-
centration by centrifugation of the swimmer population
to one end of the channel. We do not observe the spon-
taneous formation of such traveling waves here though
ours is an open system (though confined geometrically
by the assumed periodicity length) and the initial swim-
mer state is un-oriented and nearly homogeneous. The
combined effects of a confining geometry and the initial
concentration of swimmers has yet to be examined in our
theoretical system.
We have shown that the intrinsically generated fluid
flows arising from collective swimming of microorgan-
isms can modify patterns of chemotactic aggregation
and, most importantly, can limit aggregate concentra-
tion. This is unlike chemotactic models that predict
concentration blow-up or include artificial terms to cap
growth. While we have emphasized hydrodynamic ef-
fects in attractive chemotactic dynamics, it is important
to remember that ours is a dilute to semi-dilute theory
that does not capture near-interactions between swim-
mers, hydrodynamic or otherwise. In denser suspensions
swimmer size limits local swimmer density through steric
interactions though as yet well-founded models that com-
bine these with hydrodynamic interactions do not exist.
|
1501.02163 | 1 | 1501 | 2015-01-09T14:54:16 | Direct visualization of flow-induced conformational transitions of single actin filaments in entangled solutions | [
"physics.bio-ph",
"cond-mat.soft"
] | While semi-flexible polymers and fibers are an important class of material due to their rich mechanical properties, it remains unclear how these properties relate to the microscopic conformation of the polymers. Actin filaments constitute an ideal model polymer system due to their micron-sized length and relatively high stiffness that allow imaging at the single filament level. Here we study the effect of entanglements on the conformational dynamics of actin filaments in shear flow. We directly measure the full three-dimensional conformation of single actin filaments, using confocal microscopy in combination with a counter-rotating cone-plate shear cell. We show that initially entangled filaments form disentangled orientationally ordered hairpins, confined in the flow-vorticity plane. In addition, shear flow causes stretching and shear alignment of the hairpin tails, while the filament length distribution remains unchanged. These observations explain the strain-softening and shear-thinning behavior of entangled F-actin solutions, which aids the understanding of the flow behavior of complex fluids containing semi-flexible polymers. | physics.bio-ph | physics | Direct visualization of flow-induced conformational transitions of single actin
filaments in entangled solutions
Inka Kirchenbuechler,1 Donald Guu,2 Nicholas A. Kurniawan,3 Gijsje H. Koenderink,3 and M. Paul Lettinga,2,4∗
1 ICS-7, Forschungszentrum Julich, D-52425 Julich, Germany
2 ICS-3, Forschungszentrum Julich, D-52425 Julich, Germany
3 FOM Institute AMOLF, 1098XG Amsterdam, the Netherlands
KU Leuven, Celestijnenlaan 200D, Leuven B-3001, Belgium∗
4 Department of Physics and Astronomy, Laboratory for Acoustics and Thermal Physics,
5
1
0
2
n
a
J
9
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
3
6
1
2
0
.
1
0
5
1
:
v
i
X
r
a
While semi-flexible polymers and fibers are an important class of material due to their rich me-
chanical properties, it remains unclear how these properties relate to the microscopic conformation
of the polymers. Actin filaments constitute an ideal model polymer system due to their micron-sized
length and relatively high stiffness that allow imaging at the single filament level. Here we study
the effect of entanglements on the conformational dynamics of actin filaments in shear flow. We
directly measure the full three-dimensional conformation of single actin filaments, using confocal
microscopy in combination with a counter-rotating cone-plate shear cell. We show that initially en-
tangled filaments form disentangled orientationally ordered hairpins, confined in the flow-vorticity
plane. In addition, shear flow causes stretching and shear alignment of the hairpin tails, while the
filament length distribution remains unchanged. These observations explain the strain-softening
and shear-thinning behavior of entangled F-actin solutions, which aids the understanding of the
flow behavior of complex fluids containing semi-flexible polymers.
PACS numbers:
Networks of semi-flexible polymers or fibers form the
basis of many materials that we encounter in daily life.
Concentrated dispersions of wormlike micelles are used
to engineer the viscoelastic properties of industrial and
consumer products [1], while polysaccharides and self-
assembled supra-molecular structures are used in tissue
engineering [2] and smart gels [3].
In biology, eukary-
otic cells are mechanically supported by an internal cy-
toskeleton composed of protein filaments including fil-
amentous (F-)actin. These filaments can form striking
non-equilibrium patterns when they are subjected to cy-
toplasmic flows within plants [4] and animal embryos [5]
generated by molecular motor activity. The basis of un-
derstanding these phenomena is knowledge of the confor-
mational response of the stiff polymers constituting the
materials to an applied flow.
Here, we directly visualize the full three-dimensional
(3D) contour of labeled F-actin subjected to shear flow,
in order to resolve the microscopic basis of the macro-
scopic non-Newtonian flow response. Our system lies in
between the limiting cases of permanently cross-linked
networks, where filaments do not relax, and of dilute non-
interacting polymers, where filaments relax freely. Per-
manently cross-linked networks, which represent a model
for cytoskeletal networks, show remarkable viscoelastic
properties such as elasticity at low filament density and
strong strain-stiffening behavior, where the stress in-
creases with increasing strain [6 -- 8]. Theoretical mod-
els can capture this behavior on the basis of microscopic
properties of the filaments, such as the bending rigidity,
∗Corresponding author: [email protected]
length distribution, and cross-link density [9].
The rheology of F-actin solutions in the absence of
cross-links [10 -- 14] is comparatively poorly understood.
The complex behavior of such solutions is due to en-
tanglements between the filaments that form at con-
centrations above the overlap density, where diffusion
takes place by reptation within tube-like confinement
zones defined by the entanglements [15]. The tube it-
self is not a rigid object, but describes rather a con-
fining potential with a varying tube radius [16]. The
linear viscoelastic response of such entangled F-actin so-
lutions to small deformations that leave the tubes un-
affected have been described by wormlike chain mod-
els [17, 18]. When starting up shear flow entangled
F-actin solutions first display strain-stiffening, followed
by strain-softening, where stress decreases with increas-
ing strain [13, 14, 19]. In addition, the viscosity of entan-
gled F-actin decreases with increasing shear rate, known
as shear-thinning [10, 19, 20]. These are key ingredients
for the formation of flow instabilities, which often occur
in concentrated semi-flexible polymers [21] such as DNA
[22], wormlike micelles [23] and also F-actin [20]. Though
the connection between shear-thinning and flow instabil-
ities is fairly well understood [24, 25], the microscopic
mechanism of shear-thinning is still under debate mainly
because real space information for nanoscopic systems
like DNA and wormlike micelles is limited. F-actin is a
particularly useful semi-flexible model polymer because
the micron-sized lengths and relatively high stiffness al-
lows imaging of the conformation at the single filament
level by fluorescence microscopy [15, 26 -- 28].
2D filament tracking in microfluidic devices for dilute
F-actin solutions, where the filaments do not interact, has
shown that the conformational dynamics is dominated
by the interplay between the Brownian motion of the
filament ends and the shear flow [26]. Due to this com-
petition, the filaments tumble in the gradient direction
forming so-called hairpins, which contain highly curved
segments between the stretched parts of the filaments.
Here, the only relevant stress component is the shear
stress, while reorientational motion takes place in the
flow-gradient plane. The frequency of this tumbling mo-
tion in this direction was shown to decreases when the
system is entangled [31]. Since, however, stress can de-
velop in all directions for entangled systems [22, 33, 34],
knowledge of the 3D contour of the filament is crucial for
a full understanding of the mechanical response of these
systems.
We use here concentrations just below and above the
concentration where a confining tube can be defined [35].
We achieve 3D in situ imaging by employing a counter-
rotating cone-plate shear cell (Supplementary Fig. 1) in
combination with a fast confocal microscope [36]. The
advantage of this shear cell is that it induces a simple
shear flow with a linear velocity gradient, while the pres-
ence of a zero-velocity plane guarantees that the filaments
stay sufficiently long in the field of view to obtain the full
3D contour. This approach allows us to test predictions
on the structural origin of strain-softening of entangled
semi-flexible polymers [29, 30]. We quantitatively mea-
sure the effect of strain on the distribution of the local
curvature. Moreover, a detailed analysis of the orienta-
tional distributions of the filaments reveals that entangle-
ments are lost during strain deformation while hairpins
are formed that are confined in the flow-vorticity plane.
Thus we identify the mechanism for strain-softening and
shear-thinning of entangled F-actin solutions.
2
FIG. 1: Rheology data and experimental setup (a)
Stress as a function of dimensionless strain where the shear
rate is increased every five minutes at a concentration of
chigh= 0.15 mg/ml.
(b) Steady-state viscosity as obtained
from the strain window where the stress is constant (see
dashed lines in (a)) as a function of shear rate for the two
concentrations. The dashed line gives the solvent viscosity.
Error bars are due to the low torque. (c) Schematic depic-
tion of the counter-rotating cone-plate shear cell placed on an
inverted microscope. (d) Typical reconstructed 3D stack of
confocal images, showing a small fraction of labeled actin fil-
aments embedded in a dark background of unlabeled F-actin.
Green line: example of a tracked filament. Scale bar: 10 µm,
tick unit: µm.
I. RESULTS
B. Global and local information as obtained from
imaging
A. Strain-softening and shear-thinning
We subjected F-actin solutions of 0.02 and 0.15 mg/ml
to shear flow , increasing the shear rate from 0.075 to
0.3 s−1 in 4 steps of 5 minutes, while keeping track of
the total acquired strain, see section III. The stress as
a function of strain at 0.15 mg/ml is plotted in Fig. 1a.
The stress decreases with increasing strain after start-up
of the shear flow at the lowest shear rate. This behavior is
indicative for strain-softening. After about 8 strain units
the stress stays constant and the viscosity of the system
can be determined. In Fig. 1b the viscosity is plotted as a
function of shear rate for both concentrations. Both sys-
tems display a clear shear-thinning behavior. Thus we
observe both strain-softening and shear-thinning. The
viscosity at 0.02 mg/ml is an order of magnitude less
than that at 0.15 mg/ml and approaches for high shear
rates the buffer viscosity, exemplifying the effect of en-
tanglement. From now on we refer to these samples as
chigh= 0.15 mg/ml and clow= 0.02 mg/ml, where the in-
dices refer to the viscosity of the samples.
To visualize shear-induced conformational transitions
of entangled individual actin filaments, we embedded
trace amounts of fluorescently labeled actin filaments in
the host dispersion. For imaging we used a home-built
counter-rotating cone-plate cell mounted on an inverted
microscope (Fig. 1c) that was equipped with a multi-
pinhole confocal scanning head, allowing frame rates of 8
frames per second. We obtain the full 3D contour of on
average 150 filaments per analyzed dimensionless strain.
Fig. 1d shows several filaments, including one for which
the contour is fitted. The coordinate system of this fitted
contour, parametrized by rj (j is the index of the coordi-
nates along the contour), is given by the velocity direction
(along the shear direction), gradient direction (along the
gap direction) and vorticity direction. From rj we can
extract information about the global filament conforma-
tion, such as the filament contour length L and its end-to-
end vector Ree. Fig. 2a and b show the distribution of the
ratio Ree/L for chigh, which is a measure for the degree
of stretching. This distribution does not show a signifi-
cant change with strain for the short filaments (< 13 µm,
da[s-1]cbd0.10.20.3101000.15 mg/ml0.02 mg/ml[mPa.s]1101006810120.075 s-10.15 s-10.225 s-10.30 s-1σ[mPa] behavior is markedly different from the behavior observed
for filaments in dilute solutions [26], which tumble in the
gradient direction. The behavior at clow is intermediate.
3
C. Filament stretching and bending in shear flow
The typical configuration of a sheared filament as dis-
played in Fig. 4a shows a hairpin, which is characterized
by two stretched tails connected by a bent part. The for-
mation of these hairpins suggests that the distribution
of the curvature of the segments changes when entangled
semi-flexible polymers are sheared. Stretching and bend-
ing of filaments both contribute to the free energy of a
filament. Usually this is connected to Ree which mea-
sures the stretching of a filament [37]. Ree, however, can
be small for hairpins, which can thus be incorrectly taken
to suggest that the entropic contribution to the energy
is low. This is clearly a flaw because the main part of
the filament is stretched. One therefore needs to ana-
lyze the filaments at smaller length scales. We do this
by using the local curvature κj, which is related to the
scaled end-to-end vector xj = Rj/(cid:104)Ln(cid:105) (Fig. 4a). In this
analysis we only used filaments with length L > 21 µm to
assure good statistics and sufficient flexibility (L/lp > 1).
On average we analyzed about 1200 segments per image
stack, allowing us to obtain distributions for the curva-
ture P (κj).
FIG. 3: Distribution functions of κj (a) clow (b) chigh.
Open symbols: zero strain γ = 0. Solid symbols: high strain,
γ = 215 ± 12. Dashed line is fit to a log-norm distribution,
while the solid lines are fits to a Gaussian distribution.
The distribution of κj are plotted in Fig. 3 for zero
shear strain and high strain (γ = 215 ± 12.). For both
concentrations the equilibrium curvature distribution is
well-described by a Gaussian distribution. The distri-
bution in equilibrium deviates from earlier studies of un-
sheared F-actin [38], probably due to the different sample
environment and relatively high concentrations of fila-
ments that are used in this study. For the low concentra-
tion the distribution remains unchanged when straining
FIG. 2: Distribution of filament length parameters (a)
Ratio of end-to-end distance Ree over the filament length L
for short filaments L < 13 µm and (b) long filaments L >
21 µm. Black bars: zero strain and red bars strain γ = 215 ±
12. (c) Filament length distribution at different dimensionless
strain values. All distributions were determined for chigh.
Fig. 2a). For the long filaments (> 21µm, 2b), the distri-
bution widens towards small Ree/L while a pronounced
peak is formed for Ree/L → 1. This means that the
filaments become more stretched and at the same time
more bent, which is indicative for the formation of hair-
pins. Fig. 2c displays the length distribution at different
dimensionless strain values for chigh. The distribution
does not shift to lower values of filament lengths with in-
creasing strain, so that we conclude that shear-softening
and shear-thinning is not due to rupture of actin fila-
ments.
We will characterize the filament conformations by ex-
tracting the local curvature, κj, tangent vector, Tj, and
binormal vector, Bj, along the contour, see Fig. 4a. To
calculate these values we average over two neighboring
points in the contour. Thus, the contour is split up in
segments of an average length of (cid:104)Ln(cid:105)= 2.36 µm (Supple-
mentary Note 1). Fig. 4b and c display the projections
of two typical examples of filaments strained at clow (b)
and chigh (c). At clow, which corresponds to the onset of
entanglements, we find filaments with highly curved seg-
ments (region I) and segments that are stretched and ori-
entated with the flow direction (region II). At chigh, the
filament has a hairpin conformation (region III), with two
dangling ends which are stretched in the flow direction
(region IV). Strikingly, for chigh the hairpin is confined to
the flow/vorticity plane, with its binormal vectors (blue
arrows) pointing in the gradient direction Fig. 4c. This
abcab the system.
In contrast, at the high concentration we
observe that the distribution takes the form of a log-
normal distribution. Compared to the unsheared situa-
tion, the peak of the distribution shifts towards smaller
values of κj, showing that the majority of segments are
more stretched. In addition the distribution also has a
long tail with a power law form, showing that strain in-
duces curvature in the system. These observations hint
that at chigh energy is stored in the system, but not at
clow. This is consistent with the presence of entangle-
ments at chigh. However, these observations do not yet
explain the strain-softening behavior.
FIG. 4: Tracked conformations of actin filaments (a)
Example of a tracked filament (Fig. 1d) with its end-to-end
vector, Ree (black). The filament is subdivided in segments
j, allowing us to determine the local orientation in terms of
tangent Tj (red) and binormal Bj (blue) vectors, and the lo-
j. Part of
cal curvature using the local end-to-end vector Ree
the hairpin can be described a circle with curvature κj. The
relevant angles characterizing the orientation are indicated on
the right. (b,c) Typical conformations of actin filaments at
clow (b) and chigh (c). Left: views in the flow/vorticity plane.
Right: corresponding views in the flow/gradient plane. At
clow, the filament has a highly curved segment (region I), and
stretched segments that are somewhat oriented in the flow di-
rection (region II). It shows no confinement in any plane. For
chigh, the filament also has a highly curved segment (III) and
a stretched segment (IV) and is confined to the flow/vorticity
plane. Tick unit: µm.
D. Orientation of shear-induced hairpins
4
(cid:17)
(cid:16)
To quantify the orientations of the filaments, we col-
lected all local tangent and binormal vectors along the
filament contour, as plotted in Fig. 5a,d on a unit sphere,
and calculated the corresponding two-dimensional an-
gular orientational distribution functions which are
best fitted with a 2D Lorentzian given by f (θ, φ) =
( θ−∆θ
)2 + ( φ−∆φ
)2 + 1
wθ
wφ
a/
. Here the angles ϕ and θ,
which set up the vectors Tj and Bj, are defined as shown
in Fig. 4a right. As examples we show in Fig. 5 the distri-
bution for chigh after straining the sample to γ = 215±12,
separating the orientation of the stretched segments, us-
ing all segments with κj < 0.1 µm−1 (top) and bent seg-
ments using all segments with κj > 0.2 µm−1 (bottom).
The stretched segments display a strong shear-
alignment of the tangent: Fig. 5a,b shows that the seg-
ments all point in the flow direction. Note however that
the distributions are slightly biaxial, i.e. not symmetric
around the maximum. The binormal distribution has a
low degree of order and is strongly biaxial, Fig. 5c. This
can be rationalized as follows. When the tangent is well
aligned, then there is no clearly defined plane that is
spanned by two sequential segments. Thus the binormal
is ill-defined though per definition in the plane perpen-
dicular to the tangent: the distribution in Fig. 5c is very
sharp with respect to φB and very broad with respect
to θB. On the contrary, the bent segments have a pro-
nounced uniaxial ordering of the binormal pointing in
the gradient direction (Fig. 5d,f), while the distribution
of the tangent is biaxial (Fig. 5e), which is to be expected
for hairpins, see for example region I in Fig. 4b.
(cid:82) π
(cid:82) 2π
In order to quantify separately the orientational be-
havior of the stretched and bent parts of the fil-
aments, we use the distribution functions f (θ, φ)
to calculate the orientational order tensors ¯ST =
0 dθdφ sin(φ)f (θT , φT ) T T and similarly ¯SB. For
2 (3 ¯S −
0
our purpose the traceless diagonalized form ¯Q = 1
I) is particularly useful since Fig. 4 and Fig. 5b both sug-
gest that the orientational distributions can be biaxial.
¯Q may be written as
− 1
¯QT,B =
2 λT,B − ηT,B
0
0
− 1
2 λT,B + ηT,B
0
0
0
0
λT,B
,(1)
where λT,B is the orientational order parameter of the
main orientation axis and ηT,B parametrizes the biaxial-
ity of the system.
We will now discuss the behavior of these parameters
for the four features that we indicated in Fig. 4b and c:
segments with high curvatures of κj > 0.2 µm−1 for clow
(I) and chigh (III), and segments with low curvatures of
κj < 0.1 µm−1 for clow (II) and chigh (IV). The stretched
segments at chigh (IV in Fig. 4) clearly display a high
degree of ordering of the tangent with λT → 1 (solid
10203040∇v9095100105v909510010560657075∇ ×vIIIIVIVIIIbcv152025100105110v∇×vv1520252015∇v2025vIIIIIIav 5
FIG. 5: Distribution of orientational order parameters (a,d) Tangent (red) and binormal (blue) vectors on a unit sphere.
Corresponding angular distribution as given by their angles θ and φ with a 2D Lorenzian for tangent (b,e) and binormal
distribution (c,f). We used chigh= 0.15 mg/ml, filament length > 21 µm, γ = 215 ± 12. The top row are results for stretched
segments (cut-off curvature κj < 0.1 µm−1) and bottom row are results for the bent segments (κj > 0.2 µm−1).
symbols in Fig. 6b). The ordering is uni-axial (solid sym-
bols in Fig. 6f) and almost along the flow direction, as
expected (solid symbols in Fig. 7c and f). λT is higher
for chigh than for clow (compare Fig. 6a and b, II and
IV), which shows that entanglements enhance ordering
when the sample is sheared.
Interestingly, λT jumps
immediately to a high value when strain is applied for
chigh. On the contrary, λT displays a moderate increase
with increasing strain for clow, while the corresponding
eigenvector tilts towards the flow direction (solid symbols
in Fig. 7a and e). This complies with measurements on
wormlike micelles [39]. Thus, the entanglements not only
cause stretching of parts of the filaments, as seen from
Fig. 3, but also strong shear-alignment of these stretched
segments along the flow direction.
Whilst this shear-alignment is a well known phe-
nomenon, the orientational behavior of segments with
high curvature (I and III in Fig. 4) is a priori not obvi-
ous. The binormal is a particularly informative param-
eter since it is oriented along the normal of the plane
spanned by a curved segment. When there is a train of
segments with similar curvature, which is the case in the
bent part of a hairpin, then binormal vectors point in
the same direction. This can be seen for chigh, where the
blue arrows in region III of Fig. 4c on the right all point
in the gradient direction. The behavior of the binormal
order parameter λB confirms this observation.
Indeed
λB is higher for the highly curved than for the stretched
segments; compare open and solid symbols in Fig. 6d.
The eigenvectors belonging to λB point along the gradi-
ent direction (Fig. 5f and open symbols in Fig. 7d and
h) and therefore also the plane spanned by the highly
curved segments. The fact that the binormal is well de-
fined for κj > 0.2 µm−1 implies that the ordering of the
tangent is low and biaxial. This is indeed confirmed by
Fig. 6f, where the biaxiality of the tangent ηT is plotted:
for chigh the biaxiality of the tangent ηT increases with
strain when κ > 0.2 µm−1. We therefore conclude that
the plane spanned by the bent part of the hairpin, which
we labeled III in Fig. 4c, indeed is located in the flow-
vorticity plane, with its normal pointing in the gradient
direction.
−101−101−101−101−101−101abcdef The behavior of the binormal and biaxiality is distinc-
tively different for clow. No increase in λB nor in ηT is
observed with increasing strain (Fig. 6c,e). Thus, there is
no well defined orientation of hairpins at the low concen-
tration, as is exemplified in region I of Fig. 4b. In Fig. 6
we also plot the points measured about 150 s after ces-
sation of shear flow. We note a clear difference between
the relaxation times at low and high concentration: for
clow the orientation is partly lost, whereas for chigh the
orientations remain unchanged. This indicates that the
relaxation time at chigh is indeed much longer than for
clow. A time series after cessation of flow confirms these
observations: the shape is mainly lost for clow (Fig. 6g),
while at chigh the filament conserves its shape (Fig. 6h).
FIG. 6: Orientational order parameters (a,b) Tangent
and (c,d) binormal vector.
(e,f) Degree of biaxiality. The
solid symbols are the results for stretched segments (κj <
0.1 µm−1) and the open symbols are the results for the bent
segments (κj > 0.2 µm−1). The grey zone indicates data
points about 150 s after cessation of shear flow. Also shown
is the overlay of the 2D projection of filaments (g and h) in
four sequential time windows following cessation of shear flow
(time average 45 s each). The dotted white lines indicate the
confining tube for the first 45 s following cessation of shear
flow. Scale bar 5 µm. The left column is data for clow and the
right column for chigh. The lines in (b,d,f) are guides to the
eye. I-IV refer to the regions shown in Figure 4.
II. DISCUSSION
6
By directly visualizing the full 3D conformation of indi-
vidual actin filaments within entangled F-actin solutions,
we can show how entanglements influence the conforma-
tion of the filaments in response to shear flow.
First, the distribution of curvature for clow, at the on-
set of the entangled regime, remains unchanged when
applying shear flow, while the number of stretched as
well as bent segments increases for chigh when applying
shear, see Fig. 3. This explains the much higher viscos-
ity for chigh. Second, we observe at chigh that hairpins
form in shear flow, which tilt into the flow-vorticity plane.
The behavior of entangled filaments is in marked con-
trast with the dynamics of filaments in dilute solutions
which tumble in the gradient direction [26]. Our observa-
tions also explain why tumbling was previously shown to
be strongly reduced at concentrations three times higher
than chigh [31], although the shear rates used in this ref-
erence were O(102) higher. We find that the response
at clow lies between these two extremes, since there is
no well defined plane into which the filaments turn. In
this case we find shear-thinning which is purely due to
increased alignment of the filaments, as we conclude from
Fig. 6a and e.
We will now try to relate the formation of the strongly
aligned hairpins with the strain-softening in entangled so-
lutions. Hairpins are generally viewed as a signature of
an entanglement. Two entangled filaments will strongly
bend around the point where they are entangled when
they are moved in opposite directions faster than the time
they need to relax. Mechanically this leads to strain-
stiffening, while pairs of hairpins with the orientations of
the bent segments roughly perpendicular to each other
result in a very flat distribution of the binormal of the
bent segments. We observe, however, exactly the oppo-
site: the binormal of the bent segments is highly aligned
when straining the system at chigh. Moreover, there is
a strong increase in the number of stretched segments
and a marginal increase in the number of bent segments
and we predominantly find only one hairpin per filament,
see Fig. 3b. These observations strongly hint that en-
tanglements disappear as the system is strained. Since
the hairpins and their stretched tails are strictly located
in the flow-vorticity plane, they have no components in
the gradient direction that cause the shear stress, which
results in strain softening. The filaments slide over each
other facilitating lamellar flow. These findings are con-
trary to theoretical predictions that hairpins cause strain-
stiffening. However, the theory considered the contour
lengths to be much longer than the persistence length
[29, 30], while in our experiments they are of the same
order. Theory does predict an instability where shear de-
formation pushes out contacts between filaments causing
strain-softening [30]. This strain-induced loss of entan-
glements, very similar to the convective constraint release
[32], is exactly what we find.
There are no predictions of the shape of the filaments
IIIIVIabcdIIIIVIefII0.000.100.00.40.80.00.40.801002000100200//////////// 7
flow instabilities. Polymer solutions [22, 33, 34], poly-
mer melts [40] and sticky carbon nanotubes [41] are all
systems that display pronounced normal stresses as well
as flow instabilities. Flow instabilities have also been ob-
served for actin dispersions at higher concentrations than
we used [20]. We did not find any signature of such a be-
havior, scanning a significant part of the gap of the shear
cell at a fixed position from the center of the shear cell.
This could be due to the limited strain applied to the sys-
tem as well as the relatively low filament concentration
as compared to ref. [20].
In conclusion, we believe that the mechanism of stress
release we identified here may be generally valid for so-
lutions of semi-flexible polymers, including supramolec-
ular systems [23, 42 -- 44]. Thus, our findings will aid to
the understanding of the complex flow behavior of such
systems. Likewise, this mechanism could impact the self-
organization of cytoskeletal filaments in response to in-
tracellular shear flows created by processes like cytoplas-
mic streaming [4, 5].
III. METHODS
A. Protein purification and sample preparation
G-actin was isolated from rabbit skeletal muscle [45],
stored in G-buffer solution (5.0 mM Tris, 0.2 mM CaCl2,
0.2 mM ATP, 0.2 mM DTT, pH 7.4, 21◦C) at 4◦C. Flu-
orescently labeled filaments were obtained by polymer-
izing G-actin (0.2 mg/ml) for 1 h at 21◦C by adding 10
vol% 10x F-buffer solution (20 mM Tris, 2.0 mM CaCl2,
1.0 M KCl, 20 mM MgCl, pH 8.5, 21◦C, 5.0 mM ATP,
2.0 mM DTT) in the presence of an equimolar amount
of Atto647N-Phalloidin (Atto-Tec). Unlabeled G-actin
(0.2-0.3 mg/ml) was similarly polymerized, but in the
presence of unlabeled phalloidin (Sigma/P2141). These
filaments have a persistence length of lp 9-18 µm [46 --
49] Samples were prepared by diluting labeled filaments
in a 1:2000 ratio with GFS-buffer solution (10% 10x F-
buffer, 60% sucrose in G-buffer) and mixing this solution
in equal volume with unlabeled F-actin to reach final
concentrations of 0.02 and 0.15 mg/ml. The final buffer
solution thus contained 30% sucrose, which reduced the
off-rate of labeled phalloidin [50], improving the signal to
noise ratio during image acquisition. All measurements
were done at 21◦C.
B. Shear cell and Microscopy
To produce shear flows with a well-defined linear veloc-
ity gradient, we used an adapted version of the counter-
rotating cone-plane shear cell used in ref. [51] (Supple-
mentary Fig. 1). It consists of a bottom glass plate (di-
ameter 80 mm, thickness 170 µm, Menzel) which is fixed
by two Teflon rings that are pressed together between a
titanium plate at the bottom and a stainless steel block
FIG. 7: Angles of the eigenvectors Angles correspond to
the highest eigenvalue of the tangent tensor (a,b,e,f) and the
binormal tensor (c,d,g,h). The solid symbols are the results
for stretched filaments (κj < 0.1 µm−1) and the open symbols
are the results for the bent segments (κj > 0.2 µm−1). The
grey zone indicates data points 150 s after cessation of shear
flow.
at high strains after strain softening. We find experimen-
tally that the effect of the surrounding filaments is to
confine the hairpins in the flow-vorticity plane. Instead
of a confining tube as can be defined in equilibrium, there
are now confining planes, without entanglements. This
can also be appreciated from the movie (Supplementary
Movie 1), where fluctuations in the vorticity directions
are observed, but not in the gradient direction, in con-
trast to the movie of a filament at clow (Supplementary
Movie 2). While the hairpins immediately form and can
be related to the strain-softening, we also observe shear
thinning, see Fig. 1b. This behavior is likely to relate to
the small but significant increase in the orientational or-
dering of the stretched segments (solid symbols in Fig 6b)
and decrease in the biaxiality (solid symbols in Fig 6f) pa-
rameterizing the flow alignment of the segments. These
new scenarios for strain-softening and shear-thinning ex-
clude the need for scission of the F-actin filaments as a
pathway to explain non-linear rheology, which is known
for living polymers such as wormlike micelles [23].
In-
deed, we observe in Fig. 2b no change in the filament
length distribution over the full measured range of strain.
The mechanism for stress release by the formation of
disentangled sliding hairpins could be a precursor for
the formation of flow instabilities, which is often related
with local reorganization of the constituting particles
[23]. The distinct orientation of the hairpins also sug-
gests that a normal stress builds up which could lead to
////////////////8090100010020030080901000100200300θTθB-100108090100ϕTϕBabcdef at the top. In this block there is a hole where the top
cone is inserted, which is also made out of stainless steel.
Both the glass plate and steel cone can move indepen-
dently and are in our experiments moved counter clock-
wise. We used an Epiplan-Neofluar 50x/1.0 Pol objective
(Zeiss) . The shear cell was mounted on an inverted mi-
croscope (Zeiss/Axiovert 200 M), equipped with a multi-
pinhole-confocal system (VisiTech/VT-Infinity-I) and an
Epiplan-Neofluar 50x/1.0 Oil Pol Objektiv (Zeiss). An
Argon/Krypton laser (Spectra Physics/Stabilite 2250)
operating at 647 nm was used for excitation of the flu-
orescent dye. An observation area of 151 µm x 151 µm
was imaged onto an EMCCD-camera (Andor/iXon DU-
897) operated with IQ software. Confocal stacks con-
sisted of 51 frames taken at 1 µm intervals at a rate of
7.2 s per stack. The difference in the geometrical and
the optical path lenght, due to the different refractive in-
dices of the immersion oil (noil = 1.518) objective and
the actin solution (≈ nwater = 1.33) was determined by
two independent methods. First we measured the refrac-
tive indices of the immersion oil and the actin solution
and calculated the correction factor nk based on the ra-
tio of those indices. Second we filled a glass capillary of
known thickness (10 µm ±10%) with a solution of fluores-
cent beads (Ø 0.5 µm, Latex beads, Sigma) and measured
the geometrical path length, by the use of a calibrated
piezo element, calculating the correction factor nk based
on the difference of the capillary and geometrical length:
= 0.9 ± 0.04. Both methods
nk = nbuf f er
noil
gave equivalent results. Data were taken 30-80 µm into
the sample to reduce wall effects.
Our shear protocol consists of four blocks of five min-
utes where a shear rate of 0.075, 0.15, 0.225, 0.3 s−1, re-
spectively, is applied. The reason for the shear protocol
is that after sample loading and inserting the cone, the
orientation of the filaments for chigh is not well defined,
while it does not relax. This is a fact that cannot be
avoided. Thus quite some strain units are needed to re-
≈ hcapillary
hgeometrical
8
move this memory effect. For clow the sample does relax
before we start the experiment and indeed we can follow
trends with increasing strain.
Rheology data are taken with an Anton Paar MCR501,
using a cone-plate geometry with 30 mm diameter and 1
degree angle.
C. Data analysis
The 3D filament contours were tracked with the
autodepth-function of the visualization and analysis soft-
ware Imaris (Bitplane/ Imaris 6.1), and subsequent fila-
ment position analysis was done using Matlab (Version,
Mathworks). As a control for the conformation of the
filaments before the shear experiment (γ = 0) one frame
was taken before placing the cone. This frame contained
about 50 tracer filaments, that were tracked with Imaris.
For the analysis of the filament conformation at a cer-
tain strain value, 3 statistically independent stacks (circa
30 s apart) before changing the shear rate, were analyzed,
containing also about 50 tracer filaments each, corre-
sponding to a strain of γ=21± 3, 59 ± 6, 126 ± 7 and
215 ± 12. For the last data point 3 frames, 120 s, 150 s
and 180 s after cessation, were analyzed.
IV. ACKNOWLEDGMENTS
G.H.K. was supported by the Foundation for Fun-
damental Research on Matter (FOM), which is part of
NWO, and N.A.K. acknowledges support by a Marie
Curie FP7 IIF Fellowship. D.G. was supported by the
International Helmholtz Research School (BioSoft). We
thank F. MacKintosh, R. Winkler, J. Dhont, M. Giesen
and R. Merkel for fruitful discussions and Y. Jia for help
with the data analysis.
[1] Ezrahi, S., Tuval, E. & Aserin, A. Properties, main ap-
plications and perspectives of worm micelles. J. Colloid.
Interface. Sci. 128-130, 77 -- 102 (2006).
[2] Kim, I.-Y. et al. Chitosan and its derivatives for tis-
sue engineering applications. Biotechnol. Adv. 26, 1 -- 21
(2008).
[3] Hirst, A. R., Escuder, B., Miravet, J. F. & Smith, D. K.
High-tech applications of self-assembling supramolecular
nanostructured gel-phase materials:
from regenerative
medicine to electronic devices. Angew. Chem. Int. Ed.
47, 8002 -- 8018 (2008).
[4] Woodhouse, F. G. & Goldstein, R. E. Cytoplasmic
streaming in plant cells emerges naturally by microfil-
ament self-organization. Proc. Natl. Acad. Sci. U.S.A.
110, 14132 -- 14137 (2013).
[5] Ganguly, S., Williams, L. S., Palacios, I. M. & Goldstein,
R. E. Cytoplasmic streaming in Drosophila oocytes varies
with kinesin activity and correlates with the microtubule
cytoskeleton architecture. Proc. Natl. Acad. Sci. U.S.A.
109, 15109 -- 15114 (2012).
[6] Unterberger, M. J., Schmoller, K. M., Bausch, A. R. &
Holzapfel, G. A. A new approach to model cross-linked
actin networks: multi-scale continuum formulation and
computational analysis. J. Mech. Behav. Biomed. Mater.
22, 95 -- 114 (2013).
[7] Schmoller, K. M., Fern´andez, P., Arevalo, R. C., Blair,
D. L. & Bausch, A. R. Cyclic hardening in bundled actin
networks. Nat. Commun. 1, 134 (2010).
[8] Gardel, M. L. et al. Elastic behavior of cross-linked and
bundled actin networks. Science 304, 1301 -- 1305 (2004).
[9] Broedersz, C., Storm, C. & MacKintosh, F. Effective-
medium approach for stiff polymer networks with flexible
cross-links. Phys. Rev. E. 79, 061914 (2009).
[10] Sato, M. & Leimbach, G. Mechanical properties of actin.
J. Biol. Chem. 260, 8585 -- 8592 (1985).
[11] Janmey, P. A. et al. The mechanical properties of actin
gels. Elastic modulus and filament motions. J. Biol.
Chem. 269, 32503 -- 32513 (1994).
[12] Koenderink, G., Atakhorrami, M., MacKintosh, F. &
Schmidt, C. High-Frequency Stress Relaxation in Semi-
flexible Polymer Solutions and Networks. Phys. Rev.
Lett. 96, 138307 (2006).
[13] Semmrich, C. et al. Glass transition and rheological re-
dundancy in F-actin solutions. Proc. Natl. Acad. Sci.
U.S.A. 104, 20199 -- 20203 (2007).
[14] Semmrich, C., Larsen, R. J. & Bausch, A. R. Nonlinear
mechanics of entangled F-actin solutions. Soft Matter 4,
1675 -- 1680 (2008).
[15] Kaes, J., Strey, H. & Sackmann, E. Direct imaging of
reptation for semiflexible actin filaments. Nature 368,
226 -- 229, (1994).
[16] Glaser, J. et al. Tube Width Fluctuations in F-Actin
Solutions. Phys. Rev. Lett. 105, 037801 (2010).
[17] Isambert, H. et al. Flexibility of Actin Filaments Derived
from Thermal Fluctuations. J. Biol. Chem. 270, 11437 --
11444 (1995).
[18] Morse, D. C. Viscoelasticity of Concentrated Isotropic
Solutions of Semiflexible Polymers. 2. Linear Response.
Macromolecules 31, 7044 -- 7067 (1998).
[19] Maruyama, K., Kaibara, M. & Fukada, E. Rheology of
F-actin. Biochim. Biophys. Acta 371, 20 -- 29 (1974).
[20] Kunita, I. et al. Shear Banding in an F-Actin Solution.
Phys. Rev. Lett. 109, 248303 (2012).
[21] Wang, S.-Q., Ravindranath, S. & Boukany, P. E. Ho-
mogeneous Shear, Wall Slip, and Shear Banding of En-
tangled Polymeric Liquids in Simple-Shear Rheometry:
A Roadmap of Nonlinear Rheology. Macromolecules 44,
183 -- 190 (2011).
[22] Groisman, A. & Steinberg, V. Elastic turbulence in a
polymer solution flow. Nature 405, 53 -- 55 (2000).
[23] Lerouge, S. & Berret, J.-F. Shear-induced transitions and
instabilities in surfactant wormlike micelles. Adv. Polym.
Sci. 230, 1 -- 71 (2010).
[24] Olmsted, P. & Hamley, I. Lifshitz points in blends of AB
and BC diblock copolymers. Europhys. Lett. 45, 83 -- 89
(1999).
[25] Dhont, J. K. G. A constitutive relation describing the
shear-banding transition. Phys. Rev. E 60, 4534 -- 4544,
1999.
[26] Harasim, M., Wunderlich, B., Peleg, O., Kroger, M. &
Bausch, A. Direct Observation of the Dynamics of Semi-
flexible Polymers in Shear Flow. Phys. Rev. Lett. 110,
108302 (2013).
[27] Kantsler, V. & Goldstein, R. E. Fluctuations, Dynam-
ics, and the Stretch-Coil Transition of Single Actin Fila-
ments in Extensional Flows. Phys. Rev. Lett. 108, 038103
(2012).
[28] Steinhauser, D., Koster, S. & Pfohl, T. Mobility Gradi-
ent Induces Cross-Streamline Migration of Semiflexible
Polymers. ACS Macro Letters 1, 541 -- 545 (2012).
[29] Morse, D. Viscoelasticity of Concentrated Isotropic So-
lutions of Semiflexible Polymers. 3. Nonlinear Rheology.
Macromolecules 32, 5934 -- 5943 (1999).
[30] Fern´andez, P., Grosser S. & Kroy, K. A unit-cell ap-
proach to the nonlinear rheology of biopolymer solutions.
Soft Matter 5, 2047 -- 2057 (2009).
[31] Huber, B., Harasim, M., Wunderlich, B., Kroger, M.
& Bausch, A. R. Microscopic Origin of the Non-
Newtonian Viscosity of Semiflexible Polymer Solutions
in the Semidilute Regime. ACS Macro Letters 3, 136 --
9
140 (2014).
[32] Marrucci, G. Shear microscopy of the "butterfly pat-
J. Non-Newtonian Fluid
tern"in polymer mixtures.
Mech., 62, 279 (1996).
[33] Moses, E., Kume, T. & Hashimoto, T. Shear microscopy
of the "butterfly pattern"in polymer mixtures. Phys.
Rev. Lett. 72, 2037 (1994).
[34] Groisman, A. & Steinberg, V. Mechanism of elastic insta-
bility in Couette flow of polymer solutions: Experiment.
Phys. Fluids 10, 2451 -- 2463, (1998).
[35] Kaes, J. et al. F-actin, a model polymer for semiflexible
chains in dilute, semidilute, and liquid crystalline solu-
tions. Biophys. J. 70, 609 -- 625 (1996).
[36] Derks, D., Aarts, D. G. A. L., Bonn, D. & Imhof, A.
Phase separating colloid polymer mixtures in shear flow.
J. Phys.: Condens. Matter 20, 404208 (2008).
[37] Winkler, R. G. Deformation of semiflexible chains. J.
Chem. Phys. 118, 2919 -- 2928 (2003).
[38] Romanowska, M.,et al. Direct observation of the tube
model in F-actin solutions: Tube dimensions and curva-
tures. Europhys. Lett. 86, 26003 (2009).
[39] Helgeson, M. E.,et al. Rheology and spatially resolved
structure of cetyltrimethylammonium bromide wormlike
micelles through the shear banding transition. J. Rheol.
53, 727-756 (2009).
[40] Oda, K., White, J. & Clark, E. Correlation of normal
stresses in polystyrene melts and its implications. Polym.
Eng. Sci 18, 25 -- 28, 1978.
[41] Lin-Gibson, S., Pathak, J., Grulke, E., Wang, H. & Hob-
bie, E. Elastic Flow Instability in Nanotube Suspensions.
Phys. Rev. Lett. 92, 48302 (2004).
[42] van der Gucht, J., Besseling, N., Knoben, W., Bouteiller,
L. & Cohen Stuart, M. Brownian particles in supramolec-
ular polymer solutions. Phys. Rev. E 67, 051106 (2003).
[43] Lonetti, B., Kohlbrecher, J., Willner, L., Dhont, J. K.
G. & Lettinga, M. P. Dynamic response of block copoly-
mer wormlike micelles to shear flow. J. Phys.: Condens.
Matter 20, 404207 (2008).
[44] Kouwer, P. H. J.,et al. Responsive biomimetic networks
from polyisocyanopeptide hydrogels. Nature 493, 651-655
(2013).
[45] Spudich, J. A. & Watt, S. The Regulation of Rabbit
Skeletal Muscle Contraction. J. Biol. Chem. 246, 4866 --
4871 (1971).
[46] Ott, A., Magnasco, M., Simon, A. & Libchaber, A. Mea-
surement of the persistence length of polymerized actin
using Auorescence microscopy. Phys. Rev. E 48, R1642 --
R1645 (1993).
[47] Gittes, F., Mickey, B., Nettleton, J. & Howard, J. Flexu-
ral rigidity of microtubules and actin filaments measured
from thermal fluctuations in shape. J. Cell Biol. 120,
923 -- 934 (1993).
[48] Isambert, H. & Maggs, A. C. Dynamics and Rheology of
Actin Solutions. Macromolecules 29, 1036 -- 1040 (1996).
[49] van Mameren, J., Vermeulen, K. C., Gittes, F. &
Schmidt, C. F. Leveraging single protein polymers to
measure flexural rigidity. J. Phys. Chem. B 113, 3837 --
3844, 2009.
[50] De La Cruz, E. M. & Pollard, T. D. Transient kinetic
analysis of rhodamine phalloidin binding to actin fila-
ments. Biochemistry 33, 14387 -- 14392 (1994).
[51] Derks, D., Wisman, H., van Blaaderen, A. & Imhof, A.
Confocal microscopy of colloidal dispersions in shear flow
using a counter-rotating coneplate shear cell. J. Phys.:
Condens. Matter 16, S3917 -- S3927 (2004).
10
|
1608.01419 | 1 | 1608 | 2016-08-04T03:58:19 | Systems-level approach to uncovering diffusive states and their transitions from single particle trajectories | [
"physics.bio-ph",
"q-bio.QM"
] | The stochastic motions of a diffusing particle contain information concerning the particle's interactions with binding partners and with its local environment. However, accurate determination of the underlying diffusive properties, beyond normal diffusion, has remained challenging when analyzing particle trajectories on an individual basis. Here, we introduce the maximum likelihood estimator (MLE) for confined diffusion and fractional Brownian motion. We demonstrate that this MLE yields improved estimation over traditional mean square displacement analyses. We also introduce a model selection scheme (that we call mleBIC) that classifies individual trajectories to a given diffusion mode. We demonstrate the statistical limitations of classification via mleBIC using simulated data. To overcome these limitations, we introduce a new version of perturbation expectation-maximization (pEMv2), which simultaneously analyzes a collection of particle trajectories to uncover the system of interactions which give rise to unique normal and/or non-normal diffusive states within the population. We test and evaluate the performance of pEMv2 on various sets of simulated particle trajectories, which transition among several modes of normal and non-normal diffusion, highlighting the key considerations for employing this analysis methodology. | physics.bio-ph | physics |
Systems-level approach to uncovering diffusive states and their transitions from single
particle trajectories
Peter K. Koo1, 2 and Simon G. J. Mochrie1, 2, 3
1Department of Physics, Yale University, New Haven, CT
2Integrated Graduate Program in Physical and Engineering Biology,
3Department of Applied Physics, Yale University, New Haven, CT∗
Yale University, New Haven, Connecticut 06520
(Dated: August 30, 2018)
The stochastic motions of a diffusing particle contain information concerning the particle's in-
teractions with binding partners and with its local environment. However, accurate determination
of the underlying diffusive properties, beyond normal diffusion, has remained challenging when an-
alyzing particle trajectories on an individual basis. Here, we introduce the maximum likelihood
estimator (MLE) for confined diffusion and fractional Brownian motion. We demonstrate that this
MLE yields improved estimation over traditional mean square displacement analyses. We also in-
troduce a model selection scheme (that we call mleBIC) that classifies individual trajectories to a
given diffusion mode. We demonstrate the statistical limitations of classification via mleBIC us-
ing simulated data. To overcome these limitations, we introduce a new version of perturbation
expectation-maximization (pEMv2), which simultaneously analyzes a collection of particle trajec-
tories to uncover the system of interactions which give rise to unique normal and/or non-normal
diffusive states within the population. We test and evaluate the performance of pEMv2 on var-
ious sets of simulated particle trajectories, which transition among several modes of normal and
non-normal diffusion, highlighting the key considerations for employing this analysis methodology.
I.
INTRODUCTION
Single particle tracking (SPT) offers the ability to
non-invasively probe at sub-diffraction-limit resolution
the spatio-temporal motions of individual fluorescently-
labelled proteins (FPs) inside living cells. Because the
different interactions, that a FP undergoes inside a cell,
give rise to different types of diffusive motion, SPT data
encode each FP's interactions with other particles and
with its local envrionment: Biochemical binding interac-
tions can lead to different diffusivities if the FP can bind
to different substrates [6]; interactions with the cellular
medium can give rise to anomalous diffusion [1 -- 3] or can
lead to confined motions [4, 5]. Thus, important goals
of SPT measurements are (1) to infer these interactions
from an analysis of protein trajectories and (2) to deter-
mine the spatio-temporal kinetics of each interaction.
To uncover this information, the number of unique dif-
fusive states, as well as each such state's diffusion mode
and its diffusion properties, must be inferred from the
proteins' trajectories, along with the ability to classify
which portions of each trajectory correspond to a given
diffusive state, thus allowing for the determination of the
underlying transition kinetics and the spatio-temporal lo-
cations of particular diffusive states and their transitions
within the cell.
Previous work [7 -- 16] that seeks to assess dynamic het-
erogeneity in tracking data has been reviewed by us
in Ref.
[17]. The traditional approach for analyzing
the diffusive properties of individual particle trajecto-
ries is by fitting each trajectory's time-averaged mean
∗ [email protected]
square displacement (taMSD) to a corresponding diffu-
sion model [18]. However, the way in which the taMSD
is usually calculated results in an statistically-complex
representation of the underlying diffusion process, es-
pecially for short trajectories (Supplemental Materials),
rendering the taMSD unreliable. Thus, an unweighted
least squares regression against the taMSD yields sta-
tistically inefficient estimation of the diffusion model
parameters.
Improved estimation can be achieved by
analyzing longer trajectories, albeit the same interac-
tion must persist throughout the duration of the trajec-
tory, which is an increasingly unlikely condition in the
complex environment inside living cells. Alternatively,
ensemble-averaging taMSD curves across particle trajec-
tories, which share the same underlying diffusive prop-
erties, is another route for bolstering the statistics and
thus better representing the underlying diffusive behav-
ior. However, because the diffusive properties of each tra-
jectory are not known a priori, how to sort trajectories
into groups that share diffusive properties, and therefore
may be averaged together, is not straightforward.
Because of the drawbacks of taMSD analysis, a num-
ber of alternatives have emerged for determining diffusion
parameters, namely the maximum likelihood estimator
(MLE) [19], optimal least squares fitting (OLSF) [20],
and the covariance-based estimator (CVE) [21]. These
approaches have demonstrated improved estimation in
comparison with traditional taMSD analysis.
Impor-
tantly, however, to-date these approaches, which do prop-
erly account for localization noise sources, have only been
shown to be applicable to particle trajectories undergoing
normal diffusion.
Recently, systems-level analyses, namely variational
Bayes single particle tracking (vbSPT) [6] and perturba-
tion expectation-maximization (pEM) [17], have demon-
strated that the limited statistics of individual particle
trajectories can be augmented by simultaneously analyz-
ing a population of particle trajectories to uncover the
number of unique diffusive states and their correspond-
ing diffusive properties. However, both of these meth-
ods have their own limitations. While vbSPT allows for
transitions between different diffusive states, it fails to
properly account for experimental noise sources, com-
promising vbSPT's ability to reliably extract the correct
number of diffusive states and each state's diffusive prop-
erties in some situations [17]. On the other hand, while
pEM properly accounts for experimental noise sources, it
assumes that diffusive properties are constant through-
out the duration of each trajectory. Thus, pEM is only
suitable to analyze particle tracks sampled at sufficiently
short timescales that transitions between different diffu-
sive states may be neglected. In addition, both methods
make a short-time diffusion approximation, thereby effec-
tively assuming that every particle trajectory undergoes
normal diffusion. In fact, however, diffusing proteins in-
teract with the complex environment in living cells, which
can lead to non-normal diffusive behavior, including, for
example, confined diffusion within focal adhesions [5] and
membrane corals [4], in which a labelled protein is teth-
ered to a particular fixed location within a cell, and sub-
diffusive behavior in the bacterial cytoplasm [1 -- 3], which
may be the result of the complex viscoelastic properties
of this medium [1].
Thus, the short-time diffusion approximation made by
pEM and vbSPT does not necessarily hold on experimen-
tally relevant time scales.
In the present paper, we present an overall methodol-
ogy, comprising a number of advances, that overcome
these limitations: First, we extend Berglund's MLE
framework to determine the diffusion parameters for
canonical modes of non-normal diffusion, namely con-
fined diffusion and fractional Brownian motion (fBm);
Second, we introduce a model selection scheme, that we
term mleBIC, which classifies individual trajectories to a
given diffusion model; Third, we extend the pEM frame-
work to be able to uncover non-normal diffusion modes
and transitions between different diffusive states within
particle trajectories. We also give empirical guidelines
for the sort of data likely to be necessary to successfully
apply our methodology.
Specifically, in Sec. II A, we demonstrate the improved
performance of MLE against traditional taMSD analy-
sis on various sets of simulated particle trajectories un-
dergoing non-normal diffusion across a wide parameter
spectrum. Since the diffusion mode of each experimental
particle track is not known a priori, in Sec. II B, we intro-
duce a model selection scheme, based on the Bayesian in-
formation criterion (BIC), that we call mleBIC, for clas-
sifying individual trajectories to a given diffusion model.
By applying mleBIC to synthetic trajectories undergo-
ing various modes of non-normal diffusion, both without
and with localization noise, we illustrate, by example,
2
the statistical limits of mleBIC's classification. In gen-
eral, we find that, even though MLE estimation is quite
reliable for determining diffusion parameters, classifica-
tion to determine the correct underlying diffusion model
depends strongly on the length of the trajectory, and is
only accurate for sufficiently long trajectories. Moreover,
resolving the level of heterogeneity within a population
of trajectories, that realize different diffusion modes, re-
mains challenging. Consequently, the SPT analysis goals
defined above -- specifically, uncovering the number of dif-
fusive states and their properties and transitions -- cannot
generally and reliably be achieved from an analysis that
treats individual particle trajectories independently.
Therefore, in Section II C, we turn to a systems-level
analysis: We present a major extension of the pEM
framework, that we call pEM version 2 (pEMv2), that
seeks to uncover the system of diffusive behaviors arising
from distinct physical interactions by: (1) identifying the
number of unique diffusive states (normal or non-normal
diffusion modes), (2) determining the diffusive proper-
ties of each diffusive state, and (3) classifying individ-
ual trajectories to particular diffusive states to reveal the
spatio-temporal dynamics of each diffusive behavior in
reference to the cell. In addition to now being applicable
to non-normal modes of diffusion, importantly, pEMv2
eases the other important constraint on pEM, namely
that the diffusive state remain the same throughout the
trajectory. It accomplishes this by splitting long trajec-
tories into equally-sized bins of smaller trajectories, thus
enabling transitions between different diffusive states to
be accounted for. We test the performance of pEMv2 on
various sets of synthetic particle trajectories to gain bet-
ter intuition concerning its capabilities and limitations
in reference to the free parameters in the analysis. We
show that in many case pEMv2 is indeed able to uncover
and characterize normal/non-normal diffusion modes and
the transitions between them. Thus, pEMv2 represents a
powerful new analysis tool for accurately characterizing
the interactions of diffusing proteins in live cells, and it
brings us a major step closer to being able to understand
spatio-temporal biochemistry inside living cells via SPT.
II. RESULTS AND DISCUSSION
A. Maximum likelihood framework
The one-dimensional (1D) stochastic increments of a
diffusing particle undergoing a stationary Gaussian pro-
cess are given according to [23]:
x(i + 1) = x(i) + Σ(i, j)1/2W (j),
(1)
where x(i) is the x-coordinate of the particle's position
at time step i, W (j) is a standard Brownian motion with
the properties: (cid:104)W (j)(cid:105) = 0 and (cid:104)W (i), W (j)(cid:105) = δi,j,
where δi,j is the Kronecker delta, Σ(i, j) is the covariance
matrix of the particle's x-displacements at time steps i
Mode
Normal
Confined Σconf ined(i, j) = Σconf ined(i, j) +
Σconf ined(i, j) =
fBm [22]
Σf Bm(i, j) =
3
, j = i
, j = i ± 1
, otherwise
3 D∆t
−σ2 + 1
0
3 D∆t
Σnormal(i, j) =
Covariance matrix (µm2)
6
−σ2 − 1
− 1
6
6
, j = i
, j = i ± 1
, otherwise
(cid:40) 2D∆t + 2σ2 − 2
2σ2 − 1
(cid:0)2 Σconf ined(i, i) − 2 Σconf ined(i, i + 1)(cid:1)
(cid:0)2 Σconf ined(i, j) − Σconf ined(i, j − 1) − Σconf ined(i, j + 1)(cid:1)
(cid:0)2 Σconf ined(i, j) − Σconf ined(i, j − 1) − Σconf ined(i, j + 1)(cid:1)
L2
(cid:80)∞
(cid:80)∞
(cid:80)∞
2D∆tα
(α+2)(α+1) (A(1) − 2) + 2σ2
(α+2)(α+1) (A(2) − 2A(1) + 2) − σ2;
(α+2)(α+1) (A(j − i + 1) − 2A(j − i) + A(j − i − 1))
k4 (−2Φ(j − i + 1) + Φ(j − i) + Φ(j − i + 2))
where
1
k4 Φ(1)
k4 Φ(1) (2 − Φ(1))
1
, j = i
, j = i ± 1
, otherwise
6 − 16L2
−L2
12 + 8L2
π4
π4
(cid:104)−(cid:0) kπ
, j = i
, j = i ± 1
, otherwise
where Φ(n) = exp
D∆tα
D∆tα
Dn∆t
.
L
k=1,odd
k=1,odd
1
8
π4
k=1,odd
(cid:1)2
(cid:105)
where A(n) = (n + 1)α+2 + (n − 1)α+2 − 2nα+2
(cid:40) 2σ2
−σ2
0
, j = i
, j = i ± 1
, otherwise
Immobile
Σimmobile(i, j) =
TABLE I. Analytical covariance matrix of particle track displacements separated in time by ∆t for canonical diffusion models,
namely normal diffusion, confined diffusion, fractional Brownian motion, and an immobile model with localization noise
corrections, assuming that the camera exposure time equals the frame duration, ∆t. D is the diffusion coefficient, L is the
confinement size for confined diffusion, and α is the anomalous exponent for fBm.
and j. Eq. 1 employs the Einstein summation convention
in which a sum over j is implied.
It follows from Eq. 1 that the likelihood function,
P (∆xΣ), is given by a multivariate Gaussian distribu-
tion according to:
P (∆xΣ) =
1
(2π)N/2Σ1/2
exp
− 1
2
∆xT Σ−1∆x
, (2)
(cid:20)
(cid:21)
where ∆x represents the vector of the N particle track
displacements, {∆x(n)}N
n=1, and ∆xT is its transpose.
Σ is the determinant of the covariance matrix, and Σ−1
is its inverse. Eq. 2 is the likelihood function that we seek
to maximize. The dependence of the covariance matrix in
EQ. 2 on model parameters for several canonical modes
of diffusion is give in Table I.
For normal diffusion, the presence of experimental
noise sources, namely static localization noise, which
is the uncertainty due to a finite number of photons
emitted from a fluorophore during a camera's exposure
time, and dynamic localization noise, which is the un-
certainty caused by the motions of the fluorophore dur-
ing a camera's exposure time, has been shown to con-
tribute nearest-neighbor covariance terms [19].
In the
Supplemental Materials, these calculations are extended
to incorporate static localization noise into the covariance
terms for non-normal diffusion with the result that
Σstatic(i, j) =
j = i
j = i ± 1
otherwise.
(3)
2σ2
−σ2
0
Assuming that the camera exposure time equals ∆t,
which is the usual situation in SPT measurements, the
dynamic localization noise contribution to the covariance
matrix for normal diffusion and confined diffusion, may
be shown to be given approximately by:
Σdynamic(i, j) ≈
(cid:16)
− 1
6
2 Σ(i, j) − Σ(i + 1, j) − Σ(i, j + 1)
(cid:17)
(4)
.
A derivation of Eq. 4 is given in the Supplemental Mate-
rials. For fBm, the contribution of dynamic localization
noise to the covariance matrix is derived in Ref. [22]. As
also shown in the Supplemental Materials, corrections
for static localization noise, Σstatic, and dynamic local-
ization noise, Σdynamic, contribute additively to the co-
variance matrix:
Σ = Σ + Σstatic + Σdynamic,
(5)
where Σ is the covariance matrix in the absence of
noise (Appendix A). Analytical results for the covariance
matrix, incorporating localization noise corrections, for
three canonical modes of diffusion, including an immo-
bile particle model, are given in Table I. The likelihood
function is maximized numerically as described in Sec.
IV C.
To validate the performance of our maximum likeli-
hood framework, we generated various sets of synthetic
particle trajectories, corresponding to different modes of
diffusion, as described in the Methods (Sec. IV). For con-
fined diffusion, trajectories were simulated with a num-
ber of confinement sizes from 0.25 to 5 µm; for fBm,
trajectories were simulated with a number of anomalous
exponents from 0.25 to 1.75. For normal and confined
diffusion, the trajectories were simulated with a diffu-
sion coefficient of Dsim = 0.3 µm2s−1. For fBm, the
trajectories were simulated with a "diffusion coefficient"
of Dsim = 0.3 µm2s−α. Dynamic localization noise was
added by first simulating particle positions separated by
"micro" time steps of δt = ∆t/32, and then by averag-
ing blocks of 32 of these positions together to produce
positions separated by time steps of ∆t = 32 ms. The
net effect is to mimic experimental motion-blurred posi-
tions, corresponding to a camera exposure time equal to
the frame duration of ∆t = 32 ms. Static localization
noise was included by adding a normally distributed ran-
dom number with zero mean and variance, σ2
sim, to each
motion-blurred position, where σsim = 0.04 µm (Meth-
ods (Sec. IV)). For each set of diffusion parameters, we
generated sets of particle trajectories with track lengths
N = {30, 60, 120, 240} steps. To maintain the same
level of positional information across all sets of synthetic
particle trajectories, the total number of particle posi-
tions across each simulation set was constant at 12,000
total steps.
We have compared the performance of MLE and
taMSD analyses using synthetic particle trajectories both
without localization noise (Figs. S1-S2) and with local-
ization noise (Figs. S3-S4). The detailed procedures in-
volved in the MLE analysis and the taMSD analysis are
given in Methods (Sec. IV). For confined diffusion (Figs.
S1 and S3), MLE outperforms taMSD. Even though both
the MLE and the taMSD diffusivity estimates exhibit a
positive bias in their estimations of the diffusion coeffi-
cient, both the bias and the error are significantly less
for MLE than for taMSD, especially in the presence of
localization noise. When analyzing synthetic particle tra-
jectories with localization noise, taMSD-based estimates
of the confinement length are erratic. By contrast, even
with localization noise, MLE yields reasonable confine-
ment size estimates, provided the reduced confinement
size (Lreduced =
) is sufficiently small. As could
be expected, the range of reduced confinement sizes for
which MLE provides reasonable estimates increases with
L√
12D∆t
4
increasing track length, because the increased errors for
larger reduced confinement sizes are associated with each
particle's limited sampling of its confinement, that is in-
evitable for short tracks. For the MLE analyses, the
static localization noise estimate was slightly negatively
biased with a decreasing bias for increasing track length.
For fBm (Figs. S2 and S4) also, MLE is superior to
taMSD. In this case, MLE and taMSD estimates both
appear unbiased when particle tracks do not contain lo-
calization noise. However, the MLE estimates show no-
ticeably lower errors.
In the presence of localization
noise, both MLE and taMSD estimates for the diffusiv-
ity, anomalous exponent, and static localization noise be-
come biased. However, both the bias and the error are
considerably less for MLE than for taMSD. As expected,
the bias and the error are reduced the longer the trajec-
tories analyzed both without and with localization noise.
These collected results unambiguously demonstrate
that MLE improves upon taMSD estimates for non-
normal diffusion modes, in each case reliably characteriz-
ing the underlying diffusion model over a broader range
of parameter space. They also emphasize that the pres-
ence of static localization noise reduces the quality of
both taMSD- and MLE-based estimation, and in some
cases, may introduce a bias, underscoring the importance
of properly incorporating the effect of localization noise.
As expected, bias and errors are reduced for longer (but
fewer) individual trajectories, even for a fixed total num-
ber of time steps.
B. Performance of Bayesian model selection to
classify individual particle trajectories
For experimental particle trajectories, the underlying
mode of diffusion is in general unknown a priori. There-
fore, some criterion must be imposed to select the best
model, i.e. to statistically assess which diffusion model
best describes any given particle trajectory.
P (∆x)
tion constant, given by (cid:80)
(cid:82) P (∆xθ,Mk)P (θMk)dθ, where P (∆xθ,Mk) is the
According to Bayesian model selection, classifica-
tion can be made by inferring the probability of
the kth diffusion model, Mk,
from a trajectory,
P (Mk∆x), where ∆x represents the vector of dis-
placements from a particle trajectory. According to
Bayes' rule, the probability of diffusion model k is:
P (Mk∆x) = P (∆xMk)P (Mk)
, where P (Mk) is the
model prior, P (∆x) may be viewed as a normaliza-
i∈M P (∆xMi)P (Mi), and
P (∆xMk) is the model evidence given by P (∆xMk) =
likelihood distribution and P (θMk) is the prior distri-
bution of the parameters, θ, of model k.
Although the model prior, P (Mk), may be specified
to express a preference for a particular model, we elect
to take an agnostic approach and assume that all diffu-
sion models are equally probable.
In this manner, the
model evidence is the only term of interest as the nor-
malization absorbs all other contributions. However, the
5
FIG. 1.
Classification probability via mleBIC of simulated particle trajectories with localization noise for various particle
track lengths, undergoing (A) normal diffusion for various underlying diffusivities, (B) confined diffusion for various reduced
confinement sizes, and (C) fractional Brownian motion for various anomalous exponents. Each row represents a different particle
track length: (first row) N = 30 steps, (second row) N = 60 steps, (third row) N = 120 steps, and (last row) N = 240 steps.
The probability of each model was calculated on the basis of the fraction of tracks classified to that model at each point in
parameter space, and is specified by a unique marker and color: immobile (cyan cross), normal diffusion (blue circles), confined
diffusion (red square), and anomalous diffusion (green diamond). Error bars represent the observed standard deviation.
priors of each model, P (θMk), may introduce a bias
which becomes more pronounced when the peak of the
likelihood distribution is not sharp. A representative like-
lihood distribution for confined diffusion (Fig. S5), and
fBm (Fig. S6), calculated using simulated data for var-
ious track lengths, demonstrates that the likelihood dis-
tribution for these non-normal diffusion modes is indeed
broad near its global maximum.
To minimize the influence from priors, we employ a
Laplace approximation to the model evidence and assume
a broad multivariate Gaussian prior with a full rank co-
variance matrix, which leads via standard manipulations
to the Bayesian information criterion given according to
[24, 25]:
BIC = ln P(∆xMk) = ln P(∆xθ,Mk) − Nparams
ln M,
(6)
where θ are the maximum likelihood parameters of model
k, Nparams is the number of free parameters, and M is
the number of particle track displacements.
2
In summary, for a given trajectory, the MLE is found
for each candidate diffusion model, according to Meth-
ods (Sec. IV), yielding the parameter estimates and log-
likelihood value, from which the BIC can be calculated
(Eq. 6). The model probability for each diffusion model
can be subsequently calculated according to:
exp(cid:0)BICk − BIC(cid:1)
i=1 exp(cid:0)BICk − BIC(cid:1) ,
(cid:80)K
(7)
P (Mk∆x) =
where BIC is the maximum BIC value across K candi-
date diffusion models. Thus, classification is determined
by the diffusion model which yields the highest model
probability. Henceforth, this analysis pipeline is referred
to as mleBIC.
To understand the statistical limits of classification
under ideal circumstance, namely particles which have
constant diffusion properties throughout the duration of
their trajectories, we employed mleBIC across various
sets of synthetic particle trajectories with static and dy-
namic localization noise for each canonical diffusion mode
N=30N=60N=120N=240DreducedLreducedαClassification ProbabilityABC00.20.40.60.8100.20.40.60.8100.20.40.60.8110-210010200.20.40.60.811001010.40.711.31.6ImmobileDiffusionConfinedfBM (Fig. 1). For short particle trajectories undergoing nor-
mal diffusion (Fig. 1A), a normal diffusion model was fa-
vored with a high probability when Dreduced = D∆t
σ2 > 1.
When Dreduced < 1, the underlying static localization
noise dominates the underlying diffusion, which leads
mleBIC to favor an immobile model. Thus, more statis-
tics are necessary to reject the simpler immobile model.
For particle trajectories undergoing confined diffusion
(Figs. 1B), when confinement sizes are small, a confined
diffusion model is favored. As the confinement size in-
creases, a normal diffusion model becomes favored. At
this confined-to-normal crossover, a small preference for
anomalous diffusion is found. As expected, longer trajec-
tories provide more opportunities to explore the bound-
aries of confinement, resulting in a wider region of param-
eter space for which a confined diffusion model is favored.
For particle trajectories undergoing fBm (Fig. 1C),
a normal diffusion model is mostly favored when parti-
cle trajectories are short (N ≤ 60). A fBm diffusion
model is not consistently favored until trajectories con-
tain 240 steps, albeit only when the anomalous exponent
is below 0.7 or greater than 1.3. As expected, when par-
ticle trajectories contain minimal localization noise er-
rors, mleBIC yields improved estimation for fBM (Fig.
S7). Thus, the presence of localization noise requires
even longer tracks for proper classification, even though
MLE can determine reliable estimates for the underly-
ing diffusivity and anomalous exponent (Fig. S4). BIC's
built-in parsimony causes it to favor a normal diffusion
model, when there is not enough data to support a non-
normal diffusion model, even when the correct model cor-
responds to non-normal diffusion. This behavior seems
not undesirable.
Similar to taMSD analysis, mleBIC does not take into
account transitions between diffusive states. While ana-
lyzing subsets of the data may allow for different diffu-
sive states within a particle trajectory, figures 1 and S7
illustrate that accurate classification cannot be made for
wide ranges of parameter space, even in the most ideal
circumstances. As the trajectories become longer, the
statistical power grows, thereby allowing for improved
mleBIC classification over a wider parameter space, and
misclassification gradually reduces. However, longer par-
ticle trajectories which have constant diffusion properties
becomes increasingly unlikely, especially when a particle
is diffusing in a complex environment such as a living
cell. Thus, while mleBIC is certainly an improvement
over taMSD analysis, the statistical power of classifica-
tion by analyzing particle trajectories on an individual
basis remains limited.
C. Systems-level analysis of a collection of particle
trajectories
To augment the limited statistics provided by individ-
ual particle trajectories, pEM simultaneously analyzes
a collection of trajectories by employing a systems-level
M(cid:88)
(cid:40) K(cid:88)
(cid:41)
6
likelihood function to account for a finite number of
unique diffusive states, each of which we envision to arise
as a result of particular interactions within the cell. Here,
we extend the original pEM framework [17] to now in-
clude non-normal modes of diffusion, i.e. we lift the
short-time-diffusion approximation. A powerful aspect of
this new version of pEM is that it is essentially a model-
free approach, in that no prior assumptions need be made,
concerning which types of diffusion mode are present in
the data at hand.
To implement the new version of pEM, we first write
the systems-level log-likelihood function:
lnL(∆x π, Σ) =
ln
πkP (∆xmΣk)
,
(8)
m=1
k=1
where M is the total number of tracks, which collectively
realize K distinct underlying diffusive states, ∆xm rep-
resents the vector of Nm displacements for particle tra-
jectory m, ∆xm = {∆xm(n)}Nm
m=1 is
the set of M particle track displacements, π = {πk}K
k=1
is the set of variables which represent the fraction of the
population of trajectories that realize diffusive state k,
which is bounded and normalized: 0 ≤ πk ≤ 1 and
k=1 is the set of covariance
(cid:80)K
k=1 πk = 1, and Σ = {Σk}K
n=1, ∆x = {∆xm}M
matrices which defines each diffusive state.
Importantly, the theoretical covariance matrix for any
diffusion mode that undergoes a stationary Gaussian pro-
cess, including in the presence of localization noise, has
a symmetric Toeplitz form (Table I), so that element
(i, j) of the covariance matrix depends only on i − j.
We can impose the requirement that the covariance ma-
trix for each diffusive state, Σk, take on such a sym-
metric Toeplitz form by averaging the diagonal, one-
off-diagonal, two-off-diagonal, etc. elements of the em-
pirical covariance matrix for particle track m to obtain
the experimental covariance matrix elements for track m:
Cm(i, j) = Cm(i−j) = (cid:104)∆xm(l)∆xm(l+i−j)(cid:105), where
the average is taken over all possible values of l for track
m.
Furthermore, because the covariance structure of each
diffusion mode decreases rapidly to zero for increasing
separation between displacements -- i.e. with increasing
i − j -- we can reasonably restrict the number of in-
formative covariance matrix elements that we include in
the analysis by setting Cm(i − j) = 0, for i − j > f ,
where f is the number of off-diagonal covariance matrix
elements included in the analysis.
If f = 0, only the
diagonal elements of the covariance matrix are permit-
ted to be non-zero, reproducing the theoretical structure
of the covariance matrix for simple diffusion in the ab-
sence of localization noise. For f = 1, one-off-diagonal
element is included, permitting the covariance matrix to
properly account for localization noise sources. In prin-
ciple, different diffusive states, which are characterized
by unique diffusion properties, may be distinguished one
from another on the basis of different values of the co-
variance matrix elements. The inclusion of additional
off-diagonal terms introduces additional information to
help distinguish diffusive states that undergo confined
diffusion, fBm or other modes of non-normal diffusion.
Because pEM discovers the values of these covariance
matrix elements for each diffusive state, it is not neces-
sary to specify ahead of time what diffusion modes are
present, beyond specifying f . It is in this sense that this
version of pEM is model independent. In the case of K
diffusive states, insisting that the covariance matrix must
be a symmetric Toeplitz matrix and limiting the number
of off-diagonal matrix elements to f means that the num-
ber of model parameters is equal to K(1 + f ) + K − 1.
(There are K − 1 independent population fractions.)
Maximizing Eq. 8 with respect to {Σk, πk}K
k=1 nat-
urally yields the expectation-maximization (EM) algo-
rithm [26]. In the expectation step, the posterior prob-
ability, γmk, that particle trajectory m realizes diffusive
state k, given the current estimates for Σk, and πk, is
calculated according to:
(cid:80)K
πkP (∆xmΣk)
j=1 πjP (∆xmΣj)
γmk =
.
(9)
In the maximization step, the posterior probability is
used to update the parameter estimates of each diffusive
state:
M(cid:88)
m=1
Σk =
1
Mk
γmkCm
(10)
Mk
M
πk =
(cid:80)M
(11)
where Cm(i, j) = (cid:104)∆xm(i)∆xm(j)(cid:105) and Mk =
m=1 γmk. The EM algorithm solves these equations
iteratively until the change in the log-likelihood becomes
smaller than a set threshold [26].
The extension to higher dimensions than one is car-
ried out as follows. We calculate the expectation step by
averaging the posterior probability over each dimension
using the same parameter estimates. For the maximiza-
tion step, the maximized parameter estimates are calcu-
lated separately for each dimension and then averaged.
At each step in the iteration procedure, the complete
log-likelihood is calculated by summing the log-likelihood
from each dimension.
Although the EM algorithm guarantees convergence to
a maximum [26], convergence to the global maximum is
not guaranteed, depending on the initial parameter val-
ues. However, as described in detail in Ref. [17] and sum-
marized in the Methods (Sec. IV), suitably perturbing
the likelihood surface, namely pEM, is a computationally
efficient means to reach the global maximum likelihood.
Since the number of diffusive states is not known a
priori, we repeat the pEM procedure for different num-
bers of diffusive states, finding the maximum likelihood
in each case. To maintain model parsimony, we again
employ the Bayesian Information Criterion to penalize
for the inclusion of additional diffusive states, via a
7
systems level extension of Eq. 6. Specifically, we se-
lect the model with the largest value of the systems-
level BIC, where now log L is the systems-level likeli-
hood function (Eq. 8), the number of free parameters
is Nparams = K(1 + f ) + K − 1, and M is the total num-
ber of particle track displacements across the population
of tracks.
The procedure described so-far makes the assumption
that the diffusive properties remain constant through-
out the duration of each trajectory. In order to extend
pEM, so that it can be applied to trajectories contain-
ing transitions between different diffusive states, we split
each trajectory into equal-size bins, such that each bin
contains B sequential steps. The assumption of a con-
stant covariance matrix is still assumed to hold within
each such bin, but different bins can realize different dif-
fusive states. In this way, pEMv2 is able to account for
transitions between different diffusive states within the
overall trajectory. Each bin is treated as a Markovian
measurement of the diffusive state, Eq. 2. The temporal
resolution corresponds to the bin size.
To summarize, our enhanced version of pEM, which
we call pEMv2, examines a population of binned particle
trajectories, each containing B steps, to determine the
number of unique covariance matrices, contained in the
population. It accomplishes this goal by classifying each
binned trajectory to a particular diffusive state, based
on similarities in the covariance structure among trajec-
tories. Using the resultant classification, pEMv2 then
updates the parameter estimates for each diffusive state.
Iteration of this process allows pEMv2 to learn in an
unsupervised manner what unique covariance structures,
i.e. what diffusive states, are realized within the popula-
tion of binned trajectories. Since the number of diffusive
states is intrinsically handled by the BIC (Eq. 6), the
user-controllable parameters for pEMv2, are the number
of off-diagonal elements to include in the covariance ma-
trix, f , and the bin size, B.
1. Dependence on the number of covariance terms
To investigate the performance of pEMv2, we have gen-
erated a number of sets of synthetic particle trajectories
containing different numbers of diffusive states and dif-
ferent degrees of similarity between the covariance terms
across diffusive states. Table II specifies the four sets of
diffusion parameters (case 1 through case 4), which were
used to generate the synthetic data sets. There are no
transitions among different diffusive states for case 1 and
case 2, i.e. the transition probability matrix (A) is given
by A = δi,j. However, for case 3 and case 4, transitions
are permitted with the corresponding matrices of transi-
tion probabilities given by
0.995 0.001 0.004
0.001 0.995 0.004
0.015 0.015 0.970
A3 =
(12)
8
FIG. 2. Empirical probability distributions of the mean covariance matrix elements, (cid:104)C(i, j)(cid:105), for j = i, j = i ± 1, j = i ± 2,
and j = i± 6 for case 1 (top row), where states 1, 2, 3, and 4 are shown in red, green, blue, and cyan, respectively, and for case
2 (bottom row), where states 1 and 2 are shown in green and blue, respectively. Vertical dashed lines indicate the theoretical
values with a color corresponding to each diffusive state.
for case 3, and
1 − 3p
p
p
p
(13)
p
1 − 3p
p
p
p
p
1 − 3p
p
p
p
p
1 − 3p
A4 =
1
e
s
a
C
2
e
s
a
C
3
e
s
a
C
4
e
s
a
C
mode
(µm2s−1)
Dsim
k
Lsim
k
(µm)
αsim
k
σsim
k
(µm)
k
πsim
mode
(µm2s−1)
Dsim
k
Lsim
k
(µm)
αsim
k
σsim
k
(µm)
k
πsim
mode
(µm2s−1)
Dsim
k
Lsim
k
(µm)
αsim
k
σsim
k
(µm)
k
πsim
mode
(µm2s−1)
Dsim
k
Lsim
k
(µm)
αsim
k
σsim
k
(µm)
πsim
k
3
fBM
0.25
0.9
0.04
0.25
4
fBM
0.4
0.6
0.04
0.25
1
2
Confined Normal
0.05
0.13
1
0.04
0.25
0.15
1
0.04
0.25
Confined Normal
0.06
0.1
1
0.04
0.4
0.06
1
0.04
0.6
Confined Confined Normal
0.005
0.05
1
0.04
0.33
Normal
0.001
0.1
0.2
1
0.3
1
0.04
0.34
0.04
0.33
fBM Normal
0.03
0.2
1
0.04
0.25
.7
0.04
0.25
1
0.04
0.25
fBM
0.45
0.9
0.04
0.25
TABLE II. Simulation parameters for synthetic particle tra-
jectories generated for case 1, case 2, case 3, and case 4.
for case 4, where p is input into the simulation selected
from one of {0, 0.003, 0.005, 0.01, 0.015, 0.02, 0.03}.
The covariance matrix elements of different diffusive
states must be sufficiently distinct in order for pEMv2
to resolve them as separate diffusive states. First, there-
fore, we sought to explore the effect of the number of
off-diagonal covariance matrix elements (f ), that are in-
cluded in pEMv2 analysis. Figure 2 shows the measured
probability distributions of the average covariance matrix
elements, (cid:104)C(i, j)(cid:105) for i− j = 0, 1, 2, and 6, determined
from populations containing 1,500 synthetic trajectories,
realizing four diffusive states with diffusion parameters
corresponding to case 1 (top row), and 1,500 synthetic
trajectories, realizing two diffusive states with diffusion
parameters corresponding to case 2 (bottom row). To re-
capitulate the variability found experimentally, the tra-
jectory lengths were distributed according to an exponen-
tial probability distribution with a characteristic length
of 25 steps, with a minimum cut-off of 15 steps and a
maximum cut-off of 60 steps. In addition, because there
are no transitions, in our analyses of case 1 and case 2,
we analyzed each complete trajectory as a whole, as in
the original version of pEM, without splitting into bins.
Case 1 corresponds to four diffusive states, two normal
diffusion, one fBM, and one confined diffusion but their
diffusion coefficients are well-separated from each other.
Case 2 corresponds to two diffusive states with the same
diffusion coefficient, one corresponding to normal diffu-
sion and the other to confined diffusion.
For both case 1 and case 2, the means of the distribu-
tions of (cid:104)C(i, i)(cid:105) and (cid:104)C(i, i ± 1)(cid:105) for each diffusive state
are well separated. For case 1, however, the means of
(cid:104)C(i, i ± 2)(cid:105) are all very similar to each other and are
close to zero, with the exception of state 4 (cyan). By
<Ci,i><Ci,i±1><Ci,i±2><Ci,i±6>Covariance (μm2)Case 1Case 2Probability00.050.100.10.20.3-0.0200.0200.10.20.3-0.0200.0200.10.20.3-0.0200.0200.10.20.300.010.0200.050.1-0.0100.0100.050.10.15-0.0100.0100.050.10.15-0.0100.0100.050.10.15 9
FIG. 3.
Log-probability as a function of the number of
diffusive states (model size) for (A) case 1 and (B) case 2,
determined by pEMv2 analysis using f = 1, 2, 4, 6, 9, and 13
off-diagonal covariance matrix elements, shown in red, yellow,
green cyan, blue, and magenta, respectively. Each data set
consists of 5 sets of simulated particle tracks (shown as a
different curve) for each f (shown as a different color). The
inset shows a zoomed in representation near the maximum
log-probability.
contrast, for case 2, the means of (cid:104)C(i, i ± 2)(cid:105) for state
1 and state 2 remain distinguished from each other. For
case 1 and case 2, the means of (cid:104)C(i, i ± 6)(cid:105) for each dif-
fusion state are all very similar to each other and are all
close to zero, albeit their widths remain distinct.
Fig. 3 shows the log-probability of each model size,
determined on the basis of BIC score (Eq. 7), as a func-
tion of model size for different numbers of non-zero off-
diagonal covariance matrix elements between f = 1 and
13. For case 2, the correct number of diffusive states
is found (K = 2), irrespective of f . For case 1, where
K = 4, pEMv2 is able to successfully determine the cor-
rect numbers of diffusive states, as indicated by the max-
imum log-probability, except when f = 13, for which a 3
diffusive state model is favored for three out of the five
data sets analyzed.
A visual representation of how successfully pEMv2 de-
termines the correct diffusive state is given in Fig. 4,
which shows 1500 synthetic particle trajectories corre-
sponding to case 1 (top row) and case 2 (bottom row).
In the left column, each trajectory is depicted using a
color, corresponding to the known, simulated diffusive
state of the track. In the right column, each trajectory
FIG. 4.
Representations of 1500 synthetic particle trajec-
tories for case 1 (top row) and case 2 (bottom row). In the
left column, each trajectory is depicted using a color, corre-
sponding to the the known, simulated diffusive state of the
track. In the right column, each trajectory is depict using a
color, corresponding to the diffusive state, that yields maxi-
mum posterior probability, determined on the basis of pEMv2
using f = 6 off-diagonal covariance matrix elements. The
starting position of each trajectory is sequentially placed on a
2-dimensional grid, separated one from another by 1 µm (top
row) and 0.4 µm (bottom row). In both cases, the scale bar
represents 5 µm.
is depicted using a color, corresponding to the diffusive
state that realizes the maximum posterior probability for
that track, determined using f = 6 off-diagonal covari-
ance matrix elements for case 1 and case 2. Although
there are a few misclassified trajectories, the overwhelm-
ing majority of the trajectories are correctly classified,
demonstrating that pEMv2 is capable of reliably uncov-
ering the diffusive states in these cases.
Fig. 5 shows the fraction of correctly classified trajecto-
ries as a function of the number of off-diagonal covariance
matrix elements, confirming that pEMv2 reliably classi-
fies trajectories to the correct diffusive state. The clas-
sification accuracy shows only a modest dependence on
the number of off-diagonal covariance matrix elements in-
cluded in the analysis: For case 1, the accuracy of classi-
fication is uniformly high for f between 1 and 9, suggest-
ing that the first off-diagonal covariance matrix element
(f = 1) is decisive in case 1. The decrease in classifica-
tion accuracy for f = 13 may be because of the inclusion
in this case of a large number of noisy off-diagonal matrix
elements, suggesting that it is preferable to not include
too many off-diagonal covariance matrix elements. For
case 2, the accuracy noticeably improves as f increases
from 1 to 4, and remains high thereafter, suggesting that
off-diagonal covariance matrix elements up to f = 4 are
informative for classification in this case.
The classified covariance matrix elements and the clas-
Case 1Case 2Model Size12345log-Probability-6000-5000-4000-3000-2000-10000ModelSize123log-Probability-1000-800-600-400-2000123-100-80-60-40-2001246913AB12345-100-80-60-40-2009131246Case 1 Simulated TracksCase 1 Classified TracksCase 2 Simulated TracksCase 1 Classified Tracks 10
that all informative covariance matrix elements are in-
cluded in the analysis, but that unnecessary noise is ex-
cluded. We ascribe the failure to select the correct model
for f = 13 to be the result of including unnecessary noise.
For the simulations in this paper, f = 6 is a reasonable
choice.
2. Uncovering transitions by splitting tracks
A population of experimental trajectories, that real-
izes multiple diffusive states, is likely to contain at least
a subset of trajectories, which contain transitions among
the diffusive states. The prevalence of transitions de-
pends on their underlying kinetics, i.e. on the transition
rates. Our concept for extending our methodology to
permit analysis of trajectories with transitions is to split
these trajectories into shorter pieces. If the duration of
the resultant short trajectories is less than the typical
lifetimes of relevant diffusive states, then each short tra-
jectory will with high probability realize a single diffusive
state throughout, and the methods described above re-
main applicable to determine the diffusive states within
the population of these short trajectories.
To investigate the feasibility of this concept, we sim-
ulated particle tracks with three diffusive states, corre-
sponding to case 3 in Table II, that transition among
each other with transition rate matrix A3 (Eq. 12). The
protocol used for generating transitions is described in
the Methods (Sec. IV). The corresponding lifetimes of
states 1, 2, and 3 are 200 (∼6.4 s), 200 (∼6.4 s), and 33
(∼1 s) steps, respectively. We then divided the simulated
trajectories into sets of short trajectories containing 5,
10, 15, 20, 25, 30, 60, 90 or 120 steps, respectively, while
keeping the total number of steps and hence the total po-
sitional information constant at 12000 total steps across
all trajectories. We then applied the pEMv2 methods de-
scribed above to each population of different-length short
trajectories, implicitly assuming that each short trajec-
tory remains in the same diffusive state throughout. The
number of off-diagonal matrix elements used in the anal-
ysis was fixed at f = 6.
Fig. 7A shows the BIC-based log-probability of various
model sizes for simulated tracks with lengths 15, 30, 60,
90, and 120 steps. The log-probability selects the correct
number of diffusive states (K = 3) for only when N = 15
steps. For trajectories containing 30 or more steps, the
BIC-based probability incorrectly favors a four diffusive
state model, presumably in an effort to describe trajec-
tories containing transitions. Given that the three state
model is correct, Fig. 7B shows the fraction of the total
number of steps that are assigned to the correct diffu-
sive state for each set of short trajectories, plotted as a
function of the track length of each set. Evidently, the
fraction of steps correctly assigned decreases as the tra-
jectories became longer. This trend is surely due to the
fact that longer trajectories provide more opportunities
to transition, as indicated by the increasing number of
FIG. 5.
Fraction of trajectories classified into the correct
diffusive state as a function of the number of off-diagonal co-
variance matrix elements used in the pEMv2 analysis of case
1 (top) and case 2 (bottom). Error bars represent the ob-
served standard deviation across 5 different sets of simulated
particle tracks.
sified taMSD are shown in Fig. 6 for each diffusive state
corresponding to case 1.
In this instance, using either
f = 1 or f = 6 in the analysis leads to the charac-
terization of each diffusive state with high fidelity, with
the measured covariance matrix elements and measured
taMSDs for each diffusive state, shown as the data points
and the solid lines in the figure, almost exactly match-
ing the corresponding true covariance matrix elements
and true taMSDs, shown as the dashed lines, which are
very nearly coincident with the solid lines. In compar-
ison with mleBIC, which was unable to reliably classify
60-step trajectories undergoing fBm with anomalous ex-
ponents of either α = 0.9 or even α = 0.6, it is strik-
ing that pEMv2 is not only able to identify these two
diffusive states (states 3 and 4 of case 1) and to accu-
rately categorize individual trajectories into these states
(Fig. 4), pEMv2 is also able to accurately capture the
anomalous behavior of their taMSDs (Fig. 6). Thus, the
systems-level strategy employed by pEMv2 can find sub-
tle deviations from non-normal diffusive behavior, that
are statistically challenging to uncover, if trajectories are
analyzed on an individual basis.
These observations show that the particular value of f
used is not critical. In practice, we suggest that a reason-
able way to pick f is on the basis of the average covari-
ance matrix elements themselves (see Fig. 6): we suggest
picking f to correspond to the off-diagonal term of the
ensemble-averaged covariance matrix elements that has
essentially converged to zero. This choice should ensure
10.80.610.950.93711fFraction CorrectCase 1Case 2 11
FIG. 6. Average covariance matrix elements and average taMSD from maximum posterior classification as a function of time
lag for (A) states 1, 2, 3, and 4 of case 1, represented in red, green, blue and cyan, respectively, and (B) states 1 and 2 of
case 2, represented in red and blue, respectively. Each data point represents the average over five different sets of simulated
particle tracks, analyzed using pEMv2 using f = 1 (left column) and f = 6 (right column). The solid lines linking the data
points are guides-to-the-eye. The error bars correspond to the standard deviation of the mean across 5 trials. Shown as the
dashed curves are the true matrix elements and the true ensemble-averaged taMSD for each state, determined using the known
diffusive states of trajectories, while the shaded bands represent its standard deviation.
transitions per track with increasing trajectory length,
shown in Fig. 7C.
To permit pEMv2 to deal with tracks containing tran-
sitions, we implemented a procedure that splits long tra-
jectories into shorter trajectories. For the 120-step data
set, Fig. 7D shows that the log-probability of various
model sizes for tracks, split into 5, 10, 15, 20, or 30 steps,
yields the correct model (K = 3) for bin sizes less than
30 steps, in agreement with Fig. 7A. Given that the three
state model is correct, Fig. 7E shows the fraction of the
total number of steps that are assigned to the correct dif-
fusive state as a function of the bin size. Evidently, this
procedure yields a significant improvement in the frac-
tion of steps correctly assigned compared to analysis of
the 120-step data set, shown in Fig. 7B, presumably as
a result of decreasing the number of transitions per track
from ∼1.1 per track for the 120-step data set to ∼0.1
transitions per track for the bin size of 10 steps (Fig.
7). Moreover, pEMv2 is now able to provide a signifi-
cant improvement in the quality of the estimates for the
covariance elements and taMSD for each diffusive state
(Fig. 8), as well as reasonable estimates of the transition
matrix (Fig. S8).
Is there an optimal bin size? Indeed, Fig. Fig. 7E
shows that the fraction of steps correctly assigned ex-
hibits a maximum at a bin size of 10 steps and decreases
for smaller and larger bin sizes.
It turns out that us-
ing smaller bin sizes may render the results of pEMv2
more susceptible to misclassification (Fig. S9). Since
information of confinement manifests as anti-covariances
between neighboring displacements each time a particle
"bounces" off of the confinement barrier -- if the bin size
is too small, then this information is only contained in the
few bins which capture such a "bouncing" event, while
other bins would follow an apparent normal diffusion. On
the other hand, although including more steps in the bin
size allow for more anti-covariance "bouncing" events, a
large bin size also has the undesirable effect of increasing
the number of transitions per track, which can also lead
to poorer performance. Thus, the bin size should be cho-
sen to be as large possible, subject to the constraint that
the mean number of transitions per trajectory should not
be too large. In this example of case 3, satisfactory re-
sults are obtained by using a level of binning that yields
an average of 0.2 transitions per trajectory.
3. Determining the optimal bin size
To further elucidate how pEMv2's performance de-
pends on the level of transitions and how to determine the
optimal bin size in an unsupervised manner, we generated
a number of data sets containing 3,000 synthetic particle
tracks with diffusive states given according to case 4 (Ta-
ble II), and with varying mean numbers of transitions per
track (R = {0, 0.36, 0.6, 1.2, 1.8, 2.4, 3.6} transitions
per track). All of the track lengths were constant with N
= 120 steps. For each data set, we applied pEMv2 with
bin sizes ranging from 5 to 30 steps. The mean number
of transitions per trajectory for each bin size is shown as
a function of bin size in Figure 9A.
By applying pEMv2 to each of these data sets, the
BIC's log-probability found the correct model size (K =
4) when the transition rates were low (R < 0.6 transitions
f=1f=1f=6f=6Case 1Case 2Time Lags (s)Time Lags (s)Time Lags (s)Time Lags (s)taMSD (μm2)Covariance (μm2)Covariance (μm2)taMSD (μm2)AB 12
FIG. 7.
Performance of pEMv2 on particle tracks that
transition between diffusive states given by case 3. Log-
probability determined by pEMv2 analysis for 5 sets of simu-
lated particle tracks (shown as a different curve) with diffusive
states given according to case 3 with (A) track lengths of 15,
30, 60, 90, and 120 steps and (D) track lengths of 5, 10, 15,
20, and 30 steps created by splitting the 120 step data set in
(A). Each simulation set is shown in a different color. Av-
erage fraction that the classified diffusive states matches the
simulated diffusive state for (B) various track lengths and (E)
various bin sizes. For the purposes of this comparison, when
a bin contains a transition, the "true" diffusive state is cho-
sen to be the state with the highest number of displacements.
Average transition rate per track for (C) various track lengths
and (F) various bin sizes. (B,C,E,F) Error bars represent the
observed standard deviation across the 5 data sets.
per track), irrespective of the bin size used (Fig. S10).
Even when the transition rates increase (R = 2.4 and
R = 3.6 transitions per track), the BIC continues to favor
the correct four diffusive state model for smaller bin sizes.
However, the BIC favors an incorrect five-diffusive-state
model when analyzing data that uses bins containing 30
steps.
Assuming the correct model size (K = 4), figure 9B
shows the average maximum likelihood values as a func-
tion of the bin size for each data set. When transi-
tion rates are low, the average log-likelihood per step
increases monotonically with bin size, suggesting that in
these cases the optimal bin size is larger than the maxi-
mum binning used. For larger numbers of transitions per
track, however, a maximum log-likelihood per step is ob-
served within the range of bin sizes examined. Figure 9C
Average covariance matrix elements and average
FIG. 8.
taMSD from maximum posterior classification as a function
of time lag for states 1, 2, and 3, represented in red green
and blue, respectively, with a bin size of (A) 120 steps and
(B) 10 steps. Each data point represents the average over
five different sets of simulated particle tracks, analyzed us-
ing pEMv2 using f = 6. The solid lines linking the data
points are guides-to-the-eye. The error bars correspond to
the standard deviation of the mean across 5 trials. Shown as
the dashed curves are the true matrix elements and the true
ensemble-averaged taMSD for each state, determined using
the known diffusive states of trajectories, while the shaded
bands represent their standard deviations.
shows that the optimal bin size, determined as the max-
imum log-likelihood per step from Fig. 9B, decreases
as the number of transitions per track increases. Not
surprisingly, the more transitions that are present, the
smaller the bin size should be. Figures S12 and S13 shows
the pEMv2 classification of representative trajectories of
the R = 3.6 data set for various bin sizes. When the
bin size is five steps, spurious transitions are frequently
found, which we ascribe to the relatively larger statisti-
cal fluctuations that necessarily accompany smaller bin
sizes. As the bin size becomes larger, statistical fluctu-
ations are reduced. However, if the bin size becomes
too large (B = 30), the corresponding higher rate of
transitions per track limits pEMv2's ability to classify
diffusive states accurately. Evidently, the optimal bin
size balances the accuracy of the covariance matrix ele-
ments, which becomes better-determined with larger bin
sizes, against the number of transitions per track, which
mix the covariance matrix elements of different diffusive
states, leading to poorer pEMv2 performance.
Although pEMv2, using the optimal bin size, is able
to uncover the correct numbers of diffusive states and
characterize each diffusive state reliably (Fig. S11), the
overall accuracy of pEMv2's classification decreases as
the number of transitions per track increases, as is indi-
BA05010015020025000.20.40.60.81204060801001200.60.70.80.9110203000.10.20.31020300.60.70.80.91CBin Size (steps)Bin Size (steps)Track Length (steps)Track Length (steps)Fraction CorrectTransitions per trackModel Size234-3000-2000-1000234-100-80-60-40-200153060901200log-ProbabilityModel Size234-2500-2000-1500-1000-5000234-100-80-60-40-200510152030DEFABB=120Bin = 10 cated by the fraction of correctly classified steps, plotted
in Fig. 9D. Even though the optimal bin size lowers the
effective number of transitions per track, the decreased
performance may be due to the higher absolute number
of transitions for the data sets with higher R (Fig. 9A).
Notwithstanding, the ensemble behavior of each diffusive
state can still be captured accurately when the optimal
bin size determined by the maximum likelihood per dis-
placement is used (Fig. S11).
Dependence of the optimal bin size on
FIG. 9.
transition rate.
(A) Average number of transitions per
track versus bin size for various transition rates, R =
{0, 0.36, 0.6, 1.2, 1.8, 2.4, 3.6} transitions per track (each
shown in a different color). (B) Average log-likelihood value
per step versus bin size for various transition rates. (C) Opti-
mal bin size versus the mean number of transitions per track,
determined by the maximum log-likelihood per per step. (D)
Average fraction that the classified diffusive state matches the
simulated diffusive state as a function of the mean number of
transitions per track. Each error bar represents the observed
standard deviation across 5 different sets of simulated particle
tracks.
III. CONCLUSIONS
In this paper, we introduced the likelihood functions
for two canonical modes of non-normal diffusion, namely
confined diffusion, and fractional Brownian motion. We
showed that the maximum likelihood estimates provide
a significant improvement in comparison with traditional
MSD analysis. We introduced a model selection scheme,
namely mleBIC, to determine the underlying diffusion
model that best represents the motions of a diffusing
particle. We demonstrated that while mleBIC is quite
successful at classifying tracks without localization noise;
classification of tracks with localization noise was lim-
ited, especially for short trajectories. Although, in this
paper we restricted consideration to particles undergo-
13
ing normal diffusion, confined diffusion, and fBm and
immobile particles, extensions to other diffusion models
can be added facilely by incorporating these models into
mleBIC, once the likelihood functions are known.
To take advantage of a systems-level approach, we in-
troduced an updated version of pEM analysis, namely
pEMv2, that determines the number of unique covari-
ance structures within a population of particle trajecto-
ries, thereby bolstering the statistics of individual tra-
jectories. A key output from the pEMv2 algorithm is
the posterior probability, γmk, that particle trajectory m
realizes diffusive state k. For the selected model, one sim-
ple and useful way to categorize a particular trajectory
to a particular diffusive state is to assign the trajectory
to the diffusive state that realizes the largest posterior
probability, as in Fig. 4.
When analyzing simulated trajectories that transition
between different normal/non-normal diffusive states,
pEMv2 was able to determine the covariance structure
of each diffusive state quite reliably. We also demon-
strated the rationales for the selection of the free param-
eters in pEMv2, which includes the number of covariance
features and the bin size. The number of off-diagonal co-
variance matrix elements to include can be set to the
value for which the observed ensemble-averaged covari-
ance matrix element have just decayed to zero, thereby
only including informative covariance terms in the anal-
ysis. We have shown that an optimal bin size may be de-
termined by rerunning pEMv2 for various bin sizes, and
selecting the bin size that yields the highest likelihood for
a given model size. In practice, because the model size
is unknown a priori for experimental data, we envision
running pEMv2 for different model sizes and different
bin sizes to find these conditions. Importantly, pEMv2 is
rooted in physical principles of stochastic processes. Ap-
plying non-physical clustering methods to the same data,
such as k-means clustering, lead to poor characterization
of the underlying diffusive states [17].
Since pEMv2 does not make any intrinsic assumptions
of the underlying diffusion model, besides that it follows
a Gaussian process, pEMv2 is essentially a diffusion-
model-free approach. Characterization of each covari-
ance structure to determine the diffusion mode and prop-
erties can then be performed post-hoc. Specifically, tra-
ditional analyses can then be applied for each diffusive
state, such as calculation of the ensemble-average taMSD
and the ensemble-average velocity autocorrelation func-
tion. Such a procedure provides a more reliable represen-
tation of the diffusive behavior compared to individual
trajectories, which suffer from limited statistics.
One drawback to pEMv2, is that information of the
diffusive state across every bin is treated independently.
Thus, when the bin size becomes small, spurious states
may occur more frequently. A key benefit of a hidden
Markov model (HMM) approach is that spurious tran-
sitions can be intrinsically penalized by maximizing a
likelihood function which includes a transition matrix.
We envisage that, in the future, pEMv2 can be extended
Bin Size (steps)0102030Transitions per track00.40.8R=0.0R=0.4R=0.6R=1.2R=1.8R=2.4R=3.6Bin Size (steps)0102030Log-Likelihood (per step)1.41.51.6Transitions per track0123Optimal Bin (steps)15202530ABCDTransitions per track0123Fraction Correct0.80.850.90.951 to a HMM of multivariate Gaussians, to mitigate spuri-
ous transitions. In turn, this approach will improve the
temporal resolution by making it possible to reduce the
bin sizes. However, if transition rates are inhomogeneous
across the cell, any HMM approach that assumes a sin-
gle transition matrix for the entire cell, would not be
able to properly capture that inhomogeneity. Notwith-
standing, current HMMs applied to SPT data, namely
vbSPT and HMM-SPT [9], apply a HMM of univariate
Gaussians, which is equivalent to using a bin size of 2
steps (f = 0) and thus only using information of the first
covariance term. Thus, these HMM analyses overlook
localization noise, which introduces correlations between
nearest-neighbor displacements, rendering each displace-
ment non-Markovian. Moreover, neither method can
properly account for non-normal diffusion models such
as confined diffusion and fBM.
Unfortunately, there is no strict rule concerning how
many tracks are needed for pEMv2 to return accurate
results. Rather, the amount of data needed depends on
the complexity of the diffusive states involved, as dis-
cussed previously [17]. In practice, we recommend that
pEMv2 users complement their SPT analysis of exper-
imental data with an analogous analysis of simulated
tracks that recapitulate the diffusive complexity deter-
mined by pEMv2. In this way, the user can determine
the reliability of pEMv2 for the data in hand, and thereby
gain confidence in the results provided by pEMv2.
With the ability to handle normal/non-normal diffu-
sive states which contain transitions between different
diffusive states, we envision pEMv2 can help to uncover
more accurate information regarding the diffusive states
which occur inside live cells with single molecule resolu-
tion. This analysis sets the benchmark for all future sin-
gle particle tracking analysis, to begin to understand the
spatio-temporal biochemistry of diffusing particles inside
live cells with single molecule resolution.
IV. METHODS
A. Simulation procedure
Synthetic particle trajectories undergoing normal dif-
fusion are generated using the recursion given by Eq. 1,
with Σi,j = 2D∆tδi,j and x0 = 0.
To generate synthetic particle trajectories undergoing
normal diffusion confined in a finite square geometry with
size −L to L, we simulate displacements that follow nor-
mal diffusion. At each time step, if the new position falls
outside of the finite domain, then the simulated posi-
tion is set such that the difference between the proposed
position and the boundary is reflected, i.e. Neumann
boundary condition, but the total distance traveled re-
mains the same as if the wall were not present. Here, the
starting position of each trajectory is at the center of the
confinement boundary.
Synthetic particle trajectories undergoing fractional
14
Brownian motion are generated using the recursion:
given by Eq. 1 with v = 0 and Σi,j given by Appendix A.
Here, the square root of the covariance matrix is deter-
mined with the Cholesky decomposition, i.e Σ = LLT ,
where L is the Cholesky lower triangular matrix. We
then generate a vector of normally distributed random
numbers W = {Wd}D
d=1, where D is the number of dis-
placements of the particle trajectory, and apply a matrix
multiplication according to ∆x = LW. The positions are
then reconstructed by calculating the cumulative sum of
j=1 ∆xj, with x1 = 0. For
each particle trajectory, the process is carried out sepa-
rately for two spatial dimension and are then combined
to form the true two-dimensional (2D) positions of the
synthetic particle trajectory.
the displacements xi = x0 +(cid:80)i
To incorporate transitions between diffusive states
within each trajectory, we first generated a random
Markov chain, with a known transition matrix, A, to
specify the state sequence of each particle track displace-
ments. For each state, the displacements are simulated
according to the properties of the diffusive state. The
particle trajectories are then reconstructed their posi-
tions by calculating the cumulative sum of the displace-
j=1 ∆xj, with x1 = 0. Each time
the Markov state goes to a confined diffusion state, the
confinement boundaries are reset with the initial posi-
tion at the center. When the Markov state switches to
another diffusive state, information of the confinement
boundaries is forgotten.
ments xi = x0 +(cid:80)i
Dynamic localization noise is incorporated into the po-
sitions by simulating 32 micro-step displacements (δt =
∆t/32) time steps and averaging 32 successive positions.
The net effect is an exposure time equal to the frame du-
ration of 32 ms. Static localization noise is included by
adding a normally distributed random number with zero
mean and variance, σ2
sim, to each motion-blurred posi-
tion. To generate a collection of particle trajectories, the
population fractions are used to determine the number of
particle trajectories that are initialized to each diffusive
state. Here, population fractions serve as the percentage
that the initial state of each trajectory begins with.
B. MSD analysis
For a stationary sequence of T 2D particle positions,
x = {x(t), y(t)} for t = 1 through T , each separated
one from the next by a time, ∆t, the taMSD is given
according to [4, 27]:
T−∆n(cid:88)
δ(∆n, T ) =
1
(x(t + ∆n) − x(t))2
T − ∆n
+ (y(t + ∆n) − y(t))2
t=1
where δ(∆n, T ) is the taMSD for the nth time lag, ∆n =
n∆t and the bar on top of δ(∆n, T ) is used to distinguish
the time average.
We generate the taMSD for the first 14 time lags and
employ an unweighted non-linear least squares fit with
diffusion models given in Table III, where σ0 represents
the static localization noise and σ represents the com-
bined static and dynamic localization noise terms.
Mode
Normal
MSD model
0 − 2
4D∆t + 4σ2
3 D∆t
Confined L2
3 − 32L2
π4
(cid:80)∞
k=1,odd
1
k4 exp
(cid:104)−(cid:0) kπ
L
(cid:1)2
(cid:105)
Dn∆t
+ 4σ2
fBm
4D∆tα + 4σ2
TABLE III. MSD models for canonical modes of diffusion.
C. MLE analysis
For a given diffusion model, the maximum likelihood,
or equivalently the minimum negative log-likelihood, is
found by employing a constraint, gradient-based, numer-
ical optimization algorithm in MATLAB (Mathworks),
namely fmincon. At each optimization step, however,
the log-likelihood function requires the calculation of the
log-determinant and the inverse of the covariance matrix.
When particle tracks are long or the elements of the co-
variance matrix are very small, the log-determinant of
the covariance matrix can run into numerical underflow
issues. To make this optimization procedure more robust,
we employ an eigenvalue decomposition of the covariance
matrix, Σ = P ΛP T , where P is a matrix of the eigen-
vectors with their corresponding eigenvalues given along
the diagonal of Λ. The log-determinant is given by the
product of the eigenvalues or equivalently the sum of the
log eigenvalues, i.e.
ln λi.
The inverse is given by Σ−1 = P Λ−1P T .
ln det(Σ) = ln(cid:81)D
i λi = (cid:80)D
i
In summary, for a given single particle trajectory, max-
imum likelihood estimation yields the parameter esti-
mates, log-likelihood value, and Hessian for each can-
didate diffusion model. From this information, the BIC
(Eq. 6) can be calculated for each diffusion model. Once
the BIC for each diffusion model has been calculated, the
model probability for each diffusion model can be calcu-
lated according to Eq. 7. Classification is determined
by the diffusion model which yields the highest model
probability.
15
pEMv2 procedure employs the EM algorithm on the orig-
inal set of particle trajectories with random initial pa-
rameter values. pEMv2 then reemploys the EM on a
Monte Carlo bootstrap set of the original particle tra-
jectories, which serves to perturb the likelihood surface
with the aim that a local maximum may no longer be
a maximum in the perturbed likelihood surface. Upon
completion of a perturbation trial, we verify whether a
higher likelihood has truly been found by calculating the
likelihood using the pEMv2-converged parameters with
the original dataset.
If the pEMv2-converged parame-
ters indeed yield a higher likelihood, then the EM pa-
rameters are updated by reemploying the EM algorithm
initialized with the new pEMv2 parameter estimates on
the original dataset. Otherwise, the pEMv2 estimates
remain unchanged. This process is repeated until a pre-
determined number of perturbations have been executed
and yield no advance.
k}K
ity distribution, namely to(cid:80)k−1
To generate each set of random initial values, K ran-
dom numbers between 0 and 1 are drawn from a uniform
distribution. The initial population fractions, {π0
k=1,
are given by normalizing these random numbers so that
the sum is equal to 1. The first covariance values are
set using the initial randomly-chosen population frac-
tions and the empirical cumulative covariance distribu-
tion function. By dividing the cumulative distribution
function into K regions proportional to the initial pop-
ulation fractions, the initial covariance value of diffusive
state k is then picked as the diffusivity corresponding
to the midpoint of region k of the cumulative probabil-
2 . Particle tracks
are then classified to each diffusive state by their distance
to the first covariance values. The remaining covariance
values for each diffusive state is selected by averaging the
classified covariance terms. Thus, we achieve an initial-
ization that serves as a non-parametric method to ran-
domly sample from the observed distribution of diffusion
coefficients. We found this method produces better ran-
dom initializations than a k-means clustering over the
whole covariance matrix. In practice, we found k-means
clustering converges to similar values over a wide range
of parameter space. In addition, k-means tends to weed
out diffusive states with low population fractions, when
two diffusive states are close in proximity.
j + π0
j=1 π0
k
To improve pEMv2's performance, we applied 5 ran-
dom initialization trials of the EM and used the parame-
ters of the trial which yielded the highest likelihood value.
We then applied 100 perturbation trials. This was done
for each diffusive state starting from K = 1 and incre-
menting K till pEMv2 finds a lower BIC value. Upon
completion of pEMv2, the returned parameters include
the population fractions and covariance matrices of each
diffusive state, along with the posterior probabilities of
each particle trajectory.
D. pEMv2 analysis
pEMv2
written
pEMv2 analysis was performed with the Matlab script
provided in the Supplemental Materials. Briefly, our
is
and
in MATLAB
freely
available
(Math-
at
works)
https://GitHub.com/MochrieLab/pEMv2.
is
Appendix A: Covariance matrix for particle track
displacements without localization noise
ACKNOWLEDGMENTS
This work was supported by NSF PHY 1305509, and
by the Raymond and Beverly Sackler Institute for Phys-
ical and Engineering Biology.
16
[1] S. C. Weber, M. A. Thompson, W. Moerner, A. J.
Spakowitz, and J. A. Theriot, Biophysical journal 102,
2443 (2012).
[2] M. T. Valentine, P. D. Kaplan, D. Thota, J. C. Crocker,
T. Gisler, R. K. Prud'homme, M. Beck, and D. A. Weitz,
Physical Review E 64, 061506 (2001).
[3] B. R. Parry, I. V. Surovtsev, M. T. Cabeen, C. S. O'Hern,
E. R. Dufresne, and C. Jacobs-Wagner, Cell 156, 183
(2014).
[4] A. Kusumi, Y. Sako, and M. Yamamoto, Biophysical
journal 65, 2021 (1993).
[5] O. Rossier, V. Octeau, J.-B. Sibarita, C. Leduc,
B. Tessier, D. Nair, V. Gatterdam, O. Destaing, C. Al-
big`es-Rizo, R. Tamp´e, et al., Nature cell biology 14, 1057
(2012).
[14] P. J. Slator, C. W. Cairo, and N. J. Burroughs, PloS one
10, e0140759 (2015).
[15] J. Apgar, Y. Tseng, E. Fedorov, M. B. Herwig, S. C.
and D. Wirtz, Biophysical Journal 79, 1095
Almo,
(2000).
[16] Y. Tseng, T. P. Kole, and D. Wirtz, Biophysical journal
83, 3162 (2002).
[17] P. K. Koo, M. Weitzman, C. R. Sabanaygam, K. L. van
Golen, and S. G. J. Mochrie, PLoS Comput Biol 11,
e1004297 (2015).
[18] M. J. Saxton and K. Jacobson, Annual review of bio-
physics and biomolecular structure 26, 373 (1997).
[19] A. J. Berglund, Physical Review E 82, 011917 (2010).
[20] X. Michalet, Physical Review E 82, 041914 (2010).
[21] C. L. Vestergaard, P. C. Blainey, and H. Flyvbjerg, Phys-
[6] F. Persson, M. Lind´en, C. Unoson, and J. Elf, Nature
ical Review E 89, 022726 (2014).
methods (2013).
[22] M. P. Backlund, R. Joyner, and W. Moerner, Physical
[7] M. J. Saxton, Biophysical journal 72, 1744 (1997).
[8] T. Savin and P. S. Doyle, Biophysical journal 88, 623
Review E 91, 062716 (2015).
[23] D. T. Gillespie, American Journal of Physics 64, 225
(2005).
(1996).
[9] N. Monnier, Z. Barry, H. Y. Park, K.-C. Su, Z. Katz,
B. P. English, A. Dey, K. Pan, I. M. Cheeseman, R. H.
Singer, et al., Nature methods (2015).
[24] C. M. Bishop and N. M. Nasrabadi, Pattern recognition
and machine learning, Vol. 1 (Springer New York, 2006).
[25] K. P. Murphy, Machine learning: a probabilistic perspec-
[10] M. Ott, Y. Shai, and G. Haran, The Journal of Physical
tive (The MIT Press, 2012).
Chemistry B 117, 13308 (2013).
[11] R. Das, C. W. Cairo, and D. Coombs, PLoS computa-
tional biology 5, e1000556 (2009).
[26] A. P. Dempster, N. M. Laird, and D. B. Rubin, Journal
of the Royal Statistical Society. Series B (Methodologi-
cal) , 1 (1977).
[12] H. Yang, The Journal of chemical physics 129, 074701
[27] H. Qian, M. P. Sheetz, and E. L. Elson, Biophysical
(2008).
journal 60, 910 (1991).
[13] M. H. Duits, Y. Li, S. A. Vanapalli, and F. Mugele,
Physical Review E 79, 051910 (2009).
TABLE IV. Analytical covariance matrix of particle track displacements separated in time by ∆t for canonical diffusion modes,
namely normal diffusion, confined diffusion, and fractional Brownian motion, with D diffusion coefficient, L confinement size,
and α anomalous exponent. The hat, Σ, represents the covariance matrix without localization error corrections.
17
Covariance matrix (µm2s−1)
Σnormal(i, j) = 2D∆tδi,j
Mode
Normal
Confined Σ(i, j)conf ined =
L2
8
π4
6 − 16L2
−L2
12 + 8L2
(cid:80)∞
π4
π4
(cid:80)∞
(cid:80)∞
k=1,odd
k=1,odd
1
k4 (−2Φ(j − i + 1) + Φ(j − i) + Φ(j − i + 2))
k=1,odd
1
k4 Φ(1)
k4 Φ(1) (2 − Φ(1))
1
(cid:104)−(cid:0) kπ
L
(cid:1)2
(cid:105)
where Φn = exp
Dn∆t
.
, j = i
, j = i ± 1
, otherwise
fBM
Σf BM (i, j) = D∆tα(j − i + 1α + j − i − 1α − 2j − iα)
|
1312.6895 | 1 | 1312 | 2013-12-24T21:50:34 | Optimal shapes and stresses of adherent cells on patterned substrates | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.CB"
] | We investigate a continuum mechanical model for an adherent cell on two dimensional adhesive micropatterned substrates. The cell is modeled as an isotropic and homogeneous elastic material subject to uniform internal contractile stresses. The build-up of tension from cortical actin bundles at the cell periphery is incorporated by introducing an energy cost for bending of the cell boundary, resulting to a resistance to changes in local curvature. Integrin-based adhesions are modeled as harmonic springs, that pin the cell to adhesive patches of a predefined geometry. Using Monte Carlo simulations and analytical techniques we investigate the competing effects of bulk contractility and cortical bending rigidity in regulating cell shapes on non-adherent regions. We show that the crossover from convex to concave cell edges is controlled by the interplay between contractile stresses and boundary bending rigidity. In particular, the cell boundary becomes concave beyond a critical value of the contractile stress that is proportional to the cortical bending rigidity. Furthermore, the intracellular stresses are found largely concentrated at the concave edge of the cell. The model can be used to generate a cell-shape phase diagram for each specific adhesion geometry. | physics.bio-ph | physics | Optimal shapes and stresses of adherent cells on patterned substrates
Shiladitya Banerjee,1 Rastko Sknepnek,2, 3 and M. Cristina Marchetti2
1James Franck Institute, The University of Chicago, Chicago, Illinois 60637, USA
2Department of Physics and Syracuse Biomaterials Institute,
Syracuse University, Syracuse, New York 13244, USA
3School of Engineering, Physics, and Mathematics, University of Dundee, Dundee DD1 4HN, UK
We investigate a continuum mechanical model for an adherent cell on two dimensional adhesive
micropatterned substrates. The cell is modeled as an isotropic and homogeneous elastic material
subject to uniform internal contractile stresses. The build-up of tension from cortical actin bundles
at the cell periphery is incorporated by introducing an energy cost for bending of the cell boundary,
resulting to a resistance to changes in local curvature.
Integrin-based adhesions are modeled as
harmonic springs, that pin the cell to adhesive patches of a predefined geometry. Using Monte
Carlo simulations and analytical techniques we investigate the competing effects of bulk contractility
and cortical bending rigidity in regulating cell shapes on non-adherent regions. We show that the
crossover from convex to concave cell edges is controlled by the interplay between contractile stresses
and boundary bending rigidity. In particular, the cell boundary becomes concave beyond a critical
value of the contractile stress that is proportional to the cortical bending rigidity. Furthermore, the
intracellular stresses are found largely concentrated at the concave edge of the cell. The model can
be used to generate a cell-shape phase diagram for each specific adhesion geometry.
I.
INTRODUCTION
Living cells actively probe physical cues in their envi-
ronment via receptor-ligand adhesion complexes that link
the actomyosin cytoskeleton to the extracellular matrix
(ECM) [1]. The cellular microenvironment, comprising
of the ECM and of neighboring cells, imposes specific
boundary conditions that can regulate physiological pro-
cesses such as cell differentiation, division and motility,
as well as cell architecture and polarity [2]. Myosin mo-
tors generate contractile stresses in the actin cytoskeleton
that are transmitted to the substrate by focal adhesions.
The traction stresses exerted by the cells on the substrate
are thus very sensitive to the stiffness of the substrate as
well as to the adhesion geometry. Cell morphology in
turn is directly affected by traction stresses through the
tension that builds up in the actomyosin stress fibers. It
has been shown that the substrate stiffness plays a crucial
role in regulating the cell spread area, the magnitude of
traction forces and the cell morphology [3 -- 6]. Much less
explored is the role of adhesion geometry in regulating
the spatial distribution of cellular stresses. Micropattern-
ing has emerged as a powerful tool to investigate the in-
terplay of mechanics and cytoskeletal architecture in con-
trolling cell morphology by specific tuning of the geome-
try of the adhesion sites [7]. When plated on small mi-
cropatterns, cells are unable to grow, thus showing high
apoptotic rate [8]. Large adhesive patches, in contrast,
favor cell spreading and promote the assembly of con-
tractile stress fibers along the cell's perimeter [9]. These
peripheral stress fibers interconnect focal adhesions and
yield concave arcs of constant curvature in the nonadher-
ent portions of the cell boundaries. In addition, traction
forces tend to localize in regions of high curvature at the
boundary [10, 11]. The model proposed here allows to
separately study the roles of cell contractility and me-
chanical properties of peripheral cell fibers in controlling
cell shape. Future comparison with experiments where
both quantities can be perturbed using pharmacological
interventions [9, 12] may provide a quantitative under-
standing of the relative importance of boundary and bulk
properties in determining steady state cell shapes.
Various successful theoretical models of single and
multi-cell mechanics have been proposed over the past
decade that address the role of ECM elasticity in regu-
lating cell behavior [13]. Previous work has addressed
the interplay between cell mechanics and geometry by
either focusing solely on the elasticity of the cell bound-
ary [14, 15] or by considering only the bulk of the cell,
described via continuum mechanics [16 -- 19], by a cellular
Potts model [20], or as a polymer network [21]. These
models highlight the competing roles of cell contractil-
ity and substrate stiffness in regulating polymorphic cell
shapes.
Continuum models of cell mechanics have assumed
that the material constants describing the cell are spa-
tially homogeneous. Cell material properties are, how-
ever, highly heterogeneous.
In particular, experiments
have shown strong differences in the mechanical proper-
ties of the bulk and boundary regions of the cell [22]. In-
creased tension and rigidity of cell boundaries can spon-
taneously arise during adhesion as a result of the assem-
bly of peripheral stress fibers consisting of thin bundles
of semiflexible actin filaments. Due to thermal and ac-
tive forces these bundles considerably bend generating
non-uniform peripheral tensions. Cell boundary can also
resist changes in local curvature due to contact forces at
the three-phase contact line between the cell, the sub-
strate and the ambient medium. Motivated by these
observations, in this paper we couple cell contour elas-
ticity [14, 15] to a continuum description of bulk cell me-
chanics [16, 17] to investigate the cooperative roles of
cortical elasticity, bulk elasticity and active contractility
in controlling cell shapes on non-uniform adhesion pat-
3
1
0
2
c
e
D
4
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
9
8
6
.
2
1
3
1
:
v
i
X
r
a
II. CONTINUUM MECHANICAL MODEL
2
(cid:90)
We consider the mechanical equilibrium of a station-
ary cell strongly adherent to a soft elastic substrate. We
assume that the cell's interior can be described as an
isotropic homogeneous elastic material and neglect all
dissipation. We further neglect all out-of-plane deforma-
tions of the cell and assume that its thickness is uniform
throughout its entire area and remains unaffected by the
substrate-induced deformations. The bulk elastic energy
of our model cell is given by
(cid:18) ν
(cid:19)
Eel =
dA
Eh
2 (1 + ν)
A0
1 − ν
u2
γγ + u2
αβ
(1)
(α, β, γ ∈ {x, y})
with E the three-dimensional Young's modulus, h
the cell thickness, ν the Poisson's ratio, and uαβ =
1
2 (∂αuβ + ∂βuα + ∂αuγ∂βuγ)
the
strain tensor. We retain the nonlinearity in strain ten-
sor to allow for the possibility of large strains that can
arise even for small displacements [23]. The nonlinear
terms essentially describe strain stiffening which is indeed
expected in crosslinked actin networks [24]. The two-
dimensional displacement vector (cid:126)u is defined as (cid:126)u ((cid:126)r0) =
(cid:126)r − (cid:126)r0, where (cid:126)r0((cid:126)r) is a material point before (after) the
deformation. The integral is calculated over the area A0
of the undeformed (reference) state and summation over
pairs of repeated indices is assumed. Cell contractility
arising from myosin motors is modeled as a homogeneous
negative pressure resulting in an additional contribution
to the cell's energy, given by
(cid:90)
A0
Eactive = σa
dA uγγ ,
(2)
where σa > 0 is a parameter controlling active contrac-
tility, determine by concentration of myosin motors and
rate of ATP consumption. At the continuum scale, it con-
trols active contributions to the cell's surface tension [25].
Cell adhesion to the substrate is modeled via a harmonic
potential with a position-dependent rigidity parameter
Γ((cid:126)r)
Eadh =
1
2
dA Γ ((cid:126)r)(cid:126)r − (cid:126)ra2 ,
(3)
(cid:90)
with (cid:126)ra the position of focal adhesions on the substrate.
The rigidity parameter Γ is nonzero over the adhesion
region and zero elsewhere. Thus we allow for non-
uniformity in the geometry of cell-substrate adhesions
as can be realized experimentally using micropatterning
techniques [9]. The assumption of local adhesive inter-
actions with the underlying substrate strictly holds for
elastic substrates that are much thinner than the cell
perimeter or on soft microposts [26]. The rigidity param-
eter Γ depends on the elastic modulus of the underlying
substrate as well as on the stiffness kf of focal adhesions.
For an elastic substrate of shear modulus µs and thick-
ness hs, with focal adhesion density ρf , Γ is given by,
−1. Traction force density is
Γ−1 = (kf ρf )
−1 + (µs/hs)
FIG. 1. Shape phase diagram for a cell adhering to a cross
shaped micropattern (grey region of the cell images) as a func-
tion of the bending rigidity κ of the cell boundary in units of
Y (cid:96)3 and the bulk contractile stress σa in units of Y , where Y
is the cellular Young's modulus and the (cid:96) is the interparticle
distance in the triangulation. The blue dashed line is a guide
to the eye.
terns. Non-uniform tension and elasticity is incorporated
in the model by introducing a penalty for bending defor-
mations of the cell periphery [15]. Using a combination of
Monte Carlo simulation and analytical studies, we exam-
ine the interplay of bulk contractility and cortical tension
in controlling morphological transitions in adherent cells
and propose a cell shape phase diagram for specific ad-
hesion geometries. An example of such a phase diagram
for a cross-shaped adhesion pattern is shown in Fig. 1.
See section III A for details.
The paper is organized as follows.
In section II, we
describe a continuum mechanical model for a thin adher-
ent cell as an isotropic and homogeneous elastic material,
subject to a homogeneous negative pressure, embodying
active contractility. Cell-ECM adhesions are modeled as
linear springs distributed non-uniformly along the cell-
substrate interface. Cortical tension is described via a
penalty for bending deformations of the cell periphery.
In Section III we discuss the steady shapes of cells ad-
herent to different concave micropatterns obtained via
Monte Carlo simulations (numerical details are given in
the Appendix). Our simulations suggest a transition be-
tween convex and concave morphologies as a function of
cell contractility and bending rigidity, that is captured
by an analytically solvable model for the cell boundary
presented in Section IV. We conclude with a brief discus-
sion.
00.040.080.120.160.240.320.400.600.800246810sa/Yk/Yl3convexk=0.6Yl3sa=2Yconcavek=0.04Yl3sa=8Y 3
are measured in units of (cid:96) and all energies are measured
in units of Y (cid:96)2, where Y = Eh is the two-dimensional
Young's modulus. The substrate rigidity Γ has units of
Y (cid:96)2 and the bending rigidity of cortical stress fibers has
units of Y (cid:96)3. The low energy configurations are obtained
using simulated annealing Monte Carlo (see Appendix
for details).
A. Optimal shapes
We performed a series of simulations for V , U , and
cross shaped micro-patterns, corresponding to the white
dashed outlines shown in Fig. 3. The relaxed shapes for
three non-convex patterns (U , V , and cross) obtained for
a fixed value of the rigidity of adhesions Γgrey = 106,
and ν = 1/3 are shown in Fig. 3.
In all cases, the
non-adherent cell edges spanning two pinning regions are
clearly concave. The relaxed shapes can be qualitatively
compared with experiments on concave micropatterns [9].
Our simulations suggest that there is a transition between
the concave and convex morphologies as a function of σa
and κ. In Figure 1, we show a sample cell shape phase
diagram as a function of the cortical bending rigidity κ
and the active contractile stress σa for the cross shaped
micropattern. The figure indicates that the cell bound-
ary is concave at high values of the contractile stress σa,
whereas convexity is ensured at high values of bending
rigidity. The phase boundary between convex and con-
cave shapes appears to be linear in the σa − κ plane.
To justify this observation, in the next section we study
analytically the shape of the cell boundary in the non-
adhesive regions, considering small deformation about a
circular configuration.
B. Optimal Stresses
Experiments probing the distribution of traction force
density (cid:126)T ((cid:126)r) exerted by cells adhering to soft substrates
consistently show that such stresses are concentrated at
the cell edges, and strongest in region of high cell cur-
vature. Force balance requires Tα = ∂βσαβ, where σαβ
is the two-dimensional stress tensor of the bulk cellular
material, given by,
∂ (Eel + Eactive)
σαβ =
∂uαβ
(cid:18) ν
= σel
αβ + σaδαβ
(cid:19)
=
Eh
(1 + ν)
1 − ν
δαβuγγ + uαβ
+ σaδαβ . (5)
The distribution of such internal stresses can therefore be
inferred experimentally from traction force microscopy
measurements [28]. Internal stresses of adhering cells are
found to be concentrated at the cell's interior, with a
maximum value proportional to the active cell contrac-
tility, here σa. To highlight the role of patterned adhesion
on the spatial distribution of cellular stress, we display
(a) Initial configuration of the "V-shape" adhe-
FIG. 2.
sion pattern. The adhesive region where the cell is strongly
anchored to the V-shaped micro-pattern on the substrate
(Γ (cid:54)= 0) is indicated in grey, whereas the non-adherent por-
tion of the cell (Γ = 0) is indicated in yellow. (b) Zoom-in of
the upper right corner, showing the triangulation.
therefore given by, (cid:126)T = 1
Finally, we assign a bending penalty to the cell's perime-
ter, reflecting the resistance of cortical actin bundles to
changes in curvature,
h δEadh/δ(cid:126)u = 1
h Γ((cid:126)r)((cid:126)r − (cid:126)ra).
(cid:73)
Ebend = κ
ds c2,
(4)
where κ is the bending rigidity, c = γ(cid:48)(cid:48) (s) is the cur-
vature of the boundary, with γ (s) a parametric curve
describing the cell boundary, and the line integral is cal-
culated along the cell boundary.
The optimal shape of the cell is obtained by minimizing
the total mechanical energy E, that is given as the sum of
elastic, active, adhesion, and boundary bending energies,
E = Eel + Eactive + Eadh + Ebend.
III. NUMERICAL SIMULATIONS
The minimal energy cell shapes have been determined
numerically by a Monte Carlo study of a discrete rep-
resentation of the continuum model introduced in Sec-
tion II. The discrete representation of the undeformed
cell is a triangulated disk. The initial configuration is
built by randomly placing N ≈ 104 particles on a disk
of radius R0. Particles are assumed to interact pairwise
via a Weeks-Chandler-Andersen potential [27] and their
positions are equilibrated using a standard Monte Carlo
simulation with canonical (N V T ) Metropolis algorithm.
A typical equilibrated configuration is stored and the po-
sitions of the particles of that configuration are then used
as nodes to construct a Delaunay triangulation. The re-
sulting triangulation for a V-pattern is shown in Fig. 2.
We note that the initial density of points in the disk is
chosen such that even in the equilibrium state there is
a substantial overlap between neighbors, thus ensuring
a densely packed distribution of points. As a result the
equilibrium distribution of the interparticle distances is
rather narrow and its mean, denoted as (cid:96), represents a
suitable unit of length.
In the following, all distances
(a)(b) 4
FIG. 3. Relaxed shapes of (a) V -pattern with κ = 0.4 Y (cid:96)3, (b) U -pattern with κ = 0.08 Y (cid:96)3, and (c) cross-pattern with
κ = 0.4 Y (cid:96)3 and σa = 100Y . The white dashed lines indicate the boundary of the micropattern: the cell is anchored inside this
region and is free to contract outside. The color scale shows the distribution of the displacements with respect to the reference
circular configuration.
in Fig. 4 the spatial distribution of the so-called Lam´e's
stress ellipses [29] for the elastic part σel
αβ of the stress
tensor. The constant active contribution σa has been
subtracted out to highlight spatial variations. As a re-
sult, the displayed stress is largest at the cell edges. The
Lam´e's stress ellipses are obtained by computing the elas-
tic part of the two-dimensional stress tensor at a repre-
sentative subset of triangles in the Delaunay grid. This is
achieved by directly evaluating the expression in eqn (5),
excluding the active term. We then compute the low
and the high eigenvalues, σmin and σmax, respectively,
of the elastic stress tensor of a given triangle. Note that
since the stress tensor is symmetric, its eigenvalues are
always real. The length of the major and minor semi-
axes of each ellipse are then given by σmax and σmin,
respectively, whereas the orientation of the ellipse axes is
given by the directions of the corresponding eigenvectors.
As expected, and consistent with experiments [7], elastic
stresses are concentrated at the free boundaries of the
adherent cell. Boundary stresses along free edges con-
necting two adhesion points are directed normal to the
edge whereas they are oriented along the edge near the
adhesion points. This is most evident in the cross-shaped
pattern, Fig. 4(c). The large stresses in the convex re-
gions of the cell spilling outside straight portions of the
pinning regions (see, for instance, the V-shape pattern,
Fig. 4(a)) are largely an artifact of our model. They arise
because we have introduced excluded volume interactions
to prevent self-intersections of the triangulation. In other
words, we assign a hard core radius of 0.25(cid:96) to each ver-
tex of the triangulation, such that no two vertices can
came closer than 0.5(cid:96) from each other. Once this limit
has been reached, the excluded volume prevents further
collapse of the cell, thus accounting for the presence of
a sizeable portion of the cell that extends outside the
pinning region. While steric effects are present in vivo
and may describe for instance the role of structural el-
ements capable of carrying compressive loads, such as
microtubules, cells on synthetic substrates generally al-
most completely conform to the micropattern by chang-
ing their thickness. This is not possible in our strictly
two-dimensional model. As a result, the model captures
well the behavior of "free" cell edges spanning two adhe-
sion points, but has limitations for describing the behav-
ior of cell boundaries along straight pinning regions.
IV. BOUNDARY SHAPES
A. Strong pinning at adhesions
For small deformations about an initially circular con-
figuration of radius R0, the cell boundary in the non-
adhesive region can be parametrized using polar coordi-
nates as, r(θ) = R0 + ur(θ), where ur is the radial com-
ponent of the displacement field at the cell boundary,
(cid:126)u = (ur, uθ). Thus, (cid:126)u is solely a function of the angu-
lar coordinate θ.
In mechanical equilibrium, boundary
force balance along the normal and tangential directions
requires
2κ
d2c
ds2 − σijninj = 0 ,
σijtinj = 0 ,
(6a)
(6b)
where s is the arc-length parameter and (cid:126)n and (cid:126)t are unit
vectors normal and tangent, respectively, to the unper-
turbed cell boundary. Tangential force-balance in polar
coordinates reduces to σrθ = 0, which leads to the re-
lation uθ = ∂θur. Thus, the normal component of the
elastic stress is given by σel
θ ur/R0,
where λ = Y ν/(1 − ν2) is the Lam´e elastic constant.
rr (cid:39) λ∂θuθ/R0 = λ∂2
012l(a)(b)(c) 5
(a)
(b)
(c)
FIG. 4. Stress profile (Lam´e's ellipses, shown in red) at representative sets of points of contracted cells for (a) V -pattern with
κ = 0.16 Y (cid:96)3, (b) U -pattern with κ = 0.24 Y (cid:96)3, and (c) cross pattern with κ = 0.32 Y (cid:96)3; σa = 10Y for all patterns. Note
that only the elastic part of the stress tensor is shown. The active component, σa, has been removed, since it is isotropic and
much larger in magnitude than the elastic component of the stress tensor. Length of the ellipse axes is proportional to the two
eigenvalues of the stress tensor, σmin and σmax. The stress is not uniform but largest around cell's perimeter and gradually
falls off toward its interior. For clarity, the Lam´e's ellipses are computed only for a subset of all triangles selected from the
non-regular triangulation and their sizes are scaled by a factor of 0.5.
FIG. 5. Curvature, deformation and phase boundary for a pinned contractile string. (a) Radial displacement ur in units of R0
and (b) curvature profile c(θ) in units of 1/R0 for σaR3
0/κ = 40
(dotted, red), where 0 < θ < π/2, corresponding to a cross pattern of width zero. (c) Shape phase diagram for the contractile
string pinned to a cross micropattern of width w, obtained from the solution to eqn (8). Bending rigidity κ and contractile
stress σa are given in units of Y (cid:96)3 and Y respectively, corresponding to the parameters in Fig. 1, where R0/(cid:96) = 50, w/(cid:96) = 20,
and ν = 1/3.
0/κ = 10 (dashed, blue) and σaR3
0/κ = 1 (solid, black), σaR3
Furthermore, for small deformations ur, the boundary
curvature can be expanded as,
c(θ) (cid:39) 1
R0
− 1
R2
0
(ur + ∂2
θ ur) + O(u2
r) .
(7)
Using eqns (6a)-(7), and letting ds = R0dθ, we obtain an
equation for the boundary profile,
(cid:0)∂2
2κ
R3
0
θ u(cid:1) + σa + λ∂2
θ u + ∂4
θ u = 0 ,
(8)
where u = ur/R0. Without loss of generality, we can
consider solution in the interval θ ∈ [0, φ], where φ is the
angular width of the nonadherent region and depends
on the geometry of the adhesion pattern. The boundary
conditions for the case of strong pinning at adhesions
are given by: u(0) = u(φ) = ∂θ u(0) = ∂θ u(φ) = 0.
The full solution of eqn (8) is analytically tractable but
cumbersome. We instead discuss the solutions in two
limiting cases in terms of the dimensionless parameter
K = 2κ/λR3
0, reflecting the relative contributions of
bending and bulk elasticity. This parameter can also
be written as K = (ξ/R0)3 in terms of the ratio of a
length scale ξ = [2κ/λ]1/3 to the undeformed cell radius
R0. The length scale ξ described the interplay between
bulk elasticity and boundary tension in controlling the
response of the cell. When ξ (cid:29) R0 (corresponding to
K (cid:29) 1) the cel deformation is controlled by the cortical
tension at the boundary and the curvature is given by
c(θ) (cid:39) 1
R0
2 + θ2 − θφ − φ cot
σaR3
0
4κ
(cid:19)(cid:21)
(cid:18)
.
(9)
φ
2
(cid:20)
1 +
Bulk elasticity drops out and the behavior is controlled
by the ratio σaR3
0/κ of contractility to bending rigid-
ity. The curvature has a minimum at the center of the
nonadherent segment. Thus, as one increases contrac-
0.00.51.01.5\x{FFFF}0.4\x{FFFF}0.3\x{FFFF}0.2\x{FFFF}0.10.0Θu\x{FFFF}Θ\x{FFFF}\x{FFFF}R00.00.51.01.5\x{FFFF}20246Θc\x{FFFF}Θ\x{FFFF}R0(a)(b)(c)ConvexConcave0123450.00.20.40.60.8ΣaΚ tility σa, a region of negative curvature develops near
θ = φ/2, which grows upon increasing σa until convexity
is retained within a small neighborhood of the adhesion
patch. The onset of concavity is thus given by the con-
dition of reality to the solution of c(θ) = 0, which gives
the condition,
(cid:32)
(cid:33)
σa >
2κ
R3
0
1
φ2/8 + (φ/2) cot φ
2 − 1
.
(10)
Since concave shapes are commonly observed in experi-
ments [9], we now turn to estimate the critical value of σa
as predicted by our model in order to compare it with ex-
perimentally reported values for σa. The bending rigidity
of cortical stress fibers can be estimated as κ ∼ π
4 Eactr4
s,
where Eact is the Young's modulus of actin and rs is
the typical radius of the stress fibers. Using Eact (cid:39) 2.6
GPa [30] and rs ∼ 0.1 µm, we get κ ∼ 2.0 × 10−19 Nm2.
Using this value in eqn (10) for φ ∼ 2π/3, corresponding
to a thin V-pattern, we get value for the critical σa ∼2.7
nN/µm. This is indeed the order of magnitude value for
active stress or surface tension reported in experiments
for adherent epithelial cells on continuous elastic sub-
strates or endothelial cells on microposts [14, 25]. In the
opposite limit of K (cid:28) 1 the deformation is controlled by
bulk elasticity and the curvature is given by
(cid:104)
(cid:0)2 + θ2 − θφ(cid:1)(cid:105)
c(θ) (cid:39) 1
R0
1 +
σa
2λ
(cid:16)
(11)
(cid:17)
1
.
The condition of concavity is given by σa > λ
In the case when ξ is comparable to R0, a simpler solution
of eqn (8) can be obtained by neglecting the fourth order
gradient term and also the derivative boundary condi-
tions. The crossover to concave profiles can be approx-
imated by the following interpolating form between the
two limiting cases,
φ2/8−1
(cid:18)
(cid:19)(cid:34)
σa >
λ +
2κ
R3
0
1
φ2/8 + (φ/2) cot φ
2 − 1
.
(12)
(cid:35)
In the general case of eqn (8), the solution for curva-
ture and the radial displacement is given in Fig. 5a,b for
three different values of σaR3
0/κ that compares the rel-
ative strengths of contractility to bending deformations.
Furthermore, to compare numerically with the simulation
results for the shape phase diagram of the adherent cell,
we show the concave-convex phase boundary in σa − κ
plane in Fig. 5c, for a cross-shaped micropattern using
the same parameters as used in Fig. 1. The resultant
phase diagram is in good order-of-magnitude agreement
with the simulation results, and the discrepancy in nu-
merical values possibly arise from neglecting non-local
bulk elasticity in the theoretical analysis.
6
B. Soft pinning
We now consider the case of soft pinning, where the
free cell boundary is anchored to soft springs at the adhe-
sion sites. Equation (8) is now solved with the boundary
conditions u(0) = u(φ) = δ and ∂θ u(0) = ∂θ u(φ) = 0,
where we have introduced an unknown displacement δ of
the ends of the segment, which can be self consistently de-
termined by minimizing the total energy of the deformed
configuration with respect to δ. For simplicity we ignore
bulk elasticity and consider the limit K (cid:29) 1. The to-
tal energy of the deformed configuration is then given
0 dθc(θ)2 + ksδ2, where ks = ΓAf and Af
by U = κR0
is the cross-sectional area of focal adhesions. Note that
the contribution due to contractility vanishes in the fi-
nal energy due to the derivative boundary conditions on
u. The onset of concavity now depends on the substrate
stiffness ks and the condition for convex-concave transi-
tion is given by
(cid:82) φ
σa >
4κ
R3
0
κφ3/12R3
0 + ks
(cid:16)
ks
φ2/4 + φ cot φ
.
(cid:17)
2 − 2
(13)
0 (cid:29) κ, promote con-
Thus, stiffer adhesions with ksR3
cavity transition at a much higher value of contractility.
It is favorable for a cell to invaginate at the free edges
if the anchoring at adhesions is softer than the effective
bending stiffness κ/R3
0.
V. CONCLUDING REMARKS
Using a simple continuum model, coupling bulk and
contour mechanics, we investigate the equilibrium shapes
and stresses of adherent cells on substrates with various
adhesion patterns. A continuum model without contour
elasticity have been studied previously by two of us on
convex patterns [16], which was successful in capturing
distribution of traction and cellular stresses and their
dependence on substrate physical properties [26]. Here
we focus on the shape and geometry induced stresses of
non-adherent cell edges on concave micropatterns. We
demonstrate numerically and analytically that the cur-
vature of the non-adherent cell boundary can undergo
a shape transition from convex to concave morphology,
controlled by the interplay of contractility and bending
rigidity. Stiff boundaries with low contractility relax to
convex shapes, whereas at higher values of contractil-
ity, non-adherent cell edges attain a concave morphology.
Previous work has shown that contractile cable network
models are capable of reproducing the invaginated cir-
cular arc morphology of cell edges connecting strongly
adhering sites [21]. Here we demonstrate that simple
continuum whole-cell models can also predict qualita-
tively cell shape and the transition between convex and
concave cell edges, provided a bending rigidity describ-
ing cortical tension is included. For parameters realistic
to experiments (see section 4.1) our model suggest that
cells prefer to invaginate at their free edges, such that
the effective boundary stiffness on non-adhesive zones are
softer than myosin induced contractile stresses. Images
of actin from experiments on concave micropatterns do
indeed show the formation of long and thin stress fibers
that are invaginated on non-adherent edges [9], indicating
a softer cortical rigidity. In addition, elastic stresses are
found to be higher along the free cell boundaries than
in the neighborhood of adhesions, since in the absence
of mechanotransduction cellular forces along free edges
are not shared by the substrate. Previous theoretical
study with only contour elasticity indicated that sub-
strate stiffness and contractility can cooperatively con-
trol cell morphology and induce hysteresis at the onset
of convex-concave transition [15]. Here we show that
even in the presence of rigid adhesions, cell shape can
be controlled by regulating the cortical bending rigidity
and contractility. Bending rigidity can be experimentally
controlled by regulating the amount of actin cross-linking
proteins that can impact stress fiber thickness and rigid-
ity, whereas myosin based contractility can be perturbed
using the conventional inhibitor Blebbistatin. One limi-
tation of our model is that it is strictly two-dimensional
and does not allow for changes in the cell thickness. Due
to the presence of steric interactions in the finite element
simulations, the cell edges on flat adhesive segments do
not fully relax to the flat morphology, but maintain a
convex shape. This is in contrast to real cells that con-
tract to adjust to the shape of the micropattern. A fully
three-dimensional model can overcome this difficulty, and
is a natural extension of our present work.
VI. APPENDIX : SIMULATION DETAILS
A. Discrete Model
The discrete version of the elastic and active contrac-
tion energies can be expressed as a sum over triangles of
the triangulation,
(cid:26) Eh
(cid:88)
Eel =
8 (1 + ν)
T
(cid:18) ν
1 − ν
Tr FT
7
(cid:17)2
(cid:17)2(cid:19)(cid:27)
(cid:16) FT
AT ,
+ Tr
AT Tr FT ,
(14)
(15)
Eactive = σa
(cid:16)
(cid:88)
T
(cid:12)(cid:12)(cid:12)(cid:126)a × (cid:126)b
(cid:12)(cid:12)(cid:12) is the area of an undeformed triangle
where matrix F = g−1 G − I, with g ( G) being discrete
metric tensor of the reference (deformed) configuration.
AT = 1
2
spanned by two vectors (cid:126)a and (cid:126)b pointing along its sides.
The sum is carried over all triangles. Adhesion energy is
discretized as
(cid:88)
i
1
2
(cid:12)(cid:12)(cid:12)(cid:126)ri − (cid:126)r(o)
i
(cid:12)(cid:12)(cid:12)2
Eadh =
Γi
Ai,
(16)
(cid:80)
i
where Γi = 106(0) for grey (yellow) vertices in Fig. 2,
(cid:126)ri((cid:126)r(0)
) is the current (reference) position of the vertex i,
and Ai = 1
AT is the area associated to the vertex
3
(i.e., a third of the sum of areas of all triangles that share
the vertex, so-called "vertex star"). Finally, following
ref. 31, the boundary bending energy is discretized as
T∈Ωi
Ebend = 4κ
1 − cos (ϑi)
si + si+1
,
(17)
(cid:88)
i
where ϑi is the exterior angle at the boundary vertex i,
si and si+1 are lengths of two boundary edges meeting
at i, and the sum is carried over all boundary vertices.
B. Monte Carlo Sweeps
A Monte Carlo sweep consist of an attempted move
for each vertex. A randomly selected vertex is displaced
by ∆(cid:126)r where components of ∆(cid:126)r are chosen at random
with an equal probability from an interval [−0.01(cid:96), 0.01(cid:96)].
Moves were accepted according to the Metropolis rules.
Minimum energy configuration is obtained using simu-
lated annealing with linear cooling protocol. Typically
minimum energy configurations were reached to a satis-
factory precision within 106 Monte Carlo sweeps.
[1] U. S. Schwarz and M. L. Gardel, Journal of cell science,
2012, 125, 3051 -- 3060.
[2] D. Discher, P. Janmey and Y. Wang, Science, 2005, 310,
1139 -- 1143.
[3] T. Yeung, P. Georges, L. Flanagan, B. Marg, M. Ortiz,
M. Funaki, N. Zahir, W. Ming, V. Weaver and P. Janmey,
Cell Motil Cytoskel, 2005, 60, 24 -- 34.
[4] C.-M. Lo, H.-B. Wang, M. Dembo and Y.-l. Wang, Bio-
[5] M. Ghibaudo, A. Saez, L. Trichet, A. Xayaphoummine,
J. Browaeys, P. Silberzan, A. Buguin and B. Ladoux, Soft
Matter, 2008, 4, 1836 -- 1843.
[6] A. Chopra, E. Tabdanov, H. Patel, P. Janmey and
J. Kresh, Am J Physiol-Heart C, 2011, 300, H1252 --
H1266.
[7] M. Th´ery, Journal of Cell Science, 2010, 123, 4201 -- 4213.
[8] C. Chen, M. Mrksich, S. Huang, G. Whitesides and D. In-
physical journal, 2000, 79, 144 -- 152.
gber, Science, 1997, 276, 1425 -- 1428.
8
[9] M. Th´ery, A. P´epin, E. Dressaire, Y. Chen and M. Bor-
[21] P. G. Torres, I. Bischofs and U. Schwarz, Physical Review
nens, Cell Motil Cytoskel, 2006, 63, 341 -- 355.
E, 2012, 85, 011913.
[10] P. Roca-Cusachs, J. Alcaraz, R. Sunyer, J. Samitier,
R. Farr´e and D. Navajas, Biophysical journal, 2008, 94,
4984 -- 4995.
[11] A. D. Rape, W.-h. Guo and Y.-l. Wang, Biomaterials,
2011, 32, 2043 -- 2051.
[12] R. Bar-Ziv, T. Tlusty, E. Moses, S. Safran and A. Ber-
shadsky, Proc Natl Acad Sci USA, 1999, 96, 10140 --
10145.
[13] U. S. Schwarz and S. A. Safran, Reviews of Modern
Physics, 2013, 85, 1327.
[14] I. Bischofs, S. Schmidt and U. Schwarz, Phys Rev Lett,
2009, 103, 48101.
[22] S. R. Heidemann and D. Wirtz, Trends in cell biology,
2004, 14, 160 -- 166.
[23] B. Audoly and Y. Pomeau, Elasticity and geometry: from
hair curls to the non-linear response of shells, Oxford
University Press Oxford, 2010.
[24] M. Gardel, J. Shin, F. MacKintosh, L. Mahadevan,
P. Matsudaira and D. Weitz, Science, 2004, 304, 1301 --
1305.
[25] A. Mertz, S. Banerjee, Y. Che, G. German, Y. Xu, C. Hy-
land, M. Marchetti, V. Horsley and E. Dufresne, Phys
Rev Lett, 2012, 108, 198101.
[26] S. Banerjee and M. C. Marchetti, Phys Rev Lett, 2012,
[15] S. Banerjee and L. Giomi, Soft Matter, 2013, 9, 5251 --
109, 108101.
5260.
[27] J. D. Weeks, D. Chandler and H. C. Andersen, The Jour-
[16] S. Banerjee and M. C. Marchetti, EPL (Europhysics Let-
nal of chemical physics, 1971, 54, 5237.
ters), 2011, 96, 28003.
[17] C. M. Edwards and U. S. Schwarz, Physical Review Let-
ters, 2011, 107, 128101.
[18] A. Pathak, V. S. Deshpande, R. M. McMeeking and A. G.
Evans, Journal of The Royal Society Interface, 2008, 5,
507 -- 524.
[19] S. Banerjee and M. C. Marchetti, New Journal of
Physics, 2013, 15, 035015.
[28] D. T. Tambe, C. C. Hardin, T. E. Angelini, K. Rajen-
dran, C. Y. Park, X. Serra-Picamal, E. H. Zhou, M. H.
Zaman, J. P. Butler, D. A. Weitz et al., Nature Materials,
2011, 10, 469 -- 475.
[29] A. E. H. Love, A Treatise on the Mathematical Theory of
Elasticity, Cambridge University Press, Cambridge, 4th
edn, 1927.
[30] F. Gittes, B. Mickey, J. Nettleton and J. Howard, The
[20] B. Vianay, J. Kafer, E. Planus, M. Block, F. Graner and
Journal of cell biology, 1993, 120, 923 -- 934.
H. Guillou, Physical review letters, 2010, 105, 128101.
[31] K. A. Brakke, Experimental mathematics, 1992, 1, 141 --
165.
|
1911.02525 | 1 | 1911 | 2019-10-30T07:06:36 | Cosmic Rays and the Chiral Puzzle of Life | [
"physics.bio-ph",
"q-bio.BM"
] | Living organisms exhibit consistent homochirality. It is argued that the specific, binary choice made is not an accident but is a consequence of parity violation in the weak interaction expressed by cosmic irradiation. The secondary muons and pairs are spin- and magnetic moment- polarized and may introduce a small, net chiral preference when they interact electromagnetically or quantum mechanically with molecules that have made the transition to self-replication. Although this preference is likely to be very small, it may suffice to give a chirally-dominant outcome after billions of replications, especially if combined with chirally-unbiased conflict between the two choices. Examples of mechanisms that can manifest the three essential steps of polarization, preference and domination are presented and some variations and possible implications are discussed. | physics.bio-ph | physics | MNRAS 000, 1 -- ?? (2002)
Preprint 7 November 2019
Compiled using MNRAS LaTEX style file v3.0
Cosmic Rays and the Chiral Puzzle of Life
No´emie Globus1,2 and Roger D. Blandford3
1Center for Cosmology & Particle Physics, New-York University, New-York, NY10003, USA
2Center for Computational Astrophysics, Flatiron Institute, Simons Foundation, New-York, NY10003, USA
3Kavli Institute for Particle Astrophysics & Cosmology (KIPAC), Stanford University, Stanford, CA 94305, USA
Released 7 November 2019
ABSTRACT
Living organisms exhibit consistent homochirality. It is argued that the specific, binary
choice made is not an accident but is a consequence of parity violation in the weak
interaction expressed by cosmic irradiation. The secondary muons and pairs are spin-
and magnetic moment- polarized and may introduce a small, net chiral preference
when they interact electromagnetically or quantum mechanically with molecules that
have made the transition to self-replication. Although this preference is likely to be
very small, it may suffice to give a chirally-dominant outcome after billions of replica-
tions, especially if combined with chirally-unbiased conflict between the two choices.
Examples of mechanisms that can manifest the three essential steps of polarization,
preference and domination are presented and some variations and possible implications
are discussed.
Key words: cosmic-rays -- astrobiology
1 INTRODUCTION
Living organisms comprise a system of molecules organized
with specific handedness. The ribonucleic and deoxyribonu-
cleic acids (RNA and DNA) are responsible for the repli-
cation and storage of genetic information. Both are made
up of linear sequences of nucleotides that encode the lin-
ear sequence of amino acids in a specific protein. DNA is a
double helix, while RNA is a single helix. The helix is an
extremely common structural feature in biomolecules; it is
a general response to the pile up of single monomer units
into a polymer. (By monomers we refer to the helix com-
ponents, i.e., the nucleotides in DNA or the amino acids
in proteins). The helix appear to be left- or right-handed,
depending on the handedness of the monomers. Biology on
Earth has made one of the two choices of opposite handed-
ness, i.e., right-handed sugars and left-handed amino-acids
that lead to predominantly right-handed helical configura-
tions in biopolymers (Biswas et al. 2016). In what follows, we
eschew the words "left" and "right" in describing structure
and polarization because they have been used differently,
inconsistently and, consequently, confusingly over the many
subfields that contribute to our discussion; instead, we de-
note the (finally) chosen set of molecules as "live" and the
alternative set as "evil". The live choices appear to be sus-
tained, once they are established, because the presence of
single, evil monomers in live helices can de-stabilize them.
This apparent choice of handedness is often called "chi-
rality" (Kelvin 1894). This word has taken on several subtly
different meanings which should be distinguished. To Kelvin,
chirality was a geometric property of a set of points in three
© 2002 The Authors
dimensional space that could not be translated and rotated
to coincide with its inversion about a point. Biological chi-
rality is different and can encompass a larger set of phenom-
ena than what concerns us here, for the example the helical
growth of some creeping plans that can be induced by the ro-
tation of the earth relative to the direction of the sun in one
hemisphere. However, most biological chirality is believed
to be derivative of the underlying chemical chirality of the
majority of biotic molecules. Here, Kelvin's points are re-
placed by atoms connected by single sigma bonds that can
allow relative rotation of large groups. Kelvin's definition
can also apply to a crystal which is a geometric arrangements
of molecules. Most importantly, it applies to the arrange-
ment of bases in a DNA molecule which is neither periodic
nor random and contains the genetic information needed to
sustain life (e.g. Schrodinger 1945; Shannon & Weaver 1949;
Watson & Crick 1953; Shinitzky et al. 2007).
These simple, helical and consistent systems presum-
ably originated around the same time as the transition
from chemical reactions between pre-biotic molecules to self-
replication and evolution of the earliest and simplest biolog-
ical molecules -- in an environment, that we call the "fount"
of life. This fount might be found on a young Earth, another
planet, a satellite, an asteroid or a comet. Since the pioneer-
ing works of Miller (1953), and Or´o (1960, 1961), many ex-
periments have demonstrated how to synthesize amino acids
and DNA bases, by irradiation of mixtures simulating our
primitive atmosphere or the interstellar medium (see Ki-
tadai & Maruyama 2018, for a review of the different exper-
iments related to the synthesis the building blocks of life).
9
1
0
2
t
c
O
0
3
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
2
5
2
0
.
1
1
9
1
:
v
i
X
r
a
2
Ice seems to be more favorable to RNA replication than a
super-cooled solution of the same temperature (Attwater et
al. 2010). The presence of phyllosillicates is considered to
be an important factor in prebiotic chemistry (Erastova et
al. 2017). Ruff (2004) showed spectral evidence for zeolite in
the Martian dust; and recently, Ruf et al. (2017) reported
the finding of new metallo-organic compounds in meteorites.
Zeolites are generally considered to be a family of materials
with highly symmetrical structures, but many of them are
chiral and capable of enantioselective recognition (Dryzun
et al. 2009). Such a selection might have helped to form the
first chiral domains of simple monomers needed to assem-
ble the first helical biomolecules. Direct evidence of complex
prebiotic chemistry is provided by the 4.5-billion-year-old
organic materials hosted in the most primitive and least
altered meteorites (Gilmour 2003). Some of these organic
molecules come in two forms that are mirror images of each
other. Those molecules are called "enantiomers", from the
Greek χθ ρ´oς, "enemy" (Meierhenrich 2008). It was found
that in some meteorites, one of the two forms is present
in greater quantity than the other (Pizzarello 2006). These
enantiomeric excesses are an important clue to the role that
meteorites could have played in the origin of life on Earth.
Chirality is pretty much inevitable in organic chemistry
because of the peculiar atomic properties of carbon atoms
with four second shell electrons that can form three of four
non-coplanar and different bonds. Typically a chiral carbon
is a carbon with four distinct atoms or groups attached to it
(van't Hoff 1874; LeBel 1874) (the carbon is sp3 hybridized).
No other atom can enter into four robust and versatile bonds
and, for this reason, atomic life forms are generally argued
to be stable. However, this allows two sets of enantiomers to
develop along separate synchronized paths making similar
evolutionary choices in response to changes in common en-
vironments.This clearly did not happen and a single choice
was made, with an entropic price that is surely paid by the
greater facility of storing information and the higher reli-
ability of the replication (c.f. Schrodinger 1945). A precise
equilibrium between the two chiral choices seems quite un-
likely, given the high replication rate, if we accept the in-
evitability of homochirality, wherever life is found, at least
so far. The essential point is that we can imagine a biology
that makes the evil choice consistently. In this alternative
biology, everything would function in the same way except
for very small effects that are the main topic of this article.
Either there was a unique spatial and temporal source for life
where a chirality choice was made, probably randomly, and
all expressions of it have subsequently adapted and dispersed
retaining the initial choice, or the live choice was made for
a reason and preserved deterministically everywhere. This
allows life to originate independently at many different sites
and epochs with the same chirality. As there could be bil-
lions of generation of the earliest and simplest life forms, a
small bias could easily ensure and sustain homochirality.
This brings up physical chirality. Pasteur (1848) dis-
covered biological chirality when he found that tartaric acid
derived from a living source rotated the plane of linearly
polarized light. This means that circularly polarized eigen-
modes propagate with slightly different speeds. By contrast,
tartaric acid derived from chemical synthesis did not behave
in this fashion. However, Pasteur was able to separate crys-
tals of synthesized tartaric acid into two groups on the basis
of their geometrical shape and crystals from one group be-
haved like the biological material and those from the other
group rotated the plane of polarization in the opposite direc-
tion. In this way, Pasteur discovered biological homochirality
and recognized it as a consequence of some asymmetry in the
laws of nature: "If the foundations of life are dissymmetric,
then because of dissymmetric cosmic forces operating at their
origin; this, I think, is one of the links between the life on
this earth and the cosmos, that is the totality of forces in the
universe" (Quack 1989, for the translation).
Had Pasteur been alive a century later, the discovery
of parity (P) violation in the weak interaction (Lee & Yang
1956; Wu et al. 1957) would have strengthened his view. A
parity operation, P, can be thought of as a reflection in a mir-
ror plane or, equivalently, inversion through a point. Parity
violation is a signature of the weak interaction, in contrast
to strong and electromagnetic interactions. Fermions must
have "left-chirality" to interact with a W boson through the
weak interaction and their antiparticles must have "right-
chirality". In the limit that a lepton is massless, the term
"chirality" coincides with helicity (the projection of the spin
vector on to their momentum vector). In the original, stan-
dard model of particle physics, the neutrino is massless. Al-
though the discovery of small neutrino masses implies that
helicity is strictly frame-dependent, this is unimportant for
biological and chemical purposes and neutrinos are effec-
tively chiral particles. By contrast, gravitational, electro-
magnetic and strong interactions are symmetric under par-
ity change and, so, a physics-based, causal explanation of
biological chirality is almost sure to involve the weak inter-
action.
Enantiomers are truly chiral, in Kelvin's sense, and their
basic Hamiltonian does not commute with the parity oper-
ator. If we add weak neutral currents to the Hamiltonian,
there is a small energy difference (the Parity Violating En-
ergy Difference, hereafter PVED), equal in magnitude but
opposite in sign between enantiomers, making one molecule
more stable than the other (Yamagata 1966). Calculation
of PVED in aqueous solution indicates a preferred stabil-
ity of left-handed amino acid and right-handed sugars, as
observed in living organisms. However, this difference is ex-
tremely small, PVED/kT ∼ 10−17. It was concluded that
a collective effect, perhaps involving spontaneous symmetry
breaking, would be needed to explain homochirality. Salam
(1991, 1992) proposed that a phase transition into the more
stable enantiomeric form occurs at a certain critical temper-
ature. This has not been validated by laboratory experiment.
In a beautiful paper, Pierre Curie addressed the ques-
tion of chirality transfer from light to molecules, specifi-
cally involving circular polarisation (Curie 1894). As noted
above, the first indication of handedness was the observa-
tion that the refractive indices for two circularly-polarized
normal modes of propagation of light could be slightly differ-
ent, resulting in a rotation of the plane of polarisation of inci-
dent linearly polarized light (Pasteur 1848). The sense of the
rotation reflects the underlying chirality of the molecules,
though the relationship is not simple and depends upon
the wavelength of the light (Optical Rotatory Dispersion).
This rotation can be accompanied by a difference in the ab-
sorption (Circular Dichroism), consistent with the Kramers
(1927); Kronig (1926) relations. On this basis, it has been
suggested that a specific source of circularly polarized light
MNRAS 000, 1 -- ?? (2002)
Cosmic Rays and the Chiral Puzzle of Life
3
Figure 1. Left: spin-polarized muons and electrons (respectively antimuons and positrons) from charged pion decay in extensive air
showers. The cosmic-ray lodacity L =< µ · v >, always negative, is a consistent chiral environmental factor. Other environmental factors,
like an external magnetic field or a fount that might oriente the molecules, would involve Me and lead to a larger chiral bias. Right: CP
invariance in the pion decay that leads to an universal sign of the cosmic-ray lodacity L < 0.
might favour one set of enantiomers over the other (Bailey
et al. 1998).
Laboratory experiments have demonstrated that it is
possible to induce an enantiomeric excess of amino acids by
irradiation of interstellar ice analogs with UV-circularly po-
larized light (UV CPL) (De Marcellus et al. 2011). However,
this raises two problems. Firstly, Circular Dichroism is also
wavelength, pH and molecule specific (D'Hendecourt et al.
2019). It is hard to see how one sense of circular polariza-
tion can enforce a consistent chiral choice, given the large
range of conditions under which the molecules are found.
Secondly, it is often supposed that astronomical sources sup-
ply the polarization. However, optical polarimetry within
the Galaxy reveals no consistent sense of circular polarisa-
tion and the observed degrees of polarization are quite small
(Bailey 2001). This can still only produce a local chirality
as most of the light sources are achiral. CPL in the near in-
frared has been observed up to 20% in star forming regions
(Kwon et al. 2013). However the level of CPL is related to
the amount of extinction and scattering, and will be less
in the UV. If we seek a universal, chiral light source, that
consistently emits one polarization over another, then we are
again drawn to the weak interaction in order to account for a
universal asymmetry. One option is to invoke spin-polarized
particles, which can radiate one sense of circular polariza-
tion through Cerenkov radiation or bremsstrahlung and can
preferentially photolyze chiral molecules of one handedness
(Vester & Ulbricht 1959; Lahoti 1977). However, the result-
ing chiral differences are again very small. This suggests
considering, instead, the direct interaction of the particles
themselves with biological molecules.
Cosmic rays are a source of polarized particles. Proba-
bly the most relevant cosmic-ray sources are the solar corona
and hydromagnetic disturbances in the solar wind. These
produce, mostly, mildly relativistic protons. The young sun
and its wind, to which early life was exposed, are likely to
have been much more active then because the sun presum-
ably rotated faster. Galactic cosmic rays mostly produced
by supernova remnants could also be important but they
MNRAS 000, 1 -- ?? (2002)
are accelerated with a much harder spectrum, which is fur-
ther hardened by modulation by the stellar wind but are
likely to be sub-dominant.
With reference to conditions on Earth, today, we are
mostly interested in charged protons, with energies just
above the threshold for pion production, colliding with ni-
trogen and oxygen nuclei in the upper atmosphere (Fig. 1).
The resulting pions, mostly π+, created through the strong
interaction, undergo weak decay within a few meters into
(positive) anti-muons with half lives ∼ 2 µs, which decay,
in turn, also weakly, into (positive) positrons. A minority of
π− create muons and electrons. Most cosmic rays lose energy
though ionizing electrons from air molecules and they and
the daughter particles are stopped in the upper atmosphere.
However, a significant number of muons do make it down
to ground level and some even penetrate deep underground,
before they decay. In this way, the atmosphere can serve as
an effective µ filter. (We emphasize that the conditions in
the young Earth and in other locales are likely to be quali-
tatively different and may have to be specialized to induce
homochirality.)
As the weak interaction is involved and pions are spin-
less, the anti-muon and positron spins, s, are preferentially
anti-aligned with their velocity, v in order to balance the an-
tiparallel spins of the accompanying antineutrinos (Fig. 1).
By contrast, the muon and electrons spins from negative
pion decay would be preferentially aligned with their veloc-
ity. The associated magnetic dipole moments are given by
µ = γs where γ = g(e/2m) is the gyromagnetic ratio. As
they have positive (negative) charge, then g ≈ 2(−2) for the
antimuons and positrons (muons and electrons). As a conse-
quence, all of these particles have magnetic moments prefer-
entially anti-aligned with their velocity. If we introduce the
operation of charge conjugation, C, this is is an expression
of CP conservation. However the magnitude of the magnetic
moment is mµ/me ∼ 200 times smaller for µ± than for e±.
Now the µ±, e± will also undergo ionization energy loss
and this will diminish the alignment. It is normal to focus on
the helicity polarization which describes the average compo-
4
nent of spin along the velocity. However, in this paper, we
are mostly concerned with the magnetic moments of these
particles and so we introduce a quantity which we call "lo-
dacity" (after lodestone) defined by
L(T) =< µ · v >,
(1)
where we average over all cosmic rays of a given type and
kinetic energy T. The magnetic moment µ includes the gy-
romagnetic ratio and therefore the lodacity has the same
sign for the charge conjugate reaction, even if the helicity
is opposite (see Fig. 1). L (cid:38) −1 for freshly created µ, and
L (cid:38) −0.3 for new e. However, when we consider the particles
that irradiate biological molecules, L may be much smaller
(though still negative) because the magnetic moment polar-
ization of the original particles will have been degraded and
there may also be a background of additional, unpolarized,
secondary particles.
Cosmic rays have a vital role. At modest intensity,
which interests us, they promote mutation and natural ex-
ploration of biochemical and evolutionary pathways; when
the intensity is high, they will be destructive and will cre-
ate sterile environments. Now, when a cosmic ray ionizes a
biological molecule, there is a small probability that it will
liberate an electron, through classical Coulomb scattering.
Other, bound transitions are possible and can be very impor-
tant (Rosenfeld 1928). However, it is simple ionization that
we will choose as a proxy for cosmic-ray induced mutation
and a demonstration of chiral preference.
At this point, we should discuss how to describe a chiral
molecule or, more generally, a chiral unit - a small portion of
a much larger molecule that can have significant interaction
with an individual cosmic ray. We suppose that it is charac-
terized by another pseudoscalar quantity, which we call the
molecular chirality M. This has opposite signs for live and
evil molecules. Now, the chiral unit comprises a small num-
ber of nuclei surrounded by electrons with mean electron
density reflecting the disposition of inner shells and chemi-
cal bonds. If we try to capture the non-spherical charge dis-
tribution in terms of a spherical harmonic expansion, either
semi-classically or quantum mechanically, then we can intro-
duce an electric dipole moment vector, d, and a (trace-free)
electric quadrupole tensor, Q. In addition, the electrons in
the chiral unit possess orbital and spin angular momentum
and this can create a magnetic dipole moment m. Treated as
operators, d anticommutes with P, while m and Q commute.
The simplest type of molecular chirality we can describe, and
which we will emphasize, is "electromagnetic chirality"
Mem = d · m.
(2)
If this is non-zero and can couple to L, then there is a chiral
preference. There is also the possibility of electric molecular
chirality, involving just d and Q. The quadrupole tensor has
three orthogonal eigenvectors, each of either sign. Alone, this
cannot be chiral. However, if there is also an electric dipole
moment lying within one of the octants then it chooses the
three eigenvectors with which it makes an acute angle and
the triple product of these three eigenvectors, taken in the
order of their size, defines a pseudoscalar, the "electric chi-
rality"Me, which can also, in principle, couple to the cosmic
rays1.
Cosmic-rays provides a natural connection between the
weak interaction and living systems. Because of the lodac-
ity which is always of the same sign, the cosmic radiation
assure small but permanent bias (reflected by the sign of
the product M · L which is always different for live and evil
molecules) that can lead to a global symmetry-breaking as
anticipated by Pasteur.
In this paper we propose that pre-biotic chemistry pro-
duces both live and evil versions of the molecular ingre-
dients of life. At some stage in the earliest development
and evolution of living, specifically copying and reproducing,
molecules a small difference in the ionization/mutation rate,
attributable to the spin polarization of the cosmic rays, gives
a chiral preference to the live molecules over their evil coun-
terparts. To be more specific, in the simplest example the
ionization rate, a scalar, contains contributions that contain
the product M ·L. Given the large number of generations of
the simplest living organisms, a small preference suffices to
lead to chiral dominance of live molecules. (Actually, as we
shall discuss, the dominance can assert itself quite quickly if
there is also a competition between live and evil molecules.)
We suspect that this mechanism can only be relevant for the
simplest organisms; more complex forms of life will involve a
larger range of more powerful evolutionary selection effects.
The plan of the paper is as follows: in Section 2, we sum-
marize the properties of the air showers and review the dif-
ferent spin-polarized components. We show the chiral propa-
gation from cosmic-rays to biomolecules with a simple model
in Section 3. Then we discuss our evolutionary scenario that
leads to homochirality in the Section 4. We conclude with
a short discussion of variations and further implications in
Section 5.
2 COSMIC RAYS
2.1 Background levels and mutagenesis
Let us now examine the general properties of cosmic ray
showers, and their biological consequences. As recalled in
section 1, cosmic rays are continuously hitting the Earth's
atmosphere inducing extensive air showers (EAS) by succes-
sive interactions with the air molecules. EAS are not only
produced in the planet's atmospheres but also in the man-
tles of comets and other icy bodies that are exposed to solar
and Galactic cosmic rays. Cosmic rays are a necessary fac-
tor in life's evolution. A background level of radiation seems
to stimulate living systems. By contrast, radiation depriva-
tion inhibits bacterial growth (Planel 1987; Kawanishi 2012;
Castillo et al. 2015). It has also been shown that during
episodes of high cosmic-ray flux and cold climate there is an
enhancement of biological productivity (Svensmark 2006).
Some fraction of environmental mutagenesis during the
evolution of life is attributable to ionization by cosmic rays.
Radiation increases the frequency of gene mutations; this
is known since the pioneering work of Muller (1927) that
showed that the mutation rate is proportional to the ra-
diation dose. DNA is the most critical biological target be-
cause of the genetic information it contains. Damage to DNA
1 "Magnetic" chirality, only involving magnetic field, can arise in
ferromagnetic material (e.g. Grigoriev et al. 2013) but need not
concern us here.
MNRAS 000, 1 -- ?? (2002)
may be expressed in the form of mutations, the frequency
of which appears to increase as a linear function of the par-
ticle flux, M = σM F, where σM is the mutation induction
cross section (in µm2), F is the particle flux (in µm−2s−1).
An empirical approximation for the mutation cross section
is given by Kiefer et al. (2001),
σM = σ0[1 − exp(−3.68 10−3L/σ0 − 7. 10−5L2)] ,
(3)
where L is the Linear Energy Transfer and σ0 ∼ 10−5 µm2. L
is used to characterize biological damage; it is the average
amount of energy deposited per unit length of the substance
(L = dE/dx in keV/µm). It has been shown that the biolog-
ical damage is roughly proportional to the muon flux, and
that the fluence-to-dose ratio remains fairly constant with
energy (Chen 2006; Pelliccioni 2000).
On Earth, the overall globally averaged annual radia-
tion dose from natural sources is 2.4 mSv/yr, the cosmic ray
component is 0.39 mSv/yr and the muon component is 0.33
mSv/yr (Atri & Melott 2011). This is because the muon
component dominates the flux of particles on the ground
at energies above 100 MeV, contributing to 85% to the ra-
diation dose from cosmic rays. Muons arrive at sea level
with an average flux of about 1 muon per square centime-
ter per minute. Muons typically lose almost 2 GeV in pass-
ing through the atmosphere (∼2 MeV g−1cm−2). Muons are
the only biologically significant cosmic radiation with energy
sufficient to penetrate considerable depths, and they are, on
average, spin-polarized. The mean energy of muons at the
ground under contemporary conditions is ∼ 4 GeV which is
enough to penetrate a few meters of rock and several hun-
dred meters of ice.
2.2 Air shower asymmetries
2.2.1 Charge ratio
Primary cosmic rays comprise mostly positive nucleons. This
excess is transmitted via nuclear interactions to pions and
then, on to muons. The muon charge ratio is Rµ ∼ 1.25
below 1 TeV and increases to above ∼ 1.4 at higher energies
(Gaisser 2012).
Due to parity violation in the weak interaction, µ± pro-
duced from decaying pions and kaons are on average spin-
polarized. (The dominant contribution is from pion decay.)
Their daughter electrons and positrons are also, on average,
spin-polarized. The spin-polarized cosmic-rays can also pro-
duce UV CPL when propagating in the medium through
emitting Cerenkov radiation and bremsstrahlung.
2.2.2 Spin-polarized secondary particles
The spinless charged pion with a lifetime of 26 ns de-
cays at rest into a left-handed muon neutrino and a muon:
Π− → µ ¯νµ (and Π+ → µ+νµ respectively). The pion has a
mass of mπ = 140 MeV/c2, the muon has a mass mµ = 106
MeV/c2 and the neutrino is considered massless. We define
rπ = (mµ/mπ)2. In the pion rest frame (denited by ∗), the
momentum of the muon is
p∗
µ = p∗
ν =
mπc
2
(1 − rπ) ∼ 29.8 MeV/c
(4)
MNRAS 000, 1 -- ?? (2002)
(cid:113)
Cosmic Rays and the Chiral Puzzle of Life
5
µ
µ
µ
=
p∗
µ/E∗
2c2 + m2
µ ∼ 0.271 and β∗
and E∗
µc4 ∼ 109.8 MeV. The velocities of
the muon and neutrino in the rest frame of the pion are
β∗
= p∗
= 1. Let the pion move in the
laboratory with velocity vπ/c = βπ ez . Defining θ∗, the angle
of emission of the muon in the pion rest frame, we have the
following relations for the muon momentum, energy, helicity
(h = s · p/p) and angle of emission in the lab rest frame
(Lipari 1993):
νµ
h(βπ, θ
∗) =
1
βµ
pµ = γπ p∗
Eµ = γπ E∗
(cid:20) 1 − rπ + (1 + rπ) cos θ∗ βπ
µ cos θ
+ βπ γπ p∗
µ
∗ + βπ γπ E∗
µ ,
∗
µ cos θ
,
(cid:21)
1 + rπ + (1 − rπ) cos θ∗ βπ
tan θ =
µ sin θ∗
β∗
γπ(βπ + β∗
,
(5)
(6)
(7)
µ cos θ∗) .
(8)
In the limit βπ = 0, the velocity of the muon is: βµ = (1 −
rπ)/(1 + rπ) ≡ β∗
µ ∼ 0.27 and we have h = + 1 independently
from the angle of emission of the muon. The polarization
of the positive muon flux at sea level varies between ∼ 30%
and ∼ 60%, depending on the energy, and is higher than
the polarization of the negative muon flux (Lipari 1993).
The lifetime of negative muons in matter is different be-
cause the negative muons interact with the nuclei of atoms,
which will increase the charge ratio at greater depth. In the
same fashion, the electrons (and respectively positrons) from
muon decay (antimuon decay) are mostly left-handed (right-
handed) with the direction of the spin-aligned (opposite) to
their momentum: µ− → e−νµ ¯νe (µ+ → e+νe ¯νµ). The decay
probability of a positron is W(θ) = (1 + a cos θ)/(4πτµ) where
θ is the angle between the spin direction and the positron
trajectory, τµ ∼ 2.197 µs is the mean lifetime, and the asym-
metry term a is a direct consequence that the muon de-
cay is governed by the weak interaction, and depends on
the positron energy, so the positron angular distribution is
dΓ/d cos θ = W(θ). The maximum and mean positron en-
ergies resulting from the three body decay are given by:
µ) = 52.82 MeV and ¯Ee+ = 36.9
Ee+max
MeV. For a positron emitted with energy of the order of
Ee+max , we have the maximum asymmetry a = 1. When av-
eraged over all positron energies, a = 1/3.
e)c2/(2m2
= (m2
+ m2
µ
2.2.3 Circularly polarized radiation
Cerenkov radiation has a degree of polarization which is de-
pendent on the orientation of the spin of the initial par-
ticle. This is a purely relativistic quantum effect (Sokolov
1940). In the following we consider the difference between
the number of left-handed photons and right-handed pho-
tons emitted from an electron of helicity −1/2 and velocity
β ∼ 0.8 propagating in ice. Defining the ratio of the pho-
ton to the electron energies, ξ = ω/(2Ee), the velocity and
Lorentz factor of the electron β = v/c, γ = (1 − β2)−1/2, the
Cerenkov angle cos θc = [1 + ξ(n2 − 1)]/(nβ), and the function
F = cos χ(cos θc − nβ) + γ−1 sin χ cos φ sin θc where the angles
are α = π/4, χ = 0, φ = 0 for circular polarization, the num-
ber of right-handed photons N1,+ (respectively left-handed
N1,−) is (Lahoti 1977)
N1,± ∝ 0.5(β sin θc)2 + ξ2(n2 − 1) ± 0.5(β sin θc)2 cos(2α)
∓ sin(2α)ξF .
(9)
6
As an example, the ratio (N1,+ − N1,−)/(N1,+ + N1,−) emitted
by an electron of energy ∼ 0.8 MeV (β = 0.77), propagating
in ice, is ∼ 1.3 10−5 at a wavelength of 206 nm. For muons at
the same velocity (∼ 166 MeV), the ratio is 1.8 10−8 at the
same wavelength.
Longitudinally polarized β-radiation gives rise to circu-
larly polarized bremsstrahlung. Using the Born approxima-
tion, McVoy (1957) derived the following formula for circu-
lar polarization in the limit where Ee ∼ hν and the emission
angle of the photon θ = 0◦:
Pγ
Pe
=
1 +
(1 − β)(Ee + 2mc2)
(2 − β)Ee
.
(10)
(cid:19)−1
(cid:18)
Here Pe is the polarization of the electron. The polarization
transfer drops rapidly at electron energies Ee below ∼1 MeV.
2.2.4 Evolution of Spin Polarization
As we discussed in the previous section, cosmic rays are pref-
erentially positively-charged and create µ+ and e+ with mean
magnetic moments µ anti-parallel to the velocity. This asym-
metry can be degraded by two effects. The first is precession
about an external magnetic field, Bext 2; the second is de-
flection of the particle momentum in a Coulomb interaction
that also leads to energy loss while leaving the direction
of the magnetic moment unchanged. We consider these, in
turn, for antimuons/muons and for positrons/electrons.
Muons are unmagnetized (their Larmor radius exceeds
the length of their trajectory) so long as Bext (cid:46) 1 mT. By
contrast, positrons and electrons are magnetized so long as
they traverse a length L (cid:38) (p/(mec))(Bext/1 mT)−1 m, where p
is the electron Lorentz factor, a condition that is quite likely
to be satisfied.
The equation of motion for a magnetized positron is
=
e
γme
p × Bext,
dp
dt
e c2)−1/2. The magnetic moment will also
(11)
where γ = (1− p2/m2
precess about the magnetic field according to
dµ
dt
=
µ ×
e
γme
Bext +
(γ − 1)(Bext · v)v
v2
.
(12)
(cid:18)
(cid:19)
In the non-relativistic limit, which concerns us most, these
equations then imply that p and µ precess about Bext with
a common angular velocity −eBext/me. We expect the spin-
polarized daughter positrons to outnumber spin-polarized
electrons of similar momenta and to be created with a mo-
mentum distribution that is axisymmetric about a down-
g ≡ g ez . Furthermore, for each p, the
ward direction,
distribution of µ will be axisymmetric about Bext. For a
given magnetic field direction, this can lead to an average
spin/magnetic moment polarization projected perpendicular
to the velocity. However, in this case, it is only L that has
the required pseudoscalar form and the perpendicular com-
ponent leads to no bias after full averaging. The precession
contributes modest degradation of the mean polarization.
The influence of deflection can also be discussed. We
consider all the spin-polarized cosmic rays from pion decay,
2 Precession within the molecule is ignorable.
starting with the same energy. We describe here the evolu-
tion of the mean polarization (or equivalently lodacity) as
they decelerate mainly by Coulomb scattering distant elec-
trons. For a single cosmic-ray, the spin direction in space
should not change in a single encounter but the the veloc-
ity will undergo a deflection through a small angle, with the
recoiling electron removing kinetic energy from the cosmic
ray. We can regard this is a diffusion of the angle θ the spin
makes with the velocity. As distant encounters dominate, we
find that the diffusion coefficient satisfies
d(∆θ)2
d ln τ
= 4D = 1
(13)
where τ = 1 + 2mec2/T (e.g. Thorne & Blandford 2017). The
probability distribution per unit solid angle P(θ, τ) satisfies
the diffusion equation
∂P
∂ ln τ
sin θD ∂P
∂θ
1
sin θ
∂
∂θ
(14)
=
.
Multiplying by cos θ and integrating over solid angle, we
obtain
d ln < µz >
d ln τ
so that < µz >∝ τ−1/2.
= −1
2,
(15)
Now consider the evolution of the mean spin polar-
ization of all secondary cosmic-rays of same mass me. If
the cosmic-rays are created relativistically with lodacity L0,
then
(cid:18)
(cid:19)1/2
L(T) ∼ L0
T
2mec2
.
(16)
3 CHIRAL TRANSFER FROM
SPIN-POLARIZED RADIATION TO
BIOMOLECULES
3.1 Molecular Model
3.1.1 Chiral Carbon Atom
In order to understand how spin-polarized cosmic ray show-
ers might favor one chirality over the opposite choice, we
make a very simple, semi-classical model that captures little
of the actual biological, chemical and physical complexity.
However, it is sufficient for our purpose as it incorporates
the minimal and generic elements required to exhibit a chi-
ral bias. In addition, it provides a useful guide for setting
up a more realistic, general calculation and helps identify
special conditions that might lead to larger effects.
We suppose that the principal chiral unit is the Car-
bon atom. We further suppose that Carbon atoms typically
share eight electrons with four, non-coplanar, and different
neighbors. We then localize four of these bonding electrons
on a sphere of radius R ∼ 100 pm, roughly two thirds the
length of a single Carbon bond, and ignore the neighbouring
atoms including their electrons. The four bonding electrons
then screen out the Carbon nuclear charge, well beyond the
sphere. We suppose that these four electrons share a com-
mon, mean, independent ionization potential, I ∼ 0.5 IH ,
where IH ∼ 13.6 eV is the first Hydrogen ionization poten-
tial. We have explored a more realistic alternative model
that includes adjacent base pairs drawn from a long helix
MNRAS 000, 1 -- ?? (2002)
Cosmic Rays and the Chiral Puzzle of Life
7
Figure 3. Unpertubed vs. pertubed spin polarized cosmic-ray
trajectories through a chiral unit. The unpertubed trajectory is
along z. The perturbed trajectory due to the chiral molecular
field B is shown. The perturbed cosmic-ray therefore experience
a slightly different charge distribution which would lead to a dif-
ference in the ionization rate between the two enantiomers.
(The electric quadrupole introduces additional effects which
will be described elsewhere.)
3.1.5 Electromagnetic Field
The electric and magnetic fields associated with the charge
density Σ (Eq. 17 with Q = 0) and the magnetic dipole m
are:
(cid:18)
(cid:19)
E =
1
4π0
− qr
r3 − d
R3
(cid:19)
− d
r3
(cid:18) 3(d · r)r
(cid:19)
r5
, r > R,
, r > R;
(18)
(19)
1
4π0
, r < R; =
(cid:18) 3(m · r)r
B =
µ0m
2πR3 , r < R, =
µ0
4π
− m
r3
r5
3.2 Cosmic Ray Trajectory
3.2.1 Lorentz and Magnetic Dipole Forces
We now consider the path of a single cosmic ray of charge
+e, mass M, non-relativistic velocity v and impact parame-
ter b with respect to the Carbon ion, as seen in Fig.3. For
the moment, just consider one (live) molecule with a fixed,
but arbitrary, orientation. The (classical) force acting on the
cosmic ray is
(cid:20)
(cid:18)
(cid:19)(cid:21)
F = e(E + v × B) + ∇
µ ·
B − v
c2 × E
.
(20)
Figure 2. Example of an electric charge distribution (projected
on a sphere) of two biomolecules of opposite chirality, as seen
from left, from right, from above and from below. This simply
combines an electric dipole and an electric quadrupole. We need
to reflect and rotate by 180 degrees the live molecule to obtain
the evil one. The model we discuss in Section 3 is even simpler,
combining electric and magnetic dipoles.
but a single chiral unit contains the essential properties and
is simpler to describe.
3.1.2 Surface Charge Density
We describe the time-averaged surface charge density, Σ, on
the sphere with a spherical harmonic expansion up to l = 2,
5r · Q · r
3d · r
R2
+
q +
Σ =
4πR2
(17)
where r is a vector from the Carbon nucleus, q = −4e, d is
the electric dipole vector and Q is the (symmetric, trace-free)
quadrupole tensor.
2R4
,
for r = R,
(cid:18)
1
(cid:19)
3.1.3 Magnetic Dipole
In addition, we suppose that there is a magnetic dipole m
due to the motion of the electrons on the sphere, and neglect-
ing the generally somewhat larger contribution associated
with the spin of the binding electrons. The key assumption
we make in this semi-classical model is that the magnetic
dipole, as expressed by the magnetic field just outside the
molecule is not perpendicular to d, as measured by the dis-
position of the electrons within the molecule. In other words,
if for example, the electrons are more concentrated in the
southern hemisphere, then they also flow preferentially in
an equatorial direction. Assuming our model, the magnetic
field within the molecule is uniform.
3.1.4 Molecular chirality
As discussed in Sect. 1, molecular chirality M has opposite
signs for live and evil molecules. The charge density Σ alone
can exhibit electric chirality Me. This is illustrated in Fig. 2,
where Σ is showed for two biomolecules of opposite chirality.
However, we only consider in the following the simplest type
of molecular chirality, the electromagnetic chirality Mem.
The first term is the Lorentz force, which just depends upon
the electric charge and combines electric monopolar, dipolar
and magnetic dipolar components.
The second term is the cosmic ray dipole force. It de-
scribes the net force exerted by the magnetic field, B(cid:48) =
B − v × E/c2, acting on the cosmic ray dipole moment µ in
the cosmic ray rest frame, to lowest order in v.
MNRAS 000, 1 -- ?? (2002)
8
This interaction, which recalls elementary treatments of
spin-orbit coupling and the Stern-Gerlach experiment, needs
some discussion. If we treat the magnetic dipole, classically,
as a small current loop and use the requirement that ∇· B(cid:48) =
0 in an inertial frame, then the form quoted is recovered. The
force can be regarded as the spatial gradient of an interaction
energy, UM = −µ· B(cid:48). If we add the further requirement that
the magnetic field in this frame is the gradient of a potential,
which may not be true, then the alternative form, (µ · ∇)B(cid:48)
is also obtained. This is what can be derived by treating
the dipole as two equal, neighbouring, magnetic monopoles.
However, the spin is quantum mechanical, not classical, and
has to be included in a Hamiltonian formalism using UM.
The form quoted in Eq. 20 is therefore adopted.
3.2.2 Magnetic Dipole Displacement
We are interested in the interaction of the cosmic ray with
the valence electrons as it enters and as it leaves the sphere.
We introduce the coordinate z = r · v, with designating a
unit vector, so that ingress and egress are at r∓ = b + z∓ v
with z∓ = ∓(R2 − b2)1/2. Let us start with the trajectory
prior to ingress. We argued in Sec. 1 that the polarized, cos-
mic ray magnetic dipole can contribute a chiral preference.
We also explained in Sec. 2.2.4 that the average spin polar-
ization should be anti-parallel to the velocity. We therefore
restrict attention to the force F = µz∇Bz. Using ∇ × B = 0,
we find that the velocity perturbation is δv = µzB/Mv. We
are interested in the displacement perpendicular to the un-
perturbed trajectory at ingress not the displacement at a
fixed time because the cosmic ray will eventually cross the
sphere and it is the path the is important. This is given to
first order by
δr−⊥ =
µz
2T
Þ z−
(cid:34)(cid:32) 1 − (1 + η)(1 − η)1/2
−∞ dz (B − ( v · B) v) ,
(cid:32) 1 − (1 − η)1/2
(cid:33)
η
(cid:33)
=
µ0µz
8πT R2
−
η
(m · b) b − η1/2(m · v) b
(cid:35)
(m · v × b) v × b
.
(21)
where T = Mv2/2 and η = b2/R2.
The velocity perturbation immediately after ingress has
to take account of the impulse due to the current sheet in the
sphere. However, there is no additional force as the interior
magnetic field is uniform. The chiral part of the transverse
displacement at egress can then be shown to be
δr+⊥ =δr−⊥ +
v
(z+ − z−)δv+⊥
(cid:33)
(cid:34)(cid:32) 1 − (1 + 3η)(1 − η)1/2
(cid:32) 1 − (1 + 4η)(1 − η)1/2 + 6η(1 − η)
η
(cid:33)
=
µ0µz
8πT R2
−
η
(m · b) b − 7η1/2(m · v) b
(m · v × b) v × b
.
(cid:35)
(22)
3.3 Ionization Rate
3.3.1 Cross Section
(cid:18) IH
(cid:19)(cid:18) IH
(cid:19)
I
T
σion = 16πa2
0
∼ 3.5 × 10−21T−1
this ionization as a proxy for mutation.
The classical ionization cross section per Carbon is
keV m2
(23)
where a0 ∼ 0.5R is the Hydrogen Bohr radius. The probabil-
ity that a cosmic ray incident upon an atom will create an
ionization is therefore Pion ∼ σion/πR2 ∼ 0.2 T−1
keV. We take
Direct measurements below ∼ 1 keV give cross sections
lower by factors up to ten and a slower decline with increas-
ing kinetic energy (e.g. Kim & Desclaux 2002). This reflects
the fact that more tightly bound electrons can be ionized
as T increases as well as quantum mechanical effects. Again,
this is unimportant for our limited purpose. In addition to
its transverse displacement a cosmic ray will have a slightly
different energy and cross section as it crosses the sphere
and there is an associated chiral preference. This turns out
to be subdominant in our model and we ignore it although
it is likely to be significant in a more realistic description.
3.3.2 Second Order Perturbation to the Ionization Rate
We have computed the first order deflection at ingress and
egress. By itself, this leads to no net change in the ioniza-
tion rate. However, the deflection results in the cosmic ray
encountering a slightly different surface density of electrons
and producing a change in the cosmic ray energy. The sec-
ond order change in the relative ionization rate is then given
by
δ ln Pion = −δr−⊥ ·
∇⊥ ln Σ− +
− δr+⊥ ·
≈ −(δr−⊥ · ∇⊥ ln Σ− + δr+⊥ · ∇⊥ ln Σ+),
∇⊥ ln Σ+ +
eE+
t
T
(cid:19)
eE−
T
t
(cid:18)
(cid:18)
(cid:19)
,
(24)
where the perpendicular, logarithmic gradient in the relative
surface charge density at ingress and at egress is
∇⊥ ln Σ∓ =
− 3
4eR2
(cid:16)(1 − η)1/2η(d · b) b ± η1/2(d · v) b + d · v × b v × b
(25)
(cid:17)
,
after dropping the kinetic energy term as it is sub-dominant
in our particular implementation.
3.3.3 Averaging over Impact Parameter and Orientation
So far, we have considered one molecule and one cosmic ray.
The simplest assumption to make is that the cosmic ray flux
is isotropic with respect to the molecule. This still allows the
cosmic rays to be anisotropic if, as is likely, the molecules
are randomly oriented, for example in water. (We emphasize
that there are many circumstances when orientation biases
may be present and these could lead to a larger chiral pref-
erence.)
So, when a cosmic ray of fixed charge, mass, magnetic
moment and speed encounters an "atom" described by d, m,
we must average over the incident cosmic ray paths. In this
case, any term in the δ ln Pion sum that is odd in v can be
dropped as it will be canceled by the effect of a cosmic ray
MNRAS 000, 1 -- ?? (2002)
with the opposite velocity or impact parameter. We then
average the remaining terms over azimuth, perpendicular to
v using < m · b d · b >= m · d/2 etc and then average η over
a unit circle.
The final step is to average v over the surface of a unit
sphere. Averages of the form < u · v w · v > become u · w/3.
After performing these integrals and averages, we obtain
δ ln Pion =
3Kµ0µedm
32πeT R4
= 1.3 × 10−8
(26)
for a polarized positron encountering a chiral unit with d
parallel to m. The constant K evaluates to -3.06. µB is the
Bohr magneton.
(cid:18) d m
(cid:19)(cid:18) T
1 D µB
1 keV
(cid:19)−1
,
3.3.4 Chiral Preference
We assume that the relative difference in the mutation rate
M between live and evil molecules is due to the relative
difference in the ionization rates,
Mevil − Mlive
δM =
= δ ln Pion = f (M, L, O).
Mlive
(27)
• The cosmic-ray lodacity, L, depends on the nature and en-
ergy of the primary cosmic rays, the external magnetic field
Bext, and the nature of the medium where the cosmic ray
shower develops. The lodacity allows for the depolarization
of the cosmic rays as they lose energy;
• The molecular chirality, M, depends on the nature of
the chiral units and their relative configuration in the
biomolecules (helix). For a single unit, it may includes all the
possible different expressions of chirality, i.e. M = Mem +Me
(in our simple model, M = Mem = d · m);
• The molecular orientation, O, takes account of orientation
biases. It can depend on the presence of an external magnetic
field, Bext, or a chiral material in the fount (like zeolites).
The effect that we have described is directly related
to parity violation in the weak interaction. If we imagine a
cosmic ray colliding with a molecule with impact parame-
ter vector b and a mirror plane perpendicular to b passing
through the central ion then an evil atom or molecule can
always be rotated to be the reflection in this plane. If d is
not perpendicular to m, a cosmic ray with magnetic moment
anti-parallel to its velocity will be deflected towards a region
of greater charge density in one case and away from it in the
other, leading to the difference in the mutation rate. (The
non-zero value of d· m can be related to the volume integral
of E· B = −Z2
0 d· m/3πR3, where Z0 is the impedance of free
space.)
The chiral preference calculated in this model is quite
it is given by δM ∼
small. To order of magnitude,
α2(me/M)(I/T) ∼ 10−10 for ten per cent polarized, mildly
relativistic e± and ∼ 10−11, for highly polarized mildly rel-
ativistic muons. Formally, δM increases to ∼ 10−5, for e± if
T ∼ I but our semi-classical calculation is quite inappropri-
ate and the loss of lodacity (Eq. 16) is likely to reduce δM by
a further factor of 300 to 3×10−8. The presence of secondary
electrons that are produced as the cosmic ray loses energy
may also diminish the chiral preference.
A quantum mechanical calculation is necessary for
T (cid:46) 1 keV and must introduce additional spin- and charge-
dependent effects as cosmic-ray electrons anti-symmetrize
MNRAS 000, 1 -- ?? (2002)
Cosmic Rays and the Chiral Puzzle of Life
9
Figure 4. Evolution of the number of live (solid lines) and evil
(dashed lines) molecules, for different normalized proliferation
rates p ≡ (k1 − Mlive)/k2 = 105 to 10 keeping ∆M = 10−5 (up-
per panel); different ∆M/k2 = 1 to 10−9 keeping p = 10 (lower
panel). The time is the normalized time τ = k2t. We assume a
racemic initial state Nlive(t = 0) = Nevil(t = 0) = 1.
the wave function when there is an electron cosmic ray in-
teracting with a valence electron and annihilation should be
included for a cosmic ray positron. Models that better cap-
ture the helical nature of biological molecules and include
the electric quadrupole, which can be very large (Wu et al.
2017), are also likely to increase the chiral preference from
that given by this estimate.
4 EVOLUTIONARY AMPLIFICATION
4.1 From chiral preference to chiral dominance
In this section, we show that the small differences in
the propagation of cosmic rays through early biological
molecules might be amplified by replication and reproduc-
tion, to lead to the chiral dominance we see today. The mod-
eling of homochiralisation (the emergence of a single chiral
life form) has been the object of various studies (e.g. Frank
1953; Gleiser & Walker 2008). The existence of small chi-
100101102103104105time100101102103104105106107Number of moleculesNliveNevilp=10510410310210110100time100101102103104105106107Number of moleculesNliveNevil∆M/k2=110-410-9 10
ral domains (i.e. simple molecules of the same chirality) is
a prerequisite for the assembly of structurally more stable
polymers necessary for the formation of living systems; how-
ever, given the small chiral biases found in nature, it is dif-
ficult to explain the emergence of a single handedness with
a pre-biotic process (Bonner 2000; Burton & Berger 2018).
It has been also argued that the emergence of homochirality
is a consequence of a small initial enantiomeric excess am-
plified by the antagonism between the two mirror life forms
(Frank 1953). In this scenario, the proliferation rates of live
and evil molecules are identical. However, we have shown in
the previous section that the coupling between the cosmic-
ray lodacity and the molecular chirality, expressed by the
product L · M, leads to a difference in the mutation rate of
live and evil molecules, which, as we show in the following,
is sufficient to establish homochirality.
4.2 Model equations
In the following, we denote by Nlive and Nevil the number of
live and evil molecules, respectively. The evolution of the two
populations is the result of an interplay between stochastic
factors (like spontaneous mutation) and deterministic fac-
tors (the cosmic radiation). The proliferation rate of live
and evil populations can be approximated as the difference
between the natural growth rate, that we denote by k1, and
the mutation rate (natural and induced by the cosmic radia-
tion) that we denoted by M. We denote by k2 the antagonism
factor.
A simple example of antagonism is the introduction of
live monomers (naturally synthesized by live forms) in the
environment of the evil forms (and vice versa). It has been
shown that endogenous right-handed amino acids are used
as auto-regulators, inhibiting bacterial growth under low nu-
trient conditions (Cava et al. 2011). Therefore the introduc-
tion of wrong monomers (right-handed amino acids for live
systems and left-handed amino acids for evil systems) will
inhibit their respective growth. This is a simple example of
antagonism; other biological effects could be important but
our purpose is only to show a qualitative effect.
We assume that we can neglect the natural mutation
(as it is the same for live and evil molecules) and therefore,
we define Mlive = σM,liveΦ and Mevil = σM,evilΦ, where σM
is the mutation induction cross section (in µm2) given by
Eq. 3 and Φ is the particle flux (in µm−2s−1). The exposure
to the cosmic-rays radiation is a deterministic factor because
it induces a consistent chiral bias in the form of a difference
in the mutation rate δM = (Mevil − Mlive)/Mlive (see Eq. 27).
As it has been shown, δM is directly related to the sign and
amplitude of the molecular chirality M, the lodacity L, and
other factors like the molecular orientation. The factor k1
and k2 are independent of the chirality of the molecules.
The evolution of the two populations is described by the
logistic equations:
d ln Nlive
d ln Nevil
dt
dt
= k1 − Mlive − k2Nevil ,
= k1 − Mevil − k2Nlive .
(28)
(29)
Apart from δM that we have estimated in our simple
model, the other factors are unknown; we can approximate
k1 as the growth rate of bacteria on Earth. But we are left
with another unknown factor, the conflict rate between live
and evil molecules. Therefore we normalize the equations by
the conflict rate k2:
d ln Nlive
dτ
=
d ln Nevil
k1 − Mlive
k1 − Mlive
k2
(30)
− Nevil ,
− ∆M
k2
− Nlive ,
=
dτ
(31)
We introduce the normalized proliferation rate p ≡ (k1 −
Mlive)/k2, the normalized time τ = k2t and ∆M ≡ Mevil−Mlive.
Note the relation ∆M = δM Mlive.
k2
4.3 Illustrative solutions
Here we show the effect of varying the strength of the param-
eters k1, k2 and ∆M in the evolution of the two populations.
Solutions to the Eqs. 30-29 are presented in Fig. 4. We as-
sume that we start at t = 0 with a distribution of live and
evil molecules that is racemic in average (for simplicity we
assume Nlive = Nevil = 1). The time needed to reach a pure
homochiral state depends on the normalized parameters p,
∆M/k2. To illustrate the effect of these different parameters,
we show in Fig. 4 the evolution of the population number
as a function of time, varying p (upper panel) and ∆M/k2
(lower panel). As we can see qualitatively from these curves,
the cosmic-rays assure a consistent chiral bias allowing a
difference in the evolutionary path of live and evil systems,
while the selection pressure term determines the time scale
at which homochirality can be established.
The key question, that we cannot answer because we
cannot make reliable estimates,
is whether the natural
growth rate, k1 and the antagonism factor k2 are sufficient
to transform a small chiral preference into chiral dominance.
However, we note that simple bacteria can replicate in hours
and the time scale for environmental evolution on a young
planet is likely to be millions of years. There could be many
billions of generations to allow the processes described to
take effect.
5 DISCUSSION
In this paper, we have argued that homochirality is a deter-
ministic consequence of the weak interaction, expressed by
cosmic irradiation of molecules as they transition from pre-
biotic to vital components of single-celled organisms. The
choice that was made is then traceable to the preponder-
ance of baryons over antibaryons, established in the early
universe and ultimately to the symmetries of fundamental
particle interactions presenting requirements as first eluci-
dated by Sakharov (1967). In a similar fashion, we have
tried to list some of the physical and chemical factors that
seem to be necessary for such a causal biological path to
have been followed. In addition we have introduced one
specific mechanism, involving collisional ionization by sec-
ondary spin-polarized cosmic rays, that, we argue, is likely
to be more relevant than mechanisms involving circularly
polarized ultraviolet light. We have also demonstrated how
a quite small chiral preference can evolve into chiral domi-
nance, emphasizing the importance of conflict. Much more
study is needed to determine if these ingredients do, indeed,
MNRAS 000, 1 -- ?? (2002)
suffice to account for homochirality or if, instead, the choice
must be due to chance or environmental idiosyncrasy. An-
swering this question is central to understanding the origin,
prevalence and migration of life in the universe.
It should be emphasized that the purpose of this calcu-
lation is not to obtain a quantitative answer but, instead, to
demonstrate the qualitative effects involved and to highlight
the various factors that have to be included. In particular,
our restriction to the d· m interaction surely underestimates
the molecular chirality, because actual bonds have contribu-
tions from higher (cid:96) multipoles which lead to larger gradi-
ents in both the magnetic field and the charge density. For
example, it has been shown that the adsorption of chiral
molecules on specific surfaces can substantially enhance the
optical activity by several orders of magnitude (∼ 1000) be-
cause of the electric dipole - electric quadrupole interaction
(e.g. Wu et al. 2017), i.e., the electric chirality Me that we
neglected so far. We envision that, under specific conditions,
Me would lead to an enhancement of the enantioselectivity
in our model.
If we admit a deterministic path for life, then life's hand-
edness is a worldwide property (here "wordwide" has to be
taken in Giordano Bruno' sense in which our world is only
one of many worlds). So far, enantiomeric excesses of left-
handed amino acids were found in a very limited sample
of carbonaceous chondrites subject to terrestrial contamina-
tion. The search for enantiomeric excesses in amino acids
in situ is the goal of future space missions. The Hayabusa2
(Lauretta et al. 2017) and OSIRIS-REx (Yamaguchi et al.
2018) spacecrafts will return to Earth with samples in 2020
and 2023, taken from two small carbon-rich asteroids. The
future mission ExoMars 2020 (Vago et al. 2017) is planned
to return to the Earth with samples collected from Mars.
The main goal of these missions is to determine the sign of
enantiomeric excesses in the chiral molecules found in these
samples. Based on the results from pre-biotic mechanisms,
we do not expect to find large enantiomeric excesses of amino
acids in places where living systems were absent. However,
according to our model, the sign of the small excesses is ex-
pected to be the same everywhere where cosmic-rays showers
can develop.
A way to test the proposed scenario and further our
understanding of the processes we have highlighted is to
perform experiments. The mutation rate can be estimated
through several methods (e.g. Foster 2006; Pope et al. 2008).
A prediction of our model is that the mutation rate is de-
pendent upon the spin-polarization of the radiation. One
possible experiment would be to measure the mutation rate
of two cultures of bacteria under spin-polarized radiation
(either e± or µ±) of different lodacity with energy above the
threshold necessary to induce double strand breaks in DNA
(∼ 50 eV). If the coupling between lodacity and molecular
chirality is efficient in introducing a chiral bias, one of the
two cultures should exhibit a much lower mutation rate. We
emphasize that much can be learned experimentally from the
comparison of chiral molecules involved in biology and using
both signs of lodacity which can be created at accelerators.
It is not necessary to create "mirror life" to proceed. Once
the dominant processes are identified, we can have confi-
dence in our understanding of particle physics and quantum
chemistry to draw the necessary conclusions.
If these experiments show that the evolution of bacteria
MNRAS 000, 1 -- ?? (2002)
Cosmic Rays and the Chiral Puzzle of Life
11
is influenced by the spin-polarized radiation, this will be a
good indication that spin-polarized cosmic-rays might be an
important piece of the chiral puzzle of life.
ACKNOWLEDGEMENTS
The research of NG is supported by the Koret Founda-
tion, New York University and the Simons Foundation. NG
thanks Louis d'Hendecourt for helpful discussions.
REFERENCES
Attwater J., Wochner A., Pinheiro V. B., Coulson A., Holliger P.,
2010, Nat. Commun., 1
Atri D., Melott A. L., 2011, Geophysical Research Letters, 38
Avalos M., Babiano R., Cintas P., Jimenez J. L., Palacios J. C.,
2000, Tetrahedron Asymmetry 11(14), 2845
Bailey, J. 2001, Astronomical Sources of Circularly Polarized
Light and the Origin of Homochirality, Origins of Life and
Evolution of the Biosphere, 31, 167
Bailey J., Chrysostomou A., Hough J. H., et al., 1998, Sci, 281,
672
Biswas S., Sarkar S., Pandey P. R., Roy S., 2016, Phys. Chem.
Chem. Phys., 18, 5550
Bonner W., 2000, Chirality 12(3), 114
Burton A. S., Berger E. L., 2018, Life, 8, 14
Cava F., Lam H., de Pedro M. A., Waldor M. K., 2011, Cell Mol.
Life Sci., 68, 817
Castillo H., Schoderbek D., Dulal S., Escobar G., Wood J., Nelson
R. et al., 2015, Int. J. Radiat. Biol. 91, 749
Chen J., 2006, Radiat. Prot. Dosim., 118(4), 378
Curie P., 1894, J. Phys. Theor. Appl., 3 (1), 393
De Marcellus P., Meinert C., Nuevo, Filippi J. J., Danger G.,
Deboffle D., Nahon L., Le Sergeant d'Hendecourt L., Meier-
henrich U. J., 2011, Astrophys. J. 727, L27
D'Hendecourt L., Modica P., Meinert C., Nahon L., Meierhen-
rich U., 2019, Journal of Interdisciplinary Methodologies and
Issues in Science, In press. arXiv:1902.04575
Dryzun C., Mastai Y., Shvalb A., Avnir D., 2009, J. Mater. Chem.
19, 2062
Erastova V., Degiacomi M. T., Fraser D. G., Greenwell H. C.,
2017, Nat. Commun. 8, 2033
Foster, P. L., 2006, Methods Enzymol. 409, 195
Frank F. C., 1953, Biochem. Biophys. Acta, 11, 459, 463
Gaisser T. K., 2012, Astroparticle Physics, 35(12), 801
Gilmour I., 2003, in Treatise on Geochemistry, H. D. Holland, K.
K. Turekian, Eds., Vol. 1, 269
Gleiser M., Walker S. I., 2008, Orig Life Evol Biosph, 38, 293
Gohler B., Hamelbeck V., Markus T. Z., Kettner M., Hanne G.
F., Vager Z., Zacharias H., 2011, Science , 331, 894
Grigoriev S. V. et al., 2013, Phys. Rev. Lett. 110, 207201
Kawanishi M., Okuyama K., Shiraishi K., Matsuda Y., Taniguchi
R., Shiomi N., et al., 2012, J. Radiat. Res 53, 404
Kelvin, Lord, 1894, J Oxford Univ Junior Sci Club 18:25
Kiefer J., Schmidt P. Koch S., 2001, Radiat. Res. 156, 607
Kim Y.-K., Desclaux J.-P., 2002, Phys. Rev. A, 66, 012708
Kitadai N., Maruyama S., 2018, Geosci. Front., 9, 1117
Kramers H. A., 1927, Atti Cong. Intern. Fisici, (Transactions of
Volta Centenary Congress) Como., 2, 545
Kronig R. de L., 1926, J. Opt. Soc. Am. 12 (6), 547
Kwon J., Tamura M., Lucas P. W., Hashimoto J., Kusakabe N.,
Kandori R., Nakajima Y., Nagayama T., Nagata T., Hough
J. H., 2013, Astrophys. J. Lett., 765, L6
Lahoti S., Takwale G., 1977, Pranama; Vol. 9, No 2
Lauretta D. S., et al. 2017, Space Science Reviews, 212, 925
12
Lipari P., 1993, Astropart. Phys., 1, 195
LeBel J. A., 1874, Bull. Soc. Chim. Fr., 22, 337
Lee T. D., Yang C. N., 1956, Physical Review 104(1): 254
Manabe Y., Ichikawa K., Bando M., 2012, J. Phys. Soc. Jpn. 81,
104004
McVoy K. W., 1957, Phys. Rev. 106, 828
Meierhenrich U. J., 2008, Amino Acids and the Asymmetry of
Life, Springer
Miller S. L., 1953, Science, 117, 528
Muller H.J., 1927, Science 66, 84
MMX homepage: http://mmx.isas.jaxa.jp/en/index.html
Or´o J., Biochem Biophys Res Commun, 1960, 2, 407
Or´o J., Nature, 1961, 191, 1193
Pasteur L., 1848, C. R. S´eances Acad. Sci., 26, 535
Patrignani C., http://pdg.lbl.gov/2017/reviews/rpp2017-rev-
cosmic-rays.pdf
Pelliccioni M., 2000, Radiat. Prot. Dosim, 88(4), 279
Pizzarello S., 2006, Acc Chem Res 39, 231
Planel H., Soleilhavoup J. P., Tizador R., Richoilley G., Conter
A., Croute F., et al., 1987, Health Phys. 52, 571
Pope C. F., O'Sullivan D. M., McHugh T. D., Gillespie S. H.,
2008, Antimicrob. Agents Chemother. 52:1209-1214
Quack M. (1989) Angew Chem Int Ed (Engl) 28(5), 571
Rosenfeld L. , 1928, Z. Physik 52, 161
Ruf A., Kanawati B., Hertkorn N., Yin Q. Z., et al., 2017, Proc.
Natl. Acad. Sci. USA, 114, 2819
Ruff S. W., 2004, Icarus 168, 131a A¸S143
Sakharov A. D., 1967, Pisma Zh. Eksp. Teor. Fiz. 5, 32 [Sov.
Phys. JETP Lett. 5, 24 (1967)
Salam A., 1991, JMolE, 33, 105
Salam A., 1992, Phys Lett B., 288, 153
Schrodinger E., 1945, What Is Life? The Physical Aspect of Liv-
ing Cell, Cambridge University Press
Shannon C. E., Weaver W., 1949, The Mathematical Theory of
Communication, Urbana, IL: University of Illinois Press
Shinitzky M., Shvalb A., Elitzur A. C., Mastai Y., 2007, J Phys
Chem B, 111, 11004
Sokolov, A. 1940 Quantum theory of the elementary particles.
Dokl. Akad. Nauk SSSR 28, 415
Svensmark H., 2006, Astron. Nachr. 327, No. 9, 871
Thorne K. S., Blandford R. D., 2017, Modern Classical Physics,
page 1006, Princeton, NJ: Princeton Univ. Press
Vago J.L, Westall F., 2017, Astrobiology 17, 471
van't Hoff J. H., 1874, Arch. Neerl., 9, 1
Vester, F., Ulbricht, T.L.V., and Krauss, H. 1959, Naturwiss-
chaften 46, 68
Watson J. D., Crick, F. H. C., 1953, Nature 171, 737
Wu, C.S. et al., 1957, Physical Review 105(4), 1413
Wu T., Zhang W., Wang R., and Zhang X., 2017, Nanoscale 9(16),
5110
Yamagata Y., 1966, J. Theor. Biol. 11, 495
Yamaguchi T., et al., 2018, Acta Astronautica. Vol. 151, 217
MNRAS 000, 1 -- ?? (2002)
|
1002.1980 | 2 | 1002 | 2010-08-31T18:51:36 | Single-image measurements of monochromatic subdiffraction dimolecular separations | [
"physics.bio-ph"
] | Measuring subdiffraction separations between single fluorescent particles is important for biological, nano-, and medical-technology studies. Major challenges include (i) measuring changing molecular separations with high temporal resolution while (ii) using identical fluorescent labels. Here we report a method that measures subdiffraction separations between two identical fluorophores by using a single image of milliseconds exposure time and a standard single-molecule fluorescent imaging setup. The fluorophores do not need to be bleached and the separations can be measured down to 40 nm with nanometer precision. The method is called single-molecule image deconvolution -- SMID, and in this article it measures the standard deviation (SD) of Gaussian-approximated combined fluorescent intensity profiles of the two subdiffraction-separated fluorophores. This study enables measurements of (i) subdiffraction dimolecular separations using a single image, lifting the temporal resolution of seconds to milliseconds, while (ii) using identical fluorophores. The single-image nature of this dimer separation study makes it a single-image molecular analysis (SIMA) study. | physics.bio-ph | physics |
Single-image separation measurements of two
unresolved fluorophores
Shawn H. DeCenzo, Michael C. DeSantis, and Y. M. Wang∗
Department of Physics, Washington University, St. Louis, MO
63130, USA
[email protected]
1
Abstract
Measuring subdiffraction separations between single fluorescent par-
ticles is important for biological, nano-, and medical-technology studies.
Major challenges include (i) measuring changing molecular separations
with high temporal resolution while (ii) using identical fluorescent labels.
Here we report a method that measures subdiffraction separations be-
tween two identical fluorophores by using a single image of milliseconds
exposure time and a standard single-molecule fluorescent imaging setup.
The fluorophores do not need to be bleached and the separations can
be measured down to 40 nm with nanometer precision. The method
is called single-molecule image deconvolution -- SMID, and in this arti-
cle it measures the standard deviation (SD) of Gaussian-approximated
combined fluorescent intensity profiles of the two subdiffraction-separated
fluorophores. This study enables measurements of (i) subdiffraction di-
molecular separations using a single image, lifting the temporal resolution
of seconds to milliseconds, while (ii) using identical fluorophores. The
single-image nature of this dimer separation study makes it a single-image
molecular analysis (SIMA) study.
References
[1] M. Born and E. Wolf, Principles of Optics (Cambridge University Press,
Cambridge, UK, 1999).
[2] G. S. Gordon, D. Sitnikov, C. D. Webb, A. Teleman, A. Straight, R. Losick,
A. W. Murray, and A. Wright, "Chromosome and Low Copy Plasmid Seg-
regation in E. coli: Visual Evidence for Distinct Mechnisms," Cell 90 1113-
1121 (1997).
[3] A. K. Salem, P. C. Searson, and K. W. Leong, "Multifunctional nanorods
for gene delivery," Nature 2, 668-671 (2003).
[4] G. Han, P. Ghosh, M. De, and V. M. Rotello, "Drug and Gene delivery
using gold nanoparticles," Nanobiotechnolgoy 3, 40-45 (2007).
[5] A. Yildiz, M. Tomishige, R. D. Vale, and P. R. Selvin, "Kinesin walks
hand-over-hand," Science 303, 676-678 (2004).
[6] X. Qu, D. Wu, L. Mets, and N. F. Scherer, "Nanometer-localized multiple
single-molecule fluorescence microscopy," Proc. Natl. Acad. Sci. USA 101,
11298-11303 (2004).
[7] E. Betzig, G. H. Patterson, R. Sougrat, O. W. Lindwasser, S. Olenych, J.
S. Bonifacino, M. W. Davidson, J. L. Schwatz, and H. F. Hess, "Imaging
Intracellular Fluorescent Proteins at Nanometer Resolution," Science 313,
1642-1645 (2006).
[8] A. Sharonov and R. M. Hochstrasser, "Wide-field subdiffraction imaging
by accumulated binding of diffusing probes," Proc. Natl. Acad. Sci. USA
103, 18911-18916 (2006).
2
[9] T. d. Lacoste, X. Michalet, F. Pinaud, D. S. Chemla, A. P. Alivisatos, and
S. Weiss, "Ultrahigh-resolution multicolor colocalization of single fluores-
cent probes," Pro. Natl. Acad. Sci. USA 97, 9461-9466 (2000).
[10] L. S. Churchman, Z. Okten, R. S. Rock, J. F. Dawson, and J. A. Spudich,
"Single molecule high-resolution colocalization of Cy3 and Cy5 attached
to macromolecules measures intramolecular distances through time," Pro.
Natl. Acad. Sci. USA 105, 1419-1423 (2005).
[11] M. Bates, B. Huang, G. T. Dempsey, X. Zhuang, "Multicolor Super-
Resolution Imaging with Photo-Switchable Fluorescent Probes," Science
317 1749-1753 (2007).
[12] J. Behboodian, "On the modes of a mixture of two normal distributions,"
Technometrics 12 131-139 (1970).
[13] M. C. DeSantis, S. H. DeCenzo, Y. M. Wang, "Precision analysis for stan-
dard deviation measurements of single-fluorescent molecule images," Opt.
Express doc. ID 123194 (posted 15 March 2010, in press).
[14] R. E. Thompson, D. R. Larson, and W. W. Webb,"Precise nanometer lo-
calization analysis for individual fluorescent probes," Biophys. J. 82, 2775-
2783 (2002).
[15] S. Ram, E. S. Ward, R. J. Ober, "Beyond Rayleigh's criterion: A resolution
measure with application to single-colecule microscopy," Proc. Natl. Acad.
Sci. USA 103, 4457-4462 (2006).
[16] Y. M. Wang, J. Tegenfeldt, W. Reisner, R. Riehn, X.-J. Guan, L. Guo, I.
Golding, E. C. Cox, J. Sturm, and R. H. Austin, "Single-molecule studies
of repressor-DNA interactions show long-range interactions," Proc. Natl.
Acad. Sci. USA 102, 9796-9801 (2005).
[17] M. Speidel, A. Jonas, and E.-L. Florin, "Three-dimensional tracking of
fluorescent nanoparticles with subnanometer precision by use of off-focus
imaging," Opt. Lett. 28 (2003).
[18] Y. M. Wang, R. H. Austin, and E. C. Cox, "Single molecule measurements
of repressor protein 1D diffusion on DNA," Phys. Rev. Lett. 97, 048302
(2006).
1
Introduction
For conventional far-field imaging, the Rayleigh criterion determines the diffrac-
tion limit seperation below which two identical fluorescent particles cannot be
differentiated. The diffraction-limit separation is 0.61λ/N.A. ≈ 230 nm for
mean visible wavelength λ of ≈ 550 nm and microscope objective's numerical
aperture N.A. of ≈ 1.45 [1]. In biological, nano-, and medical-technology stud-
ies, many inter- and intra-molecular separations are below the diffraction limit.
3
These separations include intra-molecular separations such as the separation be-
tween different regions of a macromolecule -- an actin filament or a chromosomal
DNA molecule [2], and inter-molecular separations, e.g., between proteins and
nano-particles [3, 4].
Recent single-molecule fluorescence imaging methods circumvent the Rayleigh
criterion and measure subdiffraction molecular separations with nanometer pre-
cision. However, such precision is not achieved without tradeoffs. Some methods
bleach the fluorophores -- SHRImP [5], NALNS [6], PALM [7], and PAINT [8],
and therefore only one separation of the studied molecules can be measured;
some require photoswitchable or multicolor fluorescent labels -- Ref. [9], PALM
[7], SHREC [10], and STORM [11], rendering the fluorophore and spectral se-
lections restrictive. In addition, these methods are based on centroid location
measurements of each constituent molecule; therefore, the rate of separation
measurements is bound by the total time required to locate all constituent
molecules -- typically in seconds. The timescale of seconds is long for many
biological processes where molecular separations change faster than seconds.
Although the SHREC method allows simultaneous dimolecular separation mea-
surements in the timescale of milliseconds, labeling heterogeneous fluorophores
and overlapping two spectral distinct emission images pose additional challenges.
Here we report an alternative method to measure subdiffraction separations
of two molecules labeled with identical fluorophores, all by using a single image
of the two constituent molecules. For simplicity, in the rest of this article we
call the two separated fluorophores a dimer. The dimers were simultaneously
illuminated for milliseconds by standard laser excitation (intensity of order 1
kW/cm2) and the emitted photons were collected by a standard single-molecule
fluorescent imaging setup. The collective intensity profile of the subdiffraction
separated dimers were recorded and fit to a Gaussian function. While the point
spread function (PSF) centroids of the two constituent molecules cannot be re-
solved, the spread of the dimer intensity profile, measured as standard deviation
(SD) of the dimer's Gaussian-approximated intensity profile (in this article we
call the dimer intensity profile dimer PSF), increases with dimer separation. In
our method, the longitudinal SD (along the dimer direction) and the total de-
tected photon count of the dimer image are the principle parameters needed for
(1) differentiating dimers from monomers and (2) determining the separation of
the dimer with known precision. We have compared the experimental results
with simulations and they agree well.
The reported SD study here analyzes a convolved image of two molecules
and extracts the molecular separation information; the method is analogous
to deconvolving a complex single-molecule image for molecular properties. We
call this method of using the spread of convolved single-molecule images to
analyze molecular properties "Single-molecule image deconvolution," or SMID.
Using SMID, dimer separations can be measured by using a single image of
milliseconds exposure times, allowing improvement in temporal resolution from
seconds to milliseconds. The improvement is 10- to 1000-fold depending on
the number of photons emitted by the dimer and the dimer separation. This
improved temporal resolution is attributed to the single-image nature of the
4
SMID dimer separation measurements; here we name single-molecule studies
that use single images of millisecond exposure times the "Single-image molecular
analysis" method, or SIMA. This article applies SMID and SIMA to a simple
convolution example of two molecules, future SMID and SIMA analysis can be
extended to higher order multimers of different geometrical arrangements.
2 Results
2.1 Dimer PSF SD is an indicator of the subdiffraction
separation
When two identical fluorophores are separated by less than the diffraction limit,
the width of the dimer PSF varies with the dimer separation.
Figure 1(A) shows an array of Streptavidin-Cy3 dimers with increasing hor-
izontal subdiffraciton separations, δ, of 0, 79, 158, and 237 nm. The dimer
images were constructed from two images of an experimental monomer movie:
one of the two monomer images was displaced by 0, 1, 2, and 3 pixels (79
nm/pixel) horizontally, then the images were added. The number of detected
photons, N , of each monomer image was 511 photons. From the figures, it is
clear that the width of the dimer images changes with separation. At the 79
nm separation, the dimer is not clearly different from the constructed monomer
(dimer at 0 nm separation); at the 158 nm separation, the image becomes wider
and it resembles a dimer more than a monomer; at the 237 nm separation, which
is approximately equal to the diffraction limit of 230 nm, the image is clearly
different from a monomer and the two monomer PSF peaks may start to be
resolved.
In Fig. 1(B) the dimer's increasing width with separation is quantified. The
dimer PSFs is unimodal as the sum of two Gaussian functions [12] and fits well
to a 1D Gaussian, where all transverse pixel intensity values at each longitudi-
nal pixel were averaged. The 1D x-axis SD values of the dimer increase with
separation as 114.1, 119.7, 141.6, and 178.6 nm, respectively.
2.2 Differentiating dimers from monomers using a single
image
As shown in Fig. 1(A), for many subdiffraction separations, the dimer images
are similar to monomer images in appearance. In order to measure dimer sep-
arations, one must first determine whether an image is a monomer or a dimer.
Here we show that by using the PSF SD and photon count of a single image,
dimers can be differentiated from monomers.
For a monomer and a subdiffraction-separated dimer of the same photon
count, the image is a dimer if the dimer PSF SD (sd) exceeds the monomer PSF
SD (sm) by more than the sum of the dimer and monomer SD measurement
errors, ∆sd and ∆sm, respectively, as
sd − sm > ∆sd + ∆sm.
(1)
5
Figure 1: Dimers of different subdiffraction separations. (A) From left to right,
streptavidin-Cy3 dimers aligned in the horizontal direction separated by 0, 79,
158, and 237 nm. From the appearance, it is not clear that the image contains
two molecules until near the 237 nm separation. The scale bar is 0.5 µm. (B)
1D intensity profiles of the dimers (circles are data and lines are the Gaussian
fits) showing that the dimer SDs increase with separation.
Here the SD of a SD distribution, ∆sm or ∆sd, is the error associated with a
SD measurement of a single image.
Figure 2(A) compares the experimental and simulation SD distributions of
a constructed monomer (0 nm separated dimer) and a 158 nm separated dimer.
The two experimental distributions were constructed from one monomer movie,
which has 236 valid images with photon count of 1,100 ± 124 (mean ± SD). After
construction, the mean dimer movie photon count was 2,200, and the number
of images was 118. The simulations used the experimental monomer N and sm
distributions and the experimental 0 nm separation dimer background photon
distributions. The simulations were run for 1000 iterations, and in Fig. 2(A) the
counts were scaled to have the same amplitude as the experimental distributions
for comparison. The experimental distributions agree with the simulations.
For the experimental distributions, the Gaussian fits yield the means to be
sm = 105.2 nm and sd = 139.5 nm, and SDs to be ∆sm = 3.5 nm and ∆sd =
3.8 nm, respectively. For the simulated distributions, sm = 106.7 nm and sd
= 138.6 nm, and ∆sm = 3.8 nm and ∆sd = 4.8 nm, respectively. From the
information of Fig. 2(A), we can conclude that for an image with a photon
count above 2200 and sd above sm + ∆sd + ∆sm = 106.7 nm + 3.8 nm + 4.8
nm = 115.3 nm (the simulation results), it is a dimer. Our 158 nm separated
dimer has a mean sd of 139.5 nm > 115.3 nm, therefore the 158 nm separation
images are dimers.
The dimer SD measurement error, ∆sd, is affected by the total number
of detected photons in the image. Figure 2(B) shows the experimental and
simulated ∆sd dependence on photon counts for the 158 nm separation. ∆sd
6
Figure 2: Distinguishing dimers from monomers.
(A) Experimental (grey)
and simulated (empty bars) SD distributions of a constructed streptavidin-Cy3
monomer movie (left) and dimer movie (right) of separation 158 nm with the
same photon count distribution of N = 2200 ± 156 photons (mean ± SD).
The Gaussian fits to the experimental (solid line) and simulation data (dashed
lines) are shown. The simulations were for 1000 runs and the histograms were
scaled to be comparable to the experimental data. (B) Comparing experimental
(circles) and simulation (crosses) SD measurement error ∆sd of a 158 nm sep-
arated dimer at different detected photon counts. (C) Threshold photon count
line (spline line) for distinguishing dimers from monomers at different separa-
tions. The points on the line are simulation data; circles are experimental data
above the line, indicating differentiable dimers, and crosses are experimental
data below the line, indicating monomers or undifferentiable dimers.
7
decreases with increasing photon counts, and the experimental results agree well
with simulations (Fig. 2). There are a total of 12 monomer movies of different
photon counts used for the figure, and all experimental data in this article were
selected from these data. These monomer data are the same data as that in our
previous article ([13, ?]).
In order to differentiate dimers from monomers, Fig. 2(C) shows the thresh-
old photon count vs. separation diagram. At a certain separation, if an image's
photon counts is above the threshold value on the line, then it is a dimer; below
the line, it is not differentiable. The line was constructed by dimer simulations
of different separations and photon counts. At a particular separation, when
the condition of sd − sm = ∆sd + ∆sm was met, the photon count was the
threshold photon count for the separation. The 12 experimental data were used
to construct dimers each at a random separation. By using Fig. 2(A) distribu-
tions and Eq. 1 on each monomer data, the differentiable dimers were plotted as
circles, and undifferentiable dimers were plotted as crosses. The experimental
data agree with the diagram that all differentiable dimers were above the line
and undifferentiable dimers were below the line.
2.3 Dimer separation measurements
After determining that an image is a dimer, the next goal is to determine
the dimer separation. Figure 3 shows the experimental (circle) and simula-
tion (crosses) dimer PSF SD vs. separation diagram. The SD data are means
of dimer SD distributions, and the error bars are the SDs of the dimer SD
distributions. The experimental results at four different separations were con-
structed from a single monomer movie. The mean experimental dimer photon
count was 3050. Both experimental sd and ∆sd results agree with the simu-
lation results. For simplicity, the solid line is a fit to the simulation results
as sd = 0.001δ2 − 0.01δ + 118, and is independent of N . This diagram allows
for direct translation of dimer PSF SD to separation with known uncertainty
determined by the image's photon count, as described below.
2.4 Dimer separation measurement error
Each dimer separation measurement should have an associated precision. Fig-
ure 4 shows the dimer separation uncertainty ∆δ vs. separation δ for different
numbers of detected photons from 150 to 20,000. The separation uncertainty
∆δ was obtained from the dimer SD measurement error bars, as the ones shown
in Fig. 3 for the 3050 photon dimers. The ends of the dimer error bars were
extrapolated horizontally (horizontal grey lines) to meet with the sd vs. δ line.
The separation values at the two cross points on the sd vs. δ line (vertical grey
lines) marked the dimer's two separation measurement deviations. The differ-
ences between the cross point separation values and the mean separation are
the upper and lower errors for this separation measurement. The average of
the two errors was the dimer separation measurement error ∆δ for the specific
8
Figure 3: Dimer SD vs. separation diagram. The crosses are the mean SD
values for simulations of 1000 dimer images, and the circles are experimental
data at the separations of 0, 79, 158, and 237 nm. The error bars are the SDs
of the SD distributions. A fit to the simulation results is the solid line. The
horizontal grey lines off the extrema of the error bar of the 200 nm separation
data meet the diagram, measuring the dimer separation measurement errors by
the vertical grey lines.
9
Figure 4: Dimer separation measurement error, ∆δ vs. separation at different
photon counts of 150, 300, 400, 600, 2000, 10000, and 20,000 (slanted lines
from light to dark). The low separation termination points of the lines are
the separations below which dimers are not differentiable from monomers. The
horizontal dashed lines are the dimer separation measurement error by using
the centroid measurement method for dimer photon counts of 150 (top line)
and 20,000 (bottom line).
separation and photon count. A different sd vs. δ simulation was run for each
different photon count.
In Fig. 4, there are four properties to the lines: (1) On the left, each line
terminates at the dimer-monomer differentiation threshold for the photon count
[Eq. 1 and Fig. 2(C)]. (2) The line values decrease with increasing separation.
This is because that although the dimer SD error is approximately independent
of the separation, the sd vs. δ line flattens out at small separations, identical
SD error translates into larger separation measurement error at low separations.
As a result, the ∆δ line decreases with δ. (3) As the photon count increases,
∆δ decreases to the nanometer range. For 20,000 photons, ∆δ is ≈ 2.3 nm at
the 250 nm separation and 10 nm at the 40 nm separation. For 150 photons,
∆δ is 42 nm at the 250 nm separation and 29 nm at the 150 nm separation. (4)
Dimer separation measurement errors by using the centroid method are shown
as the two horizontal dashed lines for 150 and 20,000 photons. Comparing to the
centroid method, the precision of the SD measurement method is comparable
at high photon counts and higher at low photon counts.
The upper bound of dimer separation measurement errors by using centroid
measurements is the sum of the two centroid measurement errors, one for each
monomer at half of the detected photon count of the dimer. The centroid mea-
surement errors were calculated using the analytical expression in Ref.
[13, ?]
10
Figure 5: SMID dimer separation measurement temporal resolution diagram.
The line marks the threshold exposure time (threshold photon count) to dif-
ferentiate dimers from monomers. The data points were converted from points
in Fig. 2(C) using the conversion factor of 100 detected photons per 1 ms of
exposure time, and the line is the splined connection of the data. The SMID
timescales for all attainable subdiffraction dimer separation measurements are
milliseconds.
for the corresponding photon counts and background information, plus 40%
increase to accommodate for the discrepancy between experimental and ana-
lytical results [13, ?, 14]. For 150 photons (upper dashed line), the sum of the
two centroid measurement errors is (2 × 30.5 nm)× 1.4 = 85.3 nm. For 20,000
photons (lower dashed line), the sum of the two centroid errors is ≈ (2 × 1.6
nm) × 1.4 = 4.5 nm. The low separation end points of the centroid error lines
are the magnitude of the centroid dimer separation measurement errors, which
are 85.3 nm for 150 photons and 4.5 nm for 20,000 photons (the values agree
with another estimation in Ref. [15]).
2.5 Temporal resolution diagram
The single-image nature of the SD measurement method has determined that the
temporal resolution of the dimer separation measurements is determined by the
exposure time (or the camera's frame imaging speed, if slower). The minimum
dimer exposure time is determined by the minimum number of detected photons
required to differentiate dimers from monomers, as indicated in Fig. 2(C).
Figure 5 shows the temporal resolution vs. dimer separation diagram. In
constructing the diagram, the required photons in Fig. 2(C) were converted
into exposure times, assuming that (1) a typical laser intensity of 300 W/cm2
11
is used, (2) the detected dimer photon count is 100 photons per 1 ms exposure
time on average for our system, and (3) the number of dimer emitted photons
increases linearly with the exposure time. For the 20 ms exposure time, 2,000
photons will be detected and the dimer separation can be measured down to 75
nm with 18 nm precision. For the 100 ms exposure time, 10,000 photons will
be detected and the dimer separation can be measured down to 50 nm with 12
nm precision. Note that all the exposure times are in the milliseconds range.
3 Discussion
3.1 Advantages and limitations of SMID for subdiffrac-
tion dimer separation measurements
There are advantages and limitations to using SMID for subdiffraction dimer
separation measurements. The main advantages are that the method uses (1) a
standard single-molecule imaging system and (2) only milliseconds of imaging
time. The main limitations are that (i) the minimum measurable dimer separa-
tion only reaches 40 nm for a reasonable 200 ms exposure time. More photons
are needed to measure smaller separations; although, with increasing exposure
time the temporal resolution will be compromised; (ii) typical experimental 100
to 2000 photon counts per dimer image restricts the most useful range of mea-
surable dimer separations to be between 70 nm, the typical pixel size, and 250
nm, the Rayleigh criteria.
3.2 Comparing spatial resolution with the centroid mea-
surement method
In measuring subdiffraction dimer separations using the centroid method (SHRImP,
SHREC, PALM, and NALNS), two centroid measurements are performed (one
for each monomer, or one for the dimer and the other for the second monomer af-
ter the first monomer bleaches). Both methods reaches nanometer resolution in
measuring dimer separations. Comparing the spatial resolution of the two meth-
ods by using Fig. 4, there are following differences: (1) the precisions to dimer
separation measurements is higher for the SD method at low photon counts and
comparable at high photon counts; (2) at a fixed photon count, the centroid
separation measurement error is constant and reaches a lower measurable sep-
aration minimum, while the SD measurement error increases with decreasing
separation and terminates at a higher measurable minimum separation. If tem-
poral resolution is not a concern, future dimer subdiffraction measurements can
use Fig. 4 for choosing the appropriate method for optimal spatial resolution.
3.3 Dimer orientation effects
In this article, the two fluorophores are identical and both are located in the
imaging plane. Although the number of photon emitted per monomer differs
12
from image to image due to stochastic fluctuation [16], the emitted photon dis-
tributions of the two monomers are the same. In experiments, this is frequently
not the case and most dimers are randomly oriented in 3D, rather than both
being fixed at the imaging surface. This can (1) cause difference in sm between
the two molecules due to defocusing [17]; (2) for total internal reflection flu-
orescence (TIRF) imaging microscopy studies, the detected photon counts of
the defocused monomers will be less than that of the focused monomer due to
the decaying evanescent light intensity [18], and as a result, (3) the dimer SD
vs. separation relation in Fig. 3 will alter. Future separation measurements of
different dimer orientations should include appropriate orientation adjustments
in data analysis.
4 Conclusion
In this article we report a simple method to measure subdiffraction separations
between two identical fluorophores -- the SMID method -- that measures standard
deviation of dimer intensity profiles. This method is advantageous over existing
methods by its applicability to any fluorophores and the fast dynamic rate
in determining dimer separations by using only a single image of milliseconds
exposure times by using SIMA. Although the separation measurements only
have a lower limit of 40 nm for 20,000 photons, it will meet the demand of
many biological and nano-science applications that require dynamic separation
measurements in the range of order 70 nm to 250 nm.
5 Methods
5.1 Sample preparation and imaging
Streptavidin-Cy3 powder (0.04 nM, SA1010, Invitrogen, Carlsbad, CA) was
dissolved in 0.5X TBE buffer (45 mM Tris, 45 mM Boric Acid, 1 mM EDTA,
pH 7.0) to make the protein solution. Coverslips and fused-silica chips were
cleaned using oxygen plasma. Streptavidin-Cy3 molecules were immobilized on
fused-silica surfaces by sandwiching 6 µl of the protein solution between the
surface and a coverslip. The coverslip edges were then sealed with nail polish.
Single molecule imaging was performed using a Nikon Eclipse TE2000-S
inverted microscope (Nikon, Melville, NY) in combination with a Nikon 100X
objective (Nikon, 1.49 N.A., oil immersion). Samples were excited by prism-type
TIRF microscopy with a linearly polarized 532 nm laser line (I70C-SPECTRUM
Argon/Krypton laser, Coherent Inc., Santa Clara, CA) focused to a 40 µm × 20
µm region. The laser excitation was pulsed with illumination intervals between 1
ms and 500 ms and excitation intensities between 0.3 kW/cm2 and 2.6 kW/cm2.
By combining laser intensity and pulsing interval variations we obtained 50 to
3000 detected photons per monomer PSF. Images were captured by an iXon
back-illuminated electron multiplying charge coupled device (EMCCD) camera
(DV897ECS-BV, Andor Technology, Belfast, Northern Ireland). An additional
13
2X expansion lens was placed before the EMCCD, producing a pixel size of 79
nm. The excitation filter was 530 nm/10 nm and the emission filter was 580
nm/60 nm.
5.2 Data acquisition and selection
Typical movies were obtained by synchronizing the onset of camera exposure
with laser illumination for different intervals. The gain levels of the camera
were adjusted such that none of the pixels of a PSF reached the saturation level
of the camera. When single-molecule movies were obtained, streptavidin-Cy3
monomers were first selected in ImageJ (NIH, Bethesda, MD) by examining
the fluorescence time traces of the molecules for a single bleaching step [16].
For a selected monomer, the camera intensity values for 25 × 25 pixels centered
at the molecule were recorded. One of two 25 × 25 pixels monomer boxes were
shifted and added to the other box to create a dimer. For the summed dimer
images, the center 15 × 15 pixels were used for 2D Gaussian fitting of the 0
nm separated dimer, and the intensities from the peripheral pixels of the 0 nm
dimer box were used for experimental dimer background analysis. The center
20 × 20 pixels intensities were used for analysis of other dimer separations. By
selecting the center pixels for analysis, the non-overlapping background pixels
were avoided.
Before analysis, the camera intensity count of each pixel in an image was
converted into photon counts by using the camera-count to photon-count con-
version factor calibrated in our previous article [13, ?]. Then, the number of
detected photons, and the x-axis and y-axis dimer PSF SD were obtained. The
number of detected photons was obtained by subtracting the total photon count
of the image by the total photon count of the background; the two SD values
were parameters of a 2D Gaussian fit to the intensity profile of the image using
equation
f (x, y) = f0 exp
+ (cid:104)b(cid:105),
(2)
(cid:20)
− (x − x0)2
2s2
x
(cid:21)
− (y − y0)2
2s2
y
here f0 is the multiplication factor, sx and sy are SD values in x and y directions,
respectively, x0 and y0 are the centroid of the molecule, and (cid:104)b(cid:105) is the mean
background offset in photon counts.
The selected streptavidin-Cy3 monomers were further characterized to sat-
isfy the following conditions used for dimer SD analysis.
(1) No stage drift
detected by using centroid vs time measurements. Stage drift introduces ad-
ditional blur to each single-molecule PSF and thus affects the measured SD
values. (2) A minimum of 75 valid PSF images, each with a photon count, N ,
that fluctuated less than 20% from the experimental mean (cid:104)N(cid:105), of the monomer.
The PSF N count restriction is necessary for precise SD error analysis at N by
using a statistically sufficient number of PSFs with consistent N . (3) PSFs with
b ) larger than 2.5, where I0 is the peak PSF
photon count (total photon count minus (cid:104)b(cid:105)) and σ2
b is the background variance
in photons. (4) Mean monomer SD values, (cid:104)smx(cid:105) and (cid:104)smy(cid:105), obtained by Gaus-
signal-to-noise ratios (I0/(cid:112)I0 + σ2
14
sian fitting of the smx and smy distributions of all valid images did not differ by
more than 10 nm, or ± 5% of the mean SD value to minimize polarization effects
of Cy3. (5) The mean SD values (cid:104)smx(cid:105) and (cid:104)smy(cid:105) were between 95 nm and 135
nm to minimize defocusing effects. Images from the 12 monomer movies used
in this article satisfy the described restrictions.
5.3 Creating experimental dimer movies from monomer
movies
Experimental dimer images were constructed by adding all consecutive nonde-
generate two images of a monomer movie, with one of the two images shifted
0 - 3 pixels in the x direction. The final dimer images were reboxed to the
center 15 × 15 pixels for the 0 nm separation images and 20 × 20 pixels for
other separations. All experimental data in Fig. 2 were constructed from the 12
monomer data movies with a selected separation. For the experimental data in
Fig. 3, one monomer movie was used for all different separations.
5.4 Dimer simulations
To simulate dimers, we first generated monomers. Single-fluorescent-molecule
PSFs were generated using the Gaussian random number generator in MAT-
LAB. For simulations that later compare with experimental data (Figs. 2(A),
2(B), and 3), the simulated monomer's SD without the pixelation effect, sm0x,m0y,
was determined by the experimental means (cid:104)smx,my(cid:105) after subtracting for the
pixelation effect (Eq. 15 in Refs. [13] and [?]). The finite bandwidth of the emis-
sion filter was also taken into consideration by simulating each photon as being
drawn from a PSF whose width is varied according to a Gaussian distribution
centered about sm0x,m0y (with SD of 2 nm). The experimental N distribution
and the restriction that only photon count that fluctuated less than 20% from
the mean N were used. For simulations that do not compare with experimental
data (Figs. 2(C) and 4), the simulated monomer's SD was sm0x,m0y = 110 nm,
the standard deviation in photon count was 10% of the mean photon count and
again only randomly generated N that stayed within 20% of the mean was used
for images. The generated photons of each PSF were binned into 20 × 20 pixels
with a pixel size of 79 nm.
Using the simulated monomer movie, two nondegenerate consecutive images
were shifted and then sum to create dimers. After construction of dimer inten-
sity profile in photons, each photon in a pixel was converted into camera count
using Eq. 6 in Refs.
[13] and [?] with a conversion factor M of one. After
the dimer PSF construction in camera counts, background photon distributions
were added to the image. For Figs. 2(A), 2(B), and 3, random background pho-
tons at each pixel were generated using the corresponding experimental dimer
background distribution functions obtained by using the 0 nm separation dimer.
For Figs. 2(C) and 4, the background photon distributions had the mean values
of the experimental 0 nm dimer separation of all 12 monomer data: the mean
15
dimer background photon counts was 1.8 photons, and the mean dimer back-
ground standard deviation was 1.7 photons. From each final dimer image, the
center 20 × 20 pixels were used for 2D Gaussian analysis. For each simulated
dimer datum, 1000 iterations were performed.
Acknowledgements
Michael C. DeSantis is supported by a National Institutes of Health predoctoral
fellowship awarded under 5T90 DA022871. Shawn H. DeCenzo is supported
by an US Dept of Education "Graduate Assistance in Areas of National Need"
award under P200A090267.
16
|
1012.4782 | 1 | 1012 | 2010-12-21T20:18:02 | Fluctuation Pressure Assisted Ejection of DNA From Bacteriophage | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.BM"
] | The role of thermal pressure fluctuation excited within tightly packaged DNA prior to ejection from protein capsid shells is discussed in a model calculation. At equilibrium before ejection we assume the DNA is folded many times into a bundle of parallel segments that forms an equilibrium conformation at minimum free energy, which presses tightly against internal capsid walls. Using a canonical ensemble at temperature T we calculate internal pressure fluctuations against a slowly moving or static capsid mantle for an elastic continuum model of the folded DNA bundle. It is found that fluctuating pressure on the capsid internal wall from thermal excitation of longitudinal acoustic vibrations in the bundle may have root-mean-square values which are several tens of atmospheres for typically small phage dimensions. Comparisons are given with measured data on three mutants of lambda phage with different base pair lengths and total genome ejection pressures. | physics.bio-ph | physics | Fluctuation Pressure Assisted Ejection of DNA From
Bacteriophage
Michael J. Harrison
Department of Physics and Astronomy, Michigan State University
East Lansing, MI 48824-2320, USA
Abstract. The role of thermal pressure fluctuation excited within tightly packaged DNA prior to ejection
from protein capsid shells is discussed in a model calculation. At equilibrium before ejection we assume
the DNA is folded many times into a bundle of parallel segments that forms an equilibrium conformation
at minimum free energy, which presses tightly against internal capsid walls. Using a canonical ensemble
at temperature T we calculate internal pressure fluctuations against a slowly moving or static capsid
mantle for an elastic continuum model of the folded DNA bundle. It is found that fluctuating pressure on
the capsid internal wall from thermal excitation of longitudinal acoustic vibrations in the bundle may have
root-mean-square values which are several tens of atmospheres for typically small phage dimensions.
Comparisons are given with measured data on three mutants of lambda phage with different base pair
lengths and total genome ejection pressures.
Keywords: Fluctuation assisted ejection of DNA, internal pressure on capsid mantle
PACS numbers: 87.16dj, 87.16.-b, 87.15.ad, 87.18 Tt.
INTRODUCTION
It has been emphasized that thermal fluctuations away from equilibrium in small systems assume
increasingly important roles as the system size becomes ever smaller and approaches molecular
dimensions [1]. In the case of viral entities, where the phage size may very considerably [2,3], thermal
pressure fluctuations may become exerted internally by confined DNA that is already tightly packged in
an equilibrium conformation within a capsid mantle. Such pressure fluctuations can be expected to play a
signigicant role in the initial packaging and subsequent dynamics of DNA ejection from bacteriophage
[4,5,6,7,8] when the virus engages a host cell surface receptor and transfers its DNA into the cell.
Geometrical and topological constraints on the structure and size of viral capsids were discussed in an
early paper by Caspar and Klug [9] almost fifty years ago. More recent work [2,10,11] has discussed
highly symmetric capsid structure in the form of nanometer-sized protein shells. And viral infectivity has
been related to DNA length and capsid size [12]. It has also been concluded that very large conformal
changes may occur in some capsid shells when their viral genomes become tightly packaged [13]. We
shall adopt a simpler model of viral capsids that contain packaged genomes than has generally been
discussed [14].
CALCULATION OF MODEL ENCAPSIDATED DNA AND CAPSID
Consider a circular cylindrical bundle of folded double-stranded DNA bacteriophage segments with
1
cross-section diameter L and equal segment lengths L, packaged coaxially and situated symmetrically
within a cylindrical capsid cavity of diameter (L+2h) and length L, where h is the thickness of the empty
annular region surrounding the cylindrical DNA bundle The capsid cavity is then a circular cylinder of
diameter (L+2h) and length L. If we assume that the capsid cavity has twice the volume [15] of the DNA
bundle, then geometric considerations lead to the relation h=L / [2(1+(cid:1)2)] = 0.2071/L, so that (L+2h)
=1.414 L is the model capsid diameter. Capsids are nanometer sized in general magnitude [10], but vary
greatly [2,11] depending specifically on the DNA genomes they carry within. We shall adopt
representative capsid diameters near 50 nm [5,6,21] in our calculation of fluctuation pressures exerted
between a vibrating DNA bundle and the capsid shell to which it is tethered.
Our discussion of the role of fluctuations rests on a picture of a tightly packaged encapsidated DNA
bundle that acquires its thermal equilibrium structure through variational minimization of the free energy
[7] of the capsid-DNA system. Fluctuations of the DNA bundle about its equilibrium conformation takes
place in the form of thermally excited longitudinal acoustic vibrations. These vibratory excitations
in the DNA bundle will be regarded as taking place with the bundle bounded tightly by capsid walls which
themselves continue to maintain positions close to their thermal equilibrium geometry, and do not
significantly participate in thermal excitation on the same scale as the DNA segment bundles. The bundles
are assumed to be connected fixedly at both ends to their capsid mantles.
In order to calculate the thermal fluctuations in pressure at the locations where DNA bundles are
attached to their capsid shells we shall approximate a bundle of folded DNA segments by a continuum
elastic rod [16, 17]. Longitudinal acoustic vibrations of such a rod fixed at both ends to a static capsid
inner surface depend on the longitudinal sound velocity within the rod, which in turn depends on the rod’s
macroscopic mass density and elastic properties. These two quantities enter the speed of longitudinal
sound according to v = (cid:1) (Y/(cid:2)) , where Y is the bulk modulus and (cid:2) is the mass density of the DNA
bundle. We can relate these quantities to the microscopic force constant and total genome mass of single
double-stranded DNA molecules which have become folded into bundles of length L and circular cross-
section area (cid:3)(L/2)2 .The elastic response of individual double-stranded DNA molecules has been
measured [18]. And elastic constants have been used to calculate phase velocities of sound waves [19].
Brillouin scattering has been used to determine the longitudinal velocity of sound in B-DNA fibers over a
quarter century ago [20] with the result v = 1.9 km/s, which we shall adopt for the present calculation.
The longitudinal sound velocity in a bundle of folded DNA segments regarded as a continuum coincides
with the sound velocity of a single fiber since Y and (cid:2) have the same ratio for the bundle as for a single
constituent DNA molecule.
We introduce a coordinate system with its origin on the cylindrical bundle’s axis where it connects
with a static capsid mantle. The x-axis coincides with the bundle’s axis and extends through it to the other
end of the tightly packaged bundle on the static capsid’s internal surface a distance L away. Then with
y(x,t) representing a general longitudinal displacement field, for t > 0 in a continuum bundle we have
y(x,t) = (cid:4)n (cid:5)n(t) sin(n(cid:3)x / 2L) , 1.
which obeys boundary conditions y(0,t) = 0 and y(L,t) = 0 , where n are even integers and (cid:5)n(t) are normal
coordinates for longitudinal standing wave motion in the folded DNA bundle, regarded as a continuum.
2.
We take y(x,t) = 0 for t <0. The total hamiltonian H of the wave system is given by the sum of its kinetic
and potential energy:
L
H = (cid:6)0 dx [s(cid:2)/2] [((cid:7)y/(cid:7)t)2 + v2 ((cid:7)y/(cid:7)x)2 ] , 2.
.
where s = (cid:3)(L/2)2 is the cross-sectional area of the packaged DNA bundle, (cid:2) is the mass density and
v is the longitudinal sound velocity.
We substitute Eq.(1) into Eq.(2) and obtain
H = (s(cid:2)L/4) (cid:4)n (cid:5)n
The total energy H depends only quadratically on the (cid:5)n(t) and their time derivatives. If we now
adopt a canonical ensemble to obtain the thermal average <H> at temperature T, each quadratic
term in H has its equipartition average value kT/2 for the ensemble. We obtain:
´2 + ( v2(cid:2)s(cid:3)2/16L) (cid:4)n (cid:5)n2 n2 . 3.
‹ (cid:5)n2› = [8LkT/ (n2v2(cid:2)s(cid:3)2)] , 4.
where the brackets denote the thermal average.
At the closed end pressure antinode, x = 0, the pressure fluctuation against the constraining internal
wall of the capsid is
(cid:8)p(0,t) = - v2 (cid:2) ((cid:7)y/(cid:7)x x=0) 5.
for a displacement field y(x,t). We now take the ensemble average ‹ (cid:8)p2 › and obtain
‹(cid:8)p2 › = [v4(cid:2)2 (cid:3)2 /(4L2)] (cid:4)nm nm ‹ (cid:5)n(cid:5)m › . 6.
But ‹(cid:5)n(cid:5)m › = ‹ (cid:5)n 2› (cid:9)nm since in thermal equilibrium the normal coordinates (cid:5)n are uncorrelated with
respect to their time dependence. A single sum results:
‹ (cid:8)p2 › = [ v4(cid:2)2 (cid:3)2 /(4L2)] (cid:4)n n2 ‹ (cid:5)n2 › . 7.
Substituting Eq.(4) into Eq.(7) we obtain
‹ (cid:8)p2 › = [2v2(cid:2)kT/(Ls)] (cid:4)n 1 . 8.
There must be cut-off limits in Eq.(8) reflecting the requirement that the continuum standing waves
entering Eq.(1) have wavelengths that neither exceed the bending persistence length of DNA, 50 nm
[21,22,23], and therefore have coherence over the segment length L, nor are shorter than several base pair
separations which characterize discrete molecular structure.
3.
N 1 = N/2 = 1 9.
For any even integer n the corresponding standing wavelength is given by (cid:10) = 4L/n. Thus we require
an even integer N that leads to a wavelength (cid:10)N = 4L/N that is no smaller than twice the bundle length L in
order to have a lowest coordinate value node at x=L, which is consistent with a bundle segment length L
that is equal or less than the bend persistence length. From 4L/N (cid:11) 2L we have N=2 . From L (cid:12) bend
persistence length we require L(cid:12) 50 nm. Cited values of total capsid volume, Vcapsid [6], and
corresponding fractions of capsid volume packaged with genomes for three mutants of lambda phage [6]
enable us to calculate the respective genome volumes Vgenome = V(cid:10)cI60, VEMBL3, and V(cid:10)b221. And from these
we obtain for our surrogate cylindrical capsid model the genome volumes (Ls) and finally L = 40.38 nm,
38.34 nm and 37.13 nm for folded segment lengths of the three respective lambda phage mutants (cid:10)c160,
EMBL3, and (cid:10)b221. These values of segment length are all less than the DNA persistence length.
In the Eq,(8) summation we then have N=2, and
(cid:4)n
with the thermal fluctuation noise pressure on the capsid mantle then given by:
‹ (cid:8)p2 ›= [2v2 (cid:2) kT / (Ls)] . 10.
Taking M=(cid:2)(Ls) as the total genome mass within the capsid, we define the root-mean-square
fluctuating pressure magnitudes Prms (cid:13) (cid:1) (‹(cid:8)p2›) and obtain
Prms (cid:13) (cid:1) [2v2MkT/(Vgenome)2] , 11.
where Vgenome (cid:13) (Ls) now represents the genome volume for the respective mutant .
Since M is proportional to the genome length, Eq.(11) suggests that the fluctuating
pressure magnitude at temperature T should be proportional to the square root of the genome length and
inversely proportional to the volume it occupies within the capsid.
.
It has been observed that exposure to increasing osmotic pressure difference between the inside and
outside of a lambda bacteriophage capsid can suppress the ejection of the viral genome into a bacterial
cytoplasm with an osmotic pressure of several atmospheres [24,25]. These measurements in vitro have
shown that a sufficiently large osmotic pressure from outside of a lambda phage caspid can provide a
resisting force that balances the internal ejection force exerted by a tightly fitting DNA genome. Relevant
data given for three lambda phage mutants [6] using a parameter-free model includes their respective
base pair lengths.in addition to total capsid volume and respective fractions of capsid volume occupied by
mutant genomes.The osmotic pressures required to completely inhibit the ejection of mutant viral
genomes when the capsid is appropriately stimulated are presented [6] as well as lesser osmotic pressures
needed to partially prevent DNA ejection. In particular, the published figures indicate complete blockage
of 48.5 kbp (cid:10)cI60 ejection at 30 atm, complete blockage of 41.5 kbp EMBL3 ejection at 17 atm, and
complete blockage of 37.7 kbp (cid:10)b221 ejection at 12.5 atm osmotic counterpressures.
4.
Following [6] we shall assume these counterpressures match those arising internally within the
capsid, at least in part from tightly packaged genomes.
We have calculated the genome volumes Vgenome from the cited data [6], and the masses M of the
three mutants (cid:10)cI60, EMBL3, and (cid:10)b221 have been related to their base pair lengths using an average
base pair mass of 1.021 x 10-21 grams. Using a velocity v = 1.9 km/s [20] to evaluate Prms in Eq.(11) we
obtain for T = 310 K:
Prms,(cid:10)cI60 = 7.468 atm,
Prms,EMBL3 = 8.062 atm,
Prms,(cid:10)b221 = 8.467 atm.
The thermal fluctuation pressure component acquires greater values for the less voluminous genome, in
spite of its lesser mass and length in base pairs. However, the dependence of total ejection forces on
genome length has also been noted in bacteriophage lambda [6], and is indeed greater for longer lengths.
In consequence of the above calculations we conjecture that time-dependent fluctuation pressures
leading to Prms act like a trigger that assists the internal forces exerted on tightly packaged DNA to eject
the genome in a sequence of steps that follow stimulation by encounter with a receptor on a host cell.
I wish to thank Professor Lisa Lapidus for several stimulating conversations.
REFERENCES
1. C. Bustamente, J. Liphardt, and F. Ritort, Physics Today 58 (7), July 2005, p.43.
2. R.V. Mannige and C. L. Brooks III, PNAS 106 (21), May 26, 2009, p.8531.
3. R. Mannige and C. Brooks III, Phys. Rev. E 77, 051902.
4. M. M. Inamdar, W.M. Gelbart, and R. Phillips, Biophys. J. 91, July 2006, p.411.
5. P. K.Purohit, J. Kondev, and R. Phillips, PNAS 100 (6), March 18, 2003, p.3173.
6. P. Grayson, A. Evilevitch, M. M. Inamdar, P. K. Purohit, W. M. Gelbart, C. M. Knobler, and
R. Phillips, Virology 348, (2006), p.430
7. S. Tzlil, J. T. Kindt, W. M. Gelbart, and A. Ben-Shaul, Biophys. J. 84 (3), March 2003,
p.1616.
8. L. Ponchon, S, Mangenot, P. Boulanger, and L. Letellier, Biochim. Biophys. Acta 1724 (3),
Aug. 5, 2005, p. 255.
9. D. L. D. Caspar and A. Klug, Cold Spring Harbor Symposium on Quantitative Biology 27,
(1962), p.1
10 M. M. Gibbons and W. S. Klug, J.Mater.Sci 42,(2007), p.8995.
11. O. M. Elrad and M. F. Hagan, Nano. Lett. 8 (11), (2008), p.3850.
12..A. Evilevitch, Q. Rev. Biophys. 40, (2007), Cambridge Univ. Press, p.327.
13. I.Gertsman, L.Gan, M.Guttman, K.Lee, J.A Speir, R.L.Duda, R.W.Hendrix,
E.A.Komives, and J.E.Johnson, Nature 458, 2 April 2009, p.646.
5.
14.W. Lucas and D.M. Knipe, Encycl. Life Sci., 2002, Macmillan Pub. Ltd.
15. M.Sun and P.Serwer, Biophys. J. 72, February 1997, p.958.
16. B. Eslami-Mossallam and M.R. Ejtehadi, Phys. Rev.E 80, (2009), p. 011919.
17. R.S. Manning, J.H. Maddocks, and J.D. Kahn, J. Chem. Phys.105 (1996), p.5626.
18. S.B. Smith, Y. Cui, and C. Bustamante, Science 271, 9 February 1996, p. 795.
19. T.C. Bishop and O.O. Zhmudsky, arXiv: physics/0108008v1 [physics. bio-ph] 6 August
2001.
20. M.B. Hakim, S.M. Lindsay, and J. Powell, Biopolymers 23, (1984), p.1185.
21 Robert H. Austin, James P. Brody, Edward C. Cox, Thomas Duke and Wayne Volkmuth,
Physics Today 50, February 1997, p. 32.
22. Moukhtar J, Fontaine E, Faivre-Moskalenko C, Arneodo A. Phys Rev Lett. 98 (17), (2007),
p. 178101.
23. John A. Schellman and Stephen C. Harvey, Biophys. Chem, 55 (1995), p. 95.
24. Alex Evilevitch, Laurence Lavelle, Charles M. Knobler, Eric Raspaud, and William M.
Gelbart, PNAS 100 (16), (2003), p. 9292.
25. A. Evilevitch, J.W. Gober, M. Phillips, C.M. Knobler and W.M. Gelbart, Biophys.J. 88 (1),
(2005), p.751.
.
6.
|
1912.05735 | 1 | 1912 | 2019-12-12T02:16:12 | Growth and Elasticity of Mechanically-Created Neurites | [
"physics.bio-ph"
] | Working in the framework of morphoelasticity, we develop a model of neurite growth in response to elastic deformation. We decompose the applied stretch into an elastic component and a growth component, and adopt an observationally-motivated model for the growth law. We then compute the best-fit model parameters by fitting to force-extension curves from measurements of constant-speed uniaxial deformations of mechanically-induced neurites of rat hippocampal neurons. We find a time constant for the growth law of 0.009~s$^{-1}$, similar to the diffusion rate of actin in a cell. Our results characterize the kinematics of neurite growth and establish new limits on the growth rate of neurites. | physics.bio-ph | physics | Growth and Elasticity of Mechanically-Created Neurites
Madeleine Anthonisen∗ and Peter Grutter
Physics Department, McGill University, 3600 rue Universit´e,
Montr´eal, Quebec, Canada H3A 2T8
(Dated: December 13, 2019)
Working in the framework of morphoelasticity, we develop a model of neurite growth in response
to elastic deformation. We decompose the applied stretch into an elastic component and a growth
component, and adopt an observationally-motivated model for the growth law. We then compute
the best-fit model parameters by fitting to force-extension curves from measurements of constant-
speed uniaxial deformations of mechanically-induced neurites of rat hippocampal neurons. We find
a time constant for the growth law of 0.009 s−1, similar to the diffusion rate of actin in a cell. Our
results characterize the kinematics of neurite growth and establish new limits on the growth rate of
neurites.
PACS numbers:
I.
INTRODUCTION
Neurons are cells specialized for information process-
ing. They have long, tube-like extensions of diameter
∼ 1 µm, termed neurites, that connect the cell bodies to
other neurons and enable the exchange of information via
chemical and electrical signals. Neurites are classified as
axons, signal transmitters, or dendrites, signal receivers.
Mechanical elongation of neurites has been widely
studied (see e.g.
[1, 2] for reviews). These experi-
ments have lead to the identification of tension as a
driver of neurite growth and development [3 -- 6]; e.g. , "a
pulled axon grows as though the nerve cell contained tele-
scopic machinery prefabricated for elongation" [4]. Re-
cent work, [7 -- 9], has shown that this telescopic growth
also occurs in axon-like structures initiated from parent
axons or dendrites, see Fig. 1. However, the mechanisms
responsible for this surprising mass-accretion and the role
of tension in limiting this process remain outstanding
mysteries [2 -- 4, 10].
A natural question is the extent to which elongation
can be attributed to growth, i.e. the addition of new cel-
lular material, versus elastic stretching of existing con-
stituents. In this paper, we answer this question.
Working in the theoretical framework of morphoelas-
ticity described in [11 -- 13], we relate the experimental
force-extension curves of neurites to the material param-
eters that describe their elastic response to deformations
and the rate as well as the rates of material added due
to growth.
In our experiments, we measure the force-extension re-
lationship of new neurites using flexible, calibrated glass
micropipettes as illustrated in Fig 1. The micropipette is
connected to the cell by a bead that is chemically func-
tionalized to induce a stable mechanical contact with the
parent axon or dendrite. When the bead-pipette complex
∗ [email protected]
is displaced relative to the cell, the growth of an auxil-
iary structure, a new neurite, is induced. We elongate
the neurite while simultaneously measuring its tension
by optically tracking the beaded tip. By calibrating the
spring constant of the pipette, we can convert this de-
flection to a force. We extend our neurites at a constant
rate, in contrast with other experiments, e.g.
[14, 15],
in which a stretch is applied in one step and maintained
constant for the duration of the experiment.
We derive an expression for the force-extension re-
lationship of neurites that incorporates an exponential
growth law. We fit experimental data to find the time
constant for exponential mass addition, which is close to
the rate of actin diffusion along a pulled neurite. We find
that the time constants for different pulling experiments
are positively skewed and follow a lognormal distribution.
This puts new limits on the mass accretion of axon-like
extensions.
The structure of this paper is as follows: In Section II
we review the principles of morphoelastic theory and in-
troduce a model to characterize the kinematics of neurite
growth. We show the contributions of elastic stretching
and growth stretching to neurite deformations in Sec-
tion III. In Section IV we justify assumptions used in II
with experiments, summarizing this paper in Section V.
II. A MODEL OF GROWTH WITH ELASTIC
DEFORMATION
A general deformation can be characterized by a ge-
ometric stretch λ, defined as the relative change in the
length of the neurite to the initial length, i.e. λ ≡ l/L,
with l = l(t) and L = l(t = 0) the length of the neurite
at time t and the initial length respectively.
We work within the framework of morphoelasticity, in
which the geometric stretch is the product of an elastic
term λe and a growth term λg [10 -- 13]:
λ = λeλg.
(1)
9
1
0
2
c
e
D
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
3
7
5
0
.
2
1
9
1
:
v
i
X
r
a
From Eq. 3, we can obtain the elastic Piola stress P e,
which can be used to obtain the Piola stress P , defined
as [11]:
2
This can be expanded in terms of λ and λg as [11],
.
(4)
P ≡ ∂Ψ
∂λ
(cid:21)(cid:34)
(cid:20)
(cid:19)2(cid:35)
(cid:18) λg
λ
P =
2
λg
c1 + c2
λg
λ
λ
λg −
.
(5)
The Piola stress captures the stress across the neurite. It
can be related directly to an external loading force F on
a neurite through the principle of virtual work [11, 16] to
give
F = P A,
(6)
where A is the cross-sectional area.
The radial dimension of the neurite is a proxy for
growth through addition of new material [11].
In the
absence of radial thickening, the transverse stretch λ⊥,
that is the ratio between the neurite radius at a time t,
r(t), and the initial radius R, is defined via the elastic
stretch,
(cid:18) 1
(cid:19)1/2
λe
λ⊥ =
(cid:18) λg
(cid:19)
λ
.
(7)
The cross sectional area of a neurite can also be written
in terms of λ, λg:
A = πR2
.
(8)
This allows the force F to be written in terms of λ and
λg, and the parameters c1 and c2, as
λg −
(cid:18) λg
(cid:19)2(cid:35)
(cid:21)(cid:34)
c1 + c2
2πR2
λg
λ
(cid:20)
F =
(9)
λ
λ
λ
.
From this one can compute not only the force at a given
deformation, but also the full time evolution F (t).
Indeed, axons under axial tension will gradually in-
crease in mass to recover some homeostatic equilibrium
state, that is the axon has been observed to have some
inherent tension [11, 12, 15, 17]. Motivated by these ob-
servations, here we adopt a growth model in which the
growth rate depends on the axial stress of the neurite.
If the neurite is perturbed from that state, mass will be
added so it can recover a particular "baseline" stress. To
model growth, we assume a functional form of λg based
on experimental observations.
A. An exponential growth law
.
Here we consider a growth law that states exponential
growth or resorption occurs until a homeostatic stress is
FIG. 1: Sketch of a neurite pulling experiment. At
time t1, the micropipette tip is fixed to the bead that
contacts the axon of a neuron. At time t2, the
micropipette-bead complex has been moved by an
amount v(t2 − t1) relative to the axon, inducing the
growth of the neurite. The tension in the neurite is
captured by recording the deflection of the
micropipette, d, and calibrating its stiffness constant to
convert the bending into a force.
We assume that stress, defined as the axial force per unit
area of the neurite cross-section, is only caused by elastic
deformation, a commonly-adopted assumption in growth
theories [11, 12]. We further assume that the elastic part
of the deformation is incompressible (volume preserving)
so that only the growth component will add volume.
In continuum mechanics, the stress-stretch relationship
of soft materials is determined experimentally and can be
derived from the strain energy density function of the de-
formation process. There are many different models to
describe the strain energy density, we find that neurites
are best described by the so-called Mooney-Rivlin model
(this choice is justified in Sec. IV 1). Under the assump-
tion of constant volume from the elastic deformation, the
Mooney-Rivlin model has the form [11]
(2)
1 λe2
2 = λe2
Ψ = c1[I e
1 − 3] + c2[I e
2 − 3],
1 = λe2
where c1, c2 are material constants and I e
2 +
3 and I e
3 , I e
λe2
3 = λ1λ2λ3 = 1
are elastic invariants in terms of the elastic stretches.
Note i = 1, 2, 3 label the spatial dimensions of the defor-
mation. In the case of incompressible uniaxial extension,
1 = λe
the neurite is pulled along a single dimension, λe
and λe
2 + λe2
3 + λe2
1 + λe2
3 = 1/(λe)1/2.
2 = λe
2 λe2
1 λe2
We can re-write the Mooney-Rivlin strain energy den-
sity function in terms of the elastic stretch and then repa-
rameterize it in terms of λ, λg [11]:
(cid:35)
(cid:34)
Ψ(λ, λg) = c1
+ 2
λg
λ
− 3
+c2
2
λ
λg +
(cid:34)(cid:18) λ
(cid:19)2
λg
(cid:35)
(cid:19)2 − 3
(cid:18) λg
λ
(3)
recovered. This model has been used to describe axonal
growth in [13]. Work from our lab indicates that the
trajectory of actin (one of the principal constituents of
neurites) entering the pulled neurite follows an exponen-
tial relation [18].
We consider a law of the form,
∂λg
∂t
= kλg (λe − λ∗) Θ(λe − λ∗),
(10)
where k is a constant, λ∗ is a critical stretch associated
with the homeostatic stress σ∗ that the neurite is trying
to recover, and Θ(x) is a Heaviside theta function: Θ(x)
is 1 for x > 0 and 0 otherwise.
For λe > λ∗, Eq. 10 is
∂λg
∂t
= k(λ − λgλ∗).
(11)
Solving for the functional form of λg(t) with the initial
condition λg(0) = 1, we obtain
λg(t) =
k(L + vt) + v(e−kt − 1)
kL
,
(12)
where we have used λ = (L + vt)/L, v is the (constant)
speed at which neurites are extended. We set λ∗ = 1, an
assumption we justify in a later section.
In what follows, we will experimentally measure force-
extension curves, and from this obtain the best-fit values
of the parameters k, c1 and c2 . Example experimentally-
obtained force-extension curves are shown in Fig. 2. Here
we have fit the curves to the functional form Eq. 9 with
Eq. 12 inserted.
B. Material parameters
We fit
twenty-one experimentally-obtained force-
extension curves with Eq. 9 and Eq. 12 as described in
Section II A. The growth rate parameter k is shown in
Fig. 3a for different pull speeds. We find the mean value
of k to be 0.009 s−1, see Table I. The data is skewed
to large values of k, with a SD of 0.01 s−1. The mean
is of a similar order of magnitude as the time constant
describing the movement of actin along pulled neurites
found in [18], which is 0.001 s−1. The lower bound of
our data matches the value found in [10] for the axonal
growth rate, 2×10−5 s−1.
In Fig. 3b, we plot the cu-
mulative density function of k values and show that it is
well-characterized by a lognormal distribution with pa-
rameters µ = −5.31± 0.01 and σ = 1.52± 0.01 (standard
errors from fit). This is confirmed by a Chi-Square good-
ness of fit test at the 5% significance level. Here µ and
σ are the mean and standard deviation of the natural
logarithm of k. The skewness, which captures the asym-
metry of the distribution, can be obtained from σ and is
33. Although this is significantly higher than 0.9, which
is the sample skewness obtained from the definition of
3
FIG. 2: Examples of force-extension curves of induced
neurites. Each curve (red) is from a single pulling
experiment and can be due to 1 or more
mechanically-induced neurites. The force is the loading
force applied axially to the neurite(s) as measured by
the bending of the micropipette and the stretch λ is the
length the neurites have been pulled relative to their
initial length. The shaded regions of the curves
represent the measurement error and are calculated as
described in [9]. The dashed blue lines are fits of Eq. 9.
Pearson's moment coefficient of skewness, the two mea-
sures of skewness are consistent in describing the data as
moderately to highly skewed.
We investigate the mass addition of new neurites as
they are pulled, and find that k is independent of pull
speed. This is confirmed by the Kruskal-Wallis test,
which tests whether samples, grouped by pull speed, are
drawn from the same distribution. This isconsistent with
previous work, [9, 19], which found that neither the force-
extension relationships nor cross-sectional areas of neu-
rites depend on mechanical pull speed. This is surprising
given a greater than 10-fold increase in pull speed. In-
terpreting k as an exponential growth rate, it is reason-
able that it should be the same across pull speeds as it
could be constrained either by the properties of the cell
(the speed with which it can manufacture and transport
certain constituents) or by physical properties such as
diffusion.
An open question is what causes the large variance
in the k-values we extract, assuming this is not a fea-
ture that vanishes with more statistics. We postulate
that this is related to the mechanisms underlying mass
addition. Cell growth is typically modelled as a combina-
tion of active and passive processes. As mentioned, our
timescales are consistent with those reported for actin
in [18]. Motivated by this, we consider the behaviour of
actin filaments and myosin motors (the proteins respon-
sible for polymerizing actin filaments) in a simplistic 1D
diffusion model. We calculate the time it takes actin
filaments and myosin motors, which together have an ef-
fective diffusion coefficient of D = 0.01 µm2s−1 [20], to
diffuse along a length equivalent to L for each neurite.
We find a mean rate of actin diffusion of 0.01 s−1 across
neurites with a standard deviation of 0.01 s−1 that ex-
actly matches our results for k. This is suggestive of an
important role for actin diffusion in the addition of mass
to new neurites.
It also indicates that the variance in
our reported values of k could be explained by different
initial neurite lengths.
(a)
(b)
FIG. 3: (a) Time constants characterizing neurite
growth plotted versus pull speed. This is the fit
parameter k from Eqs. 9&12. The Kruskal-Wallis test
confirms the hypothesis that k values for each pull
speed were all drawn from the same distribution. The
mean value of k across pull speeds is 0.009 with a
standard deviation of 0.01. (b) Cumulative density
function for k values (black line), fit with a lognormal
distribution (red line) that captures the skewness of the
data. Parameters for the lognormal distribution are
µ = −5.31 and σ = 1.52.
4
TABLE I: Fit parameters from Eqs. 9&12.
Parameter
k
c1
c2
Mean
0.009 s−1
204 kPa
-13 kPa
Standard deviation
0.01 s−1
385 kPa
302 kPa
The other material parameters, which characterize the
neurite response to elastic stretch in the Mooney-Rivlin
model, are c1 = 204 ± 385 kPa and c2 = −13 ± 302 kPa
(mean±SD). These are summarized in Table I. Unlike
k, the physical reason these parameters vary over orders
of magnitude and, in the case of c2, by sign is unclear.
The Mooney-Rivlin model and the Neo-Hookean model,
which is a specific case of the MR Eq. 2 [21], have been
widely used to model other types of brain tissue [22 -- 24],
including axons [10]. While this family of models is suc-
cesful in describing brain tissue under diverse experimen-
tal conditions, these other works also contain the feature
that the material parameters are phenomenological and
vary over orders of magnitude [11]. In [25], it is shown
analytically that the Mooney-Rivlin model, applied with
different relative parameter signs, captures experimen-
tal trends observed in soft biological tissues under both
shear and compression conditions. Our results add to the
experimental evidence that the Mooney-Rivlin model is
suitable to describe brain tissue. This indicates that the
mechanical behaviour of newly induced neurites is very
similar to that of naturally grown axons. While we lack a
satisfying mechanistic interpretation of these parameters,
quantifying single-cell behaviour with the Mooney-Rivlin
model is an important step to multiscale modelling of the
brain which could in turn ultimately clarify the physical
significance of these results.
III. NEURITE GROWTH
In Fig. 4 we plot the component of neurite stretch
that is due to added mass, λg, versus time. Each curve
represents one pulling experiment and we plot Eq. 12
with the k corresponding to the force-extension curve.
These curves show the rapidity with which new mate-
rial is added to neurites as they are being pulled. They
represent the volume growth of neurites.
With the exception of v = 0.05 µm/s, the λg curves
take on a range of values for a single pull speed. As dis-
cussed in Section II B, this could be due to the different
initial lengths of neurites, which for v = 0.05 µm/s were
in the 52nd, 90th and 95th percentiles of the data respec-
tively; this is consistent with the idea that mass addition
is less extreme if the diffusion path for material along the
neurite is longer.
In Fig. 5, we show the different contributions to neu-
rite deformation, λ, from the elastic stretch λe and the
growth stretch λg. We see that for all the pull speeds,
initially there is no growth and the entire deformation
5
FIG. 4: Volume growth versus time for different pull
speeds. In each plot, the differing curves correspond to
differing initial neurite length.
is elastic, λ(0) = λe. With time, the neurite grows ac-
cording to Eq. 11 and we see λg → λ and λe → 1. This
reflects the the fact that the elastic stretch is evolving
to recover a homeostatic equilibrium value.
Interest-
ingly, after ∼175 s, the elastic stretches for all speeds
collapse to the same values approaching 1. This indi-
cates that for the range of speeds studied, there is a
point past which the elastic response is independent of
pull speed. This timescale is associated with the mech-
anisms of mass addition: As the cell has more time to
add mass to the neurite, the stretch response of the ex-
isting neurite-components becomes less significant. For
each speed, there is a time at which λe = λg. This time
is inversely proportional to the pull speed meaning neu-
rites are very flexible in their responses to deformation
and able to accommodate loading forces applied over a
large range of speeds.
We see from the form of Eq. 12 that λg tracks λ with
a pull-speed-dependant exponential term. We pull at
extremely fast speeds relative to physiological ones and
relative to other pulling experiments in the literature.
Assuming our growth model, Eq. 12, is realistic, these
results show new limits of mass addition for axon-like
extensions.
IV. RESULTS FROM EXPERIMENTS
Before we conclude, we address and experimentally
verify the validity of the assumptions made in this anal-
ysis.
FIG. 5: Stretch contributions of neurites. The different
colours correspond to different pull speeds and the
different line types (dashed, solid, dotted) are associated
with different types of stretch (λ is dashed, λg is solid
and λe is dotted). λ for each pull speed (dashed lines) is
calculated with an initial neurite length of L = 1 µm.
The volume addition λg (solid lines) for each pull speed
k = 0.009 s−1 from all pulling experiments. We see that
these track λ, approaching it for later times. λe = λ/λg
are computed with Eq. 12 taking the mean
are plotted with dotted lines and show the elastic
response of the neurite for different pull speeds. Initially
all λe track λ then all collapse to values approaching 1.
1. Model selection
We have used the Mooney-Rivlin model to characterize
the neurite response to deformation. However there do
exist alternative choices.
To model the mechanical response of neurites to
stretch, we compare a series of widely-used consti-
tutive models,
including viscoelatic and hyperelastic
relations[11, 13, 21, 26]. We obtain constitutive rela-
tionships from strain-energy density functions, Ψ, and fit
these to each curve. We determine the best fit by min-
imizing the Akaike Information Criterion (AIC)[27, 28].
Fig. 6 is a bar graph showing the frequency of 'wins'
of each constitutive relationship, that is the number of
curves for which that relationship gave the minimal AIC.
This demonstrates that a Mooney-Rivlin relation best de-
scribes the data.
6
neurite is trying to recover. Although we do not have
sufficient statistics to concretely state λ∗ = 1, we have
indication that this is a reasonable choice, see Fig. 7.
In our experiments, neurites are initiated by initially
pulling the bead very slowly for 1 − 5 µm, see first snap-
shot of Fig. 7c [30]. In some cases, the neurite was pulled
even further (∼ 10− 20 µm, see Fig. 7a-b) to ensure that
we could pull at greater rates without the bead becoming
dislodged by other cells or debris on the coverslip. This
slow initiation length was taken to be the initial length
L in the analysis described above. This is the starting
point for rapid pulling.
In some instances, during pulls at high pull speeds
(0.05 − 1.8 µm/s), the neurite was stronger than the
suction applied to fix the bead to the tip of the mi-
cropipette. In these cases, the bead would return to its
initial position -- indicating that the neurite has a critical
tension it is trying to restore.
V. CONCLUSION
In this work we have developed a model that links ten-
sion in extending neurites to the rate of mass addition.
This lets us quantify the role of tension as a driver of
neurite growth. The fact that the mechanical behaviour
of induced neurites is similar to naturally-grown axons
under stretch indicates that our pulling experiments are
relevant to questions of axon growth [11, 13, 21, 26] . We
quantify a new capacity for growth through the addition
of new material.
Using a Mooney-Rivling model, we identify the contri-
bution of hyper-elastic stretching to neurite deformation
under loading. We find the material constants c1 and c2
vary over orders of magnitude without a satisfying reason
as to why. However, we add a Mooney-Rivlin characteri-
zation of new structures to the existing body of literature.
In future, this could be used in multi-cell models of the
brain.
Motivated by previously-reported observations [18], we
adopt an exponential growth law to model mass addition.
We find that the time constants k are distributed lognor-
mally. The mean value of k is close to the time constant
of diffusion of actin in neurites, which could indicate the
importance of diffusion in the growth process.
ACKNOWLEDGEMENTS
Both authors would like to acknowledge funding from
the National Sciences and Research Council of Canada
(NSERC) and the Fonds de recherche du Qu´ebec --
Nature et technologies (FRQNT). We fervently thank
Evan McDonough for invaluable discussions and recom-
mendations for analysis.
FIG. 6: Relationships that best fit the data as
determined by minimizing the AIC. Of the models
commonly used to characterize the stress-strain
relationship, the Mooney-Rivlin model was most
frequently the best fit.
2. Added volume
The derivation of Eq. 6 assumes A is homogeneous
along the axial length of the neurite [11].
In [19], a
method for extracting radii of neurites below the optical-
diffraction limit is developed and it is shown that neurites
have a constant radius along their length a short time af-
ter they are pulled.
Together, radius measurements and our analysis con-
If volume were
firm volume growth along the neurite.
conserved during neurite deformations, then λ⊥ = 1/λ1/2
[29]. In our framework, λ (cid:54)= λe at later times so volume
is not conserved.
From the form of λe (Fig. 5), we see that initially λ⊥
decreases to accommodate stretch since mass flow is lim-
ited on very short timescales. With time, λ⊥ increases,
tending towards 1. Ref. [17] reported radial thinning then
thickening along the axon but on much longer timescales
(several hours). We apply stress at much faster rates
than [17] and our neurites are on average ∼ 5× smaller
than axons. These factors could potentially explain the
faster mass accretion rates observed here. Faster rates of
applied stress could trigger a faster response and thinner
neurites can increase their relative mass more quickly.
3. Critical stretch
Here we present experimental evidence of a critical
stretch, λ∗, associated with a critical stress, σ∗ that the
7
FIG. 7: (a-c) Snapshots of experiments where the beads detach from the pipettes and the neurites (white arrows)
return to their initial lengths. Thick scale bars are 10 µm. Thin scale bars represent different initial lengths; these
are absent in (c) where the neurite is obscured by the bead. The second image in each series is the frame before the
bead detaches.
[1] K. Franze, P. A. Janmey, and J. Guck, Annual review
of biomedical engineering 15, 227 (2013).
[2] D. M. Suter and K. E. Miller, Progress in neurobiology
[15] T. J. Dennerll, P. Lamoureux, R. E. Buxbaum, and S. R.
Heidemann, The journal of cell biology 109, 3073 (1989).
[16] A. F. Bower, Applied mechanics of solids (CRC press,
94, 91 (2011).
2009).
[3] A. I. M. Athamneh and D. M. Suter, Frontiers in cellular
neuroscience 9, 359 (2015).
[4] S. R. Heidemann and D. Bray, Frontiers in cellular neu-
roscience 9, 316 (2015).
[5] M. O'Toole, P. Lamoureux, and K. E. Miller, Biophysical
journal 94, 2610 (2008).
[6] J. Zheng, P. Lamoureux, V. Santiago, T. Dennerll, R. E.
and S. R. Heidemann, Journal of Neuro-
Buxbaum,
science 11, 1117 (1991).
[7] F. Suarez, P. Thostrup, D. Colman,
and P. Grutter,
Developmental neurobiology 73, 98 (2013).
[8] M. H. Magdesian, G. M. Lopez-Ayon, M. Mori,
D. Boudreau, A. Goulet-Hanssens, R. Sanz, Y. Miya-
hara, C. J. Barrett, A. E. Fournier, Y. De Koninck, et al.,
Journal of Neuroscience 36, 979 (2016).
[9] M. Anthonisen, M. Rigby, M. H. Sangji, X. Y. Chua,
and P. Grutter, Journal of the Mechanical Behavior of
Biomedical Materials 98, 121 (2019).
[10] M. A. Holland, K. E. Miller, and E. Kuhl, Annals of
biomedical engineering 43, 1640 (2015).
[11] A. Goriely, S. Budday,
and E. Kuhl, in Advances in
Applied Mechanics, Vol. 48 (Elsevier, 2015) pp. 79 -- 139.
[12] A. Goriely and D. Moulton, New Trends in the Physics
and Mechanics of Biological Systems: Lecture Notes of
the Les Houches Summer School: Volume 92, July 2009
92, 153 (2011).
[17] P. Lamoureux, S. R. Heidemann, N. R. Martzke, and
K. E. Miller, Developmental neurobiology 70, 135 (2010).
[18] M. Rigby, M. Anthonisen, X. Chua, A. Kaplan,
A. Fournier, and P. Grutter, AIP Advances 9, 075009
(2019).
[19] M. Anthonisen, Y. Zhang, M. H. Sangji, and P. Grutter,
"Quantifying neurite morphology below the diffraction li-
imit of an optical microscope using out-of-focus images,"
(2019).
[20] E. Hannezo, B. Dong, P. Recho, J.-F. Joanny,
and
S. Hayashi, Proceedings of the National Academy of Sci-
ences 112, 8620 (2015).
[21] R. de Rooij and E. Kuhl, Applied Mechanics Reviews 68,
010801 (2016).
[22] L. E. Bilston, Z. Liu, and N. Phan-Thien, Biorheology
38, 335 (2001).
[23] M. Hrapko, J. Van Dommelen, G. Peters, and J. Wis-
mans, Biorheology 43, 623 (2006).
[24] B. Rashid, M. Destrade, and M. D. Gilchrist, Journal of
the mechanical behavior of biomedical materials 28, 71
(2013).
[25] L. A. Mihai, L. Chin, P. A. Janmey, and A. Goriely, J.
Royal Soc. Interface 12, 20150486 (2015).
[26] R. P. Mondaini and P. M. Pardalos, Mathematical mod-
elling of biosystems, Vol. 102 (Springer Science & Busi-
ness Media, 2008).
[13] A. Goriely, The mathematics and mechanics of biological
[27] E.-J. Wagenmakers and S. Farrell, Psychonomic bulletin
growth, Vol. 45 (Springer, 2017).
& review 11, 192 (2004).
[14] R. Bernal, P. A. Pullarkat, and F. Melo, Physical review
[28] M. Snipes and D. C. Taylor, Wine Economics and Policy
letters 99, 018301 (2007).
3, 3 (2014).
t=0st=38st=48st=0st=54st=59st=0st=45st=53s(a)(b)(c) [29] A. Fan, A. Tofangchi, M. Kandel, G. Popescu,
and
of Visualized Experiments) , e55697 (2017).
T. Saif, Scientific reports 7, 14188 (2017).
[30] M. H. Magdesian, M. Anthonisen, G. M. Lopez-Ayon,
X. Y. Chua, M. Rigby, and P. Grutter, JoVE (Journal
8
|
1710.02684 | 1 | 1710 | 2017-10-07T13:34:36 | The role of vimentin in regulating cell-invasive migration in dense cultures of breast carcinoma cells | [
"physics.bio-ph",
"q-bio.CB"
] | Cell migration and mechanics are tightly regulated by the integrated activities of the various cytoskeletal networks. In cancer cells, cytoskeletal modulations have been implicated in the loss of tissue integrity, and acquisition of an invasive phenotype. In epithelial cancers, for example, increased expression of the cytoskeletal filament protein vimentin correlates with metastatic potential. Nonetheless, the exact mechanism whereby vimentin affects cell motility remains poorly understood. In this study, we measured the effects of vimentin expression on the mechano-elastic and migratory properties of the highly invasive breast carcinoma cell line MDA231. We demonstrate here that vimentin stiffens cells and enhances cell migration in dense cultures, but exerts little or no effect on the migration of sparsely plated cells. These results suggest that cell-cell interactions play a key role in regulating cell migration, and coordinating cell movement in dense cultures. Our findings pave the way towards understanding the relationship between cell migration and mechanics, in a biologically relevant context. | physics.bio-ph | physics | The role of vimentin in regulating cell-invasive
migration in dense cultures of breast carcinoma cells
Y. Messica1, A. Laser-Azogui1, T. Volberg2, Y. Elisha2, K. Lysakovskaia3,4,5, R. Eils3,4, E.
Gladilin3,4,6, *, B. Geiger2,*, R. Beck1,*
1 Raymond and Beverly Sackler School of Physics and Astronomy, Tel Aviv University, Tel
Aviv, Israel; 2 Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot,
7610001, Israel; 3 Division of Theoretical Bioinformatics, German Cancer Research Center,
69120 Heidelberg, Germany; 4 BioQuant and IPMB, University of Heidelberg, 69120
Heidelberg, Germany; 5 International Max Planck Research School for Molecular
Biology, Georg-August-University Göttingen, 37077 Göttingen, Germany; 6 Leibniz Institute of
Plant Genetics and Crop Plant Research, 06466 Seeland, Germany.
1
Abstract
Cell migration and mechanics are tightly regulated by the integrated activities of the various
cytoskeletal networks. In cancer cells, cytoskeletal modulations have been implicated in the loss
of tissue integrity, and acquisition of an invasive phenotype. In epithelial cancers, for example,
increased expression of the cytoskeletal filament protein vimentin correlates with metastatic
potential 1,2. Nonetheless, the exact mechanism whereby vimentin affects cell motility remains
poorly understood. In this study, we measured the effects of vimentin expression on the
mechano-elastic and migratory properties of the highly invasive breast carcinoma cell line
MDA231. We demonstrate here that vimentin stiffens cells and enhances cell migration in dense
cultures, but exerts little or no effect on the migration of sparsely plated cells. These results
suggest that cell-cell interactions play a key role in regulating cell migration, and coordinating
cell movement in dense cultures. Our findings pave the way towards understanding the
relationship between cell migration and mechanics, in a biologically relevant context.
Keywords: Cell motility, cell mechanics, biophysics, vimentin, collective motility, metastasis
2
Mechanosensing of the pericellular environment plays a vital role in regulating diverse cellular
functions, including growth, division, differentiation, morphogenesis, and migration 3–5. These
mechanical cues play key homeostatic roles, and their malfunction is associated with diverse
pathological states such as developmental disorders, tumorigenicity and metastasis 6,7.
Researchers have recently taken particular interest in the mechano-elastic properties of cancer
cells, showing that cancer cells tend to be softer than benign cells 8. This same correlation can
also be used to distinguish between metastatic and non-metastatic cancer cells, metastatic cells
being softer still 9. The naïve explanation for this correlation is that cell compliance and
deformability promote invasion and metastasis by enabling cells to pass through narrow tissue
barriers 9.
The cytoskeleton, consisting of microfilaments, microtubules and intermediate filaments (IFs)
is the cell's major stress-bearing component, determining its deformability in response to
external forces. Microfilament and microtubule networks are primarily composed of F-actin and
tubulin, respectively, while the building blocks of IFs comprise versatile groups of proteins that
are expressed in a highly tissue-specific and cell type-specific manner 10. Thus, epithelial cells
express diverse cytokeratins, muscle cells express desmin, nerve cells express neurofilaments,
glial cells express the glial fibrillary acidic protein, and mesenchymal cells express vimentin 11.
Vimentin is also upregulated in epithelial cells undergoing epithelial-to-mesenchymal transition
(EMT) 1,12, alongside downregulation of keratin IFs 13–15. During EMT, epithelial cells also lose
their apical cell–cell junctions and, consequently, their apical-basal polarity, thereby acquiring a
migratory capacity, and other mesenchymal characteristics 12. EMT, which is essential for
fundamental developmental processes such as embryogenesis and organogenesis, is also
3
involved in malignant transformation, enabling tumor cells to invade surrounding tissues and
eventually form distant metastases 12,16.
The cytokeratin-to-vimentin transition is one of the most prominent and functionally
significant processes associated with EMT: this switch underlies the reduction in cell stiffness,
an increase in deformability, and augmentation of invasive migration 7,17, all hallmarks of
malignant transformation 18. The direct involvement of vimentin in cell migration and
invasiveness is supported by the demonstration that its overexpression in the vimentin-negative,
non-invasive MCF-7 breast cancer cell line alters the integrin profile in the cells, and increases
their invasiveness 19,20. Moreover, vimentin expression in epithelial cells induces cell elongation,
and loss of cell-cell adhesion 2.
Ample evidence linking vimentin expression to metastasis raises several intriguing questions:
What is the mechanism underlying the effect of vimentin on cell migration? Does vimentin
directly modulate the migratory machinery of the cells, or is its effect indirect, altering the cell's
response to environmental cues?
Using direct physical characterizations, we demonstrate that vimentin expression, combined
with cell density, regulates the mechanical properties of MDA231 cancer cells, resulting in their
invasive phenotype. The cell density dependence underlying vimentin's effects would suggest
that vimentin expression mediates physical and environmental cues, rather than merely affecting
the cell motility apparatus. Although some studies show a positive correlation between cell
deformability and metastatic potential (migration rate and invasiveness) 9,21, this was not the case
with MDA231 cells. We demonstrate that despite being softer than control MDA231 cells,
4
vimentin-deficient cells (MDA231vim-) display decreased, rather than elevated, motile and
invasive capabilities.
To investigate the possible mechanisms underlying vimentin's regulation of cell mechanics
and motility, we performed a series of experiments aimed at evaluating the effects of its
expression in various cellular settings, focusing mainly on assays conducted at different cell
densities. We studied the effects of vimentin on two pairs of cells two cell lines: MDA-MB-231
and MCF7 cells. MDA-MB-231 (referred to here as MDA231) cells are breast adenocarcinoma
epithelial metastatic cells, which naturally co-express both vimentin and keratin IFs (Fig. 1A)
20,22. MDA231 cells possess a mesenchymal shape, and are commonly used as a model system
for a stable EMT state 23. The second cell line, MCF7, typically expresses keratin IFs only and
was used to further examine the effects of vimentin on keratin networks.
To determine the contribution of vimentin expression to cytoskeletal organization within
MDA231 cells, we knocked down vimentin expression by means of viral shRNA infection (see
Materials and Methods in Supporting Information). The knockdown clones (MDA231vim-)
display reduced vimentin levels (approximately 0.2% compared to controls; see Supporting
Information), with the vimentin being scattered throughout the perinuclear area (Fig. 1B).
Notably, the tubulin and actin cytoskeletons remained largely intact in the MDA231vim- cells,
similar to those seen in the MDA231 parental cells. Western blot experiments quantitatively
confirmed the knockdown of vimentin and, moreover, that the actin network was unaffected
(Supporting Information, Section I). In contrast, the cytokeratin cytoskeletal network, which was
widely distributed throughout the cytoplasm of MDA231 cells, collapsed into the perinuclear
region in MDA231 cells lacking vimentin (MDA231vim-), as shown in Fig. 1A.
5
In the second cell line studied (MCF7), the cytokeratin organization in parental cells forms a
typical cage-like network. Following induction of vimentin expression by doxycycline over a 24-
hour period (MCF7vim+), an extensive vimentin network formed, and the perinuclear keratin
network was radically transformed into an extended filamentous structure distributed throughout
the cytoplasm (Fig. 1C-D). These results indicate that vimentin expression leads to the dispersal
of cytokeratin filaments into the cell periphery. Furthermore, in vimentin's absence (e.g., in
epithelial cells, which do not naturally express vimentin, or in cells which underwent EMT and
were induced to undergo vimentin knockdown), the cytokeratin network mainly centers around
the nucleus.
Vimentin knockdown also reduced the projected area of focal adhesions (FAs), visualized by
labeling the cells for several FA-associated proteins (α-actinin, zyxin and vinculin; see Fig. 1E).
In addition, we found reduced levels of Snail1 mRNA (Fig. 1F), a transcription factor involved
in downregulating E-cadherin and claudins, and upregulating vimentin and fibronectin during
EMT 24. Surprisingly, N-cadherin mRNA levels are increased in vimentin knockdown clones,
even though N-cadherin mRNA is associated with EMT and metastatic potential 25–27. The
mRNA levels of other EMT markers; e.g., fibronectin, HMGA2, Zeb1 and Slug, were not
significantly altered following vimentin knockdown.
6
7
Figure 1. Immunostaining of cytoskeletal proteins in MDA231 and MCF7 cells, in which
vimentin is either expressed, or not expressed. (A-B) In both MDA231 and MDA231
cells,
tubulin and actin networks appear similar (upper panels). In MDA231 cells, vimentin and
cytokeratin networks span the entire cytoplasm, while in MDA231vim- cells, vimentin expression
is diminished, and the keratin network collapses into the perinuclear region. Scale bar: 10 μm. (C-
D) Cytokeratin organization in parental MCF7 cells displays a typical cage-like network
surrounding the nucleus. Following induction of vimentin expression by doxycycline (namely,
vim-
vim+
MCF7
cells), the vimentin filaments form an extensive network, and the keratin network is
altered from a cage-like to a dispersed filamentous structure. Addition of doxycycline to parental
MCF7 cells had no structural effect on their cytoskeletons. (E) Immunostaining for the adhesion-
associated proteins α-actinin, zyxin and vinculin. (F) mRNA levels of EMT markers. Mean values
± standard error of the mean (SEM) are shown. MDA231: n=3; MDA231vim-: n=9.
A more direct effect of vimentin knockdown and keratin network collapse could be observed
by directly measuring the cells' rigidity, by means of atomic force microscopy (AFM) and a
microfluidics optical stretcher (MOS). Using AFM, the Young's modulus (E) of individual cells
is extracted by indenting them, and measuring the cell's resistance to the indentation. Comparing
the rigidity of MDA231vim- cells to that of control cells indicated that vimentin knockdown
induced about 30% reduction in the cells' Young's modulus value, as shown in Fig. 2A.
Additional experiments using an MOS confirmed that the vimentin-deficient cells are more
compliant (Fig. 2B). In the MOS experiments, the entire cell is stretched, and its overall response
to the strain is measured. At the end of the creep phase (denoted by a box), a relative elongation
of 2% was detected in MDA231 cells, while in MDA231vim- cells, the relative elongation
increased by 25-50%, compared to the MDA231 cells. Both cell lines displayed no significant
difference in recovery rate. Similarly, MCF7vim+ cells were less stretched by about 15%,
compared with parental MCF7 cells (Fig. 2B). These results suggest that the vimentin network,
reinforced by the extended keratin network, contributes to cell resistance to deformation.
8
The next trait we compared in the vimentin knockdown cell lines was cell migration. For this
purpose, cells were seeded at low density, and single-cell trajectories were tracked using semi-
automated image analysis (see Supporting Information, Materials and Methods, and
Supplementary Movie 1). This analysis indicated that there is no statistical difference between
the magnitude of migration velocities displayed by the vimentin-containing and the vimentin-
lacking cell lines (Fig. 2C), and that cell velocities are exponentially distributed for both cell
lines (see Supporting Information, Section II).
Another assay commonly used for assessing cell migration is a "wound-healing" experiment.
Briefly, a scratch is mechanically inflicted on a confluent monolayer of cells, and the rate of wound
closure is measured. The parameter that is measured here is collective, rather than single-cell
migration. In these experiments, the difference between MDA231 and MDA231vim- cells was
profound (unlike the migration rates in sparse cultures), with the MDA231 cells closing the wound
50% faster than MDA231vim- cells (Fig. 2D). Notably, we found no difference in proliferation rates
between the MDA231 and MDA231vim- cell lines, ruling out the possibility that the difference in
wound healing rates could be attributed to differential proliferation (Supporting Information,
Section III). Additional migration and invasion assays, which measure the transwell migration of
cells plated at high density on filters coated with basement membrane matrix proteins, support this
finding; namely, that depletion of vimentin hinders the migration of densely-plated cells (Fig. 2E).
Notably, in these assays, the number of cells which successfully migrate through 8 μm pores was
counted; yet despite the fact that MDA231vim- cells are softer, and thus should be able to deform
more readily, they were slower to migrate and invade the matrix and filter. Our high-density
migration assay findings are in agreement with previous reports showing that vimentin enhances
densely-plated cell migration rates 18,28.
9
At this point, it would appear that the migration rate of breast cancer cells in dense cultures
(but not sparse cultures) is highly dependent on vimentin expression. Given this sensitivity to
culture density, we monitored single-cell motility as a function of cell density by tracking the
cells' trajectories in an extended time-lapse microscopy image series. In our analysis, we drew a
distinction between two density quantities: global density; i.e., the number of cells captured
within the field of the view; and local density (𝜌𝑙𝑜𝑐𝑎𝑙), which encompasses the cell's "nearest
neighbor" cells, within a range of about two cell diameters (see Supporting Information, Section
IV for further details).
10
Figure 2. Elasticity, migration and invasiveness assays. (A-B) Cell elasticity measurements.
(A) Young's modulus distribution, measured by AFM. The Young's modulus in MDA231vim-
cells is lower than in their vimentin-expressing counterparts. Mean ± SEM values: MDA231:
𝐸 = 317 ± 26 𝑃𝑎 (n=51); MDA231vim-: 𝐸 = 239 ± 12 𝑃𝑎 (n=86). (B) Deformability
measurements by MOS. MDA231vim- cells stretch approximately 50% more than MDA231 cells.
Similarly, inducing vimentin expression in MCF7 cells stiffens them by 13%, compared to the
MCF7 cells lacking vimentin. Control MCF7vim+ cells, prior to induction of vimentin by
doxycycline, display no such stiffening. Between 300-600 cells were measured per each clone.
(C) Velocities of cells seeded at low density (60
Median values and standard error calculated using the bootstrap method: MDA231: 𝑣𝐿𝐷 =
𝑚𝑚2) for MDA231 and MDA231vim- clones.
𝑐𝑒𝑙𝑙𝑠
11
18.43 ± 0.50
𝜇 𝑚
density (>400
ℎ𝑟
𝑐𝑒𝑙𝑙𝑠
(n=1842); MDA231vim-: 𝑣𝐿𝐷 = 17.41 ± 0.52
(n=1368). (D-E) High-
𝑚𝑚2) bulk migration assays of MDA231 cells and MDA231vim- clones. (D)
𝜇 𝑚
ℎ𝑟
Wound-healing assay. Each curve represents averaging of 6 different experiments – MDA231
(blue) and 4 clones of MDA231vim- (red). (E) Transwell migration and invasion assays. Each
experiment was repeated 3 times, and consisted of imaging 5-10 different fields of view per each
clone. These assays, together with the wound–healing assays, show that MDA231 cells are
significantly more motile, compared to MDA231vim- cells plated at high density. (F) Velocity
distribution at high density, normalized by the velocities at low density. MDA231 cells move
= 0.941 ± 0.016, n=2396), while the MDA231vim-
slightly more slowly (median value of
𝑣𝐻𝐷
𝑣𝐿𝐷
cells slow down significantly (
𝑣𝐻𝐷
𝑣𝐿𝐷
= 0.772 ± 0.017, n=1356).
For each experiment, cells were seeded at low density (~60
𝑐𝑒𝑙𝑙𝑠
𝑚𝑚2), and imaged until the plate
reached full confluence. Comparing the ratio between the velocity magnitudes at the end (high
density, 𝑣𝐻𝐷) with those at the beginning of the experiment (low density, 𝑣𝐿𝐷), we found that
MDA231vim- cells slowed by a rate of 23%, while MDA231 cells roughly maintained their
velocity (Fig. 2F). In addition, at a relatively high cell density (~400
𝑐𝑒𝑙𝑙𝑠
𝑚𝑚2), the local density
strongly correlated with the velocity amplitude for MDA231 cells, as if they were attempting to
"escape the crowd". On the other hand, for the MDA231vim- cells, no such correlation was found
(Fig. 3A). Supporting evidence comes from evaluating the covariance of the velocity and the
local density, < 𝑣𝜌𝑙𝑜𝑐𝑎𝑙 >, as shown in Fig. 3B. Here, < > represents ensemble and time average
for a given experiment. The velocity and local density fields in single frames also exemplify the
difference between the cell lines (Fig. 3C, and Supplementary Movie 2). Moreover, in high
density plates, we noted that MDA231vim- cells tended to approach their neighbors in closer
proximity. However, the difference between the cell lines was found to be rather minor (see
Supporting Information, Section V). Notably, even at high density, both cell lines did not form
permanent clusters, but rather were found to migrate individually.
12
𝑐𝑒𝑙𝑙𝑠
𝑚𝑚2). MDA231 cells show a positive
Figure 3. Single-cell velocity-local density dependence in MDA231 and MDA231 vimentin
knockdown (MDA231vim-) cells. (A) Velocity as a function of local density in individual
experiments using MDA231 cells at high density (>400
velocity-local density correlation, while MDA231vim- cells display no such correlation. To
produce the graph, (velocity, 𝜌𝑙𝑜𝑐𝑎𝑙) data points were binned according to 𝜌𝑙𝑜𝑐𝑎𝑙 values, and the
median velocity in each binned group was taken. (B) Covariance of velocity and local density for
the individual experiments shown in (A). MDA231 cells show positive covariance, while
MDA231vim- cells do not display a clear trend. (C) Velocity-local density correlation for
MDA231 (left) and MDA231vim- cells (right) in a single, high density frame. For each frame
(top), the local density field (middle) and the velocity field (bottom) are shown. The velocity
fields are interpolated according to the momentary velocity of each cell identified in the frame.
Scale bar: 50 μm.
13
Another property relevant to cell migration is the persistence of the movement. Persistence
corresponds to the time duration (or distance) a cell continues to move in a given direction. To
extract this quantity, we define the directionality persistence after a time lag Δ𝑡, thus: 𝐶(Δ𝑡) =
〈cos(𝛼(𝑡 + Δ𝑡) − 𝛼(𝑡))〉. Here, 𝛼(𝑡) is the angle of the cell's velocity vector at time 𝑡 (see Fig.
4A, inset for illustration), and < > stands for averaging over time and the ensemble of the cells.
Conceptually, 𝐶(Δ𝑡) decays from unity at Δ𝑡 = 0 to zero after a typical time, corresponding to
the time it takes for single cells to reorient, and lose their directionality. To obtain a numerical
estimate of the persistence time, we fit a double exponential distribution: 𝐶(Δ𝑡) = 𝐴𝑒
−
Δ𝑡
𝜏1 +
(1 − 𝐴)𝑒
−
Δ𝑡
𝜏2, as previously determined to fit experimental data from HaCaT cells 29. We found
that at all global densities, MDA231 cells demonstrate longer persistence time, with 𝜏2𝑀𝐷𝐴231 >
𝜏2𝑀𝐷𝐴231𝑣𝑖𝑚− and 𝜏2 being the longer characteristic time. Importantly, the difference in
persistence time becomes more pronounced at higher global densities (Fig. 4, Table 1). This
property can also prove crucial to the advantage of MDA231 over MDA231vim- cells in bulk
assays, as cells with longer persistence time will display greater net movement, compared to cells
that move more erratically.
14
Figure 4. Persistence time measurements. Directionality over lag time for MDA231 and
𝑐𝑒𝑙𝑙𝑠
MDA231 vimentin knockdown (MDA231vim-) cells plated at (A) low (<120
𝑚𝑚2) and (B)
𝑐𝑒𝑙𝑙𝑠
𝑚𝑚2) densities. At low density, data points comprise averages of 6
medium (200-300
experiments for MDA231 cells, and 5 experiments for MDA231vim- cells. At medium density,
data points comprise averages of 3 experiments for each cell line. Dashed lines: Double
exponential fits for the averaged curves. Inset: Schematic of calculation for directionality
analysis, derived from measured velocity directions at various time points, separated by intervals
of 𝛥𝑡.
Table 1. Fit parameters for MDA231 and MDA231vim- angle autocorrelation data displaying a
double exponential decay: 𝐶(Δ𝑡) = 𝐴𝑒
−
Δ𝑡
𝜏1 + (1 − 𝐴)𝑒
−
Δ𝑡
𝜏2
MDA231 – sparse
MDA231 – medium
MDA231vim- - sparse
MDA231vim- - medium
𝝉𝟏 [min]
𝝉𝟐 [min]
A
5.5
8.2
3.1
4.7
57.3
54.1
42.3
32.0
0.68
0.71
0.78
0.77
MDA231 cells are more persistent in their movement over time than MDA231vim- cells, as
demonstrated by the longer decay times (𝜏1, 𝜏2). The difference between the two cell lines is
more pronounced at higher densities.
15
The results presented above clearly imply that vimentin knockdown in MDA231 cells
significantly impairs the cells' migratory activity. To further test the involvement of vimentin in
the regulation of cell invasiveness, we conducted an ex vivo infiltration assay, in which two
monolayers, one of MDA231 cells (vimentin-expressing or -knockdown) and the other of
stromal fibroblasts (Fig. 5A and Supporting Information, Supplementary Movie 3), were seeded
separately in two parallel compartments. The monolayers were allowed to migrate and expand,
eventually reaching each other (closing the gap), by collective migration. Measurement of
migration and invasion speeds indicated that prior to their encounter with the fibroblasts, the
MDA231 cells migrated towards the stromal cells ~1.5-2 times faster than their MDA231vim-
counterparts (Fig. 5B and Fig. S6), in line with the wound-healing results. After reaching
contact, both cell lines slowed their rate of movement; yet the migration of MDA231vim- cells
nearly stopped at the boundary between the two cell populations, while the MDA231 cells
efficiently infiltrated the stromal monolayer.
Results from our infiltration assays indicate that the motility-density dependence of MDA231
and MDA231vim- cells is not specific to environments involving cells of the same type; rather,
that the motility of MDA231vim- cells is impaired in any dense cellular environment.
The experimental results presented in this study show that variations in vimentin levels can alter
both the rigidity and the motile behavior of breast cancer cells. However, in contrast with previous
reports 30, our findings demonstrate that while the expression of vimentin does not significantly
alter the migration of sparsely-plated MDA231 breast cancer cells (Fig. 2C), it significantly
enhances the collective migration of densely-plated cells (Fig. 2 D-F), suggesting that cell-cell
interactions play key roles in regulating migration 31,32.
16
Figure 5. Infiltration of MDA231 and MDA231vim- cells into stromal fibroblast monolayers.
(A) Snapshots of two fields of MDA231 (green in upper panels) and MDA231vim- (green in
lower panels) cells, taken at the initial stage of migration, the time of contact with the stromal
monolayer (denoted in blue), and 25 hours post-contact. Magenta and cyan lines mark the
monolayer edges as automatically identified (see Materials and Methods). After establishing
contact with the fibroblasts, the MDA231 cells infiltrated the stromal cells, while the
MDA231vim- cells progressed at a considerably slower rate. Scale bar: 50 μm. (B) Leading-edge
progression over time of the MDA231/MDA231vim- (blue/red curves, respectively) and stromal
monolayers (green curve). Brown triangles indicate the contact point. Dashed lines are linear fits
to the pre-contact step (t=0 up to contact with slope 𝑣1) and post-contact progression (from
contact up to 8 hours later, with slope 𝑣2). The MDA231 cells migrate faster prior to reaching
the stromal monolayer, in comparison to the MDA231vim- cells, with a marked difference in rates
of persistence seen post-contact. Leading-edge velocities (mean ± SEM; n=8 for both cell lines):
MDA231 = 16.0 ± 1.0
v1
MDA231vim−
v2
distribution).
(see also Supporting Information, Fig. S6 for leading-edge velocity
hr
= 4.2 ± 1.2
MDA231 = 7.2 ± 0.6
= 10.8 ± 0.6
MDA231vim−
μm
hr
, v2
μm
hr
,
μm
, v1
μm
hr
The direct mechanisms underlying this effect of vimentin on cell migration remain unclear; yet
several recent studies presented results that are in line with the data reported here, and proposed
possible molecular mechanisms underlying the vimentin-dependent enhancement of cell
17
migration. For example, Chu et al. suggested that co-expression of vimentin and keratin induces
increased cytoskeletal interactions at focal adhesions with the extracellular matrix (ECM) which,
eventually, affect cell motility 33. Vuoriluoto et al. proposed that vimentin expression during EMT
upregulates the receptor tyrosine kinase Axl which, in turn, contributes to extravasation of breast
cancer cells in mice34.
Reduced intercellular adhesion in vimentin-expressing cells may also result from an indirect
effect of vimentin expression on keratin reorganization (Fig. 1A), and consequently on desmosome
formation 35. Danuser's group further demonstrated that vimentin stabilizes the microtubule
network, thereby enhancing persistent migration 36, a phenomenon that might be related to the
enhanced persistence time we observed in vimentin-expressing cells (Fig. 4).
Another proposed mechanism of vimentin-dependent regulation of cell motility might involve
its modulation of the Rac1/RhoA pathway 37, which plays a key role in switching between
mesenchymal and amoeboid types of cell migration 38. Recently, vimentin has been shown to
enhance Jagged-mediated Notch signaling 39, which promotes tumor invasion and metastasis 40.
Obviously, some or all of these mechanisms might function in tandem. These previously published
studies are in line with the results presented here; yet they do not explain the unexpected
dependence of migration rates on local culture density, as presented for vimentin-expressing and
knockdown cells.
Given the huge complexity of the processes regulating cell migration and invasion, we wish to
propose an alternative, purely physical mechanism underlying vimentin-dependent motility-cell
density relationships. Since vimentin knockdown softens MDA231 cells (Fig. 2A-B), MDA231vim-
cells are, by definition, more susceptible to deformation by external forces. Therefore, each
migrating cell from an MDA231vim- culture can easily deform its neighboring cells, rather than
18
reorienting to conform to a vacant intercellular space (see Fig. 6 for a schematic visualization).
This characteristic can impair MDA231vim- cell migration under conditions of high cell density, as
we indeed observed experimentally. Furthermore, this susceptibility to deformation can account
for MDA231vim- cells' loss of polarity, and thus direction of movement, faster than MDA231 cells
(Fig. 4). As cells constitute active systems, interactions between cells can give rise to non-trivial
dynamics, as also discussed in the context of density-dependent motility 32. Ultimately, differences
in migration speeds of vimentin-containing and vimentin-lacking cells would not be detected at
low densities, where cells rarely interact with their neighbors.
Figure 6. A scheme of the suggested mechanism underlying vimentin-regulated motility.
MDA231vim- cells are more compliant and readily deformed, compared to MDA231 cells; thus, a
migrating MDA231vim- cell can deform surrounding cells, rather than reorienting itself and
migrate towards a vacant space.
Moreover, in this study we ruled out the possibility that a coherent mode of migration is a
distinguishing trait between the cell lines (Supporting Information, Section VII). Thus, motility of
MDA231 cells is enhanced by their neighbors in a manner unrelated to coherent movement (i.e.,
19
pulling each other in a common direction). These results indicate that neighboring cells enable
each other to develop higher velocities, regardless of their locomotion direction. It should be noted
that the average velocity of MDA231 cells does not increase with elevated plate density (Fig. 2F);
therefore, additional mechanisms must exist to keep the average velocity at similar values.
Previous studies have shown that the rigidity of a cell's environment affects various cellular
properties, motility among them 41–43. However, the reference to "environment" usually focuses
on the ECM. The results presented here show that mechanical cues from neighboring cells can
play key roles in environmental signaling. Consequently, the mechanical properties of cells play a
dual role as both sensors and inducers of mechanical stress. A body of evidence indicates that
increasing substrate stiffness slows the rate of cell migration 3,41,44, rendering the analogy between
the mechanical properties of neighboring cells and those of the underlying substrate non-trivial.
Tumor growth can produce compressive mechanical stress 45,46 that has been shown to elevate the
rates of motility of malignant cell lines, including MDA231 47. Density-enhanced motility for the
PC-3 prostate cancer cell line was also previously reported 48. Interestingly, the PC-3 cell line also
expresses vimentin, similar to the MDA231 system we studied here. Further research is still needed
to evaluate whether the phenomenon of density-velocity correlation in cells is, indeed, widespread,
and whether it could originate from sources other than vimentin expression.
In summary, our investigations into the effects of vimentin on cell mechanics and, consequently,
the cell's migratory and metastatic potential serve to convincingly illustrate the influence of cell
environment on cellular behavior. Our results suggest that the rigidity of the surrounding cells
affects the motility of the single cell. Future research should illuminate the mechanical dynamics
between cells as they migrate individually; namely, by probing inter-cellular forces and the
resulting deformations. There are many interesting biological principles that may be gleaned from
20
such dynamics. Here, these principles give rise to cell density-dependent velocities, as previously
reported in other contexts 31,32. Significantly, vimentin expression regulates this unconventional,
density-dependent motility phenotype. A quantitative approach, such as that demonstrated here,
should shed light on the mechanisms governing cell motility in different environments.
Supporting Information
Materials and Methods, Supporting Sections I-VII, Movies 1-3
Acknowledgements
This work was supported by the DKFZ-MOST Program, the Israel Science Foundation (Grant
no 550/15), and the Abramson Center for Medical Physics, Tel Aviv University. BG holds the
Erwin Neter Professorial Chair in Cell and Tumor Biology. We thank Y. Roichman, Y. Shokef
and N. Gov for fruitful discussions. We are also grateful to D. Sprinzak and D. Kaganovich for
their kind assistance with the microscopy experiments.
Corresponding Authors
Evgeny Gladilin: [email protected]
Benny Geiger: [email protected]
Roy Beck: [email protected]
21
References
(1) Kokkinos, M. I.; Wafai, R.; Wong, M. K.; Newgreen, D. F.; Thompson, E. W.; Waltham,
M. Cells. Tissues. Organs 2007, 185 (1–3), 191–203.
(2) Mendez, M. G.; Kojima, S.-I.; Goldman, R. D. FASEB J. 2010, 24 (6), 1838–1851.
(3) Discher, D. E.; Janmey, P.; Wang, Y.-L. Science 2005, 310 (5751), 1139–1143.
(4) Geiger, B.; Spatz, J. P.; Bershadsky, A. D. Nat. Rev. Mol. Cell Biol. 2009, 10 (1), 21–33.
(5) Guck, J.; Lautenschläger, F.; Paschke, S.; Beil, M.; Guck, J. R.; Käs, J. A.; Bilby, C.; Guck,
J.; Hoffmann, H.; Penzel, R.; Gdynia, G.; Ehemann, V.; Schnabel, P. A.; Kuner, R.; Huber,
P.; Schirmacher, P.; Breuhahn, K. Integr. Biol. 2010, 2 (11–12), 575–583.
(6) Wirtz, D.; Konstantopoulos, K.; Searson, P. C. Nat. Rev. Cancer 2011, 11 (7), 512–522.
(7) Suresh, S. Acta Biomater. 2007, 3 (4), 413–438.
(8) Lekka, M. Bionanoscience 2016, 6 (1), 65–80.
(9) Guck, J.; Schinkinger, S.; Lincoln, B.; Wottawah, F.; Ebert, S.; Romeyke, M.; Lenz, D.;
Erickson, H. M.; Ananthakrishnan, R.; Mitchell, D.; Kas, J.; Ulvick, S.; Bilby, C. Biophys
J 2005, 88 (5), 3689–3698.
(10) Toivola, D. M.; Tao, G.-Z.; Habtezion, A.; Liao, J.; Omary, M. B. Trends Cell Biol. 2005,
15 (11), 608–617.
(11) Fuchs, E.; Weber, K. Annu. Rev. Biochem. 1994, 63, 345–382.
(12) Kalluri, R.; Weinberg, R. A. J. Clin. Invest. 2009, 119 (6), 1420–1428.
22
(13) Paccione, R. J.; Miyazaki, H.; Patel, V.; Waseem, A.; Gutkind, J. S.; Zehner, Z. E.; Yeudall,
W. A. Mol. Cancer Ther. 2008, 7 (9), 2894–2903.
(14) Fortier, A.-M.; Asselin, E.; Cadrin, M. J. Biol. Chem. 2013, 288 (16), 11555–11571.
(15) Polioudaki, H.; Agelaki, S.; Chiotaki, R.; Politaki, E.; Mavroudis, D.; Matikas, A.;
Georgoulias, V.; Theodoropoulos, P. A. BMC Cancer 2015, 15 (1), 399.
(16) Thiery, J. P. Nat. Rev. Cancer 2002, 2 (6), 442–454.
(17) Katira, P.; Zaman, M. H.; Bonnecaze, R. T. Phys. Rev. Lett. 2012, 108 (2), 28103.
(18) Gilles, C.; Polette, M.; Zahm, J. M.; Tournier, J. M.; Volders, L.; Foidart, J. M.; Birembaut,
P. J. Cell Sci. 1999, 112, 4615–4625.
(19) Satelli, A.; Li, S. Cell. Mol. Life Sci. 2011, 68 (18), 3033–3046.
(20) Hendrix, M. J.; Seftor, E. A.; Seftor, R. E.; Trevor, K. T. Am. J. Pathol. 1997, 150 (2), 483–
495.
(21) Xu, W.; Mezencev, R.; Kim, B.; Wang, L.; McDonald, J.; Sulchek, T. PLoS One 2012, 7
(10), e46609.
(22) Joosse, S. A.; Hannemann, J.; Spötter, J.; Bauche, A.; Andreas, A.; Müller, V.; Pantel, K.
Clin. Cancer Res. 2012, 18 (4), 993–1003.
(23) Chaffer, C. L.; San Juan, B. P.; Lim, E.; Weinberg, R. A. Cancer Metastasis Rev. 2016, 35
(4), 645–654.
(24) Kaufhold, S.; Bonavida, B. J. Exp. Clin. Cancer Res. 2014, 33 (1), 62.
23
(25) Islam, S.; Carey, T. E.; Wolf, G. T.; Wheelock, M. J.; Johnson, K. R. J. Cell Biol. 1996, 135
(6), 1643–1654.
(26) Nieman, M. T.; Prudoff, R. S.; Johnson, K. R.; Wheelock, M. J. J. Cell Biol. 1999, 147 (3),
631–643.
(27) Hazan, R. B.; Phillips, G. R.; Qiao, R. F.; Norton, L.; Aaronson, S. A. J. Cell Biol. 2000,
148 (4), 779–790.
(28) Rogel, M. R.; Soni, P. N.; Troken, J. R.; Sitikov, A.; Trejo, H. E.; Ridge, K. M. FASEB J.
2011, 25 (11), 3873–3883.
(29) Selmeczi, D.; Mosler, S.; Hagedorn, P. H.; Larsen, N. B.; Flyvbjerg, H. Biophys. J. 2005,
89 (2), 912–931.
(30) Liu, C.-Y.; Lin, H.-H.; Tang, M.-J.; Wang, Y.-K. Oncotarget 2015, 6 (18), 15966–15983.
(31) Puliafito, A.; Hufnagel, L.; Neveu, P.; Streichan, S.; Sigal, A.; Fygenson, D. K.; Shraiman,
B. I. Proc. Natl. Acad. Sci. U.S.A. 2012, 109 (3), 739–744.
(32) D'Alessandro, J.; Solon, A. P.; Hayakawa, Y.; Anjard, C.; Detcheverry, F.; Rieu, J.-P.;
Rivière, C. Nat Phys 2017, DOI: 10.1038/nphys4180.
(33) Chu, Y. W.; Seftor, E. A.; Romer, L. H.; Hendrix, M. J. Am. J. Pathol. 1996, 148 (1), 63–
69.
(34) Vuoriluoto, K.; Haugen, H.; Kiviluoto, S.; Mpindi, J. P.; Nevo, J.; Gjerdrum, C.; Tiron, C.;
Lorens, J. B.; Ivaska, J. Oncogene 2011, 30 (12), 1436–1448.
(35) Kröger, C.; Loschke, F.; Schwarz, N.; Windoffer, R.; Leube, R. E.; Magin, T. M. J. Cell
24
Biol. 2013, 201 (5), 681–692.
(36) Gan, Z.; Ding, L.; Burckhardt, C. J.; Lowery, J.; Zaritsky, A.; Sitterley, K.; Mota, A.;
Costigliola, N.; Starker, C. G.; Voytas, D. F.; Tytell, J.; Goldman, R. D.; Danuser, G. Cell
Syst. 2016, 3 (3), 252–263.e8.
(37) Havel, L. S.; Kline, E. R.; Salgueiro, A. M.; Marcus, A. I. Oncogene 2015, 34 (15), 1979–
1990.
(38) Huang, B.; Lu, M.; Jolly, M. K.; Tsarfaty, I.; Onuchic, J.; Ben-Jacob, E. Sci. Rep. 2015, 4
(1), 6449.
(39) Antfolk, D.; Sjöqvist, M.; Cheng, F.; Isoniemi, K.; Duran, C. L.; Rivero-Muller, A.; Antila,
C.; Niemi, R.; Landor, S.; Bouten, C. V. C.; Bayless, K. J.; Eriksson, J. E.; Sahlgren, C. M.
Proc. Natl. Acad. Sci. U.S.A. 2017, 114 (23), E4574–E4581.
(40) Leong, K. G.; Niessen, K.; Kulic, I.; Raouf, A.; Eaves, C.; Pollet, I.; Karsan, A. J. Exp. Med.
2007, 204 (12), 2935–2948.
(41) Shukla, V. C.; Higuita-Castro, N.; Nana-Sinkam, P.; Ghadiali, S. N. J. Biomed. Mater. Res.
Part A 2016, 104 (5), 1182–1193.
(42) Lange, J. R.; Fabry, B. Exp. Cell Res. 2013, 319 (16), 2418–2423.
(43) Tzvetkova-Chevolleau, T.; Stéphanou, A.; Fuard, D.; Ohayon, J.; Schiavone, P.; Tracqui,
P. Biomaterials 2008, 29 (10), 1541–1551.
(44) Pelham, R. J.; Wang, Y.-L. Proc. Natl. Acad. Sci. U.S.A. 1997, 94, 13661–13665.
(45) Helmlinger, G.; Netti, P. A.; Lichtenbeld, H. C.; Melder, R. J.; Jain, R. K. Nat. Biotechnol.
25
1997, 15, 778–783.
(46) Cheng, G.; Tse, J.; Jain, R. K.; Munn, L. L.; Edwards, G. PLoS One 2009, 4 (2), e4632.
(47) Tse, J. M.; Cheng, G.; Tyrrell, J. A.; Wilcox-Adelman, S. A.; Boucher, Y.; Jain, R. K.;
Munn, L. L. Proc. Natl. Acad. Sci. U.S.A. 2012, 109 (3), 911–916.
(48) Jin, W.; Shah, E. T.; Penington, C. J.; McCue, S. W.; Chopin, L. K.; Simpson, M. J. J.
Theor. Biol. 2016, 390, 136–145.
26
Supporting Information
The role of vimentin in regulating cell invasive migration in dense
cultures of breast carcinoma cells
Y. Messica, A. Laser-Azogui, T. Volberg, Y. Elisha, K. Lysakovskaia, R. Eils, E. Gladilin, B.
Geiger, R. Beck
Materials and Methods
Cell tracking using a Holomonitor M4 microscope. For tracking single cell migration, a
Holomonitor M4 device [Phase Holographic Imaging PHI AB (publ), Sweden] was used. Cells
were seeded at the desired density, and images were taken automatically at intervals of 4 minutes.
The accompanying software, HStudio M4, was used to extract coordinates of individual cells,
further analysed by software written in-house. To minimize errors due to noise in image
acquisition, velocities were calculated from displacements between points at 8-minute intervals,
hereby from down-sampled data. Similarly, the angle autocorrelation, local density and flow field
were also calculated from the down-sampled data.
Transwell migration and invasion assays. Cell transwell migration and invasion kits,
CytoSelect™ Cell Migration Assay and CytoSelect™ 24-Well Cell Invasion Assay with basement
membrane (Cell Biolabs, Inc., San Diego, CA, USA) were used. In both assays, the plate on which
cells were seeded contains polycarbonate membrane inserts with 8 μm pores. Migratory cells are
able to pass through the pores of the membrane. In the invasion assay, the upper surface of the
membrane was coated with a uniform layer of basement membrane matrix proteins. This basement
27
membrane layer serves as a barrier to discriminate between invasive and non-invasive cells.
Invasive cells are able to degrade the matrix proteins, and ultimately pass through the pores of the
membrane. In both assays, in the upper well (i.e., insert) we added 105 cells with low-serum (1%)
growth medium; fresh complete cell growth medium was added to the bottom well. After 48 hours,
cells that migrated through the membrane were counted, following FDA staining.
Atomic Force Microscopy (AFM). Cells were cultured in 35-mm culture plates overnight.
A Petri Dish Heater (JPK Instruments, Berlin, Germany) was used to maintain cultures at 37oC,
and a flow of 5% CO2/air over the device was used to maintain proper gas exchange.
Measurements were performed using NanoWizardIII AFM (JPK Instruments) mounted on an
Olympus IX81inverted microscope.
Soft AFM probes (spring constant K=0.01-0.08 N/m) with colloidal tips (silicon-dioxide) were
used for indentation. Force-distance curves were obtained from several different points on each
cell. The local elasticity (Young's modulus) at each point was calculated by fitting the data to
Sneddon's solution for a spherical indenter 1. The Young's moduli for all points on each cell were
averaged, to yield the Young's modulus of the cell.
Immunofluorescent staining. Cells were cultured on FN-coated cover slips overnight. Where
indicated, expression of vimentin in MCF7vim+ cells was induced with doxycycline (dox), 5µg/ml
for 24 hr. Cells were fixed in either methanol (-20°C) for 2 min, or paraformaldehyde-
glutaraldehyde at room temperature for 5 min. Following fixation, cells were permeabilized with
0.5% Triton X-100 for 10 min. Cells were then processed for immunofluorescence with the
following antibodies: Anti-Cytokeratin C-11 conjugated to Alexa Fluor 488, 1:50; Anti-Vimentin
EPR3776 conjugated to Alexa Fluor 594, 1:100; Anti-tubulin, 1:200; Anti-actinin, 1:100; Anti-
28
vinculin, 1:100; Anti-catenin, 1:100; Anti-zyxin, 1:100. Actin staining was performed with FITC-
conjugated palloidin, 1:100. Following 75 min incubation with primary antibodies (/conjugated
primary antibodies/ conjugated phalloidin), cells were incubated with secondary antibodies for 1
hr if the primary antibody was unconjugated. Samples were then washed, mounted on microscope
slides, and subjected to fluorescence microscopy.
Wound-healing assays. Assays were performed on an Incucyte zoom system (Essen
BioScience, Ann Arbor, MI, USA) in 96-well ImageLock Plates (Essen BioScience) inside a CO2
incubator. A 96-pin wound-making tool (Essen BioScience) was used to make uniform scratch
wounds in all wells. 40,000 cells per well were cultured overnight. Following wound-making,
images were automatically acquired every 2 hrs. A Cell Migration Analysis software module
(Essen BioScience) was used for analysis.
Infiltration assays. Tissue culture dishes were pre-coated with 10 µg/ml FN. MDA-MB-231-
GFP WT/vim-, and CCD1069SK-RFP fibroblast cells (2∙104) were grown in the two wells of a
silicon insert (Ibidi®, Martinsried, Germany), until reaching confluence. The insert was then
removed, and the area between the wells was subjected to live-cell imaging, with 15 min intervals
between frames, for a total of 64 hrs, using a 10X0.3NA objective.
After obtaining the time-lapse images, the leading edges of the MDA231/MDA231vim- and the
stroma monolayers were detected. Each channel was converted to a binary image by thresholding.
Image intensity was binned to slices in the horizontal axis. The leading edge of the
MDA231/MDA231vim- monolayers was defined as the rightmost slice with an intensity greater
than 10% of the maximum intensity, and the leading edge of the stroma was defined as the leftmost
slice with an intensity greater than 50% of the maximum intensity.
29
Generation of stable vimentin knockdown MDA231 cell clones. We knocked down
vimentin in MDA231 cells (MDA231vim-), using an shRNA technique. The resulting cells
expressed significantly reduced levels of vimentin. To generate MDA231vim- cells, we constructed
lentiviral vectors containing an shRNA sequence, designed using Invitrogen's RNAi designer tool:
5′-GGAAGAGAACTTTGCCGTTGA-3'; the loop sequence between the sense and antisense
sequences: 5′-TTCAAGAGA-3'.
The shRNA sequences were cloned under control of a U6 in a pLL3.7 backbone. Viral particles
containing either pLL3.7-shVIM or pLL3.7-puro (control) were prepared, and parental MDA231
cells were infected with one of them. Following selection with puromycin, single-cell clones were
isolated, and reduced vimentin expression was verified (Figs. 1B, S1).
Generation of induced vimentin expression MCF7 cell clones. We successfully generated
MCF7 inducible vimentin cells (MCF7vim+), which express vimentin upon induction with
doxycycline. To generate MCF7vim+ cells, we cloned vimentin cDNA into a FuW-TetO lentiviral
vector, under control of the tetracycline operator and a minimal CMV promoter. Viral particles
containing FUW-TetO-hVimentin and FUW-M2rtTA (a constitutive FUW lentivirus carrying the
tetracycline controllable transactivator) were prepared, and parental MCF7 cells were co-infected
with the two viruses. Following selection with zeocin, single-cell clones were isolated, and
vimentin expression in the presence of doxycycline was verified.
XTT cell proliferation assays. An Xtt-based cell proliferation kit was purchased from
Biological Industries (Kibbutz Beit-Haemek, Israel). Assays were performed according to the
manufacturer's protocol. MDA231 and MDA231vim- cells were seeded at 2.5∙103, 5∙103 and 104
cells per well in 96-well plates, with 6 repeats for each cell type/concentration. Cells were allowed
30
to proliferate for 24 hrs; a reaction solution was then added for 3 hours. Dye absorbance intensity
was then measured using a plate reader at 500 nm.
Western blot analysis. Cells were lysed in modified RIPA buffer (50 mM Tris pH 7.4, 150
mM NaCl, 1 mM EDTA (AppliChem, Darmstadt, Germany), 10 mM NaF, 1% (v/v) NP40, 0.1%
sodium deoxycholate, 2 pg/ml aprotinin and 200 pg/ml AEBSF). Cell extracts were cleared by
centrifugation at 18,000 g for 10 min at 4°C, and the protein concentration of each sample was
measured by the BCA protein assay (Thermo Fisher Scientific, Waltham, MA USA). Equal
amounts of lysates were subjected to 10% SDS – PAGE, and subsequently transferred to
nitrocellulose membrane (Millipore, Billerica, MA USA). Blots were blocked with blocking buffer
for infrared immunoblotting (LI-COR #927-40000) for 1 hr, and co-incubated with primary
antibodies against vimentin (Cell Signaling #5741) and actin (Sigma Aldrich #A5441) overnight
at 4°C. Secondary antibodies coupled to IRDye infrared dyes (LI-COR #926-32211 and #926-
68070) were used for detection with an infrared Odyssey imager (LI-COR). Signal quantification
was performed using an ImageQuant system (GE Healthcare, Little Chalfont, UK).
qPCR for detection of mRNA levels of EMT proteins. mRNAs were isolated from the cells
using an RNeasy Plus Kit (Qiagen, Hilden, Germany) and reverse transcribed to cDNAs with
random primers and MultiScribe Reverse Transcriptase (Thermo Fisher, Waltham, MA USA), in
accordance with the manufacturer's protocols. cDNAs were used for gene expression analysis in
a qPCR reaction with Light Cycler 480 Probes Master (Roche Applied Science, Basel,
Switzerland) for 50 cycles. For detection of specific products, the Universal Probe Library (UPL)
platform (Roche Applied Science) was used. Glyceraldehyde-3-phosphate dehydrogenase
(GAPDH), Glucuronidase Beta (GUSB), and Glucose-6-phosphate dehydrogenase (G6PD) were
used as housekeeping genes. Gene expression levels were calculated using the 2−∆∆Ct method.
31
High-throughput cell deformability measurements using a microfluidic optical stretcher.
Measurements of cellular deformability were performed using the microfluidic optical stretcher
(MOS) system 2 provided by RS Zelltechnik GmbH (Leipzig, Germany). In brief, the MOS
generates stretching forces on the boundaries between optically different media, such as the
medium and the cell membrane, by exposing suspended cells to two opposing rays of infrared laser
light (wavelength=1060nm; power=800mW). The successive deformation of the cell is captured
in a time series of 2D phase contrast images. Every cell is monitored 1 sec before and after the 2
sec rectangular laser pulse, resulting in a sequence of totally 120 images. Subsequently, the outer
cell contour is segmented, and the diameter of the largest cell axis 𝑑(𝑡) is determined for every
time step. On the basis of the smoothed time series 𝑑𝑠(𝑡), the relative stretch of the cell is computed
as 𝜀(𝑡) =
𝑑𝑠(𝑡)
𝑑𝑢
− 1, where 𝑑𝑢 denotes the diameter of the largest axis of unloaded cells.
Comparative analysis of population measurements is performed using the parameter-free bootstrap
approach 3, which relies on calculation of sample means and corresponding confidence intervals
of empirical cell deformability distributions. In our previously published study 4, drug-treated
actin- and vimentin-deficient NK cells displayed, on average, 41% and 20% higher optical
deformability, in comparison to untreated control populations.
Statistical analyses. For the velocity and MOS data, confidence intervals were calculated
using the bootstrap method 3. As the cell tracking experiments showed great variability in cell
velocities, median bootstrapped distributions were simulated by pulling an equal amount of data
points from each experiment. Statistical significance for AFM data was established, using the
Mann-Whitney U-test. P-values for Western blots and proliferation assay data were calculated
using the Student's t-test. Box charts (e.g., Fig. 2A) display the 25th-75th percentiles in the box,
32
the median by the dividing line, the mean by the small inner box, the standard deviation by the
whiskers, and the 1st and 99th percentiles by the letter X.
33
I. Western blots for vimentin knockdown
To quantify the effects of vimentin knockdown on protein expression by the shRNA vector,
Western blots for vimentin and actin were performed (Fig. S1A). The knockdown of vimentin was
successful, decreasing its expression by 2 orders of magnitude (Fig. S1B). Actin expression levels
did not change significantly (p=0.41, Fig. S1C).
Figure S1. Western blots. (A) Western blot gel bands for MDA231 and the various MDA231vim-
clones. (B-C) Quantitative analysis of the vimentin (B) and actin bands (C), normalized by the
median read of the MDA231 cells. Error bars represent the standard error of the mean (SEM). n=6
for MDA231 and n=18 for MDA231vim-.
34
II. Exponential velocity distribution and cell heterogeneity
Velocity distribution was assessed from the cell tracking data. The complementary cumulative
distribution function 𝐹(𝑣) = ∫ 𝑃(𝑣′)𝑑𝑣′
𝑣
0
, 𝑃(𝑣) being the velocity probability distribution
function, is displayed in Fig. S2A-B for various experiments. 1 − 𝐹(𝑣) shows a good fit to an
exponential distribution (equivalent to 𝑃(𝑣) being exponentially distributed, since in that case:
𝑣
𝐹(𝑣) = ∫
0
1
𝑣
e−v′/ 𝑣𝑑𝑣′
= 1 − e−v/ 𝑣) in a robust manner for both the MDA231 and MDA231vim-
cell lines, as well as for the whole density range tested in our experiments (70-600 cells/mm2).
Furthermore, velocity distributions of single cells exhibited the same functional dependence,
indicating that the exponential distribution is inherent for the studied cells, rather than reflecting
heterogeneity in cell populations (Fig. S2C). These findings are in agreement with previous results
for various cell lines, including wild-type MDA231 5. However, we determined that during their
lifetimes, cells fluctuate around an average velocity unique to each cell, rather than around the
ensemble average. Fig. S2D shows the autocorrelation of the deviation of cell speed from the
ensemble average: 𝐶Δ𝑣(𝑡) = 〈∆𝑣𝑖(𝜏)∆𝑣𝑖(𝜏 + 𝑡)〉, where ∆𝑣𝑖(𝜏) = 𝑣𝑖(𝜏) − 𝑣(𝜏), and < > is the
ensemble and time average. 𝐶Δ𝑣(𝑡) remains positive over the entire lifetime of a cell
(approximately 1 day=1440 min). This means that in addition to the environmentally mediated
effects on cell speed discussed in this work, there might be further long-term intrinsic processes
that determine cell speed.
35
Figure S2. Cell Velocity Distribution. (A-B) Velocity distribution functions of single
experiments, at low (<120 cells/mm2) (A) and medium-high (200-600 cells/mm2) densities (B).
Both MDA231 and MDA231vim- cell lines display exponentially distributed velocities, regardless
of plate density. (C) Velocity distribution functions of 5 individual cells in a low density MDA231
experiment. The velocities of single cells also distribute exponentially, even though data is noisier,
due to a smaller sample size. (D) Time autocorrelation of velocity fluctuations of single cells from
the ensemble average. The autocorrelation does not decay to zero during the scale of a cell's
lifetime (1440 min), indicating that being fast or slow is an intrinsic cell property.
36
III. Cell proliferation rate
Differences in division rate between the MDA231 and MDA231vim- cell lines could impact
assays such as the wound-healing assay, where a faster division rate could enable cells to reach
confluency in the wound. Therefore, XTT–based proliferation assays were performed. Our results
rule out such differences, indicating that no significant difference exists between the two cell lines
(Fig. S3).
Figure S3. Cell proliferation analysis. MDA231 and MDA231vim- cells were seeded at 3 different
concentrations, and allowed to proliferate for 24 hrs. An assay reaction solution was then added
for 3 hours, and absorbance intensity was measured. No difference in proliferation rate between
MDA231 and MDA231vim- was measured. Error bars represent the SEM.
37
IV. Local density calculation
In order to capture the density heterogeneity within a plate, we calculated the local density at
every point. The local density can be thought of as the sum of additive fields 𝜙(𝑟) generated by
each cell, 𝑟 being the distance of each cell from the point. 𝜙(𝑟) should decay to zero at a large
distance from the cell; the relevant calculation (see also Fig. S4) is described by the equation:
𝜙(𝑟) =
1
𝑑𝑎
𝑟
− 1
0
2
{
𝑟 ≤ 𝑑𝑎
𝑑𝑎 < 𝑟 < 2𝑑𝑎
𝑟 ≥ 2𝑑𝑎
Here, we use 𝑑𝑎 = 50𝜇𝑚, which is approximately one cell diameter. Thus, the local density at
point 𝑟⃗ can be calculated as:
𝜌𝑙𝑜𝑐𝑎𝑙(𝑟⃗) = ∑ 𝜙(𝑟⃗ − 𝑟𝑖⃗⃗⃗)
𝑖
where the summation is over the entire population of cells within the frame, 𝑟𝑖⃗⃗⃗ being the
coordinate of each cell.
38
Figure S4. Local density field. Local density field, for a cell located at a distance r from the
considered point. The local density field saturates at 𝑟 = 50𝜇𝑚 (approximately one cell diameter),
and decays to zero at𝑟 = 100𝜇𝑚. The local density at a point 𝜌𝑙𝑜𝑐𝑎𝑙 is the sum of the fields
generated by all cells within the frame.
V. Arrangement of cells at high density
The arrangement of cells in a plate can be evaluated by the pair-correlation function, defined as:
𝑔(𝑟) = ⟨
𝑉
𝑁 − 1
∑
𝑖
𝛿(𝑟 − 𝑟𝑖)
2𝜋𝑟𝑑𝑟
⟩
Here, for each cell we sum over the rest of the cells distanced at 𝑟𝑖, 𝛿 is Kronecker's delta
function, and 𝑟 is the radial coordinate, with the chosen cell set as the origin. Thus, the number of
cells in the annulus 𝑟 ≤ 𝑅 ≤ 𝑟 + 𝛿𝑟 around the center cell is obtained, and it is normalized by the
annulus' area (2𝜋𝑟𝑑𝑟). A further time and ensemble average, denoted by < >, is performed over
N center cells for the total V imaged area. As shown in Fig. S5, 𝑔(𝑟) peaks at a typical length scale
for both cell lines, with the length being shorter for MDA231vim- cells (47 μm for MDA231, vs.
39
42 μm for MDA231vim-). This result suggests that MDA231vim- cells prefer to adhere to each other
more tightly. Alternatively, it could reflect the fact that MDA231 cells are more motile, as after
cell division, daughter cells can separate more rapidly from each other. This could lead the
MDA231 cells to form a more homogenous monolayer over time. On the other hand, a smaller
typical distance could represent a stand-alone phenomenon, due to tighter cell-cell adhesions.
Figure S5. Pair correlation function at dense plates. Averaged g(r) (over 4 curves for both
MDA231 and MDA231vim-) for experiments at high density (>400
𝑐𝑒𝑙𝑙𝑠
𝑚𝑚2). Data is marked by circles,
lines are calculated by FFT smoothing on a 5.48 µm window. For both cell lines,𝑔(𝑟)remains 0,
up to about 10 mm. Arrows mark peaks at 47 µm and 43 µm for MDA231 and MDA231vim-,
respectively, indicating the preferred cell-to-cell distance.
40
VI. Infiltration assay leading edge velocities
Figure S6. Infiltration assay
leading edge velocities. Leading edge velocities of
MDA231/MDA231vim- monolayers at infiltration assays (see Fig. 5 caption for a description of the
calculation procedure). 𝑣1, 𝑣2 correspond to the migration before and after contact with the stroma,
respectively. Mean ± SEM (n=8 for both cell lines): v1
MDA231 = 16.0 ± 1.0
μm
hr
MDA231vim−
, v1
=
10.8 ± 0.6
μm
hr
, v2
MDA231 = 7.2 ± 0.6
MDA231vim−
μm
hr
, v2
= 4.2 ± 1.2
μm
.
hr
VII. Coherent cell motility
Coherent cell migration can enhance cell motility in dense cell cultures 6. Merely reorienting
toward a coherent direction of migration prevents neighboring cells from obstructing each other's
movements. Therefore, one hypothesis we considered is that MDA231vim- cells lose this capacity
to some extent, impairing their motility at high density. However, searching for signatures of
coherent migration by means of flow field analysis described by Szabó et al. 7 (see Fig. S7A for a
schematic explanation) shows that coherent migration is very short-ranged (in terms of distance),
and similarly weak in MDA231 and MDA231vim- cell lines (Fig. S7B; Fig. S8). These results imply
that cells of both cell lines migrate individually, without correlation with their neighbors'
41
movements, and that inter-cellular cooperation does not constitute a distinctive factor between
these cell lines.
Figure S7. Flow field calculation. (A) Each cell is chosen as the origin of the axes. The frame of
reference is rotated so that the origin cell's velocity lies in the positive x direction. The neighboring
42
cells' velocities are associated with their relative displacements from the origin cell. Averaging
over cells and frames gives the flow field, which is a vector field of velocities likely to be seen
when moving with a cell's frame of reference. The results show short-range coherent movement
in the axis of movement. (B) Maps of the x,y (parallel and perpendicular) components of the flow
field. For 𝑣𝑥, the short-range correlation is noticeable at a narrow strip up to ±50 µm from the cell,
in the axis of movement. For 𝑣𝑦, no correlation is found.
Figure S8. Flow field analysis results. 𝑣𝑥 is the projection of a neighboring cell's velocity on the
direction of the origin cell's velocity. 𝑣𝑥 is associated with the 𝑥 coordinate, a projection of the
displacement between the two cells in the origin cell's heading direction. 𝑣𝑥 is normalized by the
median cell velocity in the plate 〈𝑣〉. The ensemble and time average produce the flow field profile
of an experiment. Curves show averaged data for 3 MDA231 and 2 MDA231vim- high density
experiments; data was taken from the narrow strip 𝑦 = [−31.25μ𝑚, 31.25μ𝑚]. Both cell lines
display a weak coherent movement component, decaying in the short range of approximately one
cell diameter (50 µm), with a slightly more significant component for the MDA231 cells.
43
Movie 1. Cell tracking with the Holomonitor M4 device. The fields shown are at low cell
density, and correspond to 400 min.
44
Movie 2. Velocity and local density evolution over time for MDA231 and MDA231vim- cells.
Local density (left) is calculated as detailed in Supporting Information, Section IV. Velocity field
(right) is calculated using MATLAB® biharmonic spline interpolation for the momentary velocity
of each cell. Hot/cool colors indicate high/low values (color map is the same as in Fig. 3C).
45
Movie 3. Infiltration to stroma assay. MDA231/MDA231vim- cells (colored in green on the left)
are seeded opposite to fibroblast stroma cells (colored in blue on the right). Magenta and cyan lines
show the edge layers, as detected automatically by an algorithm.
46
Supplementary References
(1) Sneddon, I. N. Int. J. Eng. Sci. 1965, 3 (1), 47–57.
(2) Guck, J.; Ananthakrishnan, R.; Mahmood, H.; Moon, T. J.; Cunningham, C. C.; Käs, J.
Biophys. J. 2001, 81 (2), 767–784.
(3) Efron, B.; Tibshirani, R. An introduction to the bootstrap; Chapman & Hall, 1994.
(4) Gladilin, E.; Gonzalez, P.; Eils, R. J. Biomech. 2014, 47 (11), 2598–2605.
(5) Czirók, A.; Schlett, K.; Madarász, E.; Vicsek, T. Phys. Rev. Lett. 1998, 81 (14), 3038–3041.
(6) Nardini, J. T.; Chapnick, D. A.; Liu, X.; Bortz, D. M. J. Theor. Biol. 2016, 400, 103–117.
(7) Szabó, a; Unnep, R.; Méhes, E.; Twal, W. O.; Argraves, W. S.; Cao, Y.; Czirók, A. Phys.
Biol. 2010, 7 (4), 46007.
47
|
1511.08475 | 2 | 1511 | 2015-12-19T21:10:49 | Homeostatic Fluctuations of a Tissue Surface | [
"physics.bio-ph",
"cond-mat.soft",
"physics.flu-dyn",
"q-bio.TO"
] | We study the surface fluctuations of a tissue with a dynamics dictated by cell-rearrangement, cell-division, and cell-death processes. Surface fluctuations are calculated in the homeostatic state, where cell division and cell death equilibrate on average. The obtained fluctuation spectrum can be mapped onto several other spectra such as those characterizing incompressible fluids, compressible Maxwell elastomers, or permeable membranes in appropriate asymptotic regimes. Since cell division and cell death are out-of-equilibrium processes, detailed balance is broken, but a generalized fluctuation-response relation is satisfied in terms of appropriate observables. Our work is a first step toward the description of the out-of-equilibrium fluctuations of the surface of a thick epithelium and its dynamical response to external perturbations. | physics.bio-ph | physics |
Homeostatic Fluctuations of a Tissue Surface
Thomas Risler,1,2 Aur´elien Peilloux,1,2 and Jacques Prost1,2,3
1Laboratoire Physico Chimie Curie, Institut Curie,
PSL Research University, CNRS, 26 rue d'Ulm, 75005 Paris, France
2Sorbonne Universit´es, UPMC Univ Paris 06, CNRS,
Laboratoire Physico Chimie Curie, 75005 Paris, France and
3Mechanobiology Institute, National University of Singapore, 5A Engineering Drive 1, 117411 Singapore
We study the surface fluctuations of a tissue with a dynamics dictated by cell-rearrangement, cell-
division, and cell-death processes. Surface fluctuations are calculated in the homeostatic state, where
cell division and cell death equilibrate on average. The obtained fluctuation spectrum can be mapped
onto several other spectra such as those characterizing incompressible fluids, compressible Maxwell
elastomers, or permeable membranes in appropriate asymptotic regimes. Since cell division and cell
death are out-of-equilibrium processes, detailed balance is broken, but a generalized fluctuation-
response relation is satisfied in terms of appropriate observables. Our work is a first step toward
the description of the out-of-equilibrium fluctuations of the surface of a thick epithelium and its
dynamical response to external perturbations.
PACS numbers: 87.10.Mn, 05.40.Ca, 87.10.Ca, 87.18.Tt
A standard way by which surfaces and interfaces are
brought out of equilibrium is the generation of a flux,
either orthogonal or parallel to their average plane, or
a combination thereof [1]. A familiar example of a par-
allel flux is that of wind blowing on the surface of wa-
ter or more generally of an interface between two im-
miscible fluids submitted to hydrodynamic shear [2]. In
the celebrated Kardar-Parisi-Zhang model [3], which has
been vastly studied by both physicists and mathemati-
cians [4, 5], the flux is orthogonal to the growing interface.
A flux can also be generated by the surface itself. This
situation is achieved by biological membranes, in which
proteins force an ion flow from one side of the membrane
to the other, the energy source being provided by the
hydrolysis of adenosine 5'-triphosphate (ATP) [6 -- 8]. A
similar situation is obtained with a membrane catalyzing
the polymerization of a biopolymer on one of its sides [9].
Biological membranes, however, can be brought out of
equilibrium without the existence of a flux by conforma-
tional changes of some of their constituents, which break
detailed balance [8]. In the case of red blood cells, for
instance, this is achieved by ATP-dependent attachment
and detachment of proteins that link the cytoskeleton to
the phospholipids [10].
In this Letter, we propose a new way of driving an in-
terface out of equilibrium. We consider the steady-state
fluctuations of the surface of a tissue layer maintained in
its homeostatic state, where cell division and cell death
equilibrate on average [11]. We use generic equations de-
rived in Ref. [12], in which stochastic cell division and
cell death interfere with elastic stresses to provide ef-
fective viscoelastic equations. We neglect the mechani-
cal role of the intercellular fluid and treat the tissue as
a one-component system, which is valid for characteris-
tic thicknesses smaller than a centimeter and time scales
larger than a few minutes [13, 14]. The reciprocal cou-
pling between mechanical stress and cell renewal in the
bulk of the tissue drives the system out of equilibrium.
Despite their out-of-equilibrium nature, which breaks de-
tailed balance and the fluctuation-dissipation relation,
stochastic cell division and cell death can be described
as Markovian processes [12]. Within this framework, the
tissue is, therefore, expected to satisfy a generalized form
of a fluctuation-response relation, which could potentially
be tested experimentally [15].
We consider an isotropic and homogeneous tissue of
prescribed average thickness H and infinite lateral exten-
sion resting on a rigid surface on one side and in contact
with a classical fluid on the other (see Fig. 1). The fluid is
FIG. 1:
Illustration of the geometry of the system under
study, with a schematic representation of the constitutive pro-
cesses driving its out-of-equilibrium fluctuations. Cells in the
tissue can be quiescent (smooth, rounded shapes), dividing
(pinched, double-nucleated shape), or undergoing cell death
(deformed and dotted cell and nucleus boundaries).
of negligible viscosity and exerts a constant homogeneous
pressure on the tissue, equal to the tissue homeostatic
pressure [11]. Under these conditions and on time scales
shorter than the characteristic evolution time of the av-
zxH0(cid:98)HH erage thickness H (see the Supplemental Material [16],
Sec. I), the tissue is in a steady state characterized by
specific average values ρh and σh of the cell-number den-
sity ρ and isotropic tissue stress σ and by a vanishing
cell-velocity field vα, where the index α stands for the
spatial coordinates x, y, or z.
The cell-number density ρ and the cell-velocity field vα
obey the continuity equation
∂tρ + ∂α(ρvα) = (kp + ξc)ρ ,
(1)
where ∂t and ∂α denote the partial derivatives with re-
spect to time and spatial coordinates, with summation
over repeated indices. In this equation, kp = kd − ka is
the global mean rate of cell production, taking into ac-
count cell division and apoptosis with rates kd and ka,
and ξc is a Gaussian white noise of zero mean. Cell di-
vision and cell death being independent stochastic pro-
cesses, the variance of the noise reads (cid:104)ξc(r, t)ξc(r(cid:48), t(cid:48))(cid:105) =
[(kd +ka)/ρ]δ(r−r(cid:48))δ(t−t(cid:48)). The stress tensor σαβ obeys
the force-balance condition ∂ασαβ = 0.
Close to the homeostatic state and to linear order, the
deviation of the density from its homeostatic value δρ =
ρ−ρh is proportional to δσ = σ−σh. The cell-production
rate then reads kp = −(1/τi) (δρ/ρh) = (1/τi) χcδσ,
where τi is a characteristic time and χc is the short-
time tissue compressibility in the homeostatic state. The
stress tensor is decomposed into an isotropic and a trace-
less part σαβ = σδαβ + σαβ, where σαα = 0. Under these
conditions, the deviation δσαβ = δσδαβ + δσαβ of the
stress tensor with respect to its homeostatic value obeys
the following Maxwell model [12]:
and deviatory modes are maintained at different temper-
atures.
2
To characterize the position fluctuations δH(r, t) of
the tissue upper surface, we focus our attention on the
two-point autocorrelation function C(r − r(cid:48), t − t(cid:48)) =
(cid:104)δH(r, t)δH(r(cid:48), t(cid:48))(cid:105) and on its linear response function
χ to an externally applied pressure field δPe(r, t) de-
fined as (cid:104)δH(r, t)(cid:105) = (cid:82) χ(r − r(cid:48), t − t(cid:48))δPe(r(cid:48), t(cid:48)) dr(cid:48)dt(cid:48).
transform: f (q, ω) = (cid:82) dr dt f (r, t) exp [−i(q · x + ωt)].
We further consider the Fourier components of the dif-
ferent fields defined according to the following Fourier
In a linear perturbation close to a flat interface, the dif-
ferent Fourier modes decouple, and we can consider the
dependence on one single wave vector q without loss of
generality. Close to the homeostatic state, the equation
governing the cell-velocity field reads in frequency space,
η1∂α∂αvβ +
ζ1 +
η1
∂β(∂αvα) = ∂αξ1,αβ ,
(3)
with
and
ζ1 =
ζ
1 + iωτi
,
η1 =
η
1 + iωτa
,
(4)
ξ1,αβ =
1
1 + iωτi
ξδαβ +
1
1 + iωτa
ξαβ .
(5)
Writing the stress continuity at the upper surface with
generalized Laplace pressure and negligible shear stress,
we obtain in Fourier space and in the linear regime:
(cid:18)
(cid:19)
1
3
(cid:19)
(cid:18)
ζ1 − 2
3
η1
(1 + τi∂t) δσ = ζ∂γvγ − ξ
(1 + τa∂t) δσαβ = 2η vαβ − ξαβ .
(iqvx + ∂zvz) + 2η1∂zvz + (γq2 + κq4) δH = ξ1,zz
(2)
and η1 (iqvz + ∂zvx) = ξ1,xz .
(6)
In the isotropic part, ζ = χ−1
c τi is an effective bulk vis-
cosity, and ξ = ζξc is an effective bulk noise term of
amplitude ϑ = ζ 2(kd + ka)/ρh. The remarkable ab-
sence of any compression modulus is due to the non-
conservation of cell number [11, 12]. In the anisotropic
part, τa, η, and ξαβ are the effective relaxation time
constant, shear viscosity, and noise, a priori different
from those appearing in the isotropic part, and vαβ =
(1/2) [∂αvβ + ∂βvα − (2/3)∂γvγδαβ] is the traceless part
of the velocity-gradient tensor. As done in Ref. [12], we
choose for ξαβ a Gaussian white noise and write its second
moment by symmetry arguments: (cid:104) ξαβ(r, t) ξγδ(r(cid:48), t(cid:48))(cid:105) =
θ [δαγδβδ + δαδδβγ − (2/3)δαβδγδ] δ(r− r(cid:48))δ(t− t(cid:48)), where
θ characterizes
this
anisotropic noise is related to the stochasticity of cell
rearrangements.
In general, the two noises cannot be
cast in a way respecting detailed balance with a common
effective temperature, and the system is out of equilib-
rium. One can think of such a tissue as a system in which
the particle number is not conserved and the isotropic
the amplitude.
In practice,
Here, γ is the tissue surface tension and κ its surface
bending rigidity. To these must be added the kinematic
condition vz = iωδH. At its lower surface, the tissue is
in contact with a hard wall, and we consider a no-slip
boundary condition, such that vx = vz = 0.
General expressions for the two-point autocorrelation
function and the linear response function are given in the
Supplemental Material [16], Sec. II. Using these results,
we characterize how far from thermodynamic equilibrium
the system is operating by the introduction of an effective
temperature Teff equal to the thermodynamic tempera-
ture at equilibrium:
kBTeff (q, ω) =
ω
2
C(q, ω)
χ(cid:48)(cid:48)(q, ω)
.
(7)
Here, kB is the Boltzmann constant, and χ(cid:48)(cid:48)(q, ω) is the
imaginary part of the response function in Fourier space.
The effective temperature is, in general, frequency and
wavelength dependent, and its full expression is given in
the Supplemental Material [16], Sec. II. If we choose ther-
mal noise amplitudes ϑ = 2ζkBT and θ = 2ηkBT , which
define an equilibrium Maxwell model, the effective tem-
perature equals the thermodynamic temperature T at all
frequencies. More generally, when ϑ/θ = ζ/η, the effec-
tive temperature is constant and the system is analogous
to an equilibrium system.
We first consider the limiting case where the thickness
H of the tissue is infinite or equivalently where qH (cid:29) 1,
for which full analytic expressions can be presented here.
In this limit, the functions defined above read
ϑ1η12 + θ1
3 η1
1
C(q, ω) (cid:39) 2
q
χ(q, ω) (cid:39) − 1
q
3η12(cid:1)
(cid:0)ζ1 + 1
(cid:12)(cid:12)2 γq + κq3 + 2iω η12
3η12(cid:17)
(cid:16)(cid:12)(cid:12)ζ1 + 1
(cid:1)∗(cid:105)
(cid:0)ζ1 + 1
kBTeff (q, ω) (cid:39) ϑ1η12 + θ1
(cid:12)(cid:12)ζ1 + 4
(cid:104)
(cid:12)(cid:12)2
(cid:1)(cid:0)ζ1 + 4
3 η1
γq + κq3 + 2iω η1
3 η12 + 1
2Re
η1
3 η1
+ 1
3 η1
,(8)
with ϑ1 = ϑ/[1 + (ωτi)2], θ1 = θ/[1 + (ωτa)2] and
η1 = η1(ζ1 + η1/3)/(ζ1 + 4η1/3). The star denotes the
complex conjugate. Note that, due to the absence of an
intrinsic length scale in the system in this case, the effec-
tive temperature has no dependence on the wave vector.
In the low-frequency limit (ωτa, ωτi (cid:28) 1), these func-
tions correspond to those of a passive compressible fluid
with bulk and shear viscosities ζ and η at the temper-
ature Teff but with the originality of a vanishing com-
pression modulus. As already stated, the long-time
compression modulus vanishes in a tissue due to the
nonconservation of cell number [11, 12], a feature that
is impossible in a system with a conserved number of
particles.
Interestingly, the correspondence also holds
with an incompressible fluid with effective shear viscos-
ity η = η(ζ + η/3)/(ζ + 4η/3), since then compression
modes are infinitely fast, and only viscous terms are left
in the correlation and response functions. In the oppo-
site limit (ωτa, ωτi (cid:29) 1), the expressions are analogous to
those of a passive, one-component, compressible Maxwell
elastomer.
In the case where the thickness H of the tissue layer
is finite, as stated above, the effective temperature is
both frequency and wavelength dependent. In the low-
frequency regime, it depends on wave vector only (see the
Supplemental Material [16], Sec. III for its expression).
When ϑ/θ (cid:54)= ζ/η, it can exhibit different nontrivial be-
haviors depending on the parameter values (see Fig. 2).
Despite the existence of a frequency- and wave-vector-
dependent effective temperature, the system's dynamics
is Markovian. Proper observables, therefore, exist in
terms of which a generalized fluctuation-response rela-
tion is satisfied [15]. We choose here to illustrate this
fact in the low-frequency domain (ωτa, ωτi (cid:28) 1), where
3
FIG. 2: Wave-vector dependence of the effective temperature
for six different parameter sets.
In all six panels, we plot
the adimensional ratio ¯Teff = 2ζkBTeff /ϑ as a function of
¯H = qH, for different values of ϑ/θ and ζ/η. In panels (a) --
(d), ϑ/θ − ζ/η = 1, and ζ/η takes the values 1/3, 2/3, 1, and
In panels
4/3 in panels (a), (b), (c), and (d), respectively.
(e) and (f), ϑ/θ − ζ/η = −1, and ζ/η equals 1.01 and 4/3,
respectively. In panel (b), the asymptotic values in the low-
and large-q limits are both equal to 3/5.
we can write a single evolution equation for δH, directly
under a Markovian form. This allows us to present the
explicit construction of an observable X(q, t) that obeys
a fluctuation-response relation. Solving for the integra-
tion constants and using the kinematic condition at the
upper surface vz = ∂tδH, we get
∂tδH(q, t) = −τ (q)
−1δH(q, t) + f (q, t) + ξδH (q, t) , (9)
where τ (q) is a relaxation time constant that depends
on the wave vector, f (q, t) is an external field acting
on the tissue upper surface, and the noise ξδH (q, t) is
a scalar combination of integrals containing the matri-
cial noise ξαβ = ξδαβ + ξαβ (see the Supplemental Ma-
terial [16], Sec. IV for explicit expressions). Introducing
the equal-time autocorrelation function Σ(q) of δH(q, t)
in the absence of an external field as (cid:104)δH(q, t)δH(q(cid:48), t)(cid:105) =
Σ(q)δ(q + q(cid:48)), we define X(q, t) as
X(q, t) = −τ (q)Σ(q)
−1 δH(q, t) .
(10)
With these definitions, we have [15]
χXX (q, ω) − χXX (q,−ω) = iωCXX (q, ω) ,
(11)
where CXX and χXX are, respectively, the two-point au-
tocorrelation function of the variable X and its linear
response to the external field f , here written in Fourier
space. The corresponding explicit expressions are given
in the Supplemental Material [16], Sec. IV.
Within the low-frequency domain, the regime qH (cid:28) 1
is particularly interesting, since the expressions of the
correlation and response functions are then analogous to
those of a permeable membrane with permeation con-
¯Te↵!!!!a) ¯H01234560.40.420.440.460.4801234560.60.610.620.6301234560.660.680.70.7201234560.70.720.740.760.7801234565761656901234562.52.72.93.1¯Te↵¯H¯Hb) c) d) e) f ) stant λp = H/(ζ + 4
3 η) [17]:
C(q, ω) (cid:39) 1
H
χ(q, ω) (cid:39) −
ϑ + 4
3 θ
−2
(γq2 + κq4)2 + ω2λ
p
1
−1
γq2 + κq4 + iωλ
p
kBTeff (cid:39) 1
2
ϑ + 4
3 θ
ζ + 4
3 η
.
(12)
It is remarkable that, due to the nonconservation of cell
number and the subsequent absence of long-time com-
pression modulus, there is no lubrication regime. The
leading term in the equal-time correlation function reads
C(q, t − t
(cid:48)
= 0) (cid:39) kBTeff
γq2 + κq4 ,
(13)
and we can make direct use of the results concerning the
fluctuations of equilibrium membranes [18, 19].
(cid:112)
The divergence of the mean-square fluctuations of the
tissue thickness is of particular interest. There are three
length scales entering the problem. The first one is sim-
ply the thickness H. The second one is given by the
square root of the ratio of the bending modulus over the
membrane tension lκ =
κ/γ. Finally, the third one,
the "collision" length lc, stems from the divergence of
the mean-square fluctuations of the tissue thickness as
a function of lateral position [20, 21].
It is such that
If γH 2 (cid:28) kBTeff , lc is pro-
(cid:104)δH(0, t)δH(lc, t)(cid:105) = H 2.
portional to the thickness H and is much smaller than
lκ:
(cid:114) κ
(cid:115)
lc ∝ H
= lκ
kBTeff
γH 2
kBTeff
.
(14)
In this regime, the tissue surface fluctuations scale like
distances parallel to the interface.
In the tension-
dominated regime where γH 2 (cid:29) kBTeff , lc reads
lc ∝ lκ exp
.
(15)
(cid:18) πγH 2
(cid:19)
kBTeff
In this regime, fluctuation amplitudes scale logarithmi-
cally with the distance parallel to the interface, and lc is
much larger than lκ. The surface appears locally smooth,
although thickness variations eventually reach values of
the order of the thickness itself on length scales of order
lc.
A natural question concerns the possibility to ob-
serve such effects. The effective temperature in the low-
frequency, small-wave-vector regime can be estimated us-
ing Eq. (12). Keeping only the isotropic contribution in
the homeostatic state, we obtain kBTeff = ζkd/ρh. Using
this expression and numerical values from the literature,
we estimate that the ratio lc/lκ is typically of order one
tenth to several tens and can potentially be much larger
4
(see the Supplemental Material [16], Sec. V). This indi-
cates that the two limiting regimes mentioned here could,
in principle, be observed, and that changing the physi-
cal properties of the tissue could affect its surface struc-
ture without necessarily requiring a modification of the
cell-division rate. The surface of precancerous tissues is
usually rougher than that of healthy tissues, even when
these tissues are not in a growth phase [22, 23]. One
could speculate that some of the early cancerous muta-
tions affect certain physical properties of the tissue, not
necessarily uniquely related to cell proliferation.
In equilibrium membranes, the existence of a collision
length generates an entropic repulsive interaction. In tis-
sues, the physical meaning is likely to be very different.
In the case we discuss here, this suggests that the tissue
thickness potentially vanishes. The onset of tissue disap-
pearance is given by the characteristic time of thickness
fluctuations at the scale lc:
τd =
.
(16)
3 η(cid:1) l2
(cid:0)ζ + 4
(cid:1) H
(cid:0)γ + κl
−2
c
c
The fate of the tissue layer then depends on the future
evolution of the region of zero thickness. This, in turns,
depends on the tissue-substrate interaction and adhesion
properties, and a detailed analysis of this outcome is be-
yond the scope of this work. Note that the linear approx-
imation used in this Letter is valid as long as κ (cid:29) kBTeff .
In the opposite case, the surface would appear crumpled
with no angular coherence [19].
The tissue surface fluctuations we have studied here
exhibit a surprisingly rich behavioral diversity. One can
define a wave-vector- and frequency-dependent effective
temperature, a clear signature of the out-of-equilibrium
nature of the system. In the low-frequency regime, this
effective temperature is only wave-vector dependent. As
shown in Fig. 2, it can present different nontrivial behav-
iors, among which being monotonically varying or having
an extremum at a finite wave vector. In the large-wave-
vector, high-frequency regime, the tissue behaves like a
compressible Maxwell elastomer with nonthermal noise,
whereas in the large-wave-vector, low-frequency regime,
it behaves like an incompressible fluid at equilibrium. Fi-
nally, in the small-wave-vector, low-frequency regime, the
mapping can be made to a fully permeable membrane
with tension and curvature. This implies that, at suffi-
ciently large scales, the tissue always has thickness fluc-
tuations on the order of the average thickness. This could
signal a finite lifetime for the tissue layer. In actual ep-
ithelial tissues, however, such an outcome is usually not
observed.
In this case, cell division and cell death are
regulated by external chemical gradients. As a conse-
quence, cells are globally renewed close to the basal mem-
brane and die close to the apical surface, which leads to a
well-defined thickness [24 -- 26]. One could speculate that
this is part of the solutions developed under evolutionary
pressure to create robust epithelial tissues.
5
T. Risler, B. Cabane, D. Vignjevic, J. Prost, G. Cappello,
et al., Phys. Rev. Lett. 107, 188102 (2011).
[28] F. Montel, M. Delarue, J. Elgeti, D. Vignjevic, G. Cap-
pello, and J. Prost, New J. Phys. 14, 055008 (2012).
[29] G. Forgacs, R. A. Foty, Y. Shafrir, and M. S. Steinberg,
Biophys. J. 74, 2227 (1998).
[30] K. Guevorkian, M.-J. Colbert, M. Durth, S. Dufour, and
F. Brochard-Wyart, Phys. Rev. Lett. 104, 218101 (2010).
[31] D. A. Beysens, G. Forgacs, and J. A. Glazier, Proc. Natl.
Acad. Sci. U.S.A. 97, 9467 (2000).
[32] P. Marmottant, A. Mgharbel, J. Kafer, B. Audren, J.-P.
Rieu, J.-C. Vial, B. van der Sanden, A. F. M. Mar´ee,
F. Graner, and H. Delanoe-Ayari, Proc. Natl. Acad. Sci.
U.S.A. 106, 17271 (2009).
[33] Y.-C. Fung, Biomechanics: Mechanical Properties of Liv-
ing Tissues (Springer, New York, 1993).
[34] R. A. Foty, G. Forgacs, C. M. Pfleger, and M. S. Stein-
berg, Phys. Rev. Lett. 72, 2298 (1994).
[35] R. A. Foty, C. M. Pfleger, G. Forgacs, and M. S. Stein-
berg, Development 122, 1611 (1996).
[36] R. A. Foty and M. S. Steinberg, Dev. Biol. 278, 255
(2005).
[37] E.-M. Schotz, R. D. Burdine, F. Julicher, M. S. Steinberg,
C.-P. Heisenberg, and R. A. Foty, HFSP J. 2, 42 (2008).
[38] A. Mgharbel, H. Delanoe-Ayari, and J.-P. Rieu, HFSP J.
3, 213 (2009).
[39] D. Gonzalez-Rodriguez, K. Guevorkian, S. Douezan, and
F. Brochard-Wyart, Science 338, 910 (2012).
[1] M. E. Cates and M. R. Evans, Soft and Fragile Mat-
ter: Nonequilibrium Dynamics, Metastability and Flow
(PBK) (CRC Press, Bristol, 2000).
[2] M. Thi´ebaud and T. Bickel, Phys. Rev. E 81, 031602
(2010).
[3] M. Kardar, G. Parisi, and Y.-C. Zhang, Phys. Rev. Lett.
56, 889 (1986).
[4] T. Halpin-Healy and Y.-C. Zhang, Phys. Rep. 254, 215
(1995).
[5] M. Hairer, Ann. Math. 178, 559 (2013).
[6] J. Prost and R. Bruinsma, Europhys. Lett. 33, 321
(1996).
[7] S. Ramaswamy, J. Toner, and J. Prost, Phys. Rev. Lett.
84, 3494 (2000).
[8] D. Lacoste and P. Bassereau, in Liposome, Lipid Bilay-
ers and Model Membranes, edited by Georg Pabst et al.
(CRC Press, Boca Raton, 2014).
[9] A. Maitra, P. Srivastava, M. Rao, and S. Ramaswamy,
Phys. Rev. Lett. 112, 258101 (2014).
[10] T. Betz, M. Lenz, J.-F. Joanny, and C. Sykes, Proc. Natl.
Acad. Sci. U.S.A. 106, 15320 (2009).
[11] M. Basan, T. Risler, J.-F. Joanny, X. Sastre-Garau, and
J. Prost, HFSP J. 3, 265 (2009).
[12] J. Ranft, M. Basan, J. Elgeti, J.-F. Joanny, J. Prost, and
F. Julicher, Proc. Natl. Acad. Sci. U.S.A. 107, 20863
(2010).
[13] J. Ranft, J. Prost, F. Julicher, and J.-F. Joanny, Eur.
Phys. J. E 35, 46 (2012).
[14] M. Delarue, J.-F. Joanny, F. Julicher, and J. Prost, In-
terface Focus 4, 20140033 (2014).
[15] J. Prost, J.-F. Joanny, and J. M. R. Parrondo, Phys. Rev.
Lett. 103, 090601 (2009).
[16] See the Supplemental Material, which includes Refs. [27 --
39], for further details.
[17] J. Prost, J.-B. Manneville, and R. Bruinsma, Eur. Phys.
J. B 1, 465 (1998).
[18] R. Lipowsky and E. Sackmann, Structure and Dynamics
of Membranes: I. From Cells to Vesicles/II. Generic and
Specific Interactions (Elsevier, New York, 1995).
[19] D. R. Nelson, T. Piran, and S. Weinberg, Statistical Me-
chanics of Membranes and Surfaces (World Scientific,
Singapore, 2004).
[20] M. E. Fisher and D. S. Fisher, Phys. Rev. B 25, 3192
(1982).
[21] W. Helfrich and R.-M. Servuss, Nuovo Cimento Soc. Ital.
Fis. 3D, 137 (1984).
[22] R. A. Weinberg, The Biology of Cancer (Garland Science,
New York, 2007).
[23] F. A. Tavassoli and P. Devilee, eds., Pathology and Ge-
netics of Tumours of the Breast and Female Genital
Organs (International Agency for Research on Cancer
Press, World Health Organization, and Oxford Univer-
sity Press, Oxford, 2003).
[24] B. Young, J. S. Lowe, A. Stevens, and J. W. Heath,
Wheater's Functional Histology: A Text and Colour At-
las (Churchill Livingstone Elsevier, London, 2006), 5th
ed.
[25] M. Basan, J.-F. Joanny, J. Prost, and T. Risler, Phys.
Rev. Lett. 106, 158101 (2011).
[26] T. Risler and M. Basan, New J. Phys. 15, 065011 (2013).
[27] F. Montel, M. Delarue, J. Elgeti, L. Malaquin, M. Basan,
SUPPLEMENTAL MATERIAL
6
Homeostatic Fluctuations of a Tissue Surface
Thomas Risler,1,2 Aur´elien Peilloux,1,2 and Jacques Prost1,2,3
1Laboratoire Physico Chimie Curie, Institut Curie, PSL Research University, CNRS, 26 rue d'Ulm, 75005 Paris,
France
2Sorbonne Universit´es, UPMC Univ Paris 06, CNRS, Laboratoire Physico Chimie Curie, 75005 Paris, France
3Mechanobiology Institute, National University of Singapore, 5A Engineering Drive 1, 117411 Singapore
I. SLOW TIME EVOLUTION OF THE AVERAGE THICKNESS H
In our study, we treat the average thickness H as a constant in time.
In order to justify this
framework, we estimate here the characteristic time over which the average tissue thickness H evolves
by diffusion, due to stochastic cell-divison and cell-death processes, with average rates kd = ka.
We consider a tissue layer made of N cells of average cell-number density ρ, charateristic lateral
dimension L and average thickness H. In this volume, the cell-number standard deviation in a time
dt is δN =(cid:112)(kd + ka) N dt. For a total volume V = H L2 = N/ρ, the corresponding amplitude of
thickness fluctuations reads:
δH 2 =
(kd + ka) H
ρ L2
dt .
(17)
To diffuse over distances of order H, the tissue layer therefore takes a characteristic time τH (cid:39)
ρ H L2/(kd + ka). With a maximal rate of cell renewal of one per day, a cell-diameter of 10 µm,
a tissue thickness of 100 µm, this characteristic time of the order of several hundred years for a
tissue patch of one millimeter square, and several tens of thousand years for one of one centimeter
square. Ignoring the temporal variations of the average thickness H is therefore perfectly valid in
the framework of our study.
II. EXPRESSIONS OF THE PHYSICAL QUANTITIES FOR THE TISSUE OF FINITE THICKNESS H
The two-point autocorrelation function reads
C(q, ω) =
2
q
ϑ1Na,1 + θ1Nb,1
(γq + κq3)Da,1 + 2iωη1Db,12 ,
(18)
where
Na,1 =
Nb,1 =
+ 8 ¯H
2 + 2(ζ1 + ζ∗
+2
(cid:104)
(cid:17)
+8 ¯H
+ ζ12
e8 ¯H − 1
e8 ¯H − 1
e4 ¯H + 1
1 ) + 3ζ12]
e4 ¯H − 1
+ 2(ζ1 + ζ∗
(cid:12)(cid:12)(cid:12)2 + ζ1
(cid:12)(cid:12)(cid:12)2(cid:16)
+ 4 ¯Hζ12e2 ¯H(cid:16)
(cid:17)
(cid:16)−2 + ζ12 + 2 ¯H 2[2(ζ1 + ζ∗
e2 ¯H(cid:16)
(cid:17)
(cid:12)(cid:12)(cid:12)2(cid:18)1
(cid:18)5
(cid:19)(cid:16)
(cid:12)(cid:12)(cid:12)2 + ζ1
(cid:17) − 4 ¯Hζ12
(cid:16)
(cid:104)−ζ1(1 + ζ1)
1 )2(cid:17)
(cid:18)1
(cid:19)(cid:16)
(cid:16)
(cid:104)
1 )2(cid:17)
(cid:18)1
(cid:19)(cid:16)−2 + ζ12 + 2 ¯H 2(2(ζ1 + ζ∗
+
1 ) + 3ζ12)
(cid:17)
(cid:16)
(cid:17)
2 + (1 − 2 ¯H 2)ζ12 + 2(ζ1 + ζ∗
1 )
2 + 2ζ∗
(cid:17) − 4 ¯H ζ1 e2 ¯H(cid:105)
(1 + ζ1)(2 + ζ1)
− ζ∗
+
2 + 2ζ∗
1 + (1 + 2 ¯H 2)(ζ∗
1 + (1 + 2 ¯H 2)(ζ∗
e4 ¯H − 1
(cid:17)(cid:21)
(cid:16)
1
− ζ∗
1
3
3
(cid:16)
3
+
3
+
(cid:104)
+2
e4 ¯H
Da,1 = (1 + ζ1)
(2 + ζ1)
Db,1 = ζ1(2 + ζ1)
e4 ¯H + 1
+ 2
2 + 2ζ1 + (1 + 2 ¯H 2)ζ 2
1
e2 ¯H ,
with ζ1 = ζ1/η1 + 1/3 and ¯H = qH. We also have, as in the main text,
1 ) + (1 − 2 ¯H 2)ζ12(cid:105)
(cid:19)
(cid:17)
e2 ¯H(cid:16)
e4 ¯H + 1
(cid:17)
1 ) + ζ12
(cid:17)(cid:21)
e2 ¯H(cid:16)
(cid:17)
e4 ¯H − 1
7
e4 ¯H
(19)
(20)
ζ1 =
ζ
1 + iωτi
,
η1 =
η
1 + iωτa
, ϑ1 =
ϑ
1 + (ωτi)2 ,
θ1 =
θ
1 + (ωτa)2 .
Note that even though it is not directly visible under this form, the expression of Nb,1 above is real.
The linear response function reads
(cid:32)
(cid:33)
χ(q, ω) = −1
q
1
γq + κq3 + 2iωη1
Db,1
Da,1
We then have for the effective temperature:
kBTeff(q, ω) =
ϑ1Na,1 + θ1Nb,1
2Re(η1D∗
a,1Db,1)
.
.
(21)
(22)
We recover the thermodynamic limit of an equilibrium Maxwell model by choosing noise amplitudes
that respect detailed balance, ϑ = 2ζkBT and θ = 2ηkBT . We then have kBTeff(q, ω) ≡ kBT ,
independent of wave vector and frequency.
In general, the effective temperature given by Eq. (22) above is a function of frequency and wave
vector, as well as of the parameters defining the problem. In the low- and large-q limits, it has two
asymptotic finite limits, which have simple expressions. The asymptotic expression in the large-
wave-vector regime (qH (cid:29) 1) is given in Eq. (8) of the main text. In the small-wave-vector regime
(qH (cid:28) 1), we have
(23)
In the low-frequency domain (ωτa, ωτi (cid:28) 1), the effective temperature is just a function of the wave
vector q. Its asymptotic expressions read
.
kBTeff (cid:39) 1
2
ϑ1 + 4
3θ1
Re(ζ1 + 4
3η1)
3η2(cid:105)
(cid:104)(cid:0)ζ + 1
3η(cid:1)2 + 1
2η(cid:0)ζ + 1
3η(cid:1)(cid:0)ζ + 4
3η(cid:1)
kBTeff (cid:39) ϑη2 + θ
(24)
in the large-q limit, and
kBTeff (cid:39) 1
2
ϑ + 4
3θ
ζ + 4
3η
in the low-q limit, as given in Eq. (12) of the main text.
8
(25)
III. EXPRESSIONS OF THE PHYSICAL QUANTITIES FOR THE TISSUE OF FINITE THICKNESS H
IN THE LIMIT OF LONG TIMESCALES
In the limit of timescales much larger than τi and τa, where ωτi (cid:28) 1 and ωτa (cid:28) 1 in the frequency
domain, ζ1 and η1 are real and reduce to the bulk and shear viscosities ζ and η, respectively.
Therefore, the expressions given in section 2 above can be simplified by replacing all the variables
by their real, simpler counterparts with ωτi (cid:28) 1 and ωτa (cid:28) 1. We then obtain
C(q, ω) =
2
q
ϑNa + θNb
(γq + κq3)Da + 2iωηDb2 ,
where
Na =
Nb =
(cid:16)
(cid:16)
+2
2 + ζ
2 + ζ
¯H
(cid:104)
−8
3
2
3
+
(cid:17)
e4 ¯H − 1
e4 ¯H + 1
+ 8 ¯H
e8 ¯H − 1
(cid:17)
(cid:18)5
(cid:104)
2 + 4ζ + (1 − 2 ¯H 2)ζ 2(cid:105)
+ 4 ¯H ζ 2e2 ¯H(cid:16)
(cid:17)2(cid:16)
(cid:17)
(cid:16)−2 + ζ 2 + 2 ¯H 2 ζ(4 + 3ζ)
(cid:17)
e2 ¯H(cid:16)
(cid:19)
(cid:19)(cid:16)
(cid:17)2(cid:18)1
(cid:17)
e2 ¯H(cid:16)
(cid:17) − 4 ¯H ζ 2
(cid:104)−2 + 8ζ + (23 + 2 ¯H 2)ζ 2 + 12ζ 3 + (3 + 6 ¯H 2)ζ 4(cid:105)
10 + 4(9 + 2 ¯H 2)ζ + (31 − 6 ¯H 2)ζ 2 + 12ζ 3 + (3 + 6 ¯H 2)ζ 4(cid:105)
e2 ¯H(cid:16)
(cid:104)
(cid:16)
e4 ¯H − 1
e8 ¯H − 1
+ 4ζ + ζ 2
e4 ¯H + 1
+ ζ 2
3
(cid:16)
(cid:17)
(cid:16)
e4 ¯H
3
(cid:17)
e4 ¯H − 1
Da = (1 + ζ)
(2 + ζ)
Db = ζ(2 + ζ)
e4 ¯H + 1
+ 2
(cid:17) − 4 ¯H ζ e2 ¯H(cid:105)
2 + 2ζ + (1 + 2 ¯H 2)ζ 2(cid:17)
(cid:32)
1
e2 ¯H ,
(cid:33)
.
χ(q, ω) = −1
q
γq + κq3 + 2iωη Db
Da
with ζ = ζ/η + 1/3 and ¯H = qH. The linear response function reads
(26)
e4 ¯H
(27)
(28)
(29)
We then have for the effective temperature:
kBTeff(q, ω) =
ϑNa + θNb
2(ηDaDb)
.
We recover the constant thermodynamic temperature T of a fluid at equilibrium by choosing ϑ =
2ζkBT and θ = 2ηkBT .
IV. EFFECTIVE FLUCTUATION-RESPONSE RELATION IN THE LIMIT OF LONG TIMESCALES
A. Tissue of finite thickness H
9
The evolution equation for δH(q, t) presented in the main text in the low-frequency domain
(ωτa, ωτi (cid:28) 1) reads
∂tδH(q, t) = −τ (q)−1δH(q, t) + f (q, t) + ξδH(q, t) .
It contains the following functions:
where Da and Db have already been introduced above in Eq. (27). We also have introduced
NδH(q, t) = (2 + ζ)
e4 ¯H + 2
(2 − 2 ¯H ζ)I2 + (1 + 2 ¯H)ζJ2
e3 ¯H
(2 + ζ + 2 ¯H ζ + 2(−1 + ¯H) ¯H ζ 2)I1 + (−1 + 2 ¯H)ζJ1
e2 ¯H
+2
1
2ηq
NδH(q, t)
Db
,
(cid:105)
(cid:105)
(30)
(31)
(32)
(33)
with
and
τ (q) =
and
ξδH(q, t) =
2η
Db
Da
γq + κq3
(−2 + ζ)I1 − 2J1
(cid:104)
(cid:105)
(cid:104)
(cid:104)−4(1 + ¯H ζ))I2 + 2(2 + ζ)J2
(cid:105)
(cid:104)
+
(cid:90) 0
−H
− ζq
2
e ¯H − ζ(2 + ζ)I1 ,
−H
2
−H
(cid:90) 0
dz e−qz (ξxx − ξzz − 2iξxy)z
dz eqz (ξxx − ξzz + 2iξxy)z
(cid:90) 0
I1(q, t) = − q
2
I2(q, t) = −qe− ¯H
(cid:40) ζq
(cid:90) z
dz(cid:48) eq(2z−z(cid:48)) (ξxx − ξzz − 2iξxy)z(cid:48) + eqz(cid:104)−ξzz +
(cid:40) ζq
(cid:16)
(cid:90) z
(ξxx − ξzz − 2iξxy)z(cid:48) + e−qz(cid:104)
dz(cid:48) eq(−2z+z(cid:48)) (ξxx − ξzz + 2iξxy)z(cid:48)
(ξxx − ξzz + 2iξxy)z(cid:48)
dz
(cid:90) 0
(cid:90) 0
(cid:90) 0
z
dz(cid:48) e−qz(cid:48)
dz(cid:48) eqz(cid:48)
2
−H
2
−H
dz
−H
ξzz +
J1(q, t) = q
J2(q, t) = qe− ¯H
− ζq
2
z
(cid:41)
(cid:105)z
(cid:41)
(cid:16)
(cid:17)
1 − ζ
iξxy
(cid:17)
(cid:105)z
iξxy
1 − ζ
,
(34)
where the noise components are understood as functions of (q, z, t). Dependencies in the coordinate
z are indicated as lowerscripts. The equal-time autocorrelation function Σ(q), defined such that
(cid:104)δH(q, t)δH(q(cid:48), t)(cid:105) = Σ(q)δ(q + q(cid:48)), is then given by
Σ(q) =
ϑNa + θNb
2η(γq2 + κq4)DaDb
,
(35)
where Na, Nb, Da and Db are the functions of q introduced in Eq. (27). Finally, the observable
X(q, t) defined as
X(q, t) = −τ (q)Σ(q)−1 δH(q, t)
(36)
reads
X(q, t) = − 4η2q
ϑNa + θNb
D2
b δH(q, t) .
It has the following correlation and response functions in the time domain:
8η3
CXX(q, t) =
χXX(q, t) = − 4η2q
ϑNa + θNb
(γ + κq2)(ϑNa + θNb)
b H(t)e−t/τ (q) ,
D2
e−t/τ (q)
D3
b
Da
10
(37)
(38)
where H(t) is the Heaviside step function. In the frequency domain, they read
32η4q
CXX(q, ω) =
ϑNa + θNb
χXX(q, ω) = − 8η3q
D4
b
a + 4ω2η2D2
(γq + κq3)2D2
b
D3
b
ϑNa + θNb
(γq + κq3)Da + 2iωηDb
.
(39)
B. Tissue of infinite thickness H
In the limit qH (cid:29) 1, the expressions of the quantities defined above can be given explicitly. They
read
(cid:20)(cid:18)
(cid:19)
(cid:21)
τ (q) =
2η
γq + κq3
ζ + 1
3η
ζ + 4
3η
and
ξδH(q, t) =
1
2ηq
1 − 2
ζ
I1 − 2
ζ
J1
,
(40)
with ζ = ζ/η + 1/3, and where I1 and J1 are the integrals given by Eqs. (33) and (34) in the limit
qH (cid:29) 1. This leads to the following expression:
(41)
(42)
(43)
(cid:90) z
(cid:90) 0
−∞
dz
dz
z
dz eqz
(cid:90) 0
(cid:90) 0
(cid:90) 0
−∞
−∞
−∞
ξδH(q, t) = − q
2η
q
2η
+
+
1
2η
dz(cid:48) eqz(cid:48)
(ξxx − ξzz + 2iξxy)z(cid:48)
(cid:34)
dz(cid:48) eq(2z−z(cid:48)) (ξxx − ξzz − 2iξxy)z(cid:48)
2 − ζ
2ζ
ζ + 2
2ζ
ξzz + iξxy
ξxx +
z .
(cid:35)
We then have
and
Σ(q) =
γq2 + κq4
1
ϑη2 + θ
(cid:104)(cid:0)ζ + 1
3η2(cid:105)
3η(cid:1)2 + 1
3η(cid:1)(cid:0)ζ + 4
2η(cid:0)ζ + 1
3η(cid:1)
3η2(cid:3)qδH(q, t) ,
ϑη2 + θ(cid:2)(ζ + 1
3η)2
3η)2 + 1
4η2(ζ + 1
X(q, t) = −
which gives the following correlation and response functions in the time domain:
CXX(q, t) =
8
γ + κq2
ϑη2 + θ
e−t/τ (q)
3η2
ϑη2 + θ
3η
ζ + 4
3η
(cid:0)ζ + 1
3η(cid:1)2
(cid:105) ζ + 1
(cid:104)(cid:0)ζ + 1
3η(cid:1)2 + 1
(cid:0)ζ + 1
3η(cid:1)2
(cid:104)(cid:0)ζ + 1
(cid:105) H(t)e−t/τ (q) ,
3η(cid:1)2 + 1
3η(cid:1)4
(cid:105)(cid:0)ζ + 1
(cid:104)(cid:0)ζ + 1
(cid:0)ζ + 4
3η(cid:1)2
3η(cid:1)2 + 1
(cid:105)(cid:0)ζ + 1
3η(cid:1)3
(cid:104)(cid:0)ζ + 1
3η(cid:1)2 + 1
ζ + 4
3η
32η4q
8η3q
3η2
3η2
3η2
1
1
[(γq + κq3)2 + 4ω2 η2]
γq + κq3 + 2iω η
and
χXX(q, t) = −4q
CXX(q, ω) =
ϑη2 + θ
χXX(q, ω) = −
ϑη2 + θ
11
(44)
(45)
in the frequency domain. As in the main text, η = η(ζ + η/3)/(ζ + 4η/3).
V. ESTIMATION OF THE COLLISION LENGTH AND TIME
We now estimate the collision length lc defined in the main text in the low-frequency, small-
wave-vector domain by either Eq. (14) or Eq. (15) of the main text, depending on the value of
γH 2/(kBTeff). In the low-frequency, small-wave-vector domain, kBTeff is given by Eq. (12) of the
main text, and we start by estimating this quantity. To estimate the tissue bulk viscosity ζ, we
use experiments where the bulk growth rate kp = kd − ka of a tissue spheroid is determined as a
function of the externally applied stress thanks to a fitting procedure [27, 28]. The measurement
−1 (cid:39) 5 104 Pa·day. With a cell-division rate kd of the order of
procedure leads to ζ = − (∂kp/∂P )
one per day, we get an estimate for the isotropic elastic modulus of the spheroid in its homeostatic
state ¯E = ζkd of the order of 5 104 Pa. This is in agreement with another study, in which the short-
time responses of multicellular spheroids to an external pressure jump are measured, and which give
¯E (cid:39) 104 − 105 Pa [14]. The anisotropic counterparts of these quantities are the tissue shear viscosity
η, the shear elastic modulus E and the rate of cell-neighbor exchange kr, related by E (cid:39) ηkr.
Estimating that E is typically lower if not much lower than ¯E [29, 30] and that kr should typically
be much larger than kd, we get that the tissue bulk viscosity ζ should be much larger than the tissue
shear viscosity η. This is in agreement with experimental estimates of the shear viscosity of cellular
aggregates, which give η (cid:39) 104 − 105 Pa·s [29 -- 32]. The amplitude ϑ of the isotropic noise due to
cell-division and apoptosis is given in the main text. In the homeostatic state where kd = ka, it reads
ϑ = 2ζ 2kd/ρh = 2 ¯E2/(kdρh). Estimating a similar expression for the anisotropic noise amplitude
θ (cid:39) η2kr/ρh (cid:39) E2/(krρh), we see that kd (cid:28) kr together with ¯E (cid:38) E or ¯E (cid:29) E imposes ϑ (cid:29) θ.
With these estimates, we therefore have
kBTeff (cid:39) 1
2
ϑ
ζ
=
ζkd
ρh
.
We now turn to the proper estimation of the collision length lc. We introduce the length
w = γ/(ζkd) and the linear cell dimension a with ρh = a−3. With a typical tissue thickness
H = 10 a, we have lc (cid:39) lκ exp(100πw/a) in the tension-dominated regime and lc (cid:39) 10 lκ
in the bending-dominated regime. Within this context, the ratio lc/lκ is therefore function of the
(46)
(cid:112)(w/a)
12
adimensional parameter w/a. A reasonable scale for a is 10 µm. Tissue surface tensions have been
measured for different tissue types and using different experimental methods. The measurements
give values ranging typically from a fraction up to several Millinewton per meter [29 -- 31, 33 -- 39].
With these values and ζkd (cid:39) 104 − 105 Pa as estimated above, one finds that w/a ranges typically
between 10−4 and 10−2, at most 10−1. The lower bound of this interval corresponds to the bending-
dominated regime, which leads to lc (cid:39) 0.1 lκ. The upper bound corresponds to the tension-dominated
regime, with lc ranging from a few tens of lκ to being much larger than lκ. These order-of-magnitude
estimates therefore indicate that actual tissues could be in either of these two regimes, depending
on parameter values. In addition, in the tension-dominated regime, the collision length lc is expo-
nentially dependent on the parameters characterizing the tissue. We therefore expect that small
variations in some of the parameters entering the expression of lc, like the tissue surface tension
γ, can have dramatic effects on the structure of the tissue surface, and that this could happen
independently of alterations of the cell-division rate.
To complete our estimates, we can finally give an order of magnitude for the tissue effective
temperature that we have defined. Using the values given above, one finds kBTeff (cid:39) 10−11 − 10−10 J,
a value that is some 10 orders of magnitude larger than thermal energy. This should however not
surprise us, since the stochastic processes we study here are related to the behavior of whole cells,
which are typically five orders of magnitude larger than individual molecules.
|
1809.09335 | 1 | 1809 | 2018-09-25T06:30:29 | Kinky DNA in solution: Small angle scattering study of a nucleosome positioning sequence | [
"physics.bio-ph",
"cond-mat.stat-mech",
"q-bio.BM"
] | DNA is a flexible molecule, but the degree of its flexibility is subject to debate. The commonly-accepted persistence length of $l_p \approx 500\,$\AA\ is inconsistent with recent studies on short-chain DNA that show much greater flexibility but do not probe its origin. We have performed X-ray and neutron small-angle scattering on a short DNA sequence containing a strong nucleosome positioning element, and analyzed the results using a modified Kratky-Porod model to determine possible conformations. Our results support a hypothesis from Crick and Klug in 1975 that some DNA sequences in solution can have sharp kinks, potentially resolving the discrepancy. Our conclusions are supported by measurements on a radiation-damaged sample, where single-strand breaks lead to increased flexibility and by an analysis of data from another sequence, which does not have kinks, but where our method can detect a locally enhanced flexibility due to an $AT$-domain. | physics.bio-ph | physics |
Kinky DNA in solution: Small angle scattering study of a
nucleosome positioning sequence.
Torben Schindler,1 Adri´an Gonz´alez,2, 3 Ramachandran Boopathi,4, 5
Marta Marty Roda,3 Lorena Romero-Santacreu,3 Andrew Wildes,2
Lionel Porcar,2 Anne Martel,2 Nikos Theodorakopoulos,6, 7 Santiago
Cuesta-L´opez,3 Dimitar Angelov,4 Tobias Unruh,1 and Michel Peyrard8
1Lehrstuhl fur Kristallographie und Strukturphysik,
Friedrich-Alexander-Universitat Erlangen-Nurnberg,
Staudtstr. 3, D-91058 Erlangen, Germany
2Institut Laue-Langevin, 71 avenue des Martyrs,
CS 20156, 38042 Grenoble Cedex 9, France
3ICCRAM, University of Burgos, Science and Technology Park,
Plaza Misael Banuelos, 09001 Burgos, Spain
4Universit´e de Lyon, Laboratoire de Biologie et de Mod´elisation
de la Cellule (LBMC) CNRS/ENSL/UCBL UMR 5239,
Ecole Normale Sup´erieure de Lyon, 69007 Lyon, France
5Institut Albert Bonniot, Universit´e de Grenoble Alpes/INSERM
U1209/CNRS UMR 5309, 38042 Grenoble Cedex 9, France
6Theoretical and Physical Chemistry Institute,
National Hellenic Research Foundation,
Vasileos Constantinou 48, GR-11635 Athens, Greece
7Fachbereich Physik, Universitat Konstanz, D-78457 Konstanz, Germany
8Universit´e de Lyon, Ecole Normale Sup´erieure de Lyon,
Laboratoire de Physique CNRS UMR 5672,
46 all´ee d'Italie, F-69364 Lyon Cedex 7, France∗
(Dated: September 26, 2018)
1
Abstract
DNA is a flexible molecule, but the degree of its flexibility is subject to debate. The commonly-
accepted persistence length of lp ≈ 500 A is inconsistent with recent studies on short-chain DNA
that show much greater flexibility but do not probe its origin. We have performed X-ray and
neutron small-angle scattering on a short DNA sequence containing a strong nucleosome positioning
element, and analyzed the results using a modified Kratky-Porod model to determine possible
conformations. Our results support a hypothesis from Crick and Klug in 1975 that some DNA
sequences in solution can have sharp kinks, potentially resolving the discrepancy. Our conclusions
are supported by measurements on a radiation-damaged sample, where single-strand breaks lead to
increased flexibility and by an analysis of data from another sequence, which does not have kinks,
but where our method can detect a locally enhanced flexibility due to an AT -domain.
∗ To whom correspondence should be addressed. [email protected]
2
I.
INTRODUCTION
Remarkable progress has been achieved since the famous publication of the DNA structure
by Watson and Crick [1] but basic questions are still open. One of them was raised by Crick
and Klug in 1975 [2] when they examined the ease with which duplex DNA can be deformed
into a compact structure like chromatin. They suggested that DNA can form a sharp
bend, that they called a kink. Kinks are not merely sharp bends due to fluctuations or a
broken strand, but metastable structures which can exist without a drastic distortion of the
configuration of the backbone. They have been subsequently found in highly-constrained
environments, such as in chromatin, where other molecules force sharp bends [3, 4]. However
Crick and Klug raised a more fundamental question: could DNA kinks occur spontaneously
as a result of thermal motion?
Since then this hypothesis was never confirmed and the origin ot the flexibility of DNA is
still debated [5]. The molecule is often described as a cylinder, or worm-like chain (WLC),
that is relatively stiff with a persistence length of lp ≈ 500 A [6 -- 8]. This value is at odds
with the tight-packing of DNA in chromatin and with recent experiments on short-chain
DNA that show much shorter persistence lengths [9 -- 12]. However the studies are somewhat
limited in that they do not probe the full conformation of the molecule and cannot identify
the position or the degree of any bending.
Small-angle scattering (SAS) experiments are excellent techniques to investigate particles
with linear dimensions ∼ 1 − 100 nm and are therefore well-suited to studying short-chain
DNA. They probe the spatial distribution of the scattering length density and are very
sensitive to the overall shape and size distribution of particles.
In the dilute limit, SAS
probes the ensemble average of all the orientations and shapes that the particles adopt over
the duration of the measurement. We have carried out measurements with both X-rays
(SAXS) and neutrons (SANS) of short-chain DNA in solution, free from any molecular
construct like the addition of fluorophores [12] or gold nanoparticles [13]. The principal
experiments were performed with SAXS on a 145 base-pair sequence of DNA containing the
"601" strong positioning sequence [14] known for easy wrapping around a histone core. The
sequence, which has been investigated in structural studies of nucleosome core particles [15],
is shown in Fig. 1 and is henceforth called "Widom-601". The results have been analyzed
using a dynamical model. This analysis shows the presence of kinks at positions consistent
3
with the hypothesis of Crick and Klug. As a further test, a complementary experiment
has been made with a different DNA sequence, with 204 base pairs, derived from a piece
of λ-phage DNA, which has been modified to introduce a segment containing only A − T
pairs from sites 94 to 125. This introduces a domain with a higher local flexibility and some
weak intrinsic curvature due to intrinsically curved elements (such as AATT). The same
analysis, performed on the data collected with this control sequence did not detect kinks
but the statistics of the local bending angles showed a increased average value, and increased
fluctuations, in the domain where we introduced the AT rich segment, showing the ability
of our analysis to detect detailed features of DNA flexibility with SAS.
II. SAMPLES AND EXPERIMENTAL METHODS
FIG. 1. Sequence of the 145 base-pair DNA molecule investigated in this study. This sequence was
used to build the NCP-601 nucleosome core particle investigated in [15]. The fragment with orange
background is the strong positioning element, characteristic of this sequence. The TA fragments
considered as possible kink positions in the model are underlined and marked in yellow.
A. Sample preparation
The main sample of this study is the "Widom-601" 145 base-pair sequence of DNA shown
in Fig. 1. It contains the "601" strong positioning sequence [14], known for easy wrapping
around a histone core. Multiple repeats of this nucleic sequence were inserted into the
EcoRV site of the pGEMT easy vector and expressed in E.Coli DH5α cells. The fragments
were excised from the vector by restriction enzyme EcoRV, followed by phenol chloroform
extraction and ethanol precipitation. The excised 145bp DNA were separated from the
linearized plasmid through 5% polyacrylamide gel electrophoresis (prep cell-BioRad) and
the purity of the sample was analyzed by 1% agarose gel electrophoresis and PAGE. This
preparation method provides 100% homogeneous DNA of crystallographic-grade purity and
4
was also applied to prepare samples for nucleosome crystallization, used in other experiments
[16].
FIG. 2. Sequence of 204 base pairs studied in a complementary experiment.
In the fragment
marked in blue the original sequence part of λ-phage DNA has been modified. The G − C pairs
have been changed to A − T pairs to create a domain of 32 consecutive A − T pairs.
The same method was used to prepare the modified λ-phage sequence shown in Fig. 2,
studied in complementary experiments presented in Sec. V.
Prior to the SAXS and SANS experiments the samples were dissolved in a 2H2O buffer
containing 10mM Tris (pH 7.4), 0.1 mM EDTA and 30 mM NaCl with a DNA concentra-
tion of 2.2 mg/mL for the Widom-601 sequence and 1.27 mg/mL for the modified λ-phage
sequence, and degassed in a partial vacuum of 0.5 bar for 3 hours before being loaded into
quartz containers for measurement. The quartz cells for SANS were rectangular with a 2
mm thickness. Quartz cylindrical capillaries with an inner diameter of 1 mm and a wall
thickness of 10 µm were used for SAXS.
B. Small angle scattering
Small angle scattering of neutrons and X-rays by dilute particles in solution probes the
scattering length density of the particles. The scattering cross-section is given by [17]:
dσ
dΩ
∝
N
V *(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
ZZZV
2
β (r, Θ) exp (iq · r) d3r(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)
+
Θ
,
(1)
where the sample contains N particles in volume V , and q is the momentum transfer. The
scattering length density at position r in a particle is given by β (r) = ρ (r) b (r), with ρ (r)
being the local atomic number density. The local mean scattering length, b (r), varies as
a function of atom and isotope for neutrons [18] and as the Thomson scattering length
multiplied by the local number of electron for X-rays. The triangular brackets in equation
(1) indicate that the cross-section measures an ensemble average of all the orientations, Θ,
5
and configurations that the particles have in the sample.
1. Small angle X-ray scattering (SAXS)
The SAXS measurements were performed using the VAXTER instrument at the Friedrich-
Alexander Universitat, Germany. It uses Ga-Kα1,2 radiation (λ = 1.34 A) from a GaMet-
alJet D2 70 kV X-ray source (EXCILLUM, Kista, Sweden) with 150 mm Montel optics
(INCOATEC, Geesthacht, Germany). X-ray-absorbing diaphragms defined the incident
beam collimation. The sample-detector distance was set to 1.5 m, and the instrument uses
a Pilatus3 300 K detector. The sample temperature was controlled by an external water
bath. The data were corrected for transmission and background. The scattering from a
glassy carbon standard [19] was used to normalize the data to absolute scale. The resulting
cross-sections covered the range 0.006 ≤ Q ≤ 0.2 A−1.
2. Small angle neutron scattering (SANS)
The SANS measurements were performed using the D22 instrument at the Institut
Laue-Langevin, France. The incident beam wavelength was set to λ = 6A with ∆λ/λ =
0.1. The incident beam collimation was defined using neutron-absorbing diaphragms.
The sample temperature was controlled by an external water bath. Data were recorded
with the sample-detector distance at 5.6 and 17.6 m, with the collimation set at 5.6 m
and 17.6 m respectively for the optimal compromise between resolution and beam flux.
The GRASP data reduction suite (https://www.ill.eu/instruments-support/instruments-
groups/groups/lss/grasp/home/) was used to correct the data for instrument background,
empty cell, detector efficiency and to normalize by the direct beam intensity to obtain the
intensity in absolute scale. The data were then merged, buffer subtracted, and radially
integrated to give the cross-section in the range 0.003 ≤ Q ≤ 0.15 A−1.
6
0
5
1
0
0
1
0
5
0
0
3
ρ(θ)
5
(x10
-4
)
0
5
1
0
0
1
0
5
0
0
3
ρ(θ)
5
(x10
-4
)
FIG. 3. Experimental results and data analysis for the Widom-601 sequence. (a) Blue: results of
the SAXS experiment at room temperature. The top right inset shows the scattered intensity I(q)
versus the scattering vector q, and the main panel shows the corresponding P (r) computed by GNOM
with estimated error bars. The thick red line shows the average P (r) for the 103 conformations
with the best match to the experimental data among 12 × 107 conformations generated by Monte-
Carlo. Thin magenta line: P (r) deduced from a Monte-Carlo simulation at room temperature for
the models with a kink of angle 80◦ at site 67. The dashed black line shows P (r) for a homogeneous
Kratky-Porod model having a persistence length of 500 A at room temperature (K = 150, θ0, = 0,
C = 0). The bottom left inset shows the values of the local bending angles for the 10 model
conformations which provide the best match with the experimental P (r). Successive plots are
moved up by 50◦ to limit the overlap between the curves. (b) Histogram of the bending angle
against n for the 103 conformations that provide the best matching with the room temperature
SAXS data. For each site of the polymer model, the left part shows the number of θn values that
correspond to the value marked on the left scale (the total of these numbers for a given n is equal
to 1000, the number of conformations) with a color scale shown on the right. The right part of the
figure shows the fraction ρ of θ angles, integrated over the whole model, which belongs to a given
range of theta. The scale is truncated to ρmax = 5 × 10−4 to better show the part of the curve
which corresponds to large θ angles. (c) Dark-blue: The SAXS P (r) at 70◦C after a long exposure
to X-rays. The thick red curve shows the average P (r) for the 1000 conformations that provide the
best match with the experimental data has been obtained from a search of 70 × 107 conformations
generated by Monte-Carlo. Light blue: The SAXS P (r) at room temperature (also shown in
blue in panel (a)) for comparison. Turquoise: The SAXS P (r) at 70◦C with short exposure to
X-rays. The inset shows the values of the local bending angles for the 10 model conformations
which provide the best match with the experimental P (r) as in panel (a). (d) same as panel (b)
for the 103 conformations that provide the best matching with long-exposure SAXS data. The
7
color scale and the scale for ρ in the right part of the plot are the same as for panel (b) to allow a
quantitative comparison between the two cases.
III. EXPERIMENTAL RESULTS ON THE WIDOM-601 SEQUENCE
The SAS scattering cross-section for dilute particles in solution is given by the one-
dimensional integral:
dσ
dΩ
∝
dmax
P (r)
Z0
sin (qr)
qr
dr,
(2)
where q is the momentum transfer and dmax is the longest dimension in the object. The pair-
distribution function, P (r), correlates the scattering length densities for volume elements
separated by a distance r within the particle and is further weighted by the contrast between
the scattering length densities of the particle and the solution. The P (r) therefore differs
for X-rays and neutrons, however, for scales beyond about 40 A which do not resolve the
internal structure of DNA and are only sensitive to the conformation of the molecular axis,
both techniques are expected to bring the same information.
Mathematically, P (r) is obtained from an inverse Fourier transform of the cross-section.
We used the GNOM program [20] with manually set parameters to calculate P (r). The P (r)
derived from the SAXS data at 23◦C is shown Fig. 3-a. The measurement took one hour.
Temperature-dependent P (r) derived from the SAXS data are shown in Fig. 3-c. Two
samples were measured at high temperature. For the first, labeled "short exposure", the
sample was heated directly to 70◦C and measured for one hour. For the second, labeled
"long exposure", the data result from the final measurement of a sequence of hour-long runs
at 30◦C, 50◦C and 70◦C.
The "short exposure" P (r) is very similar to that at 23◦C, while the "long exposure" data
show dramatic differences due to X-ray radiation damage. At low X-ray flux, most of the
damage occurs as single-strand breaks due to free radicals created by the interaction of X-
rays with the water molecules [21, 22]. This is likely to be exacerbated at high temperatures.
In our SAXS experiment about 50% of the X-rays were absorbed by the solvent, and the
influence of radiation damage is clearly visible.
Prior to the SAXS measurements preliminary measurements on the Widom-601 sequence
had been performed with SANS as a function of temperature up to 79◦C. In-situ UV ab-
sorption spectroscopy recorded during the SANS measurements showed that there was no
significant thermal denaturation of double-stranded DNA up to this temperature.
8
The SANS measurements up to 79◦C did show some limited changes versus temperature
(Fig. 4) which could come from an increase of the thermal fluctuations, but the overall
shape of P (r) was only weakly modified. Radiation damage from neutrons was expected to
be negligible as they are non-ionizing, and the SANS data showed no strong change with
exposure time.
FIG. 4. Comparison of the SAXS and SANS results for the Widom-601 sequence. The SAXS data
are shown in blue and the neutron data are shown in grey (dark grey: measurements at 15◦C,
light grey: measurements at 79◦C. The inset shows the scattered intensity versus the scattering
vector q. The corresponding P (r) have been obtained by GNOM. Their normalization is such that
R P (r)dr = I(q = 0).
Figure 4 compares the SAXS and SANS results. The P (r) show common features.
However there are differences because SAXS is more sensitive to heavier elements as ρb is
proportional to the electron density for X-rays. The SAXS P (r) is thus dominated by the
correlations between the phosphates. The neutron P (r) is more sensitive to the distribution
of hydrogen in the sample. Both data sets show a peak at r ≈ 20 A. The peak in the X-ray
data is due to the phosphates, which are on the outside of the double helix and are 20 A
apart.
Due to the many protons distributed on various sites in the DNA structure, for short
distances neutron scattering is well approximated by the scattering by a bulk cylinder with
a diameter of 20 A, which also shows a peak in P (r) in the vicinity of the diameter of the
cylinder. Therefore, both for SANS and SAXS, the 20 A peak appears to be signature of
the diameter of the double helix.
For larger r distances the differences between SAXS and SANS are expected to decrease
because at this scale the differences between the atom types are no longer resolved and P (r)
essentially reflects to conformation of the molecules. For r ≥ 70 A, two curves for P (r) have
9
a similar shape. They show a roughly flat region in the 70 − 160 A range and then decrease
strongly before showing a kind of plateau in the 300 − 400 A range.
IV. DATA ANALYSIS: CONFORMATION OF THE WIDOM-601 DNA MOLECULES
IN SOLUTION
We focused our analysis of the Widom-601 data on the SAXS data which provided a
stronger and less noisy signal than the SANS experiments.
A. Standard SAS data analysis
As a first step we relied on standard SAS data analysis packages. The 23◦C SAXS data
were analyzed using the SASVIEW software [23] to model DNA as an homogeneous flexible
cylinder, giving a persistence length of lp = 97.9 A. A further analysis used the Kratky-
Porod model [24] which is the discrete version of the worm-like chain model generally used
for long DNA molecules [6, 7] and whose structure factor can be calculated exactly [25].
This calculation gave a best fit of lp = 117 A. Both values are much smaller than measured
values for double-stranded DNA [5, 26], including the generally-accepted value of lp ≈ 500
A. They are also unrealistically small considering that the double helix has a diameter of
20 A. The analysis confirms the conclusions, drawn from accurate measurements on short
DNA molecules [12], that a homogeneous WLC model does not describe the flexibility of
short-chain DNA.
However, it is important to stress that the persistence length obtained in these fits is
actually an effective persistence length which includes both the effects of the statistical
fluctuations and a contribution of the intrinsic curvature of the molecule. As pointed out
by Schellmann and Harvey [27], any kind of bend, stiff or flexible, will result in a reduced
persistence length and, as a result, permanent structural kinks can reduce the persistence
length.
Therefore, to understand what appears as an anomalously low persistence length, the
next step was to attempt to reconstruct the main features of the particle shapes from the
data. However this is not straightforward because the knowledge of P (r) is not sufficient
to unambiguously determine this shape, even for the case of a linear polymer.
Iterative
10
FIG. 5. Examples of the output of the DAMMIF program, showing possible optimized shapes of the
DNA molecules in solution
schemes, starting from a broad and random conformational search and then progressively
improving by testing likely models, have been developed. Within the ATSAS package [28],
the DAMMIF program [29, 30] starts from the P (r) given by the GNOM program and uses
simulated annealing to optimize the shape of a set of dummy atoms to retrieve the shape
of the scattering molecules. The program starts from a random configuration. A polymer
like DNA can have a broad variety of conformations in solution and multiple runs lead to
different final shapes. Nevertheless the results give hints on the DNA conformations that
fit the scattering data. When it is applied to our SAXS and SANS data for the Widom-
601 sequence, this program converges either to strongly curved structures, or to branched
solutions, as shown in Fig. 5. As partial denaturation of our sample is ruled out by the UV
measurements, we do not expect any branching. However, if the solution actually contains a
mixture of weakly curved molecules while others have a sharp bend, these DAMMIF solutions
could describe a superposition of two conformations because P (r) represents the ensemble
average over all configurations in the sample, including time-dependent fluctuations.
B. Conformational search using a polymer model
The DAMMIF results show that fluctuations must be accounted for in the modeling of the
data. Moreover the ab-initio shape reconstruction performed by DAMMIF does not include any
a priori knowledge of the molecular properties. It is more efficient to search among a set of
conformations derived from a model which take into account some of the known features of
the molecules. However, to study the flexibility of the DNA molecules, a model at the atomic
scale or a coarse-grain model describing the bases would be inefficient. Instead we describe
11
DNA with an extension of the Kratky-Porod model [24], i.e. the model, shown schematically
on Fig. 6, is not concerned with the internal structure of the DNA, but only with the
conformations that the backbones could adopt. It consists of N + 1 objects representing
base pairs, each separated by a = 3.34 A, corresponding to the base-pair distance in DNA,
with bond angles that may vary at each point. The model therefore has N bonds, or
segments. The local angle at site n, which connects segments n and n + 1, is defined as θn,
and the dihedral angle of rotation between the plane containing segments n − 1 and n and
the plane containing segments n and n + 1 is defined as φn.
Thermal fluctuations were accounted for by calculating the bending and torsional energy
for the model expressed through its Hamiltonian. A possible permanent curvature is per-
mitted by considering a local equilibrium value at each site, given by θ0,n and φ0,n, which
may be different from zero. The Hamiltonian is thus given by:
N −1
H =
Kn[1 − cos(θn − θ0,n)]
(3)
N −1
Xn=1
Xn=2
+
Cn[1 − cos(φn − φ0,n)] ,
where Kn and Cn are constants setting the scale of the bending and dihedral energies re-
spectively. Dimensionless variables were used in the calculations. Distances were measured
in units of the base-pair distance a, and temperature was measured in energy units and the
energy was expressed relative to kBT at room temperature, where kB is the Boltzmann con-
stant, so that the properties of DNA at room temperature were obtained by setting T = 1.
Equation 3 reduces to the Kratky-Porod model [24] on setting Cn = 0 and Kn = K for all
n. Setting K = 150 gives lp = 500 A, while K = 35 gives the previously-discussed best fit
to the 23◦C data with lp = 117 A.
For a given model conformation, P (r) was calculated by putting a unit scattering cen-
ter at each base-pair position and then the calculated P (r) was scaled to have the same
integrated area as the data. This gives a reasonable approximation for the SAXS data for
r & 40 A, as the X-ray scattering from a base pair is dominated by its two phosphorus atoms
whose contribution can be effectively mapped onto the center of the base pair for large r.
For small r this evaluation of P (r) is a worse approximation for the SANS data which are
far more sensitive to the protons.
12
FIG. 6. Schematic showing the polymer model used for the WIDOM-601 DNA measured in the
experiments. The top part of the figure shows the definition for the numbering and the lower part
shows the definition of the bond and torsional rotation angles.
The model with N = 144, representing the Widom-601 sequence, was used for a broad
search through the conformational space to determine the conformations which provide the
best match with the experients. At this stage of the analysis, our goal is not to design a
model of the DNA molecules in solution, but to extract information on the molecular con-
formations from the data. Therefore it is important to avoid any bias in the conformational
search, which could be introduced for instance by assuming some a-priori knowledge such
as sequence dependent bending angles. The bending constant was set to Kn = K = 150.
The torsional constant was set to Cn = C = 2 which leads to a clear dominance of φ ≈ 0,
in agreement with the torsional rigidity of DNA, but nevertheless allows large fluctuations.
The equilibrium angles were set to θ0,n = φ0,n = 0.
Up to 12 × 107 Monte Carlo-generated conformations were created and accepted with
the probability exp(−H/T ) at a temperature of T = 3, i.e. three times room temperature,
to widely explore the conformational space. The P (r) were calculated for accepted confor-
mations and compared to the SAXS P (r) at 23◦C by computing the standard deviations
between the two functions. The best 103 conformations were selected. They can be consid-
ered as representative of the conformations of the DNA molecules in solution so that their
analysis allowed us to determine the main features of the those conformations. Their average
P (r) is plotted in Fig. 3-a. The agreement with the experimental P (r) is satisfactory, and
is excellent in the region 80 A ≤ r ≤ 280 A. This is the critical range to assess the shape of
DNA molecules, which have a length of 484 A and a radius of gyration of about 100 − 150 A.
A calculation of the WLC with lp = 500 A is also shown in Fig. 3-a for comparison. It does
not resemble the experimental P (r).
13
Fig. 3-b shows a histogram of θn at the various sites for the best 103 conformations. The
vast majority of θn are less than 30◦, consistent with thermal fluctuations around θ0,n = 0.
However there is a significant concentration of large θn near the center of the molecule. The
distribution is centered at θn ≈ 125◦, and it is distinct from the distribution at lower angles.
Inspection of those conformations with large θn revealed that each had one, and only one, of
the sites with a large angle. The insert on Fig. 3-a shows the local bending angles for the 10
best conformations. The modeling suggests that the Widom-601 sequence has metastable
states with large angles in its central region, consistent with the kinks proposed by Crick
and Klug. Kinks would also help explain the branched conformations given by the DAMMIF
program if the solution contains a mixture of kinked and non-kinked DNA molecules.
C. Test of a model for kinked DNA molecules in solution
As a blind conformational seach hints that some conformations may have a permanent
kink, a second stage of our analysis was to check this hypothesis. In this stage the polymer
model is used in a different context. Instead of chosing a generic set of parameters to allow
an unbiased scan of possible conformations, we use the results of the conformational search
to select specific parameters, and then run a Monte Carlo at room temperature (T = 1 in
our dimensionless units) to test the agreement of the resulting P (r) with the experimental
results. The possible presence of a kink is tested by setting θ0,n = θκ for specific n = nκ.
Following the hypothesis of Crick and Klug, we identified A−T sites in the central region
and explored conformations with 70◦ ≤ θ0,n = θκ ≤ 110◦ at these sites. Inspection of Fig. 1
shows A−T sites at nκ = 47, 57, 67, and 88. The last site is of particular interest because it
is part of the strong histone-positioning element characteristic of this sequence. The values
for Kn, Cn, φ0,n and θ0,n were all maintained as for the conformational search, except for
the one θκ.
To quantitatively evaluate the different possibilities, we calculated
χ2 =
1
Nr − 1
Nr
X1 (cid:20) Ptheo(rj) − Pexp(rj)
σj
(cid:21)2
(4)
where rj are the points where Pexp (r) has been computed, σj the experimental error at these
points (given by the GNOM program) and Nr the number of calculation points. Instead of the
14
simpler standard deviation used for the conformational search, which is equivalent to setting
σj = 1 in Eq. (4), here we take into account the actual experimental errors to measure the
validity of the model more accurately. The summation is restricted to rj > 50 A because the
polymer model, which does not take into account the diameter of the molecules, cannot be
expected to properly describe the properties of the DNA molecules at very small distances.
nK θK (deg) χ2 kink
88
88
88
88
67
67
67
57
47
2.31
9.17
1.26
3.00
3.32
9.17
0.88
2.14
2.91
95
110
80
70
95
110
80
95
95
TABLE I. Results of Monte Carlo simulations at room temperature (T = 1 in reduced units) for
various kink positions nK and angles θK. The table lists the values of the χ2 distance between the
experimental and theoretical P (r) (Eq. (4)) for kinked DNA (3rd column). The line in bold-face is
the case which provides the best agreement with experimental data. For non-kinked DNA, the χ2
distance between the theoretical and experimental probability distributions (Eq. (4)) is χ2 = 22.86.
Figure 7 compares the SAXS structure factor with the structure factor of the polymer
model with a kink at one of the two T A steps closest to the center, nK = 88 (thin blue
curve) or nK = 67 (dashed blue curve) and with the structure factor of a homogeneous WLC
model, for two values of the persistence length lp = 500 A and lp = 224 A.
The results for different kink positions and angles are given in Table I. Although such a
simple polymer model cannot describe all the fine structure of the experimental results, in
the range where the model applies the theoretical results fit the data within the experimental
error bars in the range 50 − 300 A. For r < 50 A the model should be completed by taking
into account the diameter of the DNA molecule and its internal structure, while for r > 300 A
some extra flexibility near the ends might be necessary to take into account the fluctuations
of the twist at the free boundaries.
Table I shows that the best agreement with experiments is obtained for nK = 67, although
the position θ = 88, within the T T T AA strongly positioning sequence, gives results which
are almost equally good. Our analysis cannot rule out one of the two optimal kink sites.
15
FIG. 7. comparison between the SAXS structure factor (circles) and the structure factor of the
polymer model, at room temperatures, for different parameters: i) thin blue curve, model with a
kink of 80◦ at position 67, ii) dashed blue curve, almost identical to the previous one, kink of 80◦
at position 88, iii) full red curve, Kratky-Porod model, without kink and a persistence length of
500 A, iv) dashed red curve, Kratky-Porod model, without kink and a persistence length of 224 A
The case of site nK = 57, which also gives good results is almost symmetric of nK = 88
with respect to the center of the molecule. For molecules in solution, at the scale of the
SAXS/SANS experiments, and for the model, the two ends are indistinguishable and the 57
and 88 sites are therefore almost equivalent. Moreover, looking at the sequence of the two
strands listed in Fig. 1 of the article, one can notice that, around site 57 one finds that same
sequence of 5 base pairs as around site 88. The best agreement with experiments is obtained
for a kink angle of 80◦, in good agreement with the kink predicted by Crick an Klug [2].
The tests were not exhaustive, but they establish that models with a kink in the DNA
give far better comparisons with the experimental results than a WLC model with a sensible
lp.
D. Analysis of the long-exposure SAXS data
The results pointing to the possible existence of kinks for the Widom-601 sequence in so-
lution are supported by the analysis of the SAXS data at high temperature. A conformation
search with all equilibrium angles equal to zero was applied to the "long exposure" data.
The search was performed at higher temperature of T = 6 to allow for an even broader
range of conformations. More than 7 × 108 conformations were generated in the Monte
Carlo search, and the average P (r) from the best 103 conformations are shown in Fig.3-c
with the histogram of θn vs. n shown in Fig.3-d. The results show many sites with large θn
distributed more evenly across the sequence and over intermediate angles. The distribution
16
in θn is also broader, with no gap between the low and large θn values. The plot of the local
bending angles of the 10 best conformations, seen in the insert in Fig.3-c, shows that they
may have more than one large angle. The findings are consistent with radiation damage to
the DNA, which creates single-strand breaks that allow local sites to be far more flexible.
However, as for the search at 23◦C, a high concentration of large angles is still found around
the middle of the model with an angle distribution centered at θn ∼ 125◦. The number
of sites within this distribution is greater at 70◦C which may represent more kinks due to
increased thermal fluctuations or large bendings due to single-strand breaks, which is con-
sistent with X-ray radiation damage preferentially causing single-strand breaks at sites with
local distortion [31], in this case due to kinks. Both observations indicate that kinks are
present in Widom-601.
V. COMPLEMENTARY EXPERIMENT WITH ANOTHER DNA SEQUENCE
As a further test to validate our results, we have performed a complementary experiment
with a DNA sequence previously studied in cyclization experiments and found to be well
described by the standard WLC model of DNA[32]. This sequence, of 204 base pairs, is
shown in Fig. 2. It is derived from a segment of 200 base pairs of λ-phage DNA, starting
at site 29853, modified at its ends to make the PCR step easier in the sample preparation.
This sample had been chosen for the cyclization studies because it does not have intrinsic
curvature. However, in order to check the ability of our experiments and analysis to detect
specific features of the DNA molecules under study, we modified the sequence locally, chang-
ing the some G − C pairs into A − T in order to create a domain of 32 consecutive A − T
pairs from position 92 to position 125. Due to the larger fluctuations of the A − T pairs [33]
we expect this region to have a higher flexibility. Moreover it includes some elements, such
as AATT, which have a small intrinsic curvature.
A solution with a concentration of 1.27 mg/mL was studied by SANS, using the D22
experiment at Institut Laue Langevin as described for the Widom-601 sequence. The scat-
tering data were treated by the GNOM program to compute P (r) and then we performed a
computational search using the same program as for the Widom-601 sequence and the same
model parameters, except for the number of nodes.
Figure 8 shows the histogram of the bending angles of the the 103 conformations that
17
0
5
1
0
0
1
0
5
0
0
3
ρ(θ)
5
(x10
-4
)
FIG. 8. Equivalent of Figs 3-b and 3-d, for the complementary experiment with the sequence
shown in Fig. 2. Histogram of the bending angle against n for the 103 conformations that provide
the best matching with the room temperature SANS data. For each site of the polymer model,
the left part shows the number of θn values that correspond to the value marked on the left scale
(the total of these numbers for a given n is equal to 1000, the number of conformations) with
a color scale shown on the right. The right part of the figure shows the fraction ρ of θ angles,
integrated over the whole model, which belongs to a given range of theta. The scale is truncated
to ρmax = 5 × 10−4 to better show the part of the curve which corresponds to large θ angles.
provide the best matching with the room temperature SANS data, and its right part shows
the fraction ρ of θ angles, integrated over the whole model, which belongs to a given range of
theta. This figure is analogous to Figs. 3-b and 3-d for the Widom-601 sequence, and we used
the same scales for the plots to allow a quantitative comparison between the results for the
two sequences. The difference is striking. For the sequence taken from λ-phage DNA we do
not find the many kinked conformations detected for the nucleosome positioning sequence.
0
0
4
0
0
3
0
0
2
0
0
1
)
g
e
d
(
s
n
o
i
t
a
r
u
g
i
f
n
o
c
t
s
e
b
r
o
f
θ
0
0
50
100
n
150
200
FIG. 9. Values of θn for the 10 conformations which provide the best match to the experimental
P (r) deduced from SANS measurements at room temperature on the sequence shown in Fig. 2 .
Successive plots are moved up by 50◦ to limit the overlap between the curves.
Figure 9 for the sequence taken from λ-phage DNA is equivalent to the inset on Fig. 3-
a for the Widom-601 sequence.
It shows the values of the θn for the 10 conformations
which provide the best match to the experimental P (r) deduced from SANS measurements.
Contrary to the case of the Widom-601 nucleosome positioning sequence, the bending angles
do not show any sharp spike, associated to the presence of a kink.
18
FIG. 10. Average values the the local bending angles hθni and standard deviations of the bending
angles for the 1000 conformations which provide the best match to the experimental P (r) deduced
from SANS measurements at room temperature on the sequence shown in Fig. 2. The blue lines
show the position of the AT domain along the sequence, starting either from the left end of the
sequence (full lines) or from the right end (dashed lines).
The absence of kinks for this DNA sequence does not mean that the results of the SANS
experiments are featureless. To detect the specificities of this sequence one has to look at the
average bending angles hθni, and their standard deviations σθ. They are plotted in Fig. 10.
The averages and standard deviations have been calculated over the 1000 conformations
which provide the best matching with the experiments. Both the average bending and its
fluctuations show maxima in positions corresponding to the domain of pure AT pairs. As
the measurements do not distinguish the two ends of the molecules, the contribution of this
domain on the data shows up simultaneously on the sites 94 to 125 and on 111 to 80, which
correspond to its position measured from the other end of the sequence. Moreover, as we
have showed using another experimental approach[33], the fluctuations of a large AT domain
influence the local conformation of DNA in its vicinity. Therefore we expect to detect the
influence of the AT domain also in its vicinity. This is exactly what the analysis of the
SANS data detects, as shown in Fig 10. The fluctuations of the AT domain, which lead to
disturbance in base stacking locally reduce the bending rigidity of double-stranded DNA.
This leads to an increased standard deviation of θn and an increase in its average value due
to entropic effects, which, in this case, are also reinforced by the small intrinsic curvature
of some AT elements. Our measurements also detect the small increase in hθni and σθ due
to the free ends. Therefore Fig 10 shows that our measurements and their analysis are able
to detect fairly small effects in the conformation of DNA in solution. This reinforces our
statement about the existence of kinks in the Widom-601 positioning sequence, which are
large distortions which should be easier to detect.
19
VI. DISCUSSION
Our experimental results on the Widom-601 sequence and their analysis clearly point to
a positive answer to the question raised by Crick and Klug in 1975 [2]. Some DNA sequences
can exhibit kinks, even in the absence of strong external constraints.
A first hint was provided by the analysis of the data with standard software packages
developed for SAS data analysis. The fit by SASVIEW [23], as well as the Kratky-Porod
model [24], lead to persistence lengths lp of the order of 100 A, which, at a first glance appear
unrealistically low. However this would be the case if lp was only determined by dynamical
fluctuations. But intrinsic curvature can also contribute to reduce the effective persistence
length [27]. This effect was recently studied in details for various DNA sequences with
an elaborate coarse-grain DNA model [34] and it was found that it can bring a significant
contribution. Nevertheless, for the Widom-601 sequence, our experiments indicated that the
effect had to be quite dramatic to reduce lp so much. To proceed further and determine the
main features of the molecular shapes from the data, we used an extended Kratky-Porod
model in two stages. First, an unbiased sampling of the conformational space using generic
model parameters, without sequence dependence or intrinsic curvature, showed that the
conformations providing the best match with the data exhibited a sharp, highly localized
bend, in their central region. Moreover the statistics of the bending angles found a hump
at large angle, separated by a gap from the large peak around θ = 0. This rules out a
highly flexible point caused for instance by a nick because it would lead to a single-peaked
distribution. The second stage, assuming a non-zero equilibrium value of the bending angle
at a particular site showed that kinked conformations can indeed provide a good fit of
the data. The accuracy of the SAS experiments and of our analysis cannot formally rule
out a bending distribution extending over a few sites instead of a kink. This is however
very unlikely because, owing to the very large overall bending required to fit the data, it
would need several consecutive bends of 20◦ to 30◦ which could hardly be achieved without
fully breaking the DNA structure with a high energetic cost. This is precisely because they
considered such a configuration as unlikely that Crick and Klug [2] looked for an alternative.
Instead, as shown by the model that they built, a kink in DNA can exist while leaving all
base-pair intact and all bond distances and angles stereochemically acceptable.
Our analysis relies on the choice of a particular DNA model. We opted for a model which
20
is as simple as possible but contains the essential features required to describe the DNA
backbone. The Kratky-Porod model, often used, does not include the dihedral energy. We
added this term in the Hamiltonian because it is important to control the overall shape of the
polymer, which is probed in SAS experiments. Dihedral energy prevents the free rotation
about the bonds, which could lead to large shape changes without affecting the bond-angle
energy which enters in the Kratky-Porod model. For the first stage of our analysis, the search
for conformations that best fit the data, it is important to avoid any bias in the exploration
of the conformational space, and therefore the model has to be generic. The price to pay is
that we have to generate a huge number of conformations (≈ 108) to make sure that they
include those of the molecules in solution. This price is however bearable because the model
is sufficiently simple to allow fast calculations. In the second stage of the analysis, we try
instead to design a specific model for the molecules in solution. Even within the extended
Kratky-Porod model, the number of parameters that could be adjusted is very large. We
minimized the number of free parameters by focussing our attention on a few equlibrium
values of the bending angles, as suggested by the conformational search.
Increasing the
number of adjusted parameters could improve the agreement with experiments, at the risk
of "over-fitting" with parameters that would not be statistically significant. Nevertheless it
might be interesting to refine our analysis with improved models for DNA, such as the one
used in 34. To cover all possible conformations, the model would have to be parametrized
to allow the description of kinks and not only moderate local bending. The validity of our
analysis has neverteless been tested in a complementary experiment with another sequence.
Using the same method of analysis we did not find kinks in this sequence but demonstrated
that our approach is able to detect a fairly small effect in the conformation of DNA, validat-
ing the method. The radiation dammage, caused by a long-exposure to X-Rays, provided
another, unexpected, complementary experiment. The analysis of the data is able to detect
the single stand breaks, which create additional flexible points along the sequence, and also
modify the probability distribution of the bending angles, by removing the gap between the
small angles and the very large angles characteristic of a strong permanent bend.
Kinks in DNA are not new. They were suggested by the analysis of some cyclization
experiments [10], or detected in molecular dynamics simulations of mini-circles [35] and in
the structure of the nucleosome core particle [15]. However all these examples concerned
highly constrained DNA. Our results show that kinks can also exist for DNA samples in
21
solution without any particular constraint. Therefore this peculiar DNA structure, proposed
from a model building approach by Crick and Klug, could be more common than generally
assumed. However this is not a generic property of DNA. Kinks depend on the sequence
and appear to be present in the nucleosome-positioning Widom-601 but not in a modified
λ-page sequence. Our findings may also help to resolve the recent debate concerning the
flexibility of short-chain DNA, which has focused on differences in experimental protocols
[32] but which should also consider the intrinsic properties of the DNA sequences that were
investigated.
[1] J. Watson and F. Crick, Nature 171, 737 (1953).
[2] F. Crick and A. Klug, Nature 255, 530 (1975).
[3] M. S. Ong, T. J. Richmond, and C. A. Davey, J. Mol. Biol. 368, 10671074 (2007).
[4] E. Y. D. Chua, D. Vasudevan, B. W. Gabriela E. Davey, and C. A. Davey, Nucl. Acids Res.
40, 6338 (2012).
[5] J. D. Kahn, Biophys. J. 107, 282 (2014).
[6] J. Marko and E. Siggia, Macromolecules 28, 8759 (1995).
[7] J. Yan, R. Kawamura, and J. Marko, Phys. Rev. E 71, 061905 (2005).
[8] A. Mastroianni, D. Sivak, P. Geissler, and A. Alivisatos, Biophys. J. 97, 1408 (2009).
[9] T. E. Cloutier and J. Widom, Molecular Cell 14, 355 (2004).
[10] T. E. Cloutier and J. Widom, PNAS 102, 3645 (2005).
[11] C. Yuan, H. Chen, X. W. Lou, and L. Archer, PRL 100, 018102 (2008).
[12] R. Vafabakhsh and T. Ha, Science 337, 1097 (2012).
[13] R. S. Mathew-Fenn, R. Das, T. D. Fenn, M. Schneiders, and P. A. B. Harbury, Science 322,
446 (2008).
[14] P. T. Lowary and J. Widom, J. Mol. Biol. 276, 19 (1998).
[15] D. Vasudevan, E. Y. D. Chua, and C. A. Davey, J. Mol. Biol. 403, 1 (2010).
[16] J. Bednar, I. Garcia-Saez, R. Boopathi, A. R. Cutter, G. Papai, A. Reymer, S. H. Syed,
I. N. Lone, O. Tonchev, C. Crucifix, C. Papin, D. A. Skoufias, H. Kurumizaka, R. Lavery,
A. Hamiche, J. J. Hayes, P. Schultz, D. Angelov, C. Petosa, and S. Dimitrov, Mol. Cell 66,
384 (2017).
22
[17] D. S. Sivia, Elementary Scattering Theory (Oxford University Press, Oxford, 2011).
[18] V. F. Sears, Neutron News 3:3, 26 (1992).
[19] F. Zhang, J. Ilavsky, G. G. Long, J. P. G. Quintana, A. J. Allen, and P. R. Jemian, Metall
and Mat Trans A 41, 1151 (2010).
[20] D. Svergun, J. Appl. Cryst. 25, 495 (1992).
[21] C. von Sonntag, The Chemical Basis of Radiation Biology (Taylor and Francis, London, 1989).
[22] R. M. Abolfath, A. C. T. van Duin, and T. Brabec, J. Phys. Chem. A 115, 1045 (2001).
[23] "Sasview for small angle scattering analysis," https://www.sasview.org.
[24] O. Kratky and G. Porod, Recl. Trav. Chim Pays Bas 68, 1106 (1949).
[25] N. Theodorakopoulos, "Calculation of the structure factor of an inhomogeneous Kratky-Porod
model," (2017).
[26] P. Hagerman, Ann. Rev. Biophys. Biophys. Chem. 17, 265 (1988).
[27] J. A. Schellman and S. C. Harvey, Biophys. Chem. 55, 95 (1995).
[28] M. V. Petoukhov, D. Franke, A. V. Shkumatov, G. Tria, A. G. Kikhney, M. Gajda, C. Gorba,
H. D. T. Mertens, P. V. Konarev, and D. I. Svergun, J Appl. Cryst. 45, 342 (2012).
[29] D. Svergun, Biophys. J. 76, 2879 (1999).
[30] D. Franke and D. Svergun, J. Appl. Cryst. 42, 342 (2009).
[31] J. H. Miller, J. M. Nelson, M. Ye, C. E. Swenberg, J. M. Speicher, and C. J. Benham, Int. J.
Radiat. Biol. 59, 941 (1991).
[32] Q. Du, C. Smith, N. Shiffeldrim, M. Vologodskaia, and A. Vologodskii, PNAS 102, 5397
(2005).
[33] S. Cuesta-L´opez, H. Menoni, D. Angelov, and M. Peyrard, Nucleic Acids Res. 39, 5276 (2011).
[34] J. S. Mitchell, J. Glowacki, A. E. Grandchamp, R. S. Manning, and J. H. Maddocks, J. Chem.
Theory Comput. 13, 1539 (2017).
[35] F. Lankas, R. Lavery, and J. Maddocks, Structure 14, 1527 (2006).
ACKNOWLEDGMENTS
We thank the 15ID-D USAXS beamline at the Advanced Photon Source, USA, who
kindly provided the glassy carbon standard sample used for the absolute normalization
of the SAXS data. T.S. and T.U. gratefully acknowledge the financial support by the
23
Deutsche Forschungsgemeinschaft (DFG) through the Cluster of Excellence Engineering of
Advanced Materials (EAM) and GRK 1896 In-Situ Microscopy with Electrons, X-rays and
Scanning Probes. A.G. thanks the PhD program of the ILL for providing the financial
support for his thesis. M.M.R., L.R.S. and S.C.L. acknowledge the overall support by the
Spanish Ministry of Economy, Industry and Competitiveness (BES-2013-065453, EEBB-I-
2015-09973, FIS2012-38827). S.C.L. and UC-154 are grateful for the support of Junta de
Castilla y Leon (Spain) Nanofibersafe BU079U16. D.A. acknowledges funding from the
Agence Nationale de la Recherche through ANR-12-BSV5-0017-01 "Chrome" and ANR-17-
CE11-0019-03 "Chrom3D" grants. N. T. acknowledges support by the project Advanced
Materials and Devices (MIS 5002409, Competitiveness, Entrepreneurship and Innovation,
NSRF 2014-2020) co-financed by Greece and the European Regional Development Fund.
24
|
1111.1228 | 1 | 1111 | 2011-11-04T19:57:39 | A Physiologically-Based Flow Network Model for Hepatic Drug Elimination I: Regular Lattice Lobule Model | [
"physics.bio-ph",
"q-bio.QM",
"q-bio.TO"
] | We develop a physiologically-based lattice model for the transport and metabolism of drugs in the functional unit of the liver, called the lobule. In contrast to earlier studies, we have emphasized the dominant role of convection in well-vascularized tissue with a given structure. Estimates of convective, diffusive and reaction contributions are given. We have compared drug concentration levels observed exiting the lobule with their predicted detailed distribution inside the lobule, assuming that most often the former is accessible information while the latter is not. | physics.bio-ph | physics | A Physiologically-Based Flow Network Model for Hepatic Drug
Elimination I: Regular Lattice Lobule Model
Vahid Rezania, Department of Physical Sciences, Grant MacEwan University, Edmonton, AB, Canada T5J 4S2
Rebeccah E. Marsh, Dept. of Physics, University of Alberta, Edmonton, AB, Canada T6G 2J1
Dennis Coombe, Computer Modelling Group Ltd., Calgary, AB T2L 2A6
Jack A. Tuszynski, Dept. of Physics and Experimental Oncology, University of Alberta, Edmonton, AB, Canada
T6G 2J1
(Dated: October 25, 2011)
We develop a physiologically-based lattice model for the transport and metabolism of drugs in the functional unit
of the liver, called the lobule. In contrast to earlier studies, we have emphasized the dominant role of convection
in well-vascularized tissue with a given structure. Estimates of convective, diffusive and reaction contributions are
given. We have compared drug concentration levels observed exiting the lobule with their predicted detailed
distribution inside the lobule, assuming that most often the former is accessible information while the latter is
not.
I. INTRODUCTION
Pharmacokinetics aims to understand and predict behavior of various drugs through the body by studying the drug’s
distribution, absorption, metabolism and elimination from the body. Traditionally, such information can be
extracted by examining drug concentration in the plasma or blood at several time points and generating a
concentration-time curve that first rises based on the drug’s absorption rate and then declines after reaching its
maximum value governed by the drug’s elimination ra te. The result will be used to determine optimal dosing
regimes, potential toxicities as well as possible drug-drug interactions.
To understand and predict drug behaviour through the body, compartmental models are commonly used in
pharmacokinetics, Jacquez [1]. In general, each organ is represented by a compartment that contains a homogenous
number of drug molecules undergoing a set of chemical kinetic reactions. The compartment ’s input/outpu t is
governed by linear kinetic processes with constant rate coefficients. At the end, the whole system that may contain
several compartments is described by coupled first order differential equations whose solutions take the form of a
sum of terms that are exponential in time.
Nonlinear drug-organ interactions, however, cannot be described adequately by classical compartment models. As
observed in clinical data, the concentration-time curve shows a non-exponential time-dependency, asymptotically.
Furthermore, observation of anomalous kinetics in the experimental data also suggests that the kinetic reactions
should be occurring on or within fractal media with a time-dependent kinetic rate coefficients (Anacker and
Kopelman [2], Kopelman [3]).
To address the above shortcomings several explanations have been proposed, including a stochastic random walk
model for the drug molecules (Wise et al [4] ), convective-diffusive transit behaviour in the liver (Norwich and Siu
[5]), gamma-distributed drug residence time (Wise [6]), transient fractal kinetics (Anacker and Kopelman [2]), and
fractal Michaelis-Menten kinetics (Marsh and Tuszynski [7]). All of these studies, however, represented an organ
as a homogenous compartment that is not realistic, i.e. physiologically-based.
1
A. Lattice Models of the Liver
Alternatively, to take into account the organ heterogeneity and simulate enzyme kinetics in disordered media,
lattice models have been introduced by investigators. Berry [8] performed Monte Carlo simulations of a Michaelis-
Menten reaction on a two-dimensional lattice with a varying density of obstacles to silate the barriers to diffusion
caused by biological membranes. He found that fractal kinetics resulted at high obstacle concentrations. Kosmidis
et. al [9] performed Monte Carlo simulations of a Michaelis-Menten enzymatic reaction on a two-dimensional
percolation lattice at criticality. They found that fractal kinetics emerged at large times.
Previously [10], we developed a network model of the liver consisting of a square lattice of vascular bonds
connecting two types of sites that represent either sinusoids or hepatocytes. Random walkers explored the lattice at
a constant velocity and were removed with a set probability from hepatocyte sites. To simulate different
pathological states of the liver, random sinusoid or hepatocyte sites were removed. For a lattice with regular
geometry, it was found that the number of walkers decayed according to an exponential relationship. For a
percolation lattice with a fraction p of the bonds removed, the decay was found to be exponential for high trap
concentrations but transitioned to a stretched exponential at low trap concentrations.
The models described above are all basic random walk models, and the lattices are abstract representations of the
geometry of the space. The objective of this paper is to develop a lattice model that incorporates realistic
anatomical and physiological properties of the liver as well as the flow of blood with reacting tracers
In this paper, we introduce a physiologically-based lattice model of the functional unit of the liver that takes into
account parameters such as the distribution volume, permeability, and porosity of the liver vasculature and cells.
Instead of simple random walkers, a blood-like fluid containing drug molecules flows through the vascular space of
the lattice. This model takes into account flow-limited and diffusion-limited exchange of drug molecules into the
extravascular space from the sinusoids; sequestration of the drug molecules within liver cells with enzymatic
transformation; and exchange of the metabolized drug molecules back from liver cells to the vasculature. Estimates
and consequences of the competing flow processes are given. Furthermore, the enzymatic transformation of the
drug can be either simple or saturable. The model allows us to include the effects of the intrahepatic mixing process
on the enzymatic transformation of drug molecules.
II. THE LIVER AND DRUG KINETICS
A. Liver Architecture
At the macroscopic level, the liver consists of three vascular trees, two supply trees that originate from the portal
artery and hepatic vein, and one collecting tree that drains into the portal vein [11]. The vessels bifurcate down to
the terminal arterioles and venules, which are organized into portal tracts along with a terminal bile duct. Liver
cells, called hepatocytes, radiate outward from the terminal vessels. These plates of hepatocytes are interspersed by
sinusoids, which play the role of the capillary in the liver, and the spaces of Disse, which are the extravascular
space of the liver [12]. Finally, the blood is collected and removed by the hepatic venules.
B. Functional Unit
The functional unit of an organ is the smallest structural unit that can independently serve all of the organ's
functions [13]. Because of its complexity, there is continued debate about what the functional unit of the liver
should be. The classic lobule is a hexagonal cylinder, centered around a hepatic venule and with portal tracts
situated at the corners. The portal lobule has a similar shape but is centered about a portal tract with the hepatic
venules at the periphery [14]. The acinus is another proposed unit and is based on the pattern formed by the cords
of hepatocytes between two central venules. Matsumoto and Kawakami [15] suggested that the classic lobule can
be divided into primary lobules, which are cone-shaped and each fed by one portal tract and drained by one hepatic
venule. Teutsch and colleagues [16, 17] performed a morphological study of rat and human liver lobules, and their
results support the idea of a secondary unit made up of primary units in what they term as a modular architecture.
2
They conclude, however, that the primary unit is more polyhedral in shape than conical. Other experiments done by
Ruijter et al. [18] suggest that the primary unit is needle-shaped and that there are equal amounts of portal and
central vein associated with one unit. For this study, the primary unit is taken to be one-fourth of the classical
lobule. The relevant anatomical values are listed in Table I.
C. Elimination Kinetics
m
+
=
Here drug uptake and elimination (i.e. conversion to metabolized product) is viewed as a single-step saturable
process following Michaelis-Menten kinetics [19], such that
( )
( )
( )
(
)tC
tdC
K/tCv
max
dt
(1)
( )
tdC
Here C(t) is the local drug concentration and
is the drug metabolization rate. (In what follows, C(t) is
dt
expressed as rxi with r the fluid density and xi mole fractions of i-th species). Note we are explicitly modeling the
drug transport to an individual hepatocyte surface via our lobule lattice model and assume an effective one-step
reaction transformation beyond that point. We recognize that drug incorporation and elimination is still a multi-step
process even once the drug reaches the cell surface however. It is hoped that these approximations ignore processes
occurring on a shorter time scale than the experimental resolution. Nonetheless, we explore possible complications
via simple sensitivities to the choice of reaction time constant.
III. MODEL AND METHOD
The primary unit of the liver was approximated by a symmetry element of a 51 x 51 square lattice such that four
units make up one lobule. The architecture of the lattice consists of hepatocyte grid cells interlaced by a network of
narrower sinusoidal grid cells (Figure 1). The diameter of the sinusoid grid cells was taken to be 0.0006 cm, and the
diameter of the hepatocyte grid cells was taken to be 0.0024 cm. The length of the lattice is thus 0.0750 cm per
side. Doubling this value gives a lobule diameter of 0.150 cm, which is consistent with values listed in Table I.
Convective molar flux is modelled according to Darcy's Law [20]:
(cid:209)
p
k
x
i
J
c
ik
r=
r=
vx
i
(2)
K
k
m
ik is the i-th component of fluid flux in k-direction, r and µ are fluid molar density and viscosity, v k , Kk
where Jc
and (cid:209)kp are the Darcy velocity, permeability, and pressure gradient in direction k, respectively. The blood viscosity
µ is taken to be 3.5 mPa-s (3.5 centipoise). Blood molar density is assumed that of water, r at 55.4 moles/cm3. It is
emphasized that in this paper, following Darcy’s La w and the conventions of flow in porous media, permeability K
is defined as a measure of the transmissibility of a grid cell to the flow of a fluid, and is expressed in units of area
(e.g. cm2).
Each sinusoid grid cell represents a tubular vessel of diameter 2a. Taking the ratio of the volume of the vessel to the
volume of the grid cell yields a porosity of
p=
=p=
4
(3)
7854
f
sin
k
2
aa
3
a
.0
For a cylindrical tube, the permeability can be calculated from the tube radius [13]
=
K
sin
=
2
a
8
.1
126
micron
2
(4)
3
Here and in what follows, it is noted that porosity is dimensionless while permeability has units of a characteristic
length squared.
Each parenchymal grid cell represents a cellular (hepatocyte) component and an extracellular (space of Disse)
component. A ratio of 0.75 to 0.25 was chosen for their respective contributions to the volume, and the porosity of
the parenchymal sites was therefore 0.25.
The corresponding permeability of the tissue grid cells is estimated from a Carmen-Kozeny model [20] of flow
around a spherical object (the hepatocyte) of radius 12 microns. This states the permeability is proportional to the
object's diameter Dp and the porosity ftis as follows:
f
2
3
D
tis
p
(
-
f
1 180
tis
(5)
)2
K
=
tis
An ideal result where Dp/L=1 or R/L=0.5 with L as grid size and R as particle radius gives
)
(
p=
(
)
3
5.04
1
3
(6)
=f
tis
4764
.0
5236
f-
or
.0
=
tis
Then assuming as we do, L = Dp= 24 microns, the Carmen-Kozeny formula yields
K =
tis
micron
262
.1
2
The Carman-Kozeny expression basically states physically that the order of magnitude of the permeability scales
with the particle size squared.
The above analysis is an ideal result as the ECM around the cell particle will further reduce porosity. Thus we could
study a range of tissue effective porosities, leading to a range of effective tissue permeabilities. Table II summarizes
a range of possible values. In this paper, we will utilize the “base case ” value for tissue porosity a nd permeability,
while the effects of more extreme choices will be examined in a second paper.
The convective driving force originates from an input site corresponding to a terminal portal venule at one corner of
the lattice and an output site corresponding to a terminal hepatic venule at the opposite corner. For simplicity, the
hepatic artery blood supply, which is lower in volume and pulsatile in nature, is omitted for the current simulations.
The pressure value at the inlet and the outlet are taken to be Pin = 103 kPa and Pout = 101.8 kPa, respectively. After
subtracting the atmospheric pressure, these values are consistent with experimental values quoted by Rappaport [21],
who found that the terminal portal venule pressure was in the range 0.59 kPa to 2.45 kPa and that the terminal hepatic
venule pressure was 0.49 kPa. As we shall demonstrate, this applied pressure differential results in a convective flow
level that is determined primarily by the effective permeability of the lobule. Thus various liver damage scenarios can
be expected to affect this flow. This aspect of the modelling is of practical importance and will be explored in more
detail in a separate publication.
J
x
D
(7)
For multicomponent flow, the model tracks the compositions (molar or mass fractions) of all components in the
fluid. In addition to convective transport a diffusive flux contribution of
(
)
r(cid:209)
=
d
ik
k
ik
is considered with Jd
ik being the molar diffusive flux and Dik the diffusion constant of species i in direction k. The
estimated diffusion constant in all directions used here is based on a molecular weight rescaling of glucose
diffusion. Here
(
)
GLC
(8)
·
1.7
10
D
=
2
i
10 m
-
sec
4
=
=
=
2.4
·
10
-
10
D
PAC
(9)
With a cubic root of the molecular weight ratio of glucose to PAC used as conversion factor (Factor = (180/854)0.33
= (1/4.74)0.33 = 1/1.68), an estimated effective diffusion constant for PAC is
·
-
)
(
10
10
1.7
m
68.1
sec
Tissue effective diffusion value should be less, here we employ an order of magnitude reduction in the value of D
·
-
)
(
11
10
1.7
m
68.1
sec
Effective diffusion constants for PAC-OH are assumed identical to PAC values. These values are converted to the
simulation units of cm2/min and also summarized in Table II.
The drug paclitaxel was used as a reactive tracer, and its Phase I metabolism was modeled using the general
formula of one paclitaxel (PAC) molecule being transformed into the metabolite 6ff -hydroxypaclitaxel (PAC-OH)
by the cytochrome P450 (CYP) isozyme CYP2C8 [22].
(10)
D
PAC
·
10
2.4
2
2
=
-
11
PAC + CYP > PAC-OH + CYP (11)
The enzyme only exists in grid cells containing hepatocytes so all reaction is localized in these sites. In this paper,
saturable Michaelis-Menten kinetics are assumed, defined by a maximum rate vmax (in units of molar fraction/min)
and a half saturation value KM (in units of molar fraction). This reaction proceeds in a linear manner at a rate
characterized by k = vmax/Km (in units of min-1) when injection concentrations are much below the half saturation
value.
Reaction parameter values are based on the work of Vaclavikova et al [23] who measured directly PAC conversion
to PAC-OH kinetics without any tissue distribution issues and uptake by the cell itself, as they use microsomes
directly as the source of CYP. As such, any bottlenecks associated with drug uptake should imply that reaction
rates would be slower than those based on parameters values given by Vaclakova et al [23]. We are modeling tissue
distribution effects separately based on our lobule model. Table III summarizes the reaction parameters.
The simulations were performed using the STARS advanced process simulator designed by the Computer Modelling
Group (CMG) Ltd. in Calgary, Alberta, to model the flow and reactions of multiphase, multicomponent fluids through
porous media [24], [25], [26]. Additionally, STARS has earlier been used to model reactive flow processes in cortical
bone [27], [28], [29], [30] as well as through the intervertabral disk [31].
IV. RESULTS
Non-reactive Flow Characteristics
As discussed above, flow is induced on the regular lattice of Figure 1 by applying a pressure difference across the
inlet and outlet points. With the chosen lobule flow parameters for porosity, permeability, and blood viscosity, this
translates to a steady flow rate 2.1 cm3/min as illustrated in Figure 2. A short timescale of about 2.0x10-5 min
needed to establish this pressure gradient is also illustrated in this figure by expanding the time axis. Figure 3
demonstrates the steady state velocity profile throughout the lattice, illustrating both the diverging/converging
nature of the flow near the inlet and outlet points, as well as the orders of magnitude difference of the flows in the
sinusoids and tissues, respectively. (This plot uses a logarithmic colour scale axis).
When blood with a relative composition of 1 micro-gram paclitaxel (1.8 x10-8 mole fraction) is infused into the
lattice assuming nonreactive hepatocytes, the time required to traverse the lattice is approximately 1 min as
demonstrated in Figure 4. This production profile is convective flow dominated as the addition of diffusion
minimally alters the production profile.
5
The evolution of the paclitaxel concentration on the lattice followed a spatially homogeneous progression (Figure
5), which shows the increasing levels of injected drug after 0.01 min and 0.14 min. By 0.14 min, paclitaxel is being
seen at the outlet of the lobule. After 0.5 min, paclitaxel completely covers the lattice (not shown).
If the diffusive flow contribution is removed, however, the paclitaxel profiles on the lattice are significantly different.
As is also illustrated in Figure 5, again at 0.01 min and 0.14 min, a distinct two-scale behavior is noted, whereby the
sinusoids are first infused with the drug, and only at later times do the drug levels in the tissue approach injected
concentration levels. This behavior reflects the convective levels of flow in the sinusoids and tissues noted earlier
(Figure 3). A further comparison of Figure 5 cases reveals that the sinusoid drug concentration levels in the two cases
are similar, however, explaining the similar drug production characteristics note in Figure 4, as paclitaxel is produced
directly from the sinusoids.
Base Case Reactive Flows
The effects of paclitaxel drug metabolism by hepatocytes are next considered. Here the base case reaction
parameters of Table III are employed, and the same injected paclitaxel injected concentration (1.8e-8 mole fraction)
is considered. With the employed reaction half saturation constant value of 1.8 x10-7 mole fraction, this injection
level implies the Michaelis-Menten model reduces to a linear reaction scheme.
Figure 6 illustrates injected drug and produced drug and metabolite production for this case. Essentially at this
reaction rate, all injected paclitaxel is converted to metabolite by the lobule hepatocytes. The production profile of
PAC-OH here is identical to the production profile of PAC in the non-reaction case, as shown in Figure 4. Figure 7
shows the PAC and PAC-OH profiles across the lobule lattice at 0.01 min, 0.14 min, and 0.50 min, respectively.
The PAC concentrations in the sinusoids and the PAC-OH concentrations in the tissue are equivalent to the PAC
concentrations in both sinusoids and tissue for the non-reacted case (Figure 5). Figure 7 also shows most clearly
there is an inlet distance over which the reaction conversion time is not fast enough to convert the injected
paclitaxel.
Figure 8 illustrates injected drug and produced drug and metabolite production for the same case except that
diffusive transport has been removed. In contrast to Figure 6 with diffusion, there is now only a limited amount of
conversion of PAC to PAC-OH even at long times. The PAC and PAC-OH profiles at various times (0.01 min, 0.14
min, and 0.50 min) as shown in Figure 9, confirm this behavior where it is shown that the PAC concentration in the
sinusoids propagates throughout the lobule, while the PAC-OH concentrations in the tissue increase less rapidly
and up to a lower level.
Reactive Flow Sensitivities
In this section, the consequences of the chosen reaction parameters are illustrated. Figure 10 shows production
behavior with a 100-fold reduction in maximum reaction rate and with diffusion effects included. The small level of
produced PAC indicates that reaction rates must be reduced to about this level before any significant change in drug
production behavior can be expected. Drug distribution in the lobule for this case is shown in Figure 11 for the
times 0.01 min, 0.14 min, and 0.50 min. This figure should be contrasted with Figure 9. Here the early time results
and upstream results for PAC distributions at longer times are quite different, reflecting the reduced reaction rate.
However, the later time and downstream results for PAC-OH distribution resemble quite closely the faster reaction
limit. Here the propagation time across the lobule gives enough time to compensate for changes in reaction rate. In
summary, reaction rates larger than the base case or even 10-fold reduction from base case can be expected to
produce very similar drug production behavior and differences only in the inlet region of the lobule are to be
envisioned.
In contrast, once a critical time-scale is crossed, much more significant changes in drug distribution behavior can be
expected, both internally throughout the lobule and in terms of produced profiles. Figure 12 shows drug metabolite
production behavior with a 1000-fold reduced metabolic rate, and including diffusive mixing. Here almost equal
levels of PAC and PAC-OH are seen exiting the lobule. At 0.01 min almost no PAC-OH is converted in the lobule
at this rate (see Figure 13), while at later times (0.14 min and 0.50 min), converted PAC-OH starts to be seen at the
6
outlet regions at levels similar to PAC. Essentially, the inlet region behavior occurring at faster reaction rates now
covers the whole lobule region.
Finally, sensitivities to injected PAC concentrations were explored, utilizing injection concentrations of 1.8 x10-7
and 1.8 x10-6 mole fractions (i.e. clearly above that of the base case 1.8 x10-8). In these runs, the base case reaction
parameters were maintained. In particular, the half saturation value of 1.8 x10-7 was employed, indicating that the
linear, intermediate, and saturation levels of the Michaelis-Menten expression were being probed with the three
injected concentration levels. As illustrated in Figure 14 for the runs without diffusion, the production profiles of
PAC-OH remained unchanged for each case, as long as the production maxima were rescaled to the corresponding
injection concentrations. Apparently, with fast reaction rates, the Michaelis-Menten form had little impact on
production behavior.
V. ANALYSIS
The results we have presented can be rationalized by a comparison of process timescales.
Calculation of breakthrough times can be based on two concepts: either only the sinusoids are accessible or the
whole lobule (tissue+sinusoid) is accessible to injected species. The sinusoid pore volume in our element is 4.77
x10-7 cm3 while the complete lobule element volume is 7.344 x10-7 cm3. Because the steady state flow rate in our
model is 2.44 x10-6 cm3/min (see Figure 2), this means the breakthrough time is 4.77/24.4 = 0.20 min for just
sinusoid accessibility and 7.34/24.4 = 0.30 min for the whole lobule-sampled space. These should be viewed as two
limiting vertical lines on the time history plots as two ideal limits without any diffusion or mixing effects (physical
or numerical). It is noted for example that our time history plot of PAC production with no reaction and with or
without physical diffusion (see Figure 4) has a produced concentration of 0.9 x10-8 (i.e. half of the injected 1.8 x10-8
concentration) at 0.19 min, about what is expected. The main point here is that most of the production behavior
differences for our various cases should lie between these two ideal "half-value" limits.
The next timescale is governed by a “pressure diffu sion” coefficient
D
pres
=
K
fm
C
eff
p
(12)
eff is an effective compressibility accounting for both fluid and tissue structure effects. Fluid (water
Here Cp
compressibility) is of the order of 5e-7 kPa-1. For liver (soft tissue) structural compressibility, we have chosen 1.8
x10-5 kPa-1. Using these choices and the base case parameters of this study from Table II, we obtain
2
=
+
·
10
5.1
D pres
4 cm
min
This parameter essentially describes the time taken for pressure to come to a steady state distribution as follows.
Utilizing a characteristic distance d = 0.15 cm (the lobule element size for the pressure calculation), this time is then
(13)
T
d
=
pres D
pres
2
=
5.1
-·
10
5
min
(14)
Figure 2 also illustrates this characteristic time. This timescale is essentially a function of fluid properties and
lobule structure (through f and K). If we were to consider pulsatile flow effects caused by hepatic artery inflow,
this timescale would be much more important to the general process description and a more precise definition of
compressibility might be warranted. For the present, these numbers just indicate that steady state pressure is
achieved more quickly than other process effects.
The third timescale is determined by particle diffusion. Here we have chosen diffusion constants based on
paclitaxel size and simple estimates of tortuosity. The parameter choices used here are Dsin = 2.5 x10-4 cm2/min and
7
Dtis = 2.5 x10-5 cm2/min. Again with a choice of characteristic distance d = 10 micron = 1 x10-4 cm, the times
required for particles to diffuse are
Tdiff
(
sin
)
=
T
diff
(
tis
)
=
2
d
D
sin
2
d
D
tis
=
0.4
-·
10
3
min
(15)
=
0.4
-·
10
4
min
(16)
Our diffusion values should be viewed as highly optimistic. In particular, pactlitaxel is normally not molecularly
dissolved, but rather it is some type of micellar complex with Cremophor EL surfactant, so the effective diffusion
coefficient for this complex is probably one or more orders of magnitude smaller than what has been estimated.
Thus the limits of diffusion and non-diffusion cases are meaningful extremes of what might be expected, for small
molecules and large nanoparticles, respectively.
The final timescale is reaction rate. The base limiting reaction rate 6 x10-3 min-1 converts to a reaction time
(reaction half-life) of
(
)
5.0ln
·
-
3
6
10
This is essentially seen as instantaneous compared to the other timescales considered. Reducing this basic reaction
rate by a factor of 100 or 1000 causes the reaction process to be more similar to the other timescales and different
production profiles of PAC and PAC-OH result, as has been shown.
(17)
reacT
min
0.1
=
=
·
10
+
2
Table IV summarizes the relevant assumed timescales.
VI. CONCLUSION
In pharmacokinetics, lattice models are introduced to address the non-heterogeneity of the organs on the drug
distribution that has a significant impact on drug propagation throughout the body as shown by analysis of clinical
data.
Here we utilize the interpretation of the liver as an ensemble of islands of metabolic activity and focus on the liver
lobule itself. In contrast to most earlier studies that assumed that drug molecules only randomly propagate through
the system, we have emphasized the dominant role of convection in well-vascularized tissue with a given structure.
We have utilized an idealized representation to analyze the factors affecting drug propagation and metabolism. The
lobule is divided into hepatocyte cells that are interlaced with narrower sinusiodal grid cells. These cells are
connected by constant permeability throughout the entire system. The drug molecules convectively flow through
the sinusoidal along with blood (water here) and diffuse to the hepatocyte where metabolisms are taking place. A
sensitivity analysis of convective, diffusive and reaction parameters is performed and estimates of their
contributions are presented. We have compared the drug concentration levels observed exiting the lobule with
their predicted detailed distribution inside the lobule, assuming that most often the former is accessible information
while the latter is not.
This is the first paper of series of papers on physiologically-based lattice models for liver. In this paper, we consider
an idealized lobule lattice in order to understand the basic functionality of the unit and underlying mechanisms
through simulations and also to set a basis for future studies. A following paper will expand the analysis to include
sensitivities associated with variations in lobule structure, which could reflect extents of liver damage.
Acknowledgements
J.A.T. acknowledges funding support for this project from NSERC. The Allard Foundation and the Alberta
Advanced Education and Technology.
8
REFERENCES:
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
J.A. Jacquez, "Compartmental Analysis in Biology and Medicine", BioMedware, Ann Arbor (1996).
L.W. Anacker and R. Kopelman, "Fractal Chemical Kinetics: Simulations and Experiments", J. Chem. Phys.,
81:6402Y6403 (1984).
R. Kopelman, "Rate Processes on Fractals: Theory, Simulations, and Experiments", J. Stat. Phys.,
42:185Y200 (1986).
M.E. Wise, S.B. Osborn, J. Anderson, and R.W.S. Tomlinson, "A Stochastic Model for Turnover of
Radiocalcium Based on the Observed Power Laws", Math. Biosci., 2:199Y 224 (1968).
K.H. Norwich and S. Siu, "Power Functions in Physiology and Pharmacology", J. Theor. Biol., 95:387Y 398
(1982).
M. Weiss. "Use of Gamma Distributed Residence Times in Pharmacokinetics", Eur. J. Clin. Pharmacol.,
25:695Y 702 (1983).
R.E. Marsh and J.A. Tuszynski, "Fractal Michaelis-Menten Kinetics under Steady State Conditions,
Application to Mibefradil", Pharmaceut. Res., 23(12), 2760 (2006).
H. Berry, "Monte Carlo Simulations of Enzyme Reactions in Two Dimensions: Fractal Kinetics and Spatial
Segregation", Biophys. J., 83, 1891 (2002).
K. Kosmidis, V. Karalis, P. Argyrakis, and P. Macheras, "Michaelis-Menten Kinetics under Spatially
Constrained Conditions: Application to Mibefradil Pharamacokinetics", Biophys. J., 87, 1498 (2004).
P. Chelminiak, J.M. Dixon, J.A. Tuszynski and R.E. Marsh, "Application of a Random Network with a
Variable Geometry of Links to the Kinetics of Drug Elimination in Healthy and Diseased Livers", Phys. Rev.
E., 73, 051912 (2006).
I.M. Arias, ed., The Liver: Biology and Pathology, (Lippincott Williams and Wilkins, Philadelphia, 2001), 4th
ed.
C.A. Goresky, "Uptake in the Liver: The Nature of the Process", Int. Rev. Physiol., 21, 65 (1980).
R. Saxena, N.D. Theise, and J.M. Crawford, "Microanatomy of the Human Liver – Exploring the Hidde n
Interfaces", Hepatology, 30, 1339 (1999).
E. Bhunchet and K. Wake, "The Portal Lobule in Rat Liver Fibrosis: A Re-evaluation of the Liver Unit",
Hepatology, 27, 481 (1998).
T. Matsumoto and M. Kawakami, "The Unit Concept of Hepatic Parenchyma – A Re-examination based on
Angioarchitectural Studies", Acta Pathol. Jpn., 32, 285 (1982).
[16] H.F. Teutsch, "The Modular Microarchitecture of Human Liver", Hepatology, 42, 317 (2005).
[17] H.F. Teutsch, D. Schuerfeld and E. Groezinger, "Three-dimensional Reconstruction of Parenchymal Units in
the Liver of the Rat", Hepatology, 29, 494 (1999).
[18]
[19]
J.M. Ruijter, J. Markman, M.M. Hagoort, A.F. Moorman and W.H. Lamers, "Relative Distance: The Key to
the Shape of Hepatic Building Blocks", Image Anal. Sterol., 19, 19 (2000).
R.E. Marsh, J.A. Tuszynski, M. Sawyer and K. Vos, "Emergence of Power Laws in the Pharmacokinetics of
Paclitaxol due to Competing Saturable Processes", J. Pharm. Pharmaceut. Sci., 11(3), 77, (2008).
[20]
F.A.L. Dullien, "Porous Media: Fluid Transport and Pore Structure", Academic Press, (1979).
[21] A.M. Rappaport, "Hepatic Blood Flow: Morphologic Aspects and Physiologic Regulation", Int. Rev. Physiol.,
21, 1 (1980).
9
[22]
[23]
B. Monsarrat, E. Chatelut, I. Royer, P. Alvinerie, J. Dubois, A. Dezeus, H. Roche, S. Cros, M. Wright and P.
Canal, "Modification of Paclitaxol Metabolism in Cancer Patient by Induction of Cytochrome P450 3A4",
Drug Met. Disp., 26, 229 (1998).
R. Vaclavikova, P. Soucek, L. Svobodova, P. Anzenbacher, P. Simek, F. Guengerich and I. Gut, "Different In
Vitro Metabolism of Paclitaxel and Docetaxoel in Humans, Rats, Pigs, and Minipigs", Drug Metabolism and
Disposition, 32(6), 666 (2004).
[24]
CMG. Ltd., STARS User's Guide: Advanced Process and Thermal Reservoir Simulator, (Computer Modelling
Group Ltd., Calgary, AB, 2011).
[25] V. Oballa, D. Coombe, and W. Buchanan, "Factors Affecting the Thermal Response of Naturally Fractured
Reservoirs", JCPT, 32(8), 31 (1993).
[26] G. Darche, J. E. Grabenstetter and P. H. Sammon, "The Use of Parallel Processing with Dynamic Gridding",
paper 93023, presented at the 2005 SPE Reservoir Simulation Symposium, Houston, TX, January 31-
February 2, 2005.
[27] G.C. Goulet, N. Hamilton, D. Cooper, D. Coombe, D. Tran, R. Martinuzzi and R. Zernicke, "Influence of
Vascular Porosity on Fluid Flow and Nutrient Transport in Loaded Cortical Bone", J. Biomechanics, 41(10),
2169, (2008).
[28] G.C. Goulet, D.M.L. Cooper, D. Coombe and R.F. Zernicke, "Influence of Cortical Canal Architecture on
Lacunocanalicular Pore Pressure and Fluid Flow", Comput. Methods Biomech. Biomed. Eng., 11(4), 379
(2008).
[29] G.C. Goulet, D.M.L. Cooper, D. Coombe and R.F. Zernicke, "Poroelastic Evaluation of Fluid Movement
through the Lacunocanicular System", Annals of Biomed. Eng., 37(7), 1390 (2009).
[30] G.C. Goulet, D.M.L. Cooper, D. Coombe and R.F. Zernicke, "Validation and Application of Iterative
Coupling to Poroelastic Problems in Bone Fluid Flow", Bulletin of Applied Mechanics, 5(1), 6 (2009).
[31] K.M. Louman-Gardiner, D. Coombe and C.J. Hunter, "Computational Models Simulating Notochordal Cell
Extinction during Early Aging of an Intervertebral Disk", Comput. Methods Biomech. Biomed. Eng.,
approved for publication, 2011.
TABLES:
TABLE I: Anatomical Parameter Values for the Liver [11], [12]
Parameter
Hepatocyte diameter
Diameter of liver cell sheets
Lobule diameter
Mean sinusoid diameter
Vascular tissue component
Specific gravity of liver
Liver volume
Value
12-24 mm
25 mm
1 - 2.5 mm
7.3 mm
28 - 30%
1.05
1071 ± 228 cm 3
10
TABLE II: Lobule Regular Lattice Flow Parameters
Parameter
Characteristic Unit
STARS Unit
Sinusoid Grid Cell Size
Sinusoid Porosity
Sinusoid Permeability
Tissue Porosity (Base)
Tissue Grid Cell Size
Tissue Porosity (Ideal)
Tissue Permeability (Ideal)
Sinusoid Effective Diffusion
0.0006 cm
6 micron
0.7854
0.7854
1.125 micron2
1.140 Darcy
2.5 x10-4 cm2/min
4.2e-10 m2/sec
0.0024 cm
24 micron
0.4764
0.4764
1.230 micron2
1.246 Darcy
0.2382
0.2382
7.45x10-2Darcy
7.35 x10-2 micron2
0.1192
0.1191
6.97 x10-2 Darcy
6.883 x10-2 micron2
2.5 x10-5 cm2/min
4.2 x10-11 m2/sec
(1 Darcy = 0.9869 microns**2 in engineering permeability units).
Tissue Permeability (ECM)
Tissue Permeability (Base)
Tissue Effective Diffusion
Tissue Porosity (ECM)
TABLE III: Paclitaxel Kinetic Elimination Michaelis-Menten Parameters (converted* from Vaclavikova et al, their
Table 4)
Parameter
Characteristic Unit
Maximum Rate Vmax
Half Saturation Constant Km
STARS Unit
1.08 x10-9 molefrac/min
0.06 micronM/min
1.8 x10-7molfrac
10.0 micronM
6.0 x10-3 min-1
6.0e min-1
*Their Table 4 quotes Vmax = 61 picomol/mg_protein/min. Using their microsomal
protein concentration of 1 mg/ml, these numbers convert to Vmax = 0.06 micronM/min.
Linear Rate Vmax/Km
TABLE IV: Assumed Lobule Process Time Constants
Process
Time
Convective Transit Time (sinusoid network only)
0.200 min
1.5 x10-5 min
Pressure Relaxation Time Constant (in sinusoids)
Diffusion Relaxation Time Constant (CYP/CYP-OH in sinusoids) 4.0 x10-3 min
4..0 x10-4 min
Diffusion Relaxation Time Constant (CYP/CYP-OH in tissue)
1.0 x10+2 min
Base Case Metabolic Uptake/Elimination Time Constant
11
FIGURES
Figure 1: Homogeneous lattice. The high porosity bands represent sinusoids and the lower porosity regions
represent tissue containing hepatocytes.
12
(a)
(b)
Figure 2: (a) Steady state flow across the lobule, (b) Short time flow transients.
13
Figure 3: Steady state velocity profile across the lobule.
Figure 4: Non-reactive PAC drug propagation across the lobule, with and without diffusion effects.
14
(a)
(c)
(b)
(d)
Figure 5: Non-reactive PAC profiles across the lobule (a) at 0.01 min, with diffusion effect, (b) at 0.14 min, with
diffusion effect, (c) at 0.01 min, no diffusion effect, (d) at 0.14 min, no diffusion effect.
15
Figure 6: Reactive PAC and PAC-OH drug propagation across the lobule, with diffusion effects and base case
metabolism.
16
(a)
(b)
(d)
(e)
(c)
(f)
Figure 7: Reactive PAC and PAC-OH profiles across the lobule with diffusion effects and base case metabolism
(a) PAC at 0.01 min, (b) PAC at 0.14 min, (c) PAC at 0.50 min, (d) PAC-OH at 0.01 min, (e) PAC-OH at 0.14 min, (f)
PAC-OH at 0.50 min.
17
Figure 8: Reactive PAC and PAC-OH drug propagation across the lobule, without diffusion effects and base case
metabolism.
18
(a)
(d)
(b)
(c)
(e)
(f)
Figure 9: Reactive PAC and PAC-OH profiles across the lobule without diffusion effects and base case
metabolism (a) PAC at 0.01 min, (b) PAC at 0.14 min, (c) PAC at 0.50 min, (d) PAC-OH at 0.01 min, (e) PAC-OH at
0.14 min, (f) PAC-OH at 0.50 min.
19
Figure 10: Reactive PAC and PAC-OH drug propagation across the lobule, with diffusion effects and 100-fold
reduced metabolism.
20
(a)
(b)
(c)
(d)
(e)
(f)
Figure 11: Reactive PAC and PAC-OH profiles across the lobule with diffusion effects and 100-fold reduced
metabolism (a) PAC at 0.01 min, (b) PAC at 0.14 min, (c) PAC at 0.50 min, (d) PAC-OH at 0.01 min, (e) PAC-OH at
0.14 min, (f) PAC-OH at 0.50 min.
21
Figure 12: Reactive PAC and PAC-OH drug propagation across the lobule, with diffusion effects and 1000-fold
reduced metabolism.
22
(c)
(a)
(b)
(d)
(e)
(f)
Figure 13: Reactive PAC and PAC-OH profiles across the lobule with diffusion effects and 1000-fold reduced
metabolism (a) PAC at 0.01 min, (b) PAC at 0.14 min, (c) PAC at 0.50 min, (d) PAC-OH at 0.01 min, (e) PAC-OH at
0.14 min, (f) PAC-OH at 0.50 min.
23
Figure 14: PAC-OH metabolite production levels from various injected PAC concentrations.
24
|
1801.08635 | 1 | 1801 | 2018-01-25T23:56:53 | Laser induced topological cues shape, guide, and anchor human Mesenchymal Stem Cells | [
"physics.bio-ph"
] | This report focuses on the effect of the surface topography of the substrate on the behavior of human mesenchymal stem cells from bone marrow (MSCs) before and after co-differentiation into adipocytes and osteoblasts. Picosecond pulsed laser ablation technology was applied to generate different microstructures (microgrooves and microcavities) on poly (L-lactide) (PLLA), where orientation, cell shape and MSCs co-differentiation were investigated. On flat PLLA, the undifferentiated MSCs showed rounded or elongated shapes, the latter being randomly oriented. On PLLA microgrooves however, MSCs adapted their shape to the groove size and direction and occasionally anchored to groove edges. It was found that adipocytes, contrary to osteoblasts, are highly sensitive to topological cues. Adipocytes responded to changes in substrate height and depth, by adapting the intracellular distribution of their lipid vacuoles to these physical constraints. In addition, the modification of PLLA by laser ablation enhanced the adherence of differentiated cells to the substrate. These findings show that picosecond pulsed laser micromachining can be applied to directly manufacture 3D microstructures that guide cell proliferation, control adipocyte morphology and improve the adhesion of bone and fat tissue. | physics.bio-ph | physics | Laser
induced topological cues shape, guide, and anchor human
Mesenchymal Stem Cells
R. Ortiz a, S. Moreno-Flores b, I. Quintana a,c, MdM Vivanco d, J.R. Sarasua e, J.L. Toca-Herrera f
a Ultraprecision Processes Unit, Fundación IK4-TEKNIKER, C/Iñaki Goenaga 5, 20600, Eibar,
Gipuzcoa, Spain.
b Former address: Institute for Biophysics, Department of Nanobiotechnology, University of
Natural Resources and Life Sciences of Vienna, Muthgasse 11, 1190 Vienna, Austria
c Micro and Nanoengineering Unit, CIC microGUNE, Goiru Kalea 9, 20500, Arrasate-
Mondragón, Gipuzkoa, Spain.
d Cell Biology & Stem Cells Unit, CIC bioGUNE, Technology Park of Bizkaia, Ed. 801A, 48160
Derio, Spain
e University of the Basque Country (UPV/EHU), School of Engineering, Department of Mining and
Metallurgy Engineering & Materials Science, Alameda de Urquijo s/n, 48013 Bilbao, Spain
f Institute for Biophysics, Department of Nanobiotechnology, University of Natural Resources and
Life Sciences of Vienna, Muthgasse 11, 1190 Vienna, Austria
Correspondence should be addressed to: [email protected]; [email protected];
[email protected]
Abstract
This report focuses on the effect of the surface topography of the substrate on the behavior of human
mesenchymal stem cells from bone marrow (MSCs) before and after co-differentiation into adipocytes and
osteoblasts. Picosecond pulsed laser ablation technology was applied to generate different microstructures
(microgrooves and microcavities) on poly (L-lactide) (PLLA), where orientation, cell shape and MSCs co-
differentiation were investigated. On flat PLLA, the undifferentiated MSCs showed rounded or elongated
shapes, the latter being randomly oriented. On PLLA microgrooves however, MSCs adapted their shape to
the groove size and direction and occasionally anchored to groove edges. It was found that adipocytes,
contrary to osteoblasts, are highly sensitive to topological cues. Adipocytes responded to changes in
substrate height and depth, by adapting the intracellular distribution of their lipid vacuoles to these physical
constraints. In addition, the modification of PLLA by laser ablation enhanced the adherence of differentiated
cells to the substrate. These findings show that picosecond pulsed laser micromachining can be applied to
directly manufacture 3D microstructures that guide cell proliferation, control adipocyte morphology and
improve the adhesion of bone and fat tissue.
Key words: Picosecond pulsed laser ablation; poly-L-lactide; mesenchymal stem cells; microtopography;
contact guidance effect; adipocytes
1
1. Introduction
Adult stem cells are the main source for developing future new strategies in regenerative
medicine, cell-based therapy, and tissue engineering [1-3]. Proliferation and differentiation of
stem cells in vivo are regulated by their microenvironment, known as niche, which comprises both
cellular components and interacting signals between them [4,5,6]. These niches, in addition to
other functions, provide stem cells with physical anchors (by means of adhesion molecules) and
regulate the molecular factors that control cell number and fate [5]. Some of these factors are
influenced by cell shape, cytoskeletal tension and contractility [7, 8]. Among the biomaterial
properties that affect cell behavior, surface topography has great potential to control cell shape
and location [9]. Several researchers have observed that microscale and nanoscale topographies
in the form of pillars, grooves, pits or pores can induce the differentiation of human mesenchymal
stem cells to a certain cell lineage [10, 11, 12]. In this regard, the design of biomaterials with
architectures that mimic natural cell microenvironments may be a powerful tool to better
understand and manipulate cell function as a strategy for future cell-based therapeutics. In this
context, surface microstructuring and micromanufacturing techniques can play an important role
in the field of three-dimensional scaffold fabrication, which may enable in vitro cells mimic in vivo
cells, resembling the cellular tridimensional networks and the structural organization of human
tissues [13, 14].
Pulsed laser ablation is a well-established micropatterning tool for almost all type of materials [15-
17]. This technology represents a very versatile method for 3D micro-structuring that allows direct
and fast processing of a wide variety of geometries, including the micromachining of complex
structures on flat and non-flat surfaces. Surface modification by laser technologies is thus a
promising technique for scaffold fabrication. Liu and collaborators applied for the first time the
ultra-short pulsed laser technique for this purpose [18]. They used a femtosecond pulsed laser to
generate micro-structured (i.e. holes, grooves and grids) on substrates made of collagen, and
studied the growth, adhesion and viability of human fibroblasts and rat bone marrow MSCs on
these patterns. To the best of our knowledge only five subsequent reports have applied
femtosecond pulsed laser ablation to create 3D microstructures of biocompatible materials for cell
culture [19-23]. Despite the high quality structures that a femtosecond pulsed laser is able to
generate, some technical limitations, such as the low processing speed and low efficiency, make
2
difficult its application in industry where power scaling is a necessary requirement. In contrast,
picosecond pulsed lasers are easier to be implemented in industrial processes due to their better
cost-effectiveness and reliability, which has led to an increased presence of these lasers in the
industrial market during the last years. However, it has been scarcely used for scaffold fabrication,
which has been reported only once before our previous work. Schlie and collaborators applied
picosecond pulsed laser ablation on substrates made of silicon to test cell compatibility [24], while
previous work from our group has evaluated and applied this technology to the fabrication of
three-dimensional biodegradable scaffolds aimed for cell engineering [25]. In that study, the effect
of the so-created topographies on the morphology and proliferation of breast cancer cells cultured
on Poly-L-Lactide (PLLA) and Polystyrene (PS) was investigated. In this work, we have examined
the effect that laser-created microstructures on PLLA substrates have on cell shape, orientation
and adipocyte/osteoblasts co-differentiation of human bone marrow mesenchymal stem cells.
2. Materials and methods
2.1 Materials
Poly-L-Lactide (PLLA) was supplied by Biomer. PLLA sheets of approximately 300 µm of
thickness and a degree of crystallinity of 4% (measured with differential scanning calorimetry)
were obtained by thermoforming. Cell culture dishes were fabricated by sealing Nylon rings onto
PLLA sheets (see Figure 1).
2.2 3D Microstructuring technique
Surface micro-structuring of polymer samples was carried out with a picosecond pulse Nd:YVO4
laser (RAPID: Lumera Laser, Germany) integrated in a micromachining workstation by 3D-
Micromac. A detailed description of the experimental set-up and the optimization of the micro-
structuring process for PLLA can be found in [25, 26].
By means of pulse overlapping different trenches can be fabricated. Trench width and depth were
controlled by selecting an appropriate energy (E), frequency (f) and overlapping distance between
3
pulses (d). By using a galvanometric scanner and appropriate control strategies, any desired
topography and geometry can be generated on the workpiece. Flat PLLA substrates (Ra = 240
nm) were laser-irradiated applying an energy of 0.9 μJ at a frequency of 100 kHz, and 5 µm of
overlapping distance. After this treatment, the average surface roughness (Ra) was increased to
700 nm (Figure 2a). Microgrooves on PLLA (Figure 2b) were obtained by applying an energy of
2.3 µJ at a frequency of 250 kHz, and 2.4 µm of overlapping distance. Microcavities with different
geometrical shapes (i.e. square-like, circular, rectangular and elliptical) of areas in the range
(1.25-5)·105 µm2 were fabricated by overlapping the grooves with a distance of 2 µm at an energy
of 1 µJ (Figure c-d show some of these microcavities). The depth of these microcavities has been
set to 40 µm, 8 times as large as the cell height, to ensure cell confinement. Dimensions (width
and depth) of the different laser-generated topographies were measured by a mechanical stylus
profilometer (Dektak 8, Veeco, USA). According to DIN EN ISO 4288:1998, samples profiles of 4
mm length were considered in the measurement of the average surface roughness (Ra).
2.3 Cell culture
Human mesenchymal stem cells from bone marrow (MSCs) were provided by Promocell
(Germany). Prior to cell culture, PLLA dishes were UV-sterilized for 30 minutes. Cells were
cultured in growth medium (Promocell, Germany) at a cell density of 10000 cells/cm2. After cell
seeding, the samples were maintained at 37C and 5% CO2. In order to observe cell proliferation
and morphology on laser patterns, before seeding, cells were stained with NeuroDio, a green-
fluorescent cytoplasmic membrane stain (Promokine, Germany). To study cell differentiation,
MSCs were seeded in PLLA dishes (30000 cells/cm2), and were left to proliferate until 100%
confluence, which corresponds to 24 hours in culture. Then the growth medium was replaced by
a differentiation-induction medium, which contained a 1:1 mix of adipogenic and osteogenic
induction media (Promocell), and cells were incubated for 2 weeks changing the induction
medium every three days. After two weeks, cells were fixed in 4% formaldehyde for 5 minutes at
room temperature. Subsequently, cells were stained with Fast Blue RR Salt/Napthol solution
(Sigma-Aldrich, Germany) to tag alkaline phosphatase activity (AP staining), an early indicator of
cells that undergo osteogenesis, and with Oil Red O solution (Sigma-Aldrich, Germany) to tag the
4
fat deposits (lipid vacuoles) characteristic of adipogenesis. Cell proliferation and differentiation
were observed with an inverted fluorescence microscope Nikon Eclipse TE-2000-S (Nikon,
Japan) equipped with a Hg vapor lamp equipped, a black/white digital sight DS-Qi1MC, and filter
blocks (CFP: EX 436/20, DM 455, BA 480/40; GFP/FITC: EX 480/40, DM 505 and BA 535/50).
Bright field experiments are denoted as (BF) in the text.
2.4. Data analysis
Cell number, cell coverage and osteoblast:adipocyte differentiation ratio were obtained from
microscopy images using the freeware Image J (http://imagej.nih.gov/ij/). All data were expressed
as means ± standard deviation. Statistical analysis was carried out using the Student's t-test and
the values were considered significantly different when p<0.05.
3. Results
3.1. Microgrooves manage MSCs shape and form anchorage points.
Previously, we found that cell alignment on parallel grooves occurred to the greatest extent when
pattern density increased: that means narrow grooves and spacing (10 µm wide, and 15 µm
spacing) [25]. Therefore, this array of parallel grooves was fabricated to investigate their effect on
the shape and proliferation of undifferentiated MSCs. The center of each PLLA dish was divided
into four areas, three of which were filled with parallel grooves at different directions: 0°, 45°, and
90°. A fourth square was left untreated as control (flat PLLA). Laser-created grooves exhibited
two types of topographies: depressions with dimensions close to cell size; and protrusions formed
at groove edges by the recast material. PLLA recast is particularly prominent at these regions as
a consequence of the first-pulse effect of the laser (generation of an intense first light pulse) [27].
Figure 3 shows undifferentiated MSCs proliferating on flat (Figure 3a-c) and on PLLA grooves
(Figure 3d-i) after 1, 8 and 22 days in culture. On flat PLLA, after one day in culture, MSCs
adopted various shapes, with a mixture of rounded and elongated in different directions. After
twenty-two days and 80% cell confluency, cell bundles formed with a particular orientation, maybe
due to enhanced cell-cell interactions. In contrast, on PLLA grooves, MSCs with elongated shapes
5
predominate already after 1 day in culture. The direction of these elongated cells coincides with
the groove orientation as Figure 3g-i show, which clearly confirms the influence of substrate
topography on the early alignment of cells.
The effects of substrate topography on cell orientation were quantified. Figures 4a-c depict polar
plots of the number of cells with elongated shapes, normalized to the total cell number, as a
function of groove orientation on flat PLLA and PLLA grooves after 1, 8 and 22 days in culture.
Cells with rounded shapes were considered as non-oriented. The graphs show that at low cell
confluence the number of oriented cells and angle of orientation remain constant with time in
culture, regardless of the PLLA topography (i.e., flat PLLA or grooves oriented at 0° and 45°).
Figure 4d represents the number of oriented cells normalized to the total cell number as a function
of time in culture. At times as long as 8 days, cell alignment was clearly more frequent on PLLA
grooves than on flat PLLA. After 22 days, cell alignment on flat PLLA increased to values
comparable to those obtained on PLLA grooves, however this increment is not statistically
significant (p>0.3, according to the Student t-Test) due to the high associated error (20%). On
PLLA grooves, cell alignment remained constant at a value of 70% (± 10%).
Figure 5 shows undifferentiated MSCs in the gaps between grooved areas, anchored to the
protrusions at the groove edges after 1 day in culture. As the images indicate, cells readily spread
across the 200 µm wide interspace between patterned areas by developing filopodia-like
protrusions that seem to anchor to groove edges
3.2. Effect of surface topography on co-differentiation of human mesenchymal stem cells
In-vitro differentiation of MSCs is most efficient when it is induced at high cell confluency (i.e. 90%
[28]). Under these conditions, it has been shown that cell-cell interactions may be more relevant
than cell-substrate interactions to determine the shape and orientation of MSCs [28].
Nevertheless, when a micropatterned surface is considered, changes on the surface roughness
as well as the presence of topological barriers and cavities altered the behaviour of differentiated
cells as we will show in the following sections.
6
3.2.1. Topological barriers alter the intracellular distribution of lipid vacuoles
Descending from MSC precursors, differentiated adipocytes are round to allow for maximal lipid
storage in adipose tissue, while osteoblasts tend to spread to facilitate matrix deposition activity.
To determine the level of adipogenic differentiation, the distribution of lipid vacuoles cultured on
flat or rough PLLA for 14 days was examined. The clusters of lipid vacuoles were almost rounded
and their distribution was not affected by surface roughness on PLLA (Figure 6a-b). Similarly, AP
staining suggested that osteoblasts distribution was unaffected. However the lipid vacuoles
showed an aligned distribution along the edge between the two zones of different roughness
(Figure 6c). Similar behavior was observed on the edges of laser-microcavities, where cells found
a topological barrier of approximately 40 µm high (Figures 6d-e). Lipid vacuoles close to the edge
of these cavities lined up along the borderline irrespectively of the size of the microcavity, which
in this case was much larger than a single cell.
In contrast, on PLLA grooves, where two topological barriers are just a few micrometers apart
(i.e. 15 µm), the distribution of the lipid vacuoles was different (Figures 7-c). Lipid vacuoles
arranged in strings both inside (67 ± 14 %) and in between (29 ± 13 %) the grooves (Figure 7d).
In contrast, substrate topology did not induce any noticeable effect on the osteoblast shape.
3.2.2. Laser surface modification enhance the adherence of differentiated human mesenchymal
stem cells
Laser treatment alters surface roughness, the extent of which mainly depends on the parameters
that control laser performance. In our case, laser irradiation increased the roughness of PLLA by
3-fold. We would like to determine whether the resulting surface modification may in turn affect
the extent of differentiation of MSCs, the adipocyte:osteoblast differentiation ratio (i.e.
codifferentiation) or both.
To analyze further the potential effects of the different substrates on MSC differentiation, the area
covered by each cell type was determined. The histogram in Figure 8 shows the percentage of
area covered by undifferentiated MSCs, cells differentiated into adipocytes, and cells
differentiated into osteoblasts on different substrates: flat PLLA (control), laser-irradiated (rough)
7
PLLA and laser microstructured PLLA (grooves). The area covered by MSCs that have undergone
differentiation (approximately 90%) remained constant irrespectively of the type of substrate
(p>0.1, according to the Student t-Test). Regarding the adipocyte:osteoblast differentiation ratio,
it was observed that, on all types of substrate, the area covered by osteoblasts was larger than
that covered by adipocytes.
While laser treatment thus appears not to have an effect on the co-differentiation of MSCs, it
certainly did affect cell adherence upon differentiation as shown in Figure 9a-c. MSCs were
seeded on partially laser-treated PLLA dishes at high cell confluence to form a continuous layer
and allowed to co-differentiate for as long as 14 days. During differentiation, part of the cell layer
spontaneously detached from the substrate. The low-magnification micrographs show that only
those cells cultured on laser-treated areas remained adhered, indicating that the cell layer actually
detached from the untreated (flat) PLLA surface. The enhancement of cell-substrate interactions
may actually occur on laser-irradiated areas, regardless of the presence of microstructures
(Figures 9a-b) or microsized barriers (such as in cavities, Figure 9c).
4. Discussion
We have examined the effects that laser-created microstructures on PLLA have on MSC shape,
orientation and differentiation. MSCs cultured on PLLA micro-grooves showed the cell contact
guidance effect [29] already 24 h after seeding, and it was still noticeable 22 days later. Our results
agree with previous reports on the alignment of stem cells along micro- and nano-grooved-
patterned substrates [30-32, 23]. In addition, the protrusions generated by our technique at these
points seemed to favor cell anchoring and elongation between them. Hamilton and collaborators
observed a similar phenomenon in osteoblasts proliferating on boxes and pillars [33], which they
called gap guidance, a type of contact guidance for cell alignment that is associated to
discontinuous topographical edges. Our results on microgrooves show that these microstructures
influence MSCs in two ways: on the one hand, grooves influence cell orientation, since cells
adapted their shape to groove width and orientation (the maximal effect was observed at the first
stages of MSCs proliferation); on the other hand, the grooves promote cell adherence and
guidance at their edges. These effects appear to resemble the influence of stem cell niches in
8
vivo [5]. Further experiments would be necessary to examine the extent of the gap guidance, the
topology of the edge protrusion and its spatial distribution that altogether make groove edges
effective anchor points to MSCs and guide their shape
It was found that adipocytes, in contrast to osteoblasts, are highly sensitive to surface topography.
They responded to every type of laser surface modification, adapting their morphology to the
created topographies. Adipocyte differentiation on grooved PLLA showed the alignment and
confinement of the lipid vacuoles in strings along the grooves. Similarly, adipocytes adapted the
distribution of lipid vacuoles to follow the borderline between flat and rough PLLA, and also the
40 µm-deep edge of the microcavities. To the best of our knowledge, this is the first observation
of adipocyte compliance to topological cues. The findings suggest that surface topography is a
potential tool to control (or modify) adipocyte morphology. Mature adipocytes release a big variety
of factors that play a fundamental role in the regulation of many important functions of the body,
such as the regulation of appetite, energy homeostasis, insulin sensitivity, immunological
responses and vascular diseases [34]. The expression of these factors is, in many cases,
controlled by adipocyte size and location [35, 36]. In this context, adipocytes should be able to
change their morphology in order to store the optimum amount of fat and to perform properly their
physiological functions [37]. Therefore, it is likely that if surface topography influences adipocyte
morphology, it could also affect the expression of different factors. This may have beneficial
consequences in nano- and regenerative medicine, since surface topography could be tailored to
treat diseases related to the functions regulated by these factors. Additionally, specially designed
scaffolds can be developed for the formation of adipose tissue [38].
The effect of physical constraints on cell behavior and differentiation has been investigated in this
work. Contrary to results reported on chemically confined cells (such as adhesive/non-adhesive
treatments) [7, 39-41], in our study, there were not significant differences between cell
differentiation inside and outside the microcavities of different shapes and at these dimensions.
However, it is relevant to point out that our experiments showed some other differences compared
to the studies conducted by other authors. Under the present conditions, where no chemical
restrictions were imposed on the cells, we did not induce mechanical force gradients in the
borderline of the microcavities, which can explain the observed differences.
9
Cell adherence of differentiated cells is affected by the substrate. When cells differentiate into
adipocytes and osteoblasts on flat PLLA, the cell monolayer detached easily from the substrate;
this suggests cell-cell interactions outperform cell-substrate interactions during differentiation,
which may be related to the formation of extracellular matrix (ECM) [42]. However, the cell
monolayer remained anchored to the substrate on laser-roughened PLLA, grooves and
microcavities, suggesting that laser-induced roughness in every type of microstructure enhances
cell-substrate interactions regardless of shape or dimensions, and without significantly affecting
cell type differentiation. Therefore, laser surface modification could be applied to directly improve
cell adhesion without producing adverse effects on material biocompatibility, as observed in other
tissue adhesives [43, 44].
5. Summary and Conclusions
The present study shows that topographical features at micrometer scale are able to influence
the morphology and orientation of human mesenchymal stem cell and adipocytes, without the
interplay of chemical factors. Laser-generated microstructures induced the cell contact guidance
effect on human mesenchymal stem cells, favoring cell organization, directing cell anchorage,
and enhancing cell adherence. In addition, adipocytes, contrary to osteoblasts, are highly
sensitive to substrate microtopography, and accommodate along topological edges and barriers
at different degrees of confinement. Furthermore, laser modification on PLLA enhances the
adherence of the cell monolayer during osteogenesis and adipogenesis differentiation without
compromising the latter. In view of these results, laser microstructuring represents a potential tool
for providing a 3D cell microenvironment representative of stem cell niches in vivo, enhancing cell
adherence and organization. Although it is likely that both physical and chemical surface
modification is required to control cell differentiation and induce cell fate, our experimental
approach could contribute to enhance the biocompatibility of implants, as well as to improve
existing surgical procedures that involve healing of soft tissue.
10
Acknowledgements
This work was partially funded by the Basque Government - (UE09+/13) and the IGS BioNanoTech
Program of the Federal Ministry for Science and Research, Austria.
References
[1] Chen Y, Shao JZ, Xiang LX, Dong XJ, Zhang GR. Mesenchymal stem cells: a promising candidate
in regenerative medicine. The International J Biochem and Cell Biol 2008; 40: 815-820.
[2] Wu P, Castner DG, Grainger DWJ. Diagnostic devices as biomaterials: a review of nuclei acid and
protein microarray surface performance issues. J Biomater Sci Polym Ed 2008; 19: 725-753.
[3] Ghosh K, Ingber DE. Micromechanical control of cell and tissue development: Implications for tissue
engineering. Adv Drug Deliv Rev 2007; 59: 1306-1318.
[4] Marx V. Where stem cells call home. Nat Methods 2013; 10(2): 111-115.
[5] Peerani R, Zandstra PW. Enabling stem cell therapies through synthetic stem cell-niche
engineering. J Clin Invest 2010; 120: 60-70.
[6] Prewitz MC, Seib FP, von Bonin M, Friedrichs J, Stiβel A, Niehage C, et al. Tightly anchored tissue-
mimetic matrices as instructive stem cell microenvironment. Nat Methods 2013; 10(8): 788-794.
[7] Kilian KA, Bugarija B, Lahn BT, Mrksich. Geometric cues for directing the differentiation of
mesenchymal stem cells. Proc Nat l Acad Sci USA 2010; 107: 4872-4877.
[8] Kolind K, Leong KW, Besenbacher F, Foss M. Guidance of stem cell fate on 2D patterned surfaces.
Biomaterials 2012; 33: 6626-6633.
[9] Ross AM, Jiang Z, Bastmeyer M, Lahann J. Physical aspects of cell culture substrates:topography,
roughness, and elasticity. Small 2012; 8(3): 336-355.
[10] Béduer A, Vieu C, Arnauduc F, Sol JC, Loubinoux I, Vaysse L. Engineering of adult human neural
stem cells differentiation through surface micropatterning. Biomaterials 2012; 33: 504-514.
[11] Khang D et al. Role of subnano-, nano-, and submicron-surface features on osteoblast differentiation
of bone marrow mesenchymal stem cells. Biomaterials 2012; 33: 5997-6007.
[12] Kumar G, Tison CK, Chatterjee K, Pine PS, McDaniel JH, Salit ML, et al. The determination of stem
cell fate by 3D scaffold structures through the control of cell shape. Biomaterials 2011; 32: 9188-
9196.
[13] Liu H, Roy K. Biomimetic Three-Dimensional Cultures Significantly Increase Hematopoietic
Differentiation Efficacy of Embryonic Stem Cells. Tissue Eng A 2005; 11: 319-330.
11
[14] Morimoto Y, Takeuchi S. Three-dimensional cell culture based on microfluidic techniques to mimic
living tissues. Biomater Sci 2013; DOI: 10.1039/C2BM00117A (advanced article).
[15] Meijer J, Du K, Gillner A, Hoffmann D, Kovalenko VS, Masuzawa T et al. Laser machining by short
and ultrashort pulses, state of the art and new opportunities in the age of the photons. CIRP Ann-
Manuf Techn 2002; 51 (2): 531-550.
[16] Krüger J, Kautek W. Ultrashort pulse laser interaction with dielectrics and polymers. Adv Polym Sci
2004; 168: 247-289.
[17] Leitz KH, Redlingshöfer B, Reg Y, Otto A, Schmidt M. Metal ablation with short and ultrashort laser
pulses. Phys Procedia 2011; 12: 230-238.
[18] Liu Y, Sun S, Singha S, Cho MR, Gordon RJ. 3D femtosecond laser patterning of collagen for
directed cell attachment. Biomaterials 2005; 26: 4597-4605.
[19] Yeong WY, Yu H, Lim KP, Ng KLG, Boey YCF, Subbu VS, et al. Multiscale topological
guidance for cell alignment via direct laser writing on biodegradable polymer. Tissue Eng C:
Methods 2010; 16: 1011–21.
[20] Lee CH, Lim YC, Farson DF, Powell HM, Lannutti J. Vascular Wall Engineering Via
Femtosecond Laser Ablation: Scaffolds with Self-Containing Smooth Muscle Cell Populations.
Ann Biomed Eng 2011; 39: 3031-3041.
[21] Li-Ping Lee B, Jeon H, Wang A, Yan Z, Yu J, Grigoropoulos C, et al. Femtosecond laser
ablation enhances cell infiltration into three-dimensional electrospun scaffolds. Acta Biomater
2012; 8: 2648-2658.
[22] Li H, Wen F, Shan Wong Y, Yin Chiang Boey F, Subbu VS , Tai Leong D, Woei Ng K, Ka Lai
Ng G, Tan LP. Direct laser machining-induced topographic pattern promotes up-regulation of
myogenic markers in human mesenchymal stem cells. Acta Biomater 2012, 8: 531–539.
[23] Li H, Shan Wong Y, Wen F, Woei Ng K, Lai Ng GK, Subbu S, et al. Human Mesenchymal
Stem-Cell Behaviour On Direct Laser Micropatterned Electrospun Scaffolds with Hierarchical
Structures. Macromol Biosci 2013; 13: 299-310.
[24] Schlie S, Fadeeva E, Koroleva A, Chichkov BN. Laser-engineered topography: correlation
between structure dimensions and cell control. J Mater Sci: mater Med 2012; 23: 2813-2819.
[25] Ortiz R, Moreno-Flores S, Quintana I, Vivanco MdM, Sarasua JR, Toca-Herrera JL. Ultrafast
laser microprocessing of medical polymers for cell engineering applications. Mat Sci Eng C
2014; 37: 241-250.
[26] Ortiz R, Quintana I, Etxarri J, Lejardi A, Sarasua JR. Picosecond laser ablation of poly-L-
lactide: Effect of crystallinity on the material response. J Appl Phys 2011; 110: 094902.
12
[27] Arkin WT. Advances in optics and laser research vol. 2. In: Nova Science Publishers. New
York. 2002. pp. 1-34. ISBN 1-59033-398-5.
[28] McBeath R., Pirone D.M., Nelson C.M., Bhadriraju K., Chen C.S. Cell shape, cytoskeletal
tension, and RhoA regulate stem cell lineage commitment. Dev Cell 2004; 6: 483-495.
[29] Fadeeva E, Deiwick A, Chichkov B, Schlie-Wolter S. Impact of laser-structured biomaterial
interfaces on guided cell responses. Interface Focus 2014; 4: 20130048.
[30] Fahsai Kantawong, Karl E.V. Burgess, Kamburapola Jayawardena, Andrew Hart, Mathis O.
Riehle, Richard O. Oreffo, Matthew J. Dalby, Richard Burchmore. Effects of a surface
topography composite with puerariae radix on human STRO-1-positive stem cells. Acta
Biomater 2010; 6: 3694–3703.
[31] Evelyn K.F. Yim, Eric M. Darling, Karina Kulangara, Farshid Guilak, Kam W. Leong.
Nanotopography-induced changes
in
focal adhesions, cytoskeletal organization, and
mechanical properties of human mesenchymal stem cells. Biomaterials 2010; 31: 1299–1306.
[32] Fahsai Kantawong, Richard Burchmore, Chris D.W. Wilkinson, Richard O.C. Oreffo, Matthew
J. Dalby. Differential in-gel electrophoresis (DIGE) analysis of human bone marrow
osteoprogenitor cell contact guidance. Acta Biomater 2009; 5: 1137-1146.
[33] Hamilton DW, Wong KS, Brunette DM. Microfabricated discontinuous-edge surface
topographies influence osteoblast adhesion, migration, cytoskeletal organization, and
proliferation and enhance matrix and mineral deposition in vitro. Calcif Tissue Int 2006; 78:
314–325.
[34] Ali AT, Hochfeld WE, Myburgh R, Pepper MS. Adipocyte and adipogenesis. Eur J Cell Biol
2013; 92:229-236.
[35] Bahceci M, Gokalp D, Bahceci S, Tuzcu A, Atmaca S, Arikan S. The correlation between
adiposity and adiponectin, tumor necrosis factor –a, interleukin-6 and high sensitivity C-
reactive protein. Is adipocyte size associated with inflammation in adults? J Endocrinol Invest
2007; 30: 210-214.
[36] Meyer LK, Ciaraldi TP, Henry RR, Wittgrove AC, Phillips SA. Adipose tissue depot and cell
size dependency of adiponectin synthesis and secretion in human obesity. Adipocyte 2013;
2(4): 217–226.
[37] Carraway KL, Carraway CAC. Membrane-cytoskeleton interaction in animal cells. Biochim
Biophys Acta 1989; 988: 147-171.
[38] Casadei A, Epis R, Ferroni L, Tocco I, Gardin C, Bressan E et al. Review article: Adipose tissue
regeneration: a state of
the art. J Biomed Biotechnol 2012; 2012:462543. doi:
10.1155/2012/462543.
13
[39] Ruiz SA, Chen CS. Emergence of patterned stem cell differentiation within multicellular
structures. Stem Cells 2008; 26(11): 2921-2927.
[40] Peng R, Yao X, Ding J. Effect of cell anisotropy on differentiation of stem cells on
micropatterned surfaces through the controlled single cell adhesion. Biomaterials 2011; 32:
8048-8057.
[41] Goubko CA, Cao X. Patterning multiple cell types in co-cultures: A review. Mater Sci Eng C
2009; 29:1855-1868.
[42] Freshney RI. Biology of Cultured Cells. In: Wiley-Liss, editors. Culture of animal cells: A manual
of basic technique. New York; 2000. p 11-23.
[43] Pereira JNM, Sundback CA, Lang N, Cho WK, Pomerantseva I, Ouyang B, et al. Combined
surface micropatterning and reactive chemistry maximizes tissue adhesion with minimal
inflammation. Adv Healthcare Mat 2013. DOI: 10.1002/adhm.201300264
[44] Mizrahi B, Stefanescu CF, Yang C, Lawlor MW, Ko D, Langer R, et al. Elasticity and safety of
alkoxyethyl cyanoacrylate tissue adhesives. Acta Biomater 2011; 7(8): 3150-3157.
14
Figure captions
Figure 1. Typical PLLA-bottom dish used for these studies. The thermoformed PLLA sheet
(approximately 300 µm thick) is sealed to a Nylon ring with a biocompatible silicone (blue-green,
Picodent Twinsil, Picodent GmbH, Germany).
Figure 2. SEM-obtained images of laser-created topographies: roughened PLLA (a); grooves (10
μm wide, 7 μm deep and 15 μm spacing) (b); and circular and square-shaped microcavities (c-
d).
Figure 3. Fluorescence microscopy images of undifferentiated MSCs cultured on PLLA grooves.
Images a-i show cells on flat PLLA (a-c) and PLLA grooves (d-i) as a function of time. Images (g-
i) show cells on grooves oriented at different directions after one day in culture.
Figure 4. Polar diagrams showing the number of oriented cells as a function of the orientation
angle after 1 (a), 8 (b) and 22 (c) days in culture. Size of dots indicates the error. (d) Population
histogram of oriented cells normalized to total cell number obtained from fluorescence images as
a function of time.
Figure 5. Undifferentiated MSCs between three (a) and two (b) grooved-patterned areas were
visualized by inmunofluorescence after 1 day in culture.
Figure 6. Brightfield (BF) images of MSCs differentiated into adipocytes (presence of lipid
vacuoles in red) and osteoblasts (AP staining in blue) on flat PLLA (a), laser-roughened PLLA
(b,c), and microcavities (d, e). In figures c-e strings of lipid vacuoles are outlined.
Figure 7. BF images of MSCs differentiated into adipocytes (presence of lipid vacuoles in red)
and osteoblasts (AP staining in blue) on PLLA grooves at different directions (0°, 45°, 90°) after
14 days in culture (a-c). (d) Quantification of the proportion of clusters of lipid vacuoles confined
in the grooves and on the intergroove space (*significance level according to the Student t-Test:
p<0.005).
Figure 8. Histogram of the area (represented as percentage) covered by each cell type
(adipocytes, osteoblasts, and undifferentiated MSCs) on control (flat PLLA), rough PLLA, and
PLLA grooves.
15
Figure 9. Monolayers of differentiated MSCs remain adhered only on laser-treated areas: rough
PLLA (a), PLLA grooves (b), or microcavities (c). Micrographs taken after 14 days in culture.
16
Figure 1
(a)
(b)
(d)
17
Figure 2
(a)
(c)
(b)
(d)
18
Figure 3
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
19
Figure 4
(a)
(b)
(c)
(d)
20
Figure 5
(a)
(b)
21
Figure 6
22
Figure 7
(a)
(b)
(c)
23
Figure 8
24
Figure 9
(a)
(b)
(c)
25
|
1301.7022 | 2 | 1301 | 2013-08-24T04:06:53 | Anomalous diffusion of proteins in sheared lipid membranes | [
"physics.bio-ph",
"cond-mat.soft"
] | We use coarse grained molecular dynamics simulations to investigate diffusion properties of sheared lipid membranes with embedded transmembrane proteins. In membranes without proteins, we find normal in-plane diffusion of lipids in all flow conditions. Protein embedded membranes behave quite differently: by imposing a simple shear flow and sliding the monolayers of the membrane over each other, the motion of protein clusters becomes strongly superdiffusive in the shear direction. In such a circumstance, subdiffusion regime is predominant perpendicular to the flow. We show that superdiffusion is a result of accelerated chaotic motions of protein--lipid complexes within the membrane voids, which are generated by hydrophobic mismatch or the transport of lipids by proteins. | physics.bio-ph | physics |
Anomalous diffusion of proteins in sheared lipid membranes
Atefeh Khoshnood1,2 and Mir Abbas Jalali1∗
1Computational Mechanics Laboratory, Department of Mechanical Engineering,
Sharif University of Technology, Azadi Avenue, Tehran, Iran.
2Center of Excellence in Design, Robotics and Automation, Department of Mechanical Engineering,
Sharif University of Technology, Azadi Avenue, Tehran, Iran.
We use coarse grained molecular dynamics simulations to investigate diffusion properties of
sheared lipid membranes with embedded transmembrane proteins. In membranes without proteins,
we find normal in-plane diffusion of lipids in all flow conditions. Protein embedded membranes
behave quite differently: by imposing a simple shear flow and sliding the monolayers of the mem-
brane over each other, the motion of protein clusters becomes strongly superdiffusive in the shear
direction. In such a circumstance, subdiffusion regime is predominant perpendicular to the flow. We
show that superdiffusion is a result of accelerated chaotic motions of protein -- lipid complexes within
the membrane voids, which are generated by hydrophobic mismatch or the transport of lipids by
proteins.
PACS numbers: 87.16.dj, 87.15.Vv, 87.14.ep, 82.20.Wt
I.
INTRODUCTION
Lipid bilayers are the essential parts of any living cell.
They constitute the main body of the cell membrane
while being found in different organelles inside the cell.
The cell membrane hosts collections of proteins and lipid
rafts and it is crowded with a variety of biomolecules.
In such non-homogenous and diverse environments, the
diffusion of protein molecules in lipid bilayers plays a
vital role in different biological processes like cell signal-
ing. The diffusion of lipids and proteins is not a dis-
tinct phenomenon and depends on the environment and
neighboring molecules [1], and even changes from cell
to cell [2]. Transmembrane proteins diffuse as dynamic
complexes with lipids [3, 4], and their interactions with
lipid molecules mediates traffic in cell membranes. Ex-
periments show that the hydrophobic mismatch between
proteins and lipids controls the diffusion coefficient of
molecules inside a bilayer [5, 6].
Anomalous sub- and super-diffusion processes are more
efficient scenarios for finding a nearby target than nor-
mal diffusion [7, 8], and they enhance the formation of
protein complexes and signal propagation. According
to experiments, the mean square displacement (MSD)
of membrane channel proteins of human kidney cell ex-
hibits subdiffusion [9]. The addition of cholesterols to
lipid membranes [10], and the augmented area coverage
of membrane proteins [11] also lead to subdiffusion of
lipids and proteins. Superdiffusivity has been observed
in several physical systems and is often associated with
L´evy flights. Prominent examples are the chaotic mo-
tion of particles in a rotating laminar flow [12] with long-
range flights and horizontally vibrated grains which ex-
hibit L´evy flight with small jumps compared to their di-
ameter [13]. Nevertheless, an active component such as
∗ [email protected]
molecular motor can also be the source of superdiffusiv-
ity. A recent study by Kohler et al. [14] shows that in
a gel composed of actin filaments, fascin molecules and
myosin-II filaments, the diffusion of small actin and fascin
clusters are superdiffusive because of the work done by
molecular motors.
In many conditions membranes are under shear. When
a red blood cell (RBC) migrates through vessels smaller
in diameter than itself, the RBC membrane is under
shear. The blood flow exerts tangential shear stresses on
vascular endothelia, and initiates cellular processes like
activating G protein-coupled receptors. These receptors
are able to sense the fluid shear stress as an increase in the
lateral membrane tension, and subsequently go through
conformational changes [15]. The temporal and spatial
changes in the membrane fluidity, in response to shear
flow, have been observed experimentally [16, 17].
In this study we are interested in the diffusivity of
lipid and protein molecules in flat membranes under
shear flow, and attempt to answer three fundamental
questions using molecular dynamics (MD) simulations:
(i) Do lipid molecules have different diffusion coefficients
parallel and perpendicular to flow direction? (ii) How
does a simple shear flow influence the random motions
of transmembrane proteins? (iii) Is there any correlation
between the population of proteins and their diffusion in
the membrane?
II. MODEL AND METHODS
[18] and triple-strand rigid proteins [19]
We simulate lipid membranes utilizing a flexible lipid
model
[Fig.
1(a)]. Although different coarse grained models have
been developed over years [20], the model adopted here
has the ability to mimic the physical properties of lipid
membranes. The model is not a true coarse grained
model but can qualitatively describe phenomena related
to lipid membranes and associated transmembrane pro-
teins as we will compare some of our results with true
coarse grained and atomistic models. Our goal is to ex-
plore the effect of shear flow on the motion of lipid and
protein molecules over nanosecond time scales. We per-
form MD simulations of an N V T ensemble, where the
number of particles N , the volume V and the tempera-
ture T are held constant. Lees-Edwards boundary con-
dition is employed to generate simple shear flow with the
shear rate γ [21]. Other boundary conditions are peri-
odic. The temperature is set to 324 K so that the system
is safely above the gel to liquid phase transition temper-
ature of different phosphatidylcholine lipid bilayers. A
detailed description of the model can be found in Khosh-
nood et al. [19].
We express the position and velocity vectors of par-
ticles in the Cartesian (x, y, z) coordinate system whose
origin is located at the center of our cubic simulation
box. The x and y axes lie in the membrane plane and
the z axis is perpendicular to that. MD scales of length,
time, mass and energy are σ = 1/3 nm, τ = 1.4 ps,
Navom = 36 g/mol and Navoǫ = 2 kJ/mol, respectively.
Navo is the Avogadro's number. In all simulations, the
dimensions of the box along the coordinates axes, Lx, Ly
and Lz, are set to Lx = Ly = Lz = 28.71σ. The total
number of particles equals N = 15625, which gives a fixed
number density, ρ = 2/(3σ3), and a constant average
fluid pressure of (1.7 ± 0.1)ǫσ−3 for all simulations. Here,
we note that isotropic pressure control is not appropriate
in the simulation of lipid membranes since their volume
is constant in the laboratory and biological conditions.
However, a fixed number density will give a constant av-
erage pressure and physical properties of lipid membranes
with different number of lipids and proteins can be com-
pared. Physical quantities are measured using a run-time
of ≈ 5000τ . An important mechanical property of every
membrane is the surface tension ζ that mainly affects the
diffusion of lipid molecules. We compute the tension of
our model membranes from
ζ = [Pzz − (Pxx + Pyy) /2] Lz
(1)
where Pαα (α ≡ x, y, z) are the components of the pres-
sure tensor [21].
In this study we apply MSD to determine the diffusion
properties of randomly moving particles. The diffusion
coefficient is thus calculated using the Einstein expression
Dαα = lim
t→∞
1
2N t
N
h
Xi=1
[qiα(t) − qiα(0)]2i,
(2)
where α ∈ {x, y, z} and qiα is the displacement due to the
random motion of the ith particle in the α-direction. The
summation in Eq. (2) is taken over the particles of the
same type. From here on, we will drop the summation
sign for brevity. The operator h· · · i denotes the canonical
average. Eq. (2) describes the regular Brownian motion
when the MSD is linearly proportional to t. The diffusion
2
Solvent
Lipid Membrane
Protein
0.4
0.2
z
L
/
z
0
-0.2
-0.4
Lipid
(a)
-0.4
-0.2
Solvent
0
vx
(b)
0.2
0.4
FIG. 1.
(Color online)(a) The models of lipid and protein
molecules. Red and yellow spheres are hydrophilic and hy-
drophobic particles, respectively. (b) Velocity profile of a sam-
ple solvent -- membrane system under simple shear flow with
γ = 0.03τ −1. The dashed lines mark the average boundaries
of the solvent columns and the membrane.
process is anomalous should the MSD deviate from the
linear form, and obey the relation
h[qα(t) − qα(0)]2i = 2Da
ααta,
(3)
where a is the diffusion exponent and Da
αα is the frac-
tional diffusion coefficient. The regimes with 0 < a < 1
and a > 1 are subdiffusive and superdiffusive, respec-
tively. To obtain smooth MSD curves, we evolve sys-
tems of 80 different initial conditions and report their
ensemble-averaged diffusion coefficients and MSDs. The
diffusion of lipids is investigated by tracing the motion of
their head groups. Proteins are traced using their center
of masses.
In equilibrium models without external shearing,
there is no streaming in the solvent -- membrane system.
Therefore, the flux of particles is associated with their
random motions, and any displacement is due to thermal
fluctuations. In sheared membranes, however, there is a
combination of streaming and diffusive fluxes. We thus
need to distinguish and eliminate the streaming flux
when calculating the MSD. Let us define the actual ve-
locity components of the jth particle as vjα = hvαi + vjα
where hvαi is the average streaming velocity, and vjα
is the peculiar velocity whose time integral gives the
displacements in Eqs.
In equilibrium
models, the average velocities hvαi vanish and we obtain
(2) and (3).
qjα = R vjαdt = R vjαdt. With external shearing, the
flow is always imposed in the x-direction. Therefore,
vjy = vjy and vjz = vjz are directly integrated to find
the corresponding displacements. When the simulation
box is uniformly filled with one type of particles (let us
says solvent particles), one readily finds vjx = vjx − z γ.
In the presence of a lipid bilayer, the vertical velocity
profile in the z-direction is no longer linear [see Fig.
1(b)]. Therefore, to obtain the MSD of
lipids, we
define hvxi as the average velocity of the layer where
the head groups of phospholipds reside, and obtain
qjx = R (vjx − hvxi) dt. It is remarked that transmem-
brane proteins do not experience streaming movements,
hvαi = 0, because the shear forces exerted on their end
points from the upper and lower solvent columns are
equal and in opposite directions.
III. DIFFUSION OF LIPIDS
The lateral diffusion of lipids in equilibrium conditions
is enhanced as the membrane tension increases. This
has been observed in simulations by atomistic [23] and
coarse grained models [22] and is because by stretch-
ing the membrane, the area per lipid increases and more
space is provided for the free motion of lipids. Our sim-
ulations with this very simple model shows the same
pattern. By turning on the shear flow, lipid molecules
undergo an initial ballistic motion that transforms into
an interval of subdiffusion with a = 0.7 [Fig. 2]. The
transient anomalous state has been observed in atomistic
simulations [24] as well. After the transient anomalous
diffusion and over longer time scales a normal diffusion
with a = 1 is observed [Fig. 2]. It is noted that we have
found similar MSD profiles for lipid molecules in equilib-
rium and sheared systems, and in both cases lipids ulti-
mately develop normal diffusion. Although Kneller et al.
[25] reported a permanent subdiffusive behavior by the
atomistic simulation of lipid membranes in equilibrium,
experiments support a final regular diffusion regime, as
we do, even in the presence of obstacles [26]. We conclude
that the diffusion regime of lipid molecules is invariant
with and without external shearing.
The diffusion coefficients obtained from the normal
diffusion region of MSD plots, are larger for smaller
shear rates. For example, for a membrane of Nl = 600
lipid molecules, we find ζ = (1.4846 ± 0.2624)ǫ/σ2,
Dxx = 0.0338σ2/τ and Dyy = 0.0334σ2/τ . For the same
γ = 0.03τ −1, the mem-
system under a shear flow of
brane tension drops to ζ = (0.8163 ± 0.2727)ǫ/σ2 and
diffusion coefficients reduce to Dxx = 0.0318σ2/τ and
Dyy = 0.0332σ2/τ . The reason is that the membrane
thickness increases for higher shear rates and the ten-
sion decreases without any change in the area per lipid
[19]. Consequently, the fluidity of the membrane de-
creases and slows down the diffusion process. After ap-
plying the shear force, we find that Dxx drops for about
6% while Dyy remains almost constant with only 0.4%
change which is not statistically significant since it is less
than the mean standard error for diffusion coefficients
which is less than 1%. The difference between Dxx and
Dyy is indistinguishable in Fig. 2. We speculate that
the alignment of lipid chains with the flow breaks the
isotropy and yields Dxx 6= Dyy. Atomistic simulation of
lipid membrane [23] has shown that increasing the ten-
sion of membrane, induced by altering the area per lipid,
result in larger lateral diffusion coefficients. This change
is not linear with tension and they have reported 4 − 28%
increase in lateral diffusion coefficient. Coarse grained
simulations [22] showed that for larger tensions the in-
3
FIG. 2. (Color online) MSD of lipid molecules for the mem-
brane with 600 lipids and for ∆t = 0.01τ . The membrane
is under simple shear flow with γ = 0.03τ −1, and each lipid
molecule has been hydrated by almost 20 solvent particles.
The coordinate axes are in logarithmic scale.
crease in diffusion coefficient slows down and it depends
on the range of tension. In our simulation, shear flow in-
duces 45% change in tension and consequently result in
a different diffusion coefficient in the flow direction. In
the z-direction, perpendicular to the membrane plane,
our MSD plots always show a confined motion as is ex-
pected.
IV. DIFFUSION OF PROTEINS
We add rod-like proteins to the membrane, and sim-
ulate models with different protein concentrations that
vary significantly from cell to cell. Since proteins in-
crease the membrane tension as they perturb the distri-
bution of lipids [19], we increase the number of lipids
(proportional to proteins) to keep the membrane almost
tensionless. In equilibrium and for a membrane with a
single embedded protein with ζ = (0.1153 ± 0.1742)ǫ/σ2,
we can measure the diffusion coefficients Dαα since the
MSD of protein shows an ultimate regular diffusion. We
find Dxx = 0.0253σ2/τ and Dyy = 0.0254σ2/τ , which
are equivalent to Dxx ≈ Dyy ≈ 2 × 10−9 m2/s with less
than 1% error. These values are larger than experimental
values, by two orders of magnitude. The obvious reason
is the effect of coarse graining that has reduced the in-
terdigitation and friction between molecules, and allows
for faster movements of particles. The reduced friction
affects both membrane component and solvent motion
and consequently decreases both solvent and membrane
viscosity. Viscosity of a fluid is a determinant of mobil-
ity or diffusion in that medium. Another minor source
of discrepancy is the smaller size of our model proteins
compared to real integral proteins.
4
FIG. 3. (Color online) (a) MSD of protein molecules in a protein-embedded model with 2 and (b) 4 proteins, which correspond
to 640 and 660 lipids, respectively. We have used ∆t = 0.0005τ up to t = 0.02τ , and ∆t = 0.01τ for t > 0.02τ . In both panels
(a) and (b) the membrane is under simple shear flow with γ = 0.03τ −1, and each lipid molecule has been hydrated by almost
20 solvent particles. The coordinate axes are in logarithmic scale. (c) Zoomed upper parts of the MSD profiles with the green
(light gray) and black lines corresponding to the systems of panels (a) and (b), respectively. The profiles with square symbols
correspond to the y-direction.
We note that the average diffusive behavior of lipids
is unaffected by the presence of proteins. We computed
the average MSD profile of lipids for the above system
and found the same pattern of initial ballistic regime,
transient subdiffusive region and final normal diffusion.
The diffusion coefficients for our model proteins is smaller
than model lipid molecules by a factor of 0.75. This is
expected because of the larger size and mass of proteins
compared to lipids.
An initial ballistic motion the same as what has been
observed for lipids is recovered for proteins by using time
step as small as ∆t = 0.0005τ for t < 0.01τ . This regime
is shared by all the systems regardless of the shear rate
and the number of proteins. By putting the system un-
der simple shear flow, proteins undergo Brownian motion
when only two proteins are used [Fig. 3(a)]. One could
anticipate this result, for single proteins cannot remark-
ably perturb the distribution of lipids, and change the
diffusion properties of the membrane.
With 4 proteins, however, we observe that they form
two double-protein clusters (due to the depletion force),
and exhibit a strong superdiffusive motion parallel to the
flow. Fig. 3(b) shows how after t ∼ 100τ the normal dif-
fusion regime transforms to strong superdiffusion with
a = 1.7 in the x-direction. Interestingly, this exponent is
the same as the superdiffusion exponent found by Kohler
et al. [14] for active diffusion of protein clusters by molec-
ular motors. Because of crowding effect and increase in
the concentration of proteins [1], our results show a subd-
iffusive behavior along the y axis with a = 0.7. Weigel et
al. [9] observed a = 0.8 ± 0.1 in experiments with channel
proteins of human kidney cell, and Javanainen et al. [27]
found a = 0.75 ± 0.15 by molecular simulations of ag-
gregating NaK channel proteins. For clarity, the upper
parts of the MSD profiles in Figures 3(a) and 3(b) are
plotted together in Fig. 3(c).
a
2
1.8
1.6
1.4
1.2
1
0.8
0.6
x-direction
y-direction
10-2
10-1
100
101
t
102
103
104
FIG. 4. The variation of the diffusion exponent a for the
system of Fig. 3(b). Solid lines correspond to the mean values
and dashed lines show the error. For the y-direction, a has
been plotted only for t > 30τ because for t < 30τ diffusion
exponents of the x- and y-directions are almost identical. The
horizontal coordinate axis is in logarithmic scale.
To rule out the effect of statistical errors in the de-
velopment of anomalous behavior, we have divided the
MSD profiles of the system with 4 proteins to smaller
intervals, and separately calculated a and its error over
each interval using the curve fitting toolbox of MATLAB.
We have then assigned the mean value to the center of
the time interval and plotted the calculated diffusion ex-
ponent versus time in Fig. 4. For instance, over the
initial ballistic zone, we have obtained a = 1.989 ± 0.006
from t = 0.005τ to t = 0.01τ , and assigned this value to
t = 0.0075τ . Fig. 4 shows that a approaches 1.7 and 0.7
in the x- and y-directions, respectively.
To understand the physical mechanism behind the
observed anomaly, we conduct the following analysis.
Nh PNh
Let us define the local concentration of the head par-
ticles of lipid and protein molecules at the position r
and time t as f = 1
i=1 [H(δi) − H(δi − ∆)] where
δi(t) = ri(t) − r, and ri(t) is the position vector of
the ith head particle. Nh denotes the total number of
head particles in the monolayer, H(ξ) is the Heaviside
step function, and 2∆ is the typical size of the cross sec-
tion of a protein cluster (or a protein -- lipid complex).
Our numerical experiments show ∆ = 4σ is the best
choice. We examine the trajectories of protein molecules
and the spatial variation of the normalized distribution
f (r, ∆, t) = (f − fmin)/(fmax − fmin) to explain the
physics of observed superdiffusion. Here fmin and fmax
are the minimum and maximum values of f at a given
time t. Fig. 5 demonstrates contour plots of f for the
upper and lower monolayers at a randomly chosen time.
Fig. 6(a) demonstrates the trajectories of a single pro-
tein molecule and two double-protein clusters, in equi-
librium and sheared systems, respectively. The equilib-
rium trajectory corresponds to regular diffusion because
it covers a definite area. In the sheared system the tra-
jectories are elongated and aligned with the flow direc-
tion, indicating a fractional random walk: local isotropic
wanderings followed by small-step jumps mainly in the
flow direction. These successive jumps can be interpreted
by inspecting the contour plots of f (r, ∆, t) over a long
duration of time. The hydrophobic mismatch between
protein clusters (which have asymmetric big cross sec-
tions) and the membrane disturbs the bilayer thickness
and the arrangement of nearby lipids. Moreover, pro-
teins are able to transport their neighboring lipids with
them and behave as dynamic complexes [3, 4]. These
two effects collaborate to create transient voids whose
distribution can be described by g = 1 − f (light shades
in Fig. 5). When the bilayer is sheared, protein -- lipid
groups are pushed into the voids created by themselves
or other groups/complexes and experience accelerated,
and therefore, superdiffusive movements.
It should be
noted that during our simulations, the center of mass of
the membrane and embedded proteins remains almost at
the center of the coordinate system.
None of our samples show long-step straight motions
of protein clusters. What we have seen are small-step
jumps, which are comparable with the mean distance be-
tween protein clusters and voids [compare Figs. 5 and
6(a)]. We have computed the probability distribution
function (PDF) of protein displacements and plotted it
in Fig. 6(b). A Gaussian function has been fitted to the
data by setting its maximum to the maximum of PDF,
and its variance is found using the full width at half mini-
mum of PDF. The PDF exhibits a deviation from normal
distribution and it has tails. We have also applied the
Kolmogorov-Smirnov test [28] to confirm that the PDF
is not a normal Gaussian. This is a clear indication of
anomalous diffusion.
As we noted before, the ends of proteins are pulled in
opposite directions by the two sheared solvent columns.
An important question is why do protein clusters prefer
Upper monolayer
^
f
0
0.25
0.5
0.75
1
5
-10
-5
Lower monolayer
0
x/σ
^
f
5
10
0
0.25
0.5
0.75
1
10
5
σ
/
y
0
-5
-10
10
5
σ
/
y
0
-5
-10
-10
-5
0
x/σ
5
10
FIG. 5. (Color online) The local concentration f (r, ∆, t) of
head groups in the upper (left panel) and lower (right panel)
leaflets of a sheared membrane at a randomly selected time
for ∆ = 4σ. The head particles of proteins are shown by filled
circles. Drawn circles have radii of ∆, and their centers lie at
the centroid of the head particles of proteins. The flow with
the shear rate γ = 0.03τ −1 is in the x+ and x− directions
for upper and lower leaflets, respectively [cf. Fig. 1(b)]. The
model has 660 lipids and two double-protein clusters. Since
proteins are longer than the bilayer thickness, they are aligned
in the shear flow, and their upper and lower head particles do
not have the same coordinates.
to jump into the voids when the membrane is sheared?
We have a simple explanation for this behavior: Two
ends of proteins attract lipids from two different layers of
the membrane. Thus, the symmetry in the distribution
of the upper and lower protein-bound lipids is likely to
break. Moreover, the shear force is exerted by the solvent
on both the lipid and protein heads. The mentioned sym-
metry breaking thus leads to different force components
at the upper and lower ends of protein -- lipid complexes,
6
∆t=8 τ
∆t=16 τ
(b)
10-1
10-2
F
D
P
10-3
10-4
-1
-0.5
0
qx(t+∆t)-qx(t)
0.5
1
tc
: correlation time
(c)
0.26
0.25
0.24
C
0.23
(d)
0
100
200
300
tlag / τ
400
500
0.22
0
20
40
tlag / τ
60
80
1
0.8
0.6
0.4
A
0.2
0
-0.2
-0.4
FIG. 6. (Color online) (a) Trajectories of a single protein in equilibrium and two double-protein clusters under shear flow as
pointed by arrows. (b) PDF for the displacements of a sample two-protein cluster. Solid line shows the best Gaussian function
fitted to the data. (c) Autocorrelation function A(tlag) for the difference between the head particle populations near the two
ends of a protein cluster. The correlation time tc ≈ 23τ is defined at the point where A(tlag) abruptly drops below 10% of
its maximum. We have taken 1000 successive samples in the time domain, with increments of 1τ , to compute A. (d) Cross-
correlation function C(tlag) between the distribution of protein -- lipid complexes and their neighboring voids. The integrals in
C have been taken using a grid of 29 × 29 in the xy-plane and 1000 successive points, with steps of 1τ , in the time domain.
and they will jump into a void in the direction specified
by the broken symmetry. To quantify this process, we
take a sample two-protein cluster and define ru and rl
as the center of masses of its protein heads in the upper
and lower leaflets, respectively. We then compute
ξ(ti) = f (ru, ∆, ti) − f (rl, ∆, ti),
(4)
which is proportional to the net shear force exerted on
the cluster at the time step ti: the local effective area
in contact with a solvent column is determined by the
number of head particles, and the shear force is calculated
by multiplying the effective contact area by the shear rate
and viscosity. ξ(ti) will be zero if the concentrations of
the head particles of lipids are identical around the two
heads of the cluster. Defining ¯ξ as the average of ξ(ti),
the autocorrelation function
A(tlag) = Pi(cid:2)ξ(ti + tlag) − ¯ξ(cid:3) ·(cid:2)ξ(ti) − ¯ξ(cid:3)
,
(5)
Pi(cid:2)ξ(ti) − ¯ξ(cid:3)2
plotted in Fig. 6(c) carries interesting information about
the shear force experienced by the cluster: the correla-
tion time tc ≈ 23τ shows a sustained accelerated motion
of the cluster over t0 < t . t0 + tc, independent of the
initial time t0. Moreover, the oscillatory decaying pro-
file of A(tlag) shows a random symmetry breaking in the
sense of deterministic chaos [29, §5.3.4]. The existence
of a correlation time can also be verified by studying the
cross-correlation function
C(tlag) = Z dtZ dr f (r, ∆, t + tlag) · g(r, ∆, t)
(6)
over one of the leaflets. Fig. 4(d) shows that C(tlag)
of the upper leaflet steeply rises from tlag = 0 and
peaks at tlag ≈ 21τ , which is the earliest time span that
protein -- lipid complexes need to occupy their nearest
voids. This is quite consistent with the acceleration time
scale of protein clusters predicted by the autocorrelation
function A(tlag). For tlag & 21τ the cross-correlation
function remains almost flat because the size of a void
is always bigger than the distance that a protein cluster
travels during t ∼ O(tc).
V. CONCLUSIONS
In this work, we use a toy model of proteins and lipids
to simulate the dynamics of cell membranes undergoing
shear flow. We calculate the MSD profile of lipids and
proteins in equilibrium and sheared system and compare
our result with existing works in the literature. This sim-
ple model beautifully captures the basic regimes in the
diffusive behavior of lipid molecules: short initial ballis-
tic regime, transient subdiffusive regime and final Normal
diffusion. All-atom MD simulations show the same dif-
fusive regions in the MSD profile of lipid molecule [24].
Moreover, we show that shear flow reduces the tension of
membrane and consequently decreases the diffusion co-
efficient in the direction of flow. The fact that reducing
membrane tension slows down lipids movement has been
reported as the result of all-atom MD simulations [23]
and experimental measurements [16].
Since there were small number of protein molecules
in the system, we did extensive MD simulations to ob-
tain the final MSD profiles for proteins. Our C++ code,
which has already been calibrated [19] for membranes
under simple shear flow, is not parallel and allows only
for small-scale simulations. We have recovered our results
by LAMMPS for models in equilibrium conditions. How-
ever, LAMMPS has not the capability of imposing simple
shear flow conditions and we could not use it to run our
7
sheared systems over longer time scales. Despite these
limitations, the smooth MSD profiles obtained from our
numerous samples clearly show the distinction between
the regular diffusion of single proteins and anomalous
diffusion of protein clusters. We explain this anomaly by
deliberately examine the distribution of head particles
of lipids and proteins and introducing void generation
mechanism. Our simulations cover timescales of order of
nanoseconds.
Cell responses to stimuli are fast due to enhanced
mobility of protein receptors. Consider our findings,
superdiffusion of proteins under shear flow can play a
dominant role in the process of signaling in endothelium
cells, RBCs, liposomes used for targeted drug delivery
and other sheared membranes. We note that since
the length and shape of proteins and their ability in
attracting neighboring lipids controls the sizes of voids
-- there was no other mechanism in our models for void
generation -- superdiffusion properties reported in this
study are not universal and they highly correlate with
the properties of embedded proteins. Simulations using
detailed structures of lipids and proteins are needed to
better assess the superdiffusive behavior in realistic cell
membranes. Experimental exploration of our results can
be done using single particle tracking [30] which can give
the MSD of proteins directly.
ACKNOWLEDGMENTS
We thank anonymous referees for their insightful com-
ments that helped us to improve the presentation of our
results. A.K. was partially supported by the National
Elites Foundation of Iran. We express our sincere thanks
to Prof. Ali Meghdari for his helpful discussions and en-
couragements.
[1] J. A. Dix, and A. S. Verkman, Annu. Rev. Biophys., 37,
[9] A. V. Weigel, B. Simon, M. M. Tamkun, and D. Krapf,
247 (2008).
Proc. Natl. Acad. Sci., 108, 6438 (2011).
[2] S. Wieser, J. Weghuber, M. Sams, H. Stockinger, and
[10] J.-H. Jeon, H. M.-S. Monne, M. Javanainen, and R. Met-
G. J. Schuetz, Soft Matter, 5, 3287 (2009).
zler, Phys. Rev. Lett., 109, 188103 (2012).
[3] P. S. Niemela, M. S. Miettinen, L. Monticelli, H. Ham-
maren, P. Bjelkmar, T. Murtola, E. Lindahl, and I. Vat-
tulainen, J. Am. Chem. Soc., 132, 7574 (2010).
[11] S. Ramadurai, A. Holt, V. Krasnikov, G. van den Bo-
gaart, J. A. Killian, and B. Poolman, J. Am. Chem. Soc.,
131, 12650 (2009).
[4] A. Prasad, J. Kondev, and H. A. Stone, Phys. Fluids,
[12] T. H. Solomon, E. R. Weeks, and H. L. Swinney, Phys.
19, 113103 (2007).
[5] S. Ramadurai, A. Holt, L. V. Schafer, V. V. Krasnikov,
D. T. S. Rijkers, S. J. Marrink, J. A. Killian, and B. Pool-
man, Biophys. J., 99, 1447 (2010).
[6] Y. Gambin, M. Reffay, E. Sierecki, F. Homble, R. S.
Hodges, N. S. Gov, N. Taulier, and W. Urbach, J. Phys.
Chem. B, 114, 3559 (2010).
[7] G. Guigas and M. Weiss, Biophys. J., 94, 90 (2008).
[8] F. Bartumeus, J. Catalan, U. L. Fulco, M. L. Lyra, and
G. M. Viswanathan, Phys. Rev. Lett., 88, 097901 (2002).
Rev. Lett., 71, 3975 (1993).
[13] F. Lechenault, R. Candelier, O. Dauchot, J. P. Bouchaud,
and G. Biroli, Soft Matter, 6, 3059 (2010).
[14] S. Kohler, V. Schaller, and A. R. Bausch, Nat. Mater.,
10, 462 (2011).
[15] M. Chachisvilis, Y. Zhang, and J. A. Frangos, Proc. Natl.
Acad. Sci., 103, 15463 (2006).
[16] P. J. Butler, G. Norwich, S. Weinbaum, and S. Chien,
Am. J. Physiol. Cell Physiol., 280, C962 (2001).
8
[17] M. A. Haidekker, N. L'Heureux, and J. A. Frangos, Am.
[25] G. R. Kneller, K. Baczynski, and M. Pasenkiewicz-
J. Physiol. Heart Circ. Physiol., 278, H1401 (2000).
Gierula, J. Chem. Phys., 135, 141105 (2011).
[18] R. Goetz and R. Lipowsky, J. Chem. Phys., 108, 7397
[26] M. J. Skaug, R. Faller, and M. L. Longo, J. Chem. Phys.,
(1998).
134, 215101 (2011).
[19] A. Khoshnood, H. Noguchi, and G. Gompper, J. Chem.
Phys., 132, 025101 (2010).
[20] S. J. Marrink, H. J. Risselada, S. Yefimov, D. P Tieleman,
and A. H. de Vries, J. Phys. Chem. B, 111, 7812 (2007).
[21] M. P. Allen, and D. J. Tildesley, Computer Simulation of
Liquids, (Oxford University Press, Oxford, UK., 1991).
[22] J. Neder, B. West, P. Nielaba, and F. Schmid, J. Chem.
Phys., 132, 115101 (2010).
[23] H. S. Muddana, R. R. Gullapalli, E. Manias, and P. J.
Butler, Phys. Chem. Chem. Phys., 13, 1368 (2011).
[27] M. Javanainen, H. Hammaren, L. Monticelli, J.-H. Jeon,
M. S. Miettinen, H. Martinez-Seara, R. Metzler, and Ilpo
Vattulainen, Faraday Discuss., (2013).
[28] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B.
P. Flannery, Numerical Recipes: The Art of Scientific
Computing, 3rd Edition, (Cambridge University Press,
2007)
[29] J. H. Argyris, G. Faust, and M. Haase, An exploration
of chaos: an introduction for natural scientists and engi-
neers, (North-Holland, 1994)
[24] E. Flenner, J. Das, M. C. Rheinstadter, and I. Kosztin,
[30] J. M. Crane, and A. S. Verkman, Biophys. J., 94, 702
Phys. Rev. E, 79, 011907 (2009).
(2008).
|
1112.6190 | 1 | 1112 | 2011-12-28T22:15:52 | Toward a Theory on the Stability of Protein Folding: Challenges for Folding Models | [
"physics.bio-ph",
"math-ph",
"math-ph",
"physics.chem-ph"
] | We adopt the point of view that analysis of the stability of the protein folding process is central to understanding the underlying physics of folding. Stability of the folding process means that many perturbations do not disrupt the progress from the random coil to the native state. In this paper we explore the stability of folding using established methods from physics and mathematics. Our result is a preliminary theory of the physics of folding. We suggest some tests of these ideas using folding simulations.
We begin by supposing that folding events are related in some way to mechanical waves on the molecule. We adopt an analytical approach to the physics which was pioneered by M.V. Berry, (in another context), based upon mathematics developed mainly by R. Thom and V.I. Arnold. We find that the stability of the folding process can be understood in terms of structures known as caustics, which occur in many kinds of wave phenomena.
The picture that emerges is that natural selection has given us a set of protein molecules which have mechanical waves that propagate according to several mathematically specific restrictions. Successful simulations of folding can be used to test and constrain these wave motions. With some additional assumptions the theory explains or is consistent with a number of experimental facts about folding. We emphasize that this wave-based approach is fundamentally different from energy-based approaches. | physics.bio-ph | physics | Toward a Theory on the Stability of Protein Folding:
Challenges for Folding Models
Walter Simmons
Department of Physics and Astronomy
University of Hawaii at Manoa
Honolulu, HI 96822
Joel L. Weiner
Department of Mathematics
University of Hawaii at Manoa
Honolulu, HI 96822
1
Abstract
We adopt the point of view that analysis of the stability of the protein folding
process is central to understanding the underlying physics of folding.
Stability of the folding process means that many perturbations do not
disrupt the progress from the random coil to the native state. In this paper
we explore the stability of folding using established methods from physics
and mathematics. Our result is a preliminary theory of the physics of
folding. We suggest some tests of these ideas using folding simulations.
We begin by supposing that folding events are related in some way to
mechanical waves on the molecule. We adopt an analytical approach to
the physics which was pioneered by M.V. Berry, (in another context), based
upon mathematics developed mainly by R. Thom and V.I. Arnold. We find
that the stability of the folding process can be understood in terms of
structures known as caustics, which occur in many kinds of wave
phenomena.
The picture that emerges is that natural selection has given us a set of
protein molecules which have mechanical waves that propagate according
to several mathematically specific restrictions.
Successful simulations of folding can be used to test and constrain these
wave motions.
With some additional assumptions the theory explains or is consistent with
a number of experimental facts about folding.
We emphasize that this wave-based approach is fundamentally different
from energy-based approaches.
2
Contents
Introduction
Background Biology
Properties of Stable Caustics
The Analogy
State Variables and Control Parameters in Wave Motion
Toward a New Theory
Challenges for Models
Conclusions
Technical Sections:
The Mathematics Literature
Theory of Caustics
3
Introduction
Rogue ocean waves that appear randomly and sometimes sink ships,
bands of light seen on the bottom of a pool or stream on a sunny day,
certain seismic events, and some recently discovered electronic effects in
graphene, are highly diverse phenomena in the purview of the fields of fluid
dynamics, optics, acoustics, and quantum mechanics but all have many
aspects in common. (1) (2) (3) (4) (5) (6)
These phenomena are forms of caustics, which arise in wave propagation,
wherein the waves pile up to very high amplitude, usually along a curve.
Caustics, in contrast to other wave forms, arise only when certain specific
conditions apply to the phases of the waves. (7) (8) (9) (10)
Broadly speaking, caustics can be divided into two categories: structurally
stable versus unstable caustics. By structural stability, we mean that the
general shape of the caustic is not sensitive to small details of the
propagation process. Unstable caustics change form radically when the
details of formation are modified even slightly. An example of an unstable
caustic is a point image on a screen formed using a thin lens; any small
defect in the lens blurs the image from its ideal point shape. Examples of
stable caustics are the parallel bands of light that sometimes appear and
persist on the bottom of a swimming pool on a sunny day; their form is
relatively independent of the details of the surface motion of the water.
The study of the motion of waves is usually developed in terms of
equations of motion and their solutions. In many cases this analysis
approach is not very practical. Another approach involves treating waves
as geometrical objects of study and identifying special features that serve
to define the geometry of the waves. This approach is particularly well
suited to the study of the stability of processes involving waves.
Following upon well-known developments in mathematics by R. Thom, V.I.
Arnold, E.C. Zeeman, and others, M.V. Berry pioneered the application of
this geometric approach to stable optical caustics; this is a quantitative
4
theory. These bright curves and surfaces can be quite complicated but
they are formed from a small set of stable, geometrically distinct shapes
and the theory and experiment are in excellent, wide-ranging, agreement.
In a typical experimental arrangement, light from a single source is
scattered in all directions from some set of lenses and/or mirrors. A caustic
is a very bright curve that appears on a nearby screen. If the caustic is
structurally stable, then minor adjustments to the shapes of the lenses and
mirrors do not significantly change the shape of the curve on the screen.
As we have said, these structurally stable regions of high amplitude occur
in optics, fluid dynamics, and acoustics, as well as other fields and, in some
cases, quantitative theories have been developed; in other cases,
especially outside of the physical sciences and engineering, qualitative
theories have been presented by many authors; unfortunately, some of the
qualitative theories have attracted criticism.
Motivated by the appearance of waves of variable wave-speed in an
analytic model of protein folding, we suggested (11) that a quantitative
theory of some parts of the folding process might be developed using the
mathematical structure that underlies the physics of stable optical caustics.
Besides important empirical observations about folding, a key reason for
pursuing this approach is that the mathematics and physics places very
strong restrictions upon wave behavior at caustics. If this same
mathematics and physics is involved in folding, then it will dominate certain
folding processes and is deterministic.
Rogue water waves make a particularly interesting parallel with folding.
These large waves appear suddenly and in random places. They arise
because the wave-speed of gravity waves is frequency dependent in deep
water. The faster waves can catch up with the slower waves leading to a
sudden caustic. In a protein, we conjecture, features defined by the amino
acid sequence pick out a region of the molecule for large amplitude motion.
For example, when viewed from a stability perspective, mutations of a
stable protein have either no impact upon stability, change the protein to
5
another stable form, or make the protein unstable (and hence probably not
present in the data bases of naturally occurring proteins). The mathematics
and physics of these kinds of transitions are available to characterize both
stability and structural change.
In this paper, we re-introduce the ideas presented earlier and amplify upon
them in detail. Our intended readership is scientists who model proteins
and those who study mutations and protein structure. We will suggest
some ways that the theoretical ideas discussed here might be explored in
simulations.
6
Background Biology
The process of protein manufacture in simple cells accounts for about 75%
of the cell’s power utilization. A typical simple cell manufactures several
thousand proteins (12). The process of translating the information encoded
in the DNA into folded proteins represents the principle flow of genetic
information in simple cells and is essential in all cellular life.
The proteins are linear chains of amino acids, (twenty in number), which
are formed and extruded into the intra-cellular fluid where they fold.
The work of Anfinsen showed that unfolded proteins of length up to about
350 residues fold spontaneously in water. This demonstrates that (1) the
information needed to describe the fold is in the sequence, (2) the
mechanism of folding is present in the molecule (plus water) itself, and (3)
the details of the initial condition of the random coil are not determinant of
the final form of the native state.
The random coil and the folded protein are often separated
thermodynamically by only a small difference in the Gibbs potential. Within
this short energy separation lies a phase transition. Aside from the energy
needed to build the molecule and place it in solution, little additional energy
is required to fold it.
A common pattern is that the entropy changes rapidly at the inception of
folding. This is consistent with the idea that long segments of the chain are
pinned by intra-chain contacts. The enthalpy changes later consistent with
the formation of many bonds and detailed rearrangements.
The most striking feature of protein folding is the unique nature of the
native state and the insensitivity of the folded form to perturbations during
the folding process, including initial shapes of the random coil. In fact, this
is one of the most puzzling phenomena in science (13) (14) (15) (16) (17)
(18) (19) (20) (21) (22) (23).
7
One can easily construct a lengthy list of reasons why proteins should not
fold to unique forms in water solution. The lack of significant energy,
interaction with the water, and the vast number of possible conformations
that must be sorted out are three prominent issues. Others include the list
of small forces that are at play locally and globally, and the fact that
generally speaking the rates of chemical processes are temperature
sensitive, implying that parallel steps may complete in different order
depending upon temperature.
Nevertheless, proteins do fold to their unique final states. The best
explanations as to why all these difficulties do not block reliable folding
usually relate in some way to natural selection and the potential energy
landscape; biological molecules have been selected to fold reliably.
The folding process terminates with the formation of chemical bonds; often
this is an annealing process.
The current state of the art in modeling entails fitting empirical parameters
using known folds, applying classical statistical mechanics or molecular
dynamics, and usually making some simplifying assumptions about the
potential energy surface. For many specific molecules this approach is
quite successful. The reverse process, protein engineering, is more
straight-forward and has been highly successful with libraries of elementary
structures.
In a 2011 essay in Nature Structural and Molecular Biology, Stephen C.
Harrison wrote (24),
“Can we carry out, conceptually or computationally, the fundamental
transformation from one dimension of information to three that is embodied
in the central dogma? Can we robustly predict how a protein will fold? Can
we deduce structure and function from sequence? There are clear hints
that these questions will ultimately have positive answers, but we are no
more confident of a timeline for the answers than we were in 1971.”
8
Properties of Stable Optical Caustics
In thinking about optical caustics, it may be useful to look at the beautiful
photographs of caustics in the papers of M.V. Berry (3) (7). Wikipedia has
some photos and some diagrams showing the relationships between wave-
fronts and caustics. We also suggest examination of the electronic
caustics in Cheianov, et al, (6) figures 3B, 3C, and 3D. Those illustrate the
stability of a caustic in the presence of perturbations.
The caustics are typically three-dimensional but are usually viewed on a
screen or table top. Throughout this paper, when discussing optical
caustics, we assume the geometric optics limit in which the wavelength is
small compared to other length scales. (At wavelength scale in optics or
quantum mechanics the caustics are decorated with beautiful diffraction
patterns).
An essential property of stable caustics in physics is that, beyond the
energy required to set waves in motion, no additional energy is required to
form the caustics. The formation process is geometrical and is fully
determined in optics by the lenses and mirrors.
If the shape of the wave-front emerging from the scattering region is
regarded as an initial value specification for the formation of the caustic
down field, then the shape of the caustic is not significantly altered by small
changes in the initial wave-front. Neither is it altered significantly by small
changes along the path of propagation, such as the index of refraction.
As will be discussed in another section, this insensitivity follows from the
underlying mathematics.
According to the mathematics, the number of elementary stable caustics is
fixed; in fact the number is seven. More complicated stable caustics can
form when multiple elementary caustics appear on the same screen. It is
common for the caustics to form a network structure of elementary
caustics; the ribbon-like bright lines on the bottom of a swimming pool are
often cited as an example of this situation.
9
On the subject of the relevant physics and mathematics, Fermat’s principle
is the starting point: the time from the source to a point on the screen is
minimized (technically, extremized). The quantitative theory of optical
caustics is based upon singularity theory, of which, Thom’s Catastrophe
Theory is a sub-set. Note that this is a rather different application of
Catastrophe Theory than that which is often used in potential theory.
Finally, we note that the propagation of the light stops when it encounters
the image plane of a camera or the eye. This is absorption and differs from
the scattering (refraction and/or reflection) that creates the caustic.
10
The Analogy
We summarize here, in broad terms, the analogy between stable caustics
and protein folds.
Energy:
Caustic formation requires no additional energy beyond that required
to power the light source. Energy considerations are not
deterministic of the structure.
Protein folding requires little energy and, in fact, energy may not be
the most important physical quantity in folding.
Insensitivity to Perturbations:
Caustics can be stable against perturbations, including initial
conditions.
The native state of a protein is stable against perturbations in the
initial random coil form and against perturbations during folding.
These perturbations can be directly in the shape, such as initial
random coil shape variations, or in the mechanism of folding such as
small pH changes.
The native state is not significantly changed by temperature
variations (over some limited range) and hence thermal agitation
does not disrupt the folded form.
11
Unique Final Form:
Proteins fold to form unique final structures. Stable caustics are
unique in shape up to a coordinate transformation.
Elementary stable caustics are only seven in number. More
complicated multi-caustic networks are known to form. The more
complicated catastrophes unfold into simpler catastrophes. For
example, a cusp looks like two fold caustics with a point in common.
The folding of proteins to unique final structures is essential to the
biology.
Much research (25) (26) (27) (28) (29) supports the idea that
complicated structures are built up from simpler ones. For example,
domains built up from helixes, beta sheets, and coil.
Insensitivity to Initial Conditions
Caustics are insensitive to small perturbations in the initial wave form.
The random coil has a large number of initial conformations. All of
these progress to the unique native state.
Global Action
A singularity in caustic theory arises from an integration of wave
motion over a surface or volume.
Folding involves global and local sequence structure.
12
Termination of Formation:
Caustics formation terminates with an interaction with a screen.
Protein folding terminates when bonds form to lock in the final state.
Wave Phenomenon:
Caustics originate in wave motion but may also occur in shock
waves. Light waves propagate independently of one another but end
up at the same space-time locations.
Protein folding may be related to torsion waves of variable speed.
The proteins of an ensemble fold independently of one another but
end up with the same shape.
Changes and Mutations
When the mirrors and/or lenses are modified in the presence of a
caustic, the caustic is stable against small changes but eventually it
will undergo a major change. It may become unstable or change into
another caustic depending, in a predictable way, upon the details.
More specifically, a change in the dimensionality of the physical
arrangement makes one kind of caustic impossible but may permit
another (see below for more detail).
Mutations either make no change in structure, lead to an unstable
structure, or change to a different structure.
13
State Variables and Control Parameters in Wave Motion
There are two broad classes of variables of interest in wave motion: control
parameters and state variables. It is important to distinguish these classes.
For the purposes of this descriptive section, we do not require that these
technical terms be defined carefully.
Control parameters are usually physical quantities that an experimenter
can change. An example is the position and orientation of the screen upon
which the caustic is viewed. Caustics are viewed in the space of all
relevant control parameters.
For light, state variables directly affect the time of travel along various
possible paths. An example is the index of refraction along the possible
light paths. According to Fermat’s principle, the light will follow the path
defined by the state variables that takes the least time. In more technical
terms, Fermat requires that the optical path of a ray be stationary with
respect to variation of paths between points represented by state variables
and control parameters.
As we shall note, below, stable caustics occur only if the number of control
parameters is 5 or fewer. There can be many state variables.
14
Toward a New Theory
If natural selection has resulted in a set of biological proteins that have a
special property, derivable from the sequence, that causes them to fold to
unique structures in the presence of potentially disrupting influences, then
what is the physical nature of that ‘special property’?
From the perspective of the physics and mathematics we have suggested
that folding entails the propagation of mechanical waves, presumably
mainly torsion waves, which have variable wave-speed. The special
property we are looking for is a set of constraints on the phase (or time-
delay function) of these waves.
If the theory described here agrees with observations then it has the
potential to become a completely quantitative theory. When completed, it
might potentially relate the geometry of mechanical waves to the chemical
and physical properties of the residues.
Another advantage is that the mathematics has essentially all been worked
out. What remains to test and complete the theory is discussed in the next
section.
In this section, we discuss some of the aspects of this problem that can be
inferred using available information.
We shall begin by supposing the existence of some, perhaps multi-
dimensional, mathematical variable that is derivable from the sequence
over some length of the molecule and which effects the formation of a
caustic at another locus. For simplicity, we start by using a scalar variable
X ; (more commonly, this actually stands for a collection of state variables).
The control parameters are similarly symbolized byξ.
We denote the phase function by (
)X ξΦ
,
.
The two essential conditions for the appearance of a caustic at a specific
point are,
15
and
∂
X
∂
2
∂
X
∂
2
Φ
(
X
) 0
,
=
ξ
Φ
(
X
) 0
,
ξ
=
These conditions determine points in both variables but Thom’s theory
applies around this point, thereby making it enormously powerful in
physics, especially with respect to issues of stability. The equation of the
caustic, expressed in the control variables ξ, is obtained by eliminating the
variable X in the two equations. It is not necessary to go into how to carry
that out, as all the cases have already been worked out and catalogued.
Moreover, there are only seven solutions for stable caustics and they arise
)X and control parameters
from simple polynomials in the state variables (
( )ξ . Any mathematical form, (up to a change in coordinates), for the phase
that is not on Thom’s list is not stable.
So far, our theory looks like this: natural selection picked out molecules
that have mechanical waves whose phases depend upon state variables
and control parameters. The phases, (or propagation delay functions), are
determined by some property involving the residues and derivable from the
sequence. The shape of the resulting caustic is expressed in control
variables, ξ and this determines the shape of the protein. The possible
shapes are limited to seven possibilities. For stable caustics the maximum
number of control variables is five. An analogy with the bands seen on a
stream bed is appropriate. The dimensions of the problem pick out one
catastrophe; the details of the surface motion are not important.
We called this property, X , in the interest of simplicity. X might be
analogous to an index of refraction which is dependent upon position. It
16
might be a simple scalar derivable from, say, the polarity of the residues.
However, it is likely to be more complicated.
At present, we have only limited information about the waves of interest.
They must, of course propagate in a space of three spatial dimensions and
three dimensions of orientation at each point. We do not know that all of
these dimensions are involved in picking out a caustic; the (
)Φ Ψ angles
,
and the distance along the chain may be sufficient.
If the waves and their phases can be identified in simulations, then
mutations involving substitutions can potentially identify the precise
dependence of the phases upon the structure of the residues.
Even in the simple form just described the theory has a number of
interesting features. The fact that the native state is not sensitive to small
variations in the initial conditions of the random coil is important. Similarly,
at least qualitatively speaking, the stability of the process of the formation
of native state against perturbations also follows by the same argument.
All of the features described in the Analog section, above, are described at
either quantitatively or qualitatively by this theory.
We offer now some speculations to expand the theory.
We have not established a specific relationship between the shape of the
caustic and the shape of the molecule. However, it seems reasonable that
the higher order of the singularity, the more complicated the associated
shape.
We assume that only certain aspects of the molecule are associated with
these caustics.
Natural conjectures would be that for a fold caustic, the shape is a hairpin,
for a swallowtail catastrophe there is a crossing point and two sharp bends
so this may form the initial configuration of a helix, where the first three
intra-chain contacts are needed to get the helix formation process started.
Based upon the appearance of high-degree polynomials in the state
variables ( X , above), corresponding to the appearance of caustics, we
17
have suggested that heterogeneity of the sequence is critical in producing
the formation of a caustic and hence of forming a fold. This, in turn, hints
that there is a significant scale-length, over which the sequence is
heterogeneous, and hence suggests that homopolymers and relatively
smooth random polymers do not reliably fold in this theory. This also
follows from the fact that we are only considering waves of variable speed.
(In our analytical models, variable speeds will not arise in homogeneous
situations).
18
Challenges
The most direct way to test and limit this theory is to study the propagation
of waves during folding. We need to know how many dimensions are
relevant, how the phase is changed by the physical and chemical
properties of the residues (polarity, moment of inertia, nearest neighbor
interactions, etc.), the wavelengths, and whether traveling-waves and/or
solitons are involved.
Another approach, assuming the waves can be defined in the context of
various models, is to find out if the phase conditions, above, are, in fact,
satisfied and if the molecule assumes the shape of the caustic around such
points.
19
Conclusions
We have presented our analysis of the stability of the protein folding
process in terms of established physics and mathematics and found a
strong correspondence between caustics and protein folds. We restated
our findings as a simple statement of a theory of folding.
We propose that torsion waves of variable wave-speed conspire to form
caustics in a way similar to the way in which gravity waves of variable
wave-speed conspire to form caustics. We suggest that natural selection
has left us with molecules that have this unusual but well defined property.
It seems unlikely that waves of such potentially complicated form will ever
be understood analytically. We have therefore emphasized a geometric
approach involving caustics. Technically speaking, the wave is treated as
a geometric object and the study of its singularities is a quantitative
approach to understanding the wave behavior.
If a form of this theory is correct, then important aspects of folding can be
quantitatively determined from data on the phases of mechanical waves on
the molecule. In particular, the key data elements would be the manner in
which the phase is changed by the interaction of the wave with the residues
The most fundamental way to test and expand this theory is to find the
behavior described in equations in the Theory section, above, in waves of
variable wave-speed. This is a formidable problem in view of the large
number of forces and moving components that come into play.
Simulations of folding can be used to explore the ideas presented here by
looking for the patterns we have described: waves of variable wave-speed,
heterogeneity of the sequence when an amino acid substitution results in a
new fold, decomposition of the form of the molecule just prior to the
formation of bonds or of geometrical hindrance, and the study of nearly
homogeneous and random sequences to determine the length scales
involved.
20
Mathematics and Physics
In these technical sections we discuss material somewhat more advanced
than that used in the main part of the paper.
I.) Mathematics Literature:
A wide range of sources are available for the study of the mathematics.
For completeness, we present here a brief literature survey.
The book by P.T. Saunders is an excellent introduction of the subject of
catastrophe theory. He has only short section on optical caustics.
The book by R. Gilmore is comprehensive and has a clear section on
caustics.
The best sources on optical caustics are the papers by Berry and his
colleagues. In particular, the review of catastrophe optics by Berry and
Upstill is excellent. Various other papers by Berry and colleagues are also
excellent.
21
II.) Caustics:
In this section, we present a brief guide to the technical treatment of the
theory of optical caustics.
(Parenthetically, we comment that the treatment requires the use of vector
calculus of multiple variables but we chose our example so that elementary
calculus of one variable could be be used.)
The set-up is a standard text-book optical arrangement: a source of light
illuminates an object. Light from every point on the object reaches each
point on an image screen located some distance away. The coordinates of
points on the screen are control variables and are described here by some
equations in variables described by the Greek symbol, ξ. The state
variables, the variables such as local path and velocity that appear in the
action are written in Latin (x).
, which depends
The light has a phase, or time-delay function,
X ξ ξ
)
,
,
(
Φ
1
2
upon both control parameters and state variables. This function carries all
the information that we require about the object and about where it is
viewed.
The electric field at the image is given by the Fresnel integral, which we
simplify to
E
,
(
ξ ξ
1
2
)
=
∫
dX
i
exp(
Φ
(
X
,
,
ξ ξ
1
2
))
A key point is that such integrals get their maximum contribution from those
points where the phase function is stationary; that is to say, the gradient of
the phase function is zero.
22
We begin by stepping inside the integral sign. We shall assume that for the
particular set-up we that have, a stable caustic exists. This would be easily
identified as a very bright area on the image. (Unstable caustics will be
discussed below).
The first step in understanding the stable caustic is to set the gradient of
the phase with respect to the Latin variables (state variables) to zero and
also to set the next higher derivative with respect to these variables to zero.
(Please see the Theory section, above). The two resulting equations pick
out a set of points
)ξ ξ that correspond to the singularity.
,
(
1
2
Next, we apply Thom’s powerful theorems. From this, we learn that the
phase function must have one of seven possible forms. In order to make
this more specific, let’s write down one possibility (the cusp catastrophe):
Φ
(
X
,
,
ξ ξ
1
2
)
=
4
X
+
ξ
1
X
2
+
ξ
2
X
The calculation steps can be constructed as follows. First is the condition,
following from Fermat’s principle, that the phase (time delay) is at a
minimum (or extremum)
d
dX
Φ
(
X
,
,
ξ ξ
1
2
)
=
4
X
3
+
X
2
ξ ξ
+
1
2
≡
0
Next, the condition for the appearance of the caustic,
2
d
dX
2
Φ
(
X
,
,
ξ ξ
2
1
) 12
=
X
2
+
2
ξ
1
X
≡
0
The final step is to find the places on the screen, i.e. values of
)ξ ξ , at
(
,
1
2
which the phase is stationary and the second derivatives also vanish.
23
3
8
ξ
1
=
0
27
ξ+
2
Since both of the above equations must be simultaneously true, this can be
done algebraically by eliminating X from the above two equations with the
result,
That is the equation describing the caustic as seen on the screen. As is
well known, this is a cusp.
As an example of stability (8), let us examine what happens to the cusp
caustic, above, which arises from the variable X , for a perturbation in
another variable, Y , of the form
. We calculate, first the condition
2
Y
Y
ξ ξ+
2
1
from Fermat’s principle,
Y
2
+
ξ ξ
1
2
,
,
ξ ξ
1
2
X Y
,
Φ
=
=
0
)
2
∂
Y
∂
(
and second the condition for the appearance of a caustic in this variable.
X Y
,
Φ
,
,
ξ ξ
2
1
2
ξ
1
=
=
0
(
)
2
∂
Y
∂
2
There is no impact upon the caustic due to this other variable at an
arbitrary point on the screen.
There are seven stable catastrophes and hence seven stable optical
caustics. Other caustics are unstable; that is, any small change in the
Greek variables significantly alters the form of the caustic.
As noted before, a point image on a screen is not one of the stable
caustics; it is therefore unstable. When the phase depends upon more than
five control parameters, it produces only unstable caustics. Technically,
there are catastrophes that have a looser form of stability
(homeomorphisms) but that is beyond the scope of this paper.
24
Caustics are classified by their co-dimension, which is the dimension of the
space containing the caustic minus the dimension of the caustic itself. The
result is often just one. (There is just one equation in addition to the
condition implied by Fermat’s theorem.). Berry remarks, “Loosely stated,
[the codimension] is the minimum number of essential control parameters
of a space that contains the singularity.”
We emphasize, again, that in view of the fact that there are only seven
caustics, the mathematics has been worked out rather completely and is
available from the literature.
25
1. White, B.S. & Fornberg, B. On the Chance of Freak Waves at Sea.
Journal of Fluid Mechanics. 1998, Vol. 355, 113.
2. Brown, M.G. Space-time surface gravity wave caustics. Wave Motion.
2001, Vol. 33, 117.
3. Berry, M.V,& Upstill. Catastrophe Optics. Morppologies of Caustics and
Their Diffraction Patterns. Progress in Optics. 1980, Vol. 18, 257.
4. Berry, M.V. Singularities in waves and rays. [book auth.] R. ,Kleman, M.,
and Poirier, J.-P. Balian. Physics of Defects. Amsterdam : North Holland,
1981.
5. Chapman, C.H. & Drummond, R. Body-Wave Seismograms in
Inhomogeneous Media Using Maslov Asymptotic Theory. Bulletin of the
Seismological Society of America. 1987, Vol. 72, S277.
6. Cheianov, V.V. Fal'ko, V., and Altshuler, B.L. The Focusing of
Electron flow and a Veselago Lens in Graphene p-n Junctions . Science.
2007, Vol. 315, 1252.
7. Berry, M.V. Beyond Rainbows. Current Science . 1990, Vol. 59, 1252.
8. —. Waves and Thom's Theorem. Advances in Physics. 1976, Vol. 25, 1.
9. Gilmore, R. Catastrophe Theory for Scientists and Engineers. New
York : Wiley & Sons, 1981.
10. Saunders, P.T. Catastrophe Theory. Cambridge : Cambridge
University Press, 1980.
11. Physics of Caustics and Protein Folding: Mathematical Parallels.
Simmons, W. & Weiner, J.L. s.l. : arXiv 1108.2740 , 2011.
12. Lane, N. Energetics and genetics across the prokaryote-eukaryote
divide. Biology Direct. 2011, Vol. 6, 35.
13. Dagget, V. & Fersht,A.R. Is there a unifying mechanism for protein
folding. TRENDS in Biochemical Sciences. 2009, Vol. 28, 18.
26
14. Dill, K.A., Ozkan, S.B., Shell, M.S., and Weikl, T.T. The Protein
Folding Problem. Ann. Rev. Bioph. 2008, Vol. 37, 289.
15. Rose, G.D., Fleming, P.J., Banavar, J.R. and Maritan, A. A
backbone-based theory of protein folding. PNAS. 2006, Vol. 103, 16623.
16. Onuchic, J.N. & Wolynes, P.G. Theory of Protein Folding. Current
Opinion in Structural Biology. 2004, Vol. 14, 70.
17. Yang, J.S., Chen, W.W., Skolnick, J., and Shakhnovich, E.I. All-
Atom Ab Initio Folding of a Diverse Set of Proteins. Structure. 2007, Vol.
15, 53.
18. Dodson, E.J. Protein predictions. Nature. 2007, Vol. 450, 176.
19. Wolynes, P.G. Recent successes of teh energy landscape theory of
protein folding and function. Quarterly Reviews of Biophysics. 2004, Vol.
38, 4.
20. Shakhnovich, E. Theoretical studies of protein-folding
thermodynamikcs and kinetics. Current Opinion in Structural Biology. 1997,
Vol. 7, 29.
21. Wolynes, P.G. Onuchic, J.N. and Thirumalai, D. Navigating the
Folding Routes. Science. 1995, Vol. 267, 1619.
22. Karplus, M. & Kurlyan, J. Molecular dynamics and protein function.
PNAS. 2005, Vol. 102, 6679.
23. Hubner, I.A., Deeds, E.J., and Shakhnovich, E.I. Understanding
ensemble protein folding at atomic detail. PNAS. 2006, Vol. 17747.
24. Harrison, S.C. Forty years after. Nature Structural & Molecular Biology.
2011, Vol. 18, 1304.
25. Koonin, E.V, Wolf, Y.I. and Karev, G.P. The structure of the protein
universe and genome evolution. Nature. 2002, Vol. 420, 218.
26. Choi, I-G. & Kim, S-H. Evolution of protein structural classes and
protein sequence families. PNAS. 2006, Vol. 103, 14056.
27
27. Chothia, C., Hubard, T.m Barns, H. and Murzin, A. Protein folds in
the all beta and all alpha classes. Annu. Rev. Biophys. Biomol Structure .
1997, Vol. 26, 597.
28. Chothia, C. & Finkelstein, A. V. The Classifiction and Origins of
Protein Folding Patterns. Annu. Rev. Biochem. . 1990, Vol. 59, 1007.
29. Bashton, M. & Chothia, C. The Generation of New Protein Functions
by the Combination of Domains. Structure. 2007, Vol. 15, 85.
28
|
1311.2545 | 1 | 1311 | 2013-11-11T19:40:40 | A physicist's view of DNA | [
"physics.bio-ph",
"q-bio.BM"
] | Nucleic acids, like DNA and RNA, are molecules that are present in any life form. Their most notable function is to encode biological information. Why then would a physicist be interested in these molecules? As we will see, DNA is an interesting molecular tool for physicists to test and explore physical laws and theories, like the ergodic theorem, the theory of elasticity and information theory. DNA also has unique material properties, which attract material scientists, nanotechnologists and engineers. Among interesting developments in this field are DNA-based hybrid materials and DNA origami. | physics.bio-ph | physics | A
physicist’s
view
of
DNA
Alireza Mashaghi & Allard Katan
Cees Dekker lab, Bionanoscience Department, Kavli Institute of Nanoscience, Delft
University of Technology.
Nucleic acids, like DNA and RNA, are molecules that are present in any life form. Their
most notable function is to encode biological information. Why then would a physicist be
interested in these molecules? As we will see, DNA is an interesting molecular tool for
physicists to test and explore physical laws and theories, like the ergodic theorem1, the
theory of elasticity2 and information theory. DNA also has unique material properties,
which attract material scientists, nanotechnologists and engineers. Among interesting
developments in this field are DNA-based hybrid materials and DNA origami3-6.
DNA,
the
regular
aperiodic
molecule
In his 1945 book “What is life?” Schrodinger7 postulated that for a molecular system to
carry information, it has to be regular but aperiodic. In his view, irregular amorphous
materials are too chaotic to be able to practically carry information, but full periodicity
reduces the capacity to encode information. Was Schrodinger right?
In principle not, since we now know there are ways to encode information even in
irregular materials. But in the case of DNA, Schrodinger’s prediction was crucial to the
discovery of the structure. Not only did he inspire theoretical physicist Francis Crick and
biochemist James Watson to look in the right direction, the fingerprint of a regular
structure that was visible in X-ray diffraction data was instrumental in unraveling the
now famous double helix structure of DNA8. The DNA molecule is indeed structurally
periodic, but aperiodic at the sequence level (Figure 1A).
The
structure
of
DNA
DNA is an organic polymer of four different monomers. Each monomer is composed of a
phosphate group, a single-ring sugar and one of four bases: A, T, C and G. The bases are
planar single- or double-ring aromatic compounds. The monomers can form bonds
between the sugar and the phosphate and form into a long polymer or strand of DNA.
Such a single strand is irregular and if it consists of N monomers, there can be 4N
different combinations of bases, so the molecule carries 22N bits of information. The
bases can also interact with each other through the formation of hydrogen bonds.
Specifically, A pairs with T and G pairs with C. The energy gain of base-pairing is not
the same for these two pairs, since A and T only share 2 hydrogen bonds, while C and T
share 3. An entire strand of DNA can form base-pairs with a complementary strand to
make double-stranded DNA. The chirality of the sugars and the optimization of hydrogen
bond angles leads to the formation of a regular helical structure with 10.4 base pairs per
turn. Since the sequence of the complementary strand is entirely determined by the
template strand, double stranded DNA does not carry more information than single
stranded, even though it uses twice the amount of material. In spite of this, almost all
organisms encode their genetic information on double-stranded DNA.
Twisting
DNA
beyond
the
helix:
supercoiling
To read the base sequence, proteins use the same hydrogen bonding sites as used for
DNA pairing. Therefore reading requires breaking of base-pairing, so-called melting.
Melting unwinds the helix and allows the information to be read. When a protein walks
along the DNA and locally melts it, it displaces the helical twist. Just like a normal rope
that is twisted, DNA that is wound up will at some point collapse into supercoils, or
plectonemes as they are called.
Recently, researchers in the Cees Dekker lab have managed to visualize such supercoils9.
To do this, they attached a DNA molecule to a magnetic bead on one side, and a glass
slide on the other side. Using external magnets, they rotated the DNA, mimicking the
action of an unwinding protein and forming supercoils. By fluorescently labeling the
DNA, and pulling the magnetic bead to the side, they could see where the supercoils were
located, and follow how they are created and annihilated, or moved spontaneously along
the DNA.
Figure 1. (A) Structure of DNA double helix (B) DNA supercoiling visualized with
fluorescence and magnetic tweezers. Schematic illustrating how higher fluorescence
intensity reveals the position of a plectoneme. (C) plectoneme hopping from one position
to another and DNA. Analysis of different experimental conditions shows that the rate of
formation of new plectonemes depends on salt concentration and pulling force.
DNA
and
theory
of
elasticity
The single molecule measurements of force and extension of DNA have provided the
most rigorous test to date of theories of entropic elasticity2. When free in solution, a
flexible polymer randomly coils, wraps around itself and adopts various conformations,
leading to an average end-to-end distance much shorter than its contour length. In
contrast, when a linear polymer is stretched from its ends, fewer conformations are
available to the molecule. Pulling the molecule into a more extended chain is thus
entropically unfavorable, resulting in an entropic resistance against pulling.
Various approaches have been used to stretch DNA molecules including magnets, fluid
flow and later optical traps. The entropic force–extension behavior of double stranded
DNA agreed well with the worm like chain model. Forces exerted by molecular motors
found in cells, can stretch double stranded DNA to nearly its contour length.
In addition to its use in reading the information, the base pairing also provides
mechanical strength to the DNA molecule. The persistence length of double stranded
DNA is ~50 nm, at least a factor of five larger than the dimension of the molecular
machines that read the information on the DNA. Single stranded DNA is much more
floppy, it’s persistence length is only 1 nm.
A
Physical
way
to
read
DNA
Reading the information encoded in DNA is extremely important for biomedical
sciences. Much effort has been put in to develop technology for sequencing DNA, which
is the accurate reading of the order of bases in a DNA strand. A new technique for
sequencing, that is still in early stages of development, is based on electrical current
measurements. Simulations and theoretical studies suggest that it is possible to sequence
a DNA molecule during its translocation through a nanopore, a few nm wide hole in a
solid state membrane. For instance it is proposed10 that by passing single DNA molecules
through a nano-gap between two electrodes and measuring the electron tunneling, one
could read the DNA sequence. Electron tunneling through a nano-gap is affected
differently by the presence of different DNA bases in the gap. Besides the future promise
of sequencing, solid state nanopore measurements are already being used to determine
the presence of proteins on single DNA molecules, or detect secondary structure11, 12.
DNA
origami:
using
DNA
as
a
nanoscopic
building
material
One important difference between DNA and a typical polymer is that the latter is
synthesized via chemical means, and has a distribution of molecular weights. The
synthesis of DNA by living cells or isolated cell components is performed such that
(nearly) all synthesized molecules are exactly the same, not only with the same number
of each monomer in the polymer, but also with the same order of the monomers. This
precise manufacturing, along with the known base pairing, has led scientists to use DNA
as a building material for nanostructures called DNA origami (Figure 2). Using keenly
selected short ‘staple’ strands, a long single-stranded DNA molecule is wrapped up into a
desired shape. The nanometer scale precision attainable with these structures is nearly
impossible with conventional methods. Though the first origami structures - smiley faces
50 nm in size – were not very useful, proof-of-concept applications in drug delivery have
already been demonstrated13 and more and more complex 3D structures are being built
out of DNA. In our lab, DNA origami is used as a tool to enhance the functionality of
nanopore and AFM measurements.
Figure 2: DNA origami objects. Top left: models and AFM images of two-dimensional
DNA origami structures. Bottom left: a crystal made of DNA tiles. Center and right:
models and electron micrographs of various 3-dimensional DNA origamis. Bottom right:
model and AFM image of proteins bound to DNA origami. Figure reproduced from
Castro et al.14
Ways
to
visualize
a
single
DNA
molecule
A single DNA molecule is only 2 nm in size, too thin to be visible under a light
microscope. Fluorescently labeling the DNA, like in the experiment of Figure 1B, can
work around this. There are many uses of fluorescent imaging of DNA, but still it has
some drawbacks. Resolution is limited to that of the optical microscope, unlabeled
objects are invisible, and biological function can be compromised by the labeling.
Electron microscopes can visualize stained DNA, but they can only operate under
unnatural conditions.
The Atomic Force Microscope (AFM) is a mechanical microscope, it forms an image by
gently ‘feeling’ the surface of objects. The AFM can resolve single DNA molecules even
in a dense environment, can operate in aqueous solution and requires no labeling. The
latest developments in AFM technology have pushed the time resolution from a few
minutes to tenths of seconds. Such a high-speed AFM – one of the first in Europe- is
available in our laboratory. This allows us to make real-time ‘molecular movies’ of the
interactions between DNA and the proteins that maintain it (Figure 3).
Figure 3. A protein complex (tetrasome) hopping to a new position on a DNA molecule.
The hopping distance is only 4 nm. Images by Allard Katan, TU Delft
Exciting
opportunities
in
DNA
physics
and
engineering
More than a century has passed since the discovery of DNA in late 19th century by
Friedrich Miescher. Still there are mysteries around the physical properties of DNA. How
does DNA respond to force and negative torque? DNA in cellular nuclei is highly
constrained and confined. How does confinement affect the structure and dynamics of
DNA?
in DNA nanotechnology
important opportunities
Among most
to fabricate
is
nanomachines with useful
recently witnessed promising
functions. We have
developments in this direction, such as programmable DNA based motors that move on
linear tracks15 and DNA nanorobots that kill cancer cells13.
DNA may be the quintessential molecule of biology, but as you have seen it is also a rich
source for physicists, both as a subject in itself and as a tool to reach new capabilities.
References
1.
2.
3.
B. Liu, R. J. Baskin and S. C. Kowalczykowski, Nature, 2013, 500, 482-485.
C. Bustamante, Z. Bryant and S. B. Smith, Nature, 2003, 421, 423-427.
A. Kuzyk, R. Schreiber, Z. Y. Fan, G. Pardatscher, E. M. Roller, A. Hogele, F. C.
Simmel, A. O. Govorov and T. Liedl, Nature, 2012, 483, 311-314.
X. M. Tu, S. Manohar, A. Jagota and M. Zheng, Nature, 2009, 460, 250-253.
H. T. Maune, S. P. Han, R. D. Barish, M. Bockrath, W. A. Goddard, P. W. K.
Rothemund and E. Winfree, Nature nanotechnology, 2010, 5, 61-66.
4.
5.
P. W. K. Rothemund, Nature, 2006, 440, 297-302.
E. Schrödinger, What is life? The physical aspect of the living cell, The University
press; The Macmillan company, Cambridge Eng. New York,, 1945.
J. D. Watson and F. H. Crick, Nature, 1953, 171, 737-738.
M. T. J. van Loenhout, M. V. de Grunt and C. Dekker, Science, 2012, 338, 94-97.
T. Thundat, Nature nanotechnology, 2010, 5, 246-247.
G. F. Schneider and C. Dekker, Nat Biotechnol, 2012, 30, 326-328.
C. Dekker, Nature nanotechnology, 2007, 2, 209-215.
S. M. Douglas, I. Bachelet and G. M. Church, Science, 2012, 335, 831-834.
C. E. Castro, F. Kilchherr, D. N. Kim, E. L. Shiao, T. Wauer, P. Wortmann, M.
Bathe and H. Dietz, Nat Methods, 2011, 8, 221-229.
S. F. J. Wickham, J. Bath, Y. Katsuda, M. Endo, K. Hidaka, H. Sugiyama and A.
J. Turberfield, Nature nanotechnology, 2012, 7, 169-173.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
|
1504.02139 | 2 | 1504 | 2015-10-22T21:22:59 | Universal Protein Distributions in a Model of Cell Growth and Division | [
"physics.bio-ph",
"q-bio.SC"
] | Protein distributions measured under a broad set of conditions in bacteria and yeast were shown to exhibit a common skewed shape, with variances depending quadratically on means. For bacteria these properties were reproduced by temporal measurements of protein content, showing accumulation and division across generations. Here we present a stochastic growth-and-division model with feedback which captures these observed properties. The limiting copy number distribution is calculated exactly, and a single parameter is found to determine the distribution shape and the variance-to-mean relation. Estimating this parameter from bacterial temporal data reproduces the measured distribution shape with high accuracy, and leads to predictions for future experiments. | physics.bio-ph | physics | Universal Protein Distributions in a Model of Cell Growth and Division
Department of Chemical Engineering and Laboratory of Network Biology, Technion, Haifa 32000, Israel
Naama Brenner
NY 10012 USA and NYU-ECNU Institute of Mathematical Sciences at NYU Shanghai,
Courant Institute of Mathematical Sciences, New York,
3663 Zhongshan Road North, Shanghai 200062, China
C.M. Newman
Department of Physics and Institute of Nanotechnology and Advanced Materials, Bar-Ilan University, Ramat Gan 52900, Israel
Dino Osmanovi´c and Yitzhak Rabin
Department of Physics and Astronomy, Department of Computational and Systems Biology,
University of Pittsburgh, Pittsburgh, PA 15260 USA
Hanna Salman
Department of Physics and Courant Institute of Mathematical Sciences, New York University, New York,
NY 10012 USA and NYU-ECNU Institutes of Physics and Mathematical Sciences at NYU Shanghai,
3663 Zhongshan Road North, Shanghai, 200062, China
D.L. Stein
Protein distributions measured under a broad set of conditions in bacteria and yeast were shown
to exhibit a common skewed shape, with variances depending quadratically on means. For bacteria
these properties were reproduced by temporal measurements of protein content, showing accumu-
lation and division across generations. Here we present a stochastic growth-and-division model
with feedback which captures these observed properties. The limiting copy number distribution
is calculated exactly, and a single parameter is found to determine the distribution shape and the
variance-to-mean relation. Estimating this parameter from bacterial temporal data reproduces the
measured distribution shape with high accuracy, and leads to predictions for future experiments.
INTRODUCTION
The phenotype of a biological cell — in particular, the
types and copy numbers of its expressed proteins — fluc-
tuates from cell to cell, even among those whose geno-
types and growth environments are identical (reviewed
in [1–3]). Protein content depends on a complex in-
terplay of genetic, epigenetic and metabolic processes,
with numerous cell-specific regulatory mechanisms and
feedback loops. However, recent experiments [4] have
demonstrated that for two different types of microorgan-
ism (yeast and bacteria), each under a broad range of
conditions, the distribution of highly expressed protein
copy number appears universal : under rescaling by mean
and standard deviation, all such distributions collapse
onto a single skewed curve [5]. In the same experiments
variances were found to depend quadratically on their
means, a trend displayed also by all highly expressed pro-
teins in E. coli in a genome-wide study [6] (see also [7]).
A recent study following the protein content in indi-
vidual E. coli bacteria over roughly 70 generations has
revealed that, under the same scaling criteria, the shape
of the distribution of protein copy number sampled over
time in an individual converges to the one observed in
large populations [8]. While analogous temporal data
are currently unavailable for yeast, this is an important
property that reflects the ergodicity of the relative fluc-
tuations in protein expression in bacteria. These results
can serve as a basis for constructing a model relating bac-
terial temporal protein dynamics to their distributions.
MODEL WITHOUT FEEDBACK
Given that the universal statistical properties de-
scribed above were found for a range of experimental con-
ditions for various proteins in bacteria [4], such a model
should rely only on general coarse-grained processes. We
therefore start by assuming as little as possible given the
experimental data:
• Protein number increases as ekit during the ith gen-
eration, where the exponential growth rate ki fluc-
tuates with i [8].
• The time Ti of the ith generation (i.e., the time
between cell division at the (i− 1)st generation and
that at the ith) is also random [9–13].
• The product Xi = kiTi is a random variable, with
(positive) mean µ and variance σ2. We will refer to
Xi as the accumulation exponent.
5
1
0
2
t
c
O
2
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
2
v
9
3
1
2
0
.
4
0
5
1
:
v
i
X
r
a
(cid:16) n(cid:88)
j=1
(cid:17)
Xj
n(cid:88)
j=1
• Protein number is conserved at cell division, and
protein degradation is much slower than a typical
interdivision time [14].
Let Ni denote the copy number of a given type of pro-
tein in the cell and fi the copy number ratio between
the daughter and parent cells, both at the end of the ith
generation. Incorporating the features listed above gives
rise to the recursion relation
Ni+1 = fiNi exp (Xi+1) .
(1)
In bacteria fi is narrowly distributed about 1/2 [15]. We
take fi = 1/2 for now and discuss deviations from this
assumption later. The solution of (1) for arbitrary gen-
eration number n is
Nn = 2−nN0 exp
(2)
where N0 is the initial copy number. Then
ln(Nn/N0) = −n ln 2 +
Xj
(3)
The mean of Xj should compensate for the decrease in
protein number in the daughter cells caused by division;
otherwise copy numbers would be unstable, running off
to unsustainably large numbers or falling to zero within
a few generations. Eq. (3) can be rewritten as
ln(Nn/N0) − n(µ − ln 2)
√
nσ =
/
Xj − nµ
√
nσ .
/
(cid:105)
(cid:105)
(cid:104)
(cid:104) n(cid:88)
j=1
(4)
If the accumulation exponents Xj are independent, the
central limit theorem gives
ln(Nn/N0) − n(µ − ln 2)
/
nσ → N (0, 1) ,
(5)
(cid:105)
√
(cid:104)
that is, the LHS converges in distribution to the normal
distribution with mean zero and variance one. There
is no stationary distribution for this process; the mean
and variance of Nn vary with time (or n), even when
µ = ln 2 exactly. This conclusion holds independently of
the various distributions used; all that matters is that
fluctuations are independent between generations. This
analysis demonstrates that a stationary distribution, as
experimentally observed, can result only if some negative
feedback is present. Given this, we next introduce and
analyze a modified model with effective feedback regulat-
ing protein accumulation, and following that we discuss
its experimental justification and consequences.
MODEL WITH FEEDBACK
A given protein type in an individual cell has a well-
defined typical copy number N . Its value is nonuniver-
sal, depending on protein type, growth conditions and
2
possibly other biological factors [8, 16]. The stationary
distribution shape must therefore be independent of N .
A natural extension of the growth-and-division model
consistent with observations is the introduction of an ac-
cumulation exponent that is negatively correlated with
protein number at the start of the cycle. The experimen-
tal requirement of universality constrains the form of the
feedback term: a change in scale of N cannot alter the
functional form of the recursion relation. The only func-
tion with this property of scale invariance is the power
law; the modified recursion relation is therefore
Ni+1 = fiNi[exp(ξi+1)](Ni/N )−α
(6)
with fi defined as before, ξi+1 (which we will call the
residual accumulation exponent) the component of the
accumulation that fluctuates independently from gener-
ation to generation, and a new phenomenological param-
eter 0 < α < 1; α = 0 is the case without feedback [17].
The recursion relation of the modified model is
ln Ni+1 = ln fi + ξi+1 + (1 − α) ln Ni + α ln N .
(7)
It is not hard to check that first, there is now a limiting
stationary distribution, with (cid:104)N(cid:105) ≈ N ; and second, that
N can be scaled out of the growth equations.
The introduction of a nonzero α makes the specific
form of the limiting distribution dependent (though not
too sensitively; see below) on the distributions of fi and
ξi. Experiments on bacteria indicate that ξi is approxi-
mately normally distributed (see Fig. 2b). Using this, we
can solve for the limiting distribution exactly when the
division ratio is fixed. The limiting distribution is again
lognormal:
(cid:16)
ln N − M(cid:17)2
(cid:105)
(cid:104)−
√
1
N Σ
× exp
2π
(8)
2Σ2
P (N ) =
,
√
2α − α2.
with M = ln N + (µ − ln 2)/α and Σ = σ/
These two parameters together determine the mean and
variance of the distribution: specifically, (cid:104)N(cid:105) = exp{M+
Σ2/2}, and (cid:104)N 2(cid:105) − (cid:104)N(cid:105)2 = (eΣ2 − 1) exp{2M + Σ2}.
However, only Σ determines the shape of the distribution;
different values of M collapse on one another following
scaling by a linear transformation. Moreover, for fixed Σ
the variance scales quadratically with the mean.
In terms of the model, Eq. (8) has several important
features. First, it preserves universality under scaling
with respect to all variables that appear only in the pa-
rameter M, because the distribution shape is indepen-
dent of this additive term. Consequently, all values of N ,
µ, and division ratio yield the same distribution shape.
Second, as noted above a single composite parameter
Σ determines the shape of the distribution. Σ charac-
terizes the balance between the variance of accumulation
exponents, which tends to drive the process to diverge,
and the effective feedback parameter α, which provides a
“restoring force”. Once Σ is determined, both properties
– collapse of scaled distributions and quadratic depen-
dence of variance on mean – are preserved.
Third, it should be noted that in the setting of our
model, the limiting steady state distribution equally well
represents the time average over many generations of a
single individual or the average at a single large time over
a large population in which the individuals are evolving
independently.
The analysis above assumed a Gaussian distribution
of the accumulation exponents and a fixed value of the
division ratio. We now explore the robustness of our
conclusions. Without these assumptions, the lognormal
solution will no longer be exact, but will not be signifi-
cantly altered for a variety of unimodal distributions for
both variables. Moreover, the scaling properties within
classes characterized by Σ (defined as before and which
is now close to but not exactly equal to the standard
deviation of ln N ) still hold. Fig. 1a shows examples of
means and variances computed from many simulations,
in which the fi’s were drawn from a Gaussian distribu-
tion and the ξi’s from a gamma distribution. For each
simulation, α and N were chosen randomly, and the vari-
ance σ2 of the gamma distribution was adjusted to give
a shape parameter Σ equal to one of three values: 0.2,
0.4 or 0.6. The resulting means and variances are seen
in Fig. 1a to collapse onto three parabolas corresponding
to the three classes defined by the value of Σ. Limiting
distributions are shown in Figs. 1(b-d); while they are
all very close to lognormal, the different Σ’s lead to a
range of shapes, from nearly Gaussian (1b; Σ = 0.2), to
skewed with exponential-like tail (1c; Σ = 0.4) and finally
to highly skewed (1d; Σ = 0.6). Within each class the
distributions from all simulations collapse after rescaling
onto a single curve. Thus, both properties of distribu-
tion collapse and the quadratic dependence of variance
on mean hold once Σ is fixed, even though the condi-
tions of the exactly solvable model are relaxed.
In addition, the assumption of symmetric division can
be relaxed as long as the average accumulation compen-
sates for the loss at division. The assumption of stable
proteins can also be relaxed to include first-order protein
degradation; this would require an additional parameter
and would modify only the accumulation exponent.
COMPARISON WITH DATA
To test the assumption of negative correlation we plot-
ted experimental values of ∆ ln Ni = ln Ni+1 − ln Ni
vs. ln Ni, as measured for bacteria, in Fig. 2a. The data
points were collected from six individual trajectories nor-
malized to unit average (data from [8]).
In agreement
with Eq. (7), the data are consistent with random scatter
about an overall linear dependence, with negative slope
3
FIG. 1: Model simulations. The stochastic process
described by Eq. (6) was simulated over 20,000
generations for each run. Values for α and N were
chosen uniformly at random in the range [0,1] and
[0.25,0.75], respectively. The division ratio fi was drawn
from a Gaussian distribution with mean 1/2 and
standard deviation 0.1, and ξi from a
gamma distribution whose variance was adjusted to α
to obtain one of 3 values of the shape parameter Σ. (a)
Variance vs. mean of 18 simulations for each shape
parameter show a collapse on 3 corresponding
parabolas. (b-d) Limiting distributions of three
simulations for each class are all well fit by lognormal
but have different shape parameters and so span a
range of shapes, from approximately Gaussian (b), to
an exponential-like tail (c), to a highly skewed
distribution (d). In each class the distributions collapse
onto one another to high accuracy.
determining α to be approximately 0.37±0.04. Using this
value the residual accumulation exponents ξi can be ex-
tracted from the data using Eq. (7) and measured values
of Ni. The approximately Gaussian distribution of these
exponents is shown in Fig. 2b, and their independence
between consecutive generations is evident from Fig. 2c.
√
The parameter determining the distribution shape in
2α − α2. Estimating α and σ from
our model is Σ = σ/
Figs. 2a and 2b respectively, we find Σ ≈ 0.4 ± 0.02.
Fig. 2d shows the lognormal distribution of Eq. (8) cor-
responding to this parameter in rescaled units (black
line), together with data from a large bacterial popu-
lation (grey circles), and single cell trajectories (black
squares and red stars) that exhibits the measured univer-
sal distribution shape over several decades of probability.
We note that this is not a fit, but a model prediction
with no adjustable parameters: the single parameter de-
4
forces in the dynamics of their protein content across
time, i.e. between the variance of accumulation expo-
nents (σ) that drive the process to diverge, and the feed-
back parameter (α) that prevents divergence. Therefore,
if the variance of the exponents ξi changes, the feed-
back parameter α should change in a correlated manner.
To test this prediction, single-cell dynamical trajectories
need to be measured over a variety of conditions that
span these parameters. Another possibility consistent
with our model is that both σ and α are fixed. At the mo-
ment, experimental perturbations — for example, chang-
ing medium or temperature — can change the mean, for
example by modifying the mean cell cycle time; but their
effect on the variance of exponents is unknown [10].
Our approach shares some features with previous the-
oretical work but differs in other respects. Earlier work
focused on protein accumulation and division [18–22], or
protein accumulation and continuous dissipation [23, 24].
The recent data on protein content over multiple gener-
ations [8] shows that, due to the exponential nature of
protein accumulation, division or dissipation alone can-
not stabilize copy numbers, and reveal a correlation be-
tween variables across generations.
The classes of proteins of interest are those consisting
of high-copy-number molecules, characterized by expo-
nential accumulation between successive cell divisions.
The exponential accumulation of protein during a cell
cycle suggests that protein production reflects a coher-
ent integration of many correlated processes in the cell.
Exponential growth of the cell size between divisions,
as well as negative correlation analogous to the one re-
ported here, were measured in several recent experi-
ments [10, 11, 13, 25]. Moreover, results on trapped bac-
teria show explicitly that the exponents of cell size growth
and protein accumulation are strongly correlated on a
cycle-by-cycle basis [8]. This suggests a picture where
highly expressed proteins that are strongly coupled to cell
metabolism are components of multi-dimensional pheno-
types that covary between individual cells. This view is
supported by a model recently proposed to explain expo-
nential biomass growth as resulting from an interacting
network of reactions [26, 27]. Furthermore, our model
is mathematically related to a recently proposed model
of cell size regulation [12, 28], which finds under simi-
lar assumptions a lognormal distribution with the same
compound parameter governing its shape. For highly
expressed proteins, this may be expected since protein
production and cell growth are tightly coupled [8]. How-
ever, there are also important differences between the two
models, which we address in detail in the Appendix.
Our model addresses directly the universal behavior of
bacterial protein distribution among different biological
realizations, including expression regulation mechanisms,
growth conditions, and types of microorganism. Its in-
gredients are independent of specific biological mecha-
nisms and rely on those general aspects of cellular events
FIG. 2: Comparison with data. (a)
∆ ln Ni ≡ ln Ni+1 − ln Ni plotted vs. ln Ni in units of N .
Solid line is y = −0.37x. (b) Accumulation exponents in
consecutive generations are approximately normally
distributed with average ln2 (subtracted out in the
figure), and (c) are independent between generations.
Solid line is y = 0.0075x − 0.00036. (d) Estimating the
universality class parameter Σ for these data from (a)
and (b), the distribution shape is predicted by Eq. (8)
(black line) and compared to data. Grey circles: large
population snapshot; black squares: protein trajectories
of individual trapped bacteria; red stars: sampled
points at the end of each cell cycle. Data from [4, 8].
termining the distribution shape is computed separately
using the single-cell dynamic measurements.
DISCUSSION
Motivated by experiments that found universal protein
distributions under various conditions in yeast and bacte-
ria, and by single-cell measurements of protein accumula-
tion and division in bacteria across multiple generations,
we have presented a model based on the premise that
the combined processes of growth, division, and feedback
set the distribution shape. With fixed division ratio and
Gaussian randomness the model is exactly solvable. The
solution identifies a single parameter Σ (Eq. (8)) defining
the distribution shape: it quantifies the balance between
growth of variance and feedback that stabilizes protein
numbers. With Σ fixed, a rescaling by mean and stan-
dard deviation collapses these distributions onto a single
curve, and displays a quadratic relation between variance
and mean.
Thus, our model predicts that populations in the same
class — i.e., which share the same shape parameter —
exhibit similar set-point balance between the opposing
— exponential protein accumulation, division and feed-
back — that are likely to be common to all dividing cell
populations. This marks a significant departure from the
main current line of research on protein number varia-
tion, which investigates synthetically produced proteins
while experimentally isolating the contribution of specific
microscopic mechanisms [29–34].
In particular, we have observed that feedback must be
present, because without it the mean and variance neces-
sarily drift to larger values as time increases. Moreover,
regardless of the specific processes leading to feedback
(which may differ for different protein types and organ-
isms), the mathematical form of the feedback in a growth-
and-division model must be power law to be consistent
with universality.
Aacknowledgments We thank Erez Braun for many
useful discussions. This work was supported by the US-
Israel Binational Science Foundation (NB and HS). DLS
thanks the John Simon Guggenheim Foundation for a fel-
lowship that partially supported this research, and Bar-
Ilan University for its hospitality while this work was
initiated. YR acknowledges support by the I-CORE Pro-
gram of the Planning and Budgeting committee and the
Israel Science Foundation, and by the US-Israel Bina-
tional Science Foundation. NB acknowledges support of
the Israel Science Foundation (Grant No. 1566/11). HS
acknowledges the support of NSF Grant PHY-1401576.
CMN acknowledges the support of NSF Grant DMS-
1007524 and DLS acknowledges the support of NSF
Grant DMS-1207678.
APPENDIX: UNIVERSAL PROTEIN
DISTRIBUTIONS AND CELL GROWTH
In this appendix, we detail the differences between our
model and a mathematically similar model of cell size
regulation recently proposed by Amir [12, 28]. However,
before describing those differences, we first discuss some
nontrivial consequences following from the similarity be-
tween the models. The fact that similar mathematical de-
scriptions, albeit with different biological interpretations,
may capture the dynamics of two distinct phenotypes is a
potentially significant mathematical unification resulting
from a strong coupling among disparate biological pro-
cesses. As noted above, cell size and protein copy number
are two separate, but strongly correlated, phenotypes of
the cell. Which one controls the other, or whether both
are regulated together, is not known at this time. One
interesting implication of the work presented here is that
not only cell size, but also the total content of highly ex-
pressed proteins, is under the control of what appears to
be a global cellular feedback, supporting the viewpoint
that protein copy number variation is a global variable.
This marks a significant departure from the main current
5
line of research on protein number variation.
We now turn to a discussion of some important differ-
ences in the interpretation and consequences of the two
models. First, the requirement that the average must
scale out of the distribution shape in any model of univer-
sal distributions necessitates mathematically the power-
law form of the feedback. Second, the parameter Σ that
governs the “universality class” of the family of distribu-
tions has been identified, and its constancy leads both to
the collapse of all curves under linear scaling and to the
observed quadratic relation between variance and mean.
The latter is an especially important consequence and
has no analogue in cell size distributions. We have also
seen numerically that the division into such classes ex-
tends beyond the conditions of the analytically solvable
case, rendering this result robust with respect to a wide
class of distributions of the underlying random variables.
Perhaps most importantly, our model makes specific
predictions on the constrained changes allowed in pro-
tein trace parameters under varying conditions. Similar
predictions cannot currently be made for cell size distri-
butions.
Both our model and that of [12, 28] are also easily mod-
ified to handle asymmetric division, as in yeast. However,
until data are available that relate temporal to popula-
tion statistics, it remains to be seen to what extent the
dynamics of proteins across generations in yeast can be
described by the approach outlined in the main text.
A final important difference between the approach de-
scribed in this paper and that in [28] concerns the na-
ture of the feedback itself. Analysis of E. coli data led
to the conclusion that cell size feedback is characterized
by α = 1/2 [28], corresponding (in leading order) to the
proposal that the feedback arises from constant addition
of volume over the cell cycle.
In contrast, a different
mechanism(s) may apply for copy numbers, and α can in
principle vary among protein types: our current results
are consistent with various values of α with an average
of α ≈ 0.37. However, at this stage there is no direct
evidence that can determine which phenotype (cell size,
protein copy number, etc.) controls the division point of
the cell and thus the feedback mechanism that controls
it. In principle, it could also be a cellular state that is
defined by several phenotypes simultaneously.
We show this difference explicitly in the figure below.
Fig. 3 uses cell size data from [8] to compute the feedback
parameter α, in manner similar to that used in Fig. 2a,
for the cell size phenotype. This results in α ≈ 0.5, in
agreement with [28], but different from the value α ≈ 0.37
shown in Fig. 2a.
[1] N. Maheshri and E.K. O’Shea, Ann. Rev. Bio-
phys. Biomol. Struct. 36, 413 (2007).
6
[12] I. Soifer, L. Robert, N. Barkai,
and A. Amir,
arXiv:1410.4771v2 (2014).
[13] S. Taheri-Araghi, S. Bradde, J.T. Sauls, N.S. Hill,
P.A. Levin, J. Paulsson, M. Vergassola, and S. Jun, Cur-
rent Biology 25, 385 (2015).
[14] A degradation rate that is faster than a typical cell in-
terdivisional time would likely simply modify the accu-
mulation exponent, so such proteins would also display
exponential increase of copy number during a cell cycle
and be covered equally well by our model. However, we
exclude such proteins for now to avoid complicating fac-
tors.
[15] J. Mannik,
Jaan, F. Wu, F.J. Hol, P. Bisic-
chia, D.J. Sherratt, J.E. Keymer, and C. Dekker,
Proc. Acad. Natl. Sci. USA 109, 6957 (2012).
[16] N could in principle be modulated on long timescales
by processes not included in our model. If they occur,
such modulations would be sufficiently slow to be de-
coupled from the fast growth-and-division processes de-
scribed here. It is therefore reasonable, and consistent
with observations, to take N to be constant over a large
number of generations.
[17] A priori, it is possible that α is not a constant but varies
between protein types or individuals, or it could even
fluctuate from generation to generation for a given pro-
tein in a single individual. We defer consideration of these
possibilities, and analyze here the case of constant α spe-
cific to a given protein in a given cell.
[18] O.G. Berg, J. Theoret. Biol. 71, 587 (1978).
[19] N. Brenner, K. Farkash and E. Braun, Phys. Biol. 3, 172
FIG. 3: Cell-size analysis. Similar analysis of that
carried out for protein copy number (Fig. 2a) was
carried out for cell-size data, quantified here by the
length L of the bacterial cell in the trapping channel.
∆ ln Li = ln Li+1 − ln Li, computed from the data in [8],
is plotted vs. ln Li. The red dashed line is the linear
best fit given by ∆ ln Li = −0.5185 ln Li + 0.702. The
slope, which represents the feedback parameter α, is
approximately 1/2 as predicted by the constant volume
(2006).
addition model described in [28].
[2] A. Raj and A. van Oudenaarden, Cell 135, 216 (2008).
[3] A. Sanchez and I. Golding, Science 342, 1188 (2013).
[4] H. Salman, N. Brenner, C.-K. Tung, N. Elyahu,
E. Stolovicki, L. Moore, A. Libchaber, and E. Braun,
Phys. Rev. Lett. 108, 238105 (2012).
[5] Note that we use the term universal to describe these sta-
tistical properties of cellular protein content. This should
not be confused with the controversy over the dependence
of mRNA production on promoter architecture (see, e.g.,
L-H. So, A. Ghosh, C. Zong, L.A. Sepulveda, R. Segev
and I. Golding, Nature Genetics 43, 554 (2011) and
D.L. Jones, R.C. Brewster and R. Phillips, Science 346,
1533 (2014)). Indeed, quantitative measurements have
shown that there is almost no correlation between mRNA
and protein copy numbers in the cell [6].
[6] Y. Taniguchi, P.J. Choi, G.-W. Li, H. Chen, M. Babu,
J. Hearn, A. Emili, and X.S. Xie, Science 329, 533
(2010).
[7] C. Furasawa, T. Suzuki, A. Kashiwagi, T. Yomo and K.
Kaneko, Biophysics 1, 25-31 (2005).
[8] N. Brenner, E. Braun, A. Yoney, L. Susman, J. Rotella,
and H. Salman, Eur. Phy. J. E 38, 102 (2015).
[9] E.O. Powell and F.P. Errington, J. Gen. Microbiol. 31,
315 (1963).
[10] S. Iyer-Biswas, C.S. Wrighta, J.T. Henryb, K. Loa, S.
Burov, Y. Linc, G.E. Crooksd, S. Crossonb, A.R. Dinner
and N.F. Scherera, Proc. Natl. Acad. USA 111, 15912
(2014).
[11] M. Osella, E. Nugent
and M.C. Lagomarsino,
Proc. Natl. Acad. Sci. USA 111, 3431 (2014).
[20] N. Brenner and Y. Shokef, Phys. Rev. Lett. 99, 138102
(2007).
[21] T. Friedlander and N. Brenner, Phys. Rev. Lett. 101,
018104 (2008).
[22] K. Hosoda, T. Matsuura, H. Suzuki and T. Yomo,
Phys. Rev. E 83, 031118 (2011).
[23] J. Paulsson, Physics of Life Reviews 2, 157 (2005).
[24] N. Friedman,
and X.
L. Cai
Sunney Xie,
Phys. Rev. Lett. 97, 168302 (2006).
[25] M. Campos, I.V. Surovtsev, S. Kato, A. Paintdakhi,
B. Beltran, S.E. Ebmeier C. Jacobs-Wagner, Cell 159,
1433 (2014).
[26] S. Iyer-Biswas, G.E. Crooks, N.F. Scherer and A.R. Din-
ner, Phys. Rev. Lett. 113, 028101 (2014).
[27] R. Pugatch, Proc. Natl. Acad. Sci. USA early edition,
www.pnas.org/content/early/2015/02/06/1418738112
(2015).
[28] A. Amir, Phys. Rev. Lett. 112, 208102 (2014).
[29] A. Bar-Even, J. Paulsson, N. Maheshri, M. Carmi,
E. O’Shea, Y. Pilpel, and N. Barkai, Nat. Gen. 38, 636
(2006).
[30] K.F. Murphy, G. Balazsi and J.J. Collins, PNAS 104
12726-12731 (2007).
[31] G. Hornung, R. Bar-Ziv, D. Rosin, N. Tokurkik, D.S.
Tawfik, M. Oren and N. Barkai, Genome Research 22,
2409-2417 (2012).
[32] M. Dadiani, D. van Dijk, B. Segal, Y. Field, G. Ben-
Artzi, T. Raveh-Sadka, M. Levo, I. Kaplow, A. Wein-
berger and E. Segal, Genome Research 23, 966976(2013)
[33] L.B. Carey, D. van Dijk, P.M.A. Sloot, J.A. Kaandrop
and E. Segal, PLoS Biology 11, e1001528 (2013)
[34] E. Sharon, D. van Dijk, Y. Kalma, L. Keren, O. Manor,
Z. Yakhini and E. Segal, Genome Research 24, 16981706
(2014).
7
|
1212.6113 | 3 | 1212 | 2013-03-14T07:17:58 | In-situ determination of the mechanical properties of gliding or non-motile bacteria by Atomic Force Microscopy under physiological conditions without immobilization | [
"physics.bio-ph",
"cond-mat.mes-hall"
] | We present a study about AFM imaging of living, moving or self-immobilized, bacteria in their genuine physiological liquid medium. No external immobilization protocol, neither chemical nor mechanical, was needed. For the first time, the native gliding movements of Gram-negative Nostoc cyanobacteria upon the surface, at speeds up to 900microns/h, were studied by AFM. This was possible thanks to an improved combination of a gentle sample preparation process and an AFM procedure based on fast and complete force-distance curves made at every pixel, drastically reducing lateral forces. No limitation in spatial resolution or imaging rate was detected. Gram-positive and non-motile Rhodococcus wratislaviensis bacteria were studied as well. From the approach curves, Young modulus and turgor pressure were measured for both strains at different gliding speeds and are ranging from 20 to 105MPa and 40 to 310kPa depending on the bacterium and the gliding speed. For Nostoc, spatially limited zones with higher values of stiffness were observed. The related spatial period is much higher than the mean length of Nostoc nodules. This was explained by an inhomogeneous mechanical activation of nodules in the Nostoc. We also observed the presence of a soft extra cellular matrix (ECM) around the Nostoc bacterium. Both strains left a track of polymeric slime with variable thicknesses. For Rhodococcus, it is equal to few hundreds of nm, likely to promote its adhesion to the sample. While gliding, the Nostoc secretes a slime layer the thickness of which is in the nanometer range and increases with the gliding speed. These results open a large window on new studies of both dynamical phenomena of practical and fundamental interests such as the formation of biofilms and dynamic properties of bacteria in real physiological conditions. | physics.bio-ph | physics | |
1608.06598 | 1 | 1608 | 2016-08-23T18:12:58 | Comment concerning Leonardo's rule | [
"physics.bio-ph"
] | In this comment we propose a novel explanation for the Leonardo's rule concerning the tree branching. According to Leonardo's notebooks he observed that if one observes the branches of a tree, the squared radius of the principal branch is equal to the sum of the squared radius of the branch daughters. | physics.bio-ph | physics |
Comment concerning Leonardo's rule.
Sotolongo-Costa, O.1, Gaggero-Sager, L. M.2,
Oseguera-Manzanilla, T.3, and D´ıaz-Guerrero, D. S.1
1CInC-Autonomous University of Morelos State, Cuernavaca,
Morelos.
2CIICAp-Autonomous University of Morelos State,
Cuernavaca, Morelos.
3CIDC-Autonomous University of Morelos State, Cuernavaca,
Morelos.
September 26, 2018
Leonardo da Vinci's rule concerning tree branching states that "all the
branches of a tree at every stage of its height when put together are equal in
thickness to the trunk" [1], i.e., if L is the radius of a branch which ramifies
in two branches of radius, say, L1 and L2, then L2 ≈ L2
2. With this
obsevation Leonardo establishes a qualitative link between the morphology
of ramification and the flow of sap.
1 + L2
In [2] it is proposed that Leonardo's rule is a consequence of the self-
similarity of the tree trunk and wind-induced stress. In the mentioned paper
some ad hoc hypothesis, based in fractal theory and fracture theory, were
introduced in order to obtain Leonardo's law.
However, it is curious that Leonardo himself proposes a very simple ex-
planation of this rule based on the characteristics of fluid motion. "When
a branch grows, Leonardo argues, its thickness will depend on the amount
of sap it receives from the one below the branching point. In the tree as a
whole, there is a constant flow of sap, which rises up through the trunk and
divides between the branches flowing through succesive ramifications. Since
1
the total quantity of sap carried by the tree is constant, the quantity carried
by each branch will be proportional to its cross section, so the total cross
section at each level will be equal to that of a trunk" [4].
This argument is naive, since the flux of the sap, as viscous liquid, is not
proportional to the cross section of the branch, but to the fourth power of
its radius, if we consider branches and trunk as cylinders.
In our opinion, though Leonardo's rule is not a trivial result from fluid
mechanics, a simple and direct approach, based on mass conservation, can
be made to explain this observation.
The amount of fluid that goes through a conduit of this type is given by
the Poiseuille's Law
Q(R) =
π∇P
8η
R4. Q is the flux.
(1)
Now let us add the fact that the xilema of a tree is composed of several
conduits with radii distributed according to a given law. In each of those
conduits the flow obeys Poiseuille's law. Hence the total amount of fluid in
a branch of width L that contains a distribution of conduits per radius f (R)
is
(cid:90) L
QT =
Q(R)f (R)dR
(2)
0
and if we assume that the main part of the distribution of conduits per
radius, i. e., those that transport most of the sap, is of type f (R) ≈ R−x
(scaling), then the integral gives
QT ≈ L5−x.
(3)
Hence, in order to get Leonardo's rule we must have that x = 3.
If we rely on the assumption that branch structure is determined by flow
dynamics in a power law distribution of sap conduits, Leonardo's rule can be
understood in a very simple way. Therefore, let us make the following single
hypothesis: the distribution of conduits per size (radius) is such that their
histogram have a heavy tail.
2
(a) Histogram of the conduit radius
taken from [3]. For betula pendula roth
note the heavy tail of the distribution.
The tail obeys a scaling law R−x start-
ing from the maximum (pointing ar-
row).
(b) A curve for the heavy tail of the
histogram. Taking the frequency from
1a we adjust the R−x curve that bet-
ter fits the data in order to obtain the
value of x.
This hypothesis has experimental validation.
Indeed, in [3] the betula
pendula roth xylema distribution is presented. The distributions were mea-
sured for different parts of the trees. In all cases the distributions are far
from the gaussian, exhibiting asymmetry and a heavy tail. In this case we
are interested only in the tail of the distribution since the total flux of the
tree is determined by the larger vessels, which carry most of the sap.
In
figure 1a we reproduce one of their results, in this case for the branches of
the petiole (a slender stem that supports the blade of a foliage leaf). The
comparison of the data with the measurment gives a correlation coefficient
of 0.98.
So, it is possible to take from this distribution the corresponding one for
vessels larger than the maximum value, since the main flow occurs through
them. Indeed, in this case the vessels with radius larger than 5 µm distribute
as shown in figure 1b. The distribution clearly exhibits a fat tail characteristic
of Levy distributions. The adjustment of equation 3 with the data gives
x = 3.17. What gives according to 2 a satisfactory agreement with Leonardo's
law.
This result, at least for the case of betula pendula roth, is not casual. Other
parts if this species were measured and always the xilema distribution is such
3
that the self-similarity of the tail is evident. We have no notice of xilema
measurements in other trees, though the ubiquity of Levy distributions in
nature allows us to think that Leonardo's law have a simpler explanation
than that proposed in [2].
In our opinion, other "rules" by Leonardo could be understood starting
from basic laws of physics. Some of our results will be published elsewhere.
References
[1] Richter, J. P. The Notebooks of Leonardo da Vinci (Dover, New York,
1970).
[2] Eloy Christophe Leornardo's Rule, Self-similar, and Wind-Induced
Stresses in Trees. Phys. Rev. Lett. 107 (2011).
[3] ´Atala, C. and Lusk, C. H. Xylemanatomyof Betulapendula Roth-
saplings: Relationship To Physical Vascular Models. Gayana Bot. 65
(2008).
[4] Capra, F. Learning from Leonardo. (Berrett-Koehler Publishers, Inc.),
2013.
[5] Evert, R. F. and Eichhorn, S. E. Raven Biology of Plants. (W. H. Free-
man and Company), 2013.
4
|
1512.08589 | 1 | 1512 | 2015-12-29T03:02:23 | Quantitative analysis of the nanoscale intra-nuclear structural alterations in hippocampal cells in chronic alcoholism via transmission electron microscopy study | [
"physics.bio-ph"
] | Chronic alcoholism is known to alter morphology of hippocampal, an important region of cognitive function in the brain. We performed quantification of nanoscale structural alterations in nuclei of hippocampal neuron cells due to chronic alcoholism, in mice model. Transmission electron microscopy images of the neuron cells were obtained and the degrees of structural alteration, in terms of mass density fluctuations, were determined using the recently developed light localization analysis technique. The results, obtained at the length scales ranging from 33 to 195 nm, show that the 4-week alcohol fed mice have higher degree of structural alteration in comparison to the control mice. The degree of structural alterations starts becoming significantly distinguishable around 100 nm sample length, which is the typical length scale of the building blocks of cells, such as DNA, RNA, etc. Different degrees of structural alterations at such length scales suggest possible structural rearrangement of chromatin inside the nuclei. | physics.bio-ph | physics | Quantitative analysis of the nanoscale intra-nuclear
structural alterations in hippocampal cells in chronic
alcoholism via transmission electron microscopy study
Peeyush Sahaya, Pradeep Shuklab, Hemendra M. Ghimirea, Huda Almabadia,
Vibha Tripathia, Samarendra K. Mohantyc, Radhakrishna Raob, Prabhakar Pradhana*
aDepartment of Physics, BioNanoPhotonics Laboratory, University of Memphis, Memphis, TN, USA
bDepartment of Physiology, University of Tennessee Health Science Center, Memphis, TN, USA
cBiophysics and Physiology Group, Department of Physics, University of Texas, Arlington, TX, USA
*Corresponding author: [email protected]
ABSTRACT
Chronic alcoholism is known to alter morphology of hippocampal, an important region of cognitive
function in the brain. We performed quantification of nanoscale structural alterations in nuclei of
hippocampal neuron cells due to chronic alcoholism, in mice model. Transmission electron microscopy
images of the neuron cells were obtained and the degrees of structural alteration, in terms of mass density
fluctuations, were determined using the recently developed light localization analysis technique. The
results, obtained at the length scales ranging from 33 – 195 nm, show that the 4-week alcohol fed mice
have higher degree of structural alteration in comparison to the control mice. The degree of structural
alterations starts becoming significantly distinguishable ~100 nm sample length, which is the typical
length scale of the building blocks of cells, such as DNA, RNA, etc. Different degrees of structural
alterations at such length scales suggest possible structural rearrangement of chromatin inside the nuclei.
INTRODUCTION
Hippocampus is an important part of human brain, which deals with brain functions, such as
storing memories, emotional responses, navigation, etc.1-3. Morphological and chemical changes in
hippocampal cells/tissues can significantly affect brain functions. External factor such as consumption of
alcohol, including short term as well as long term, are known to bring such changes in the hippocampal
cells/tissues, which has both, temporal and chronic adverse effects. One of the important cell categories in
the hippocampal cells is the neural cell. Studies show that, as an immediate effect of the consumption of
alcohol impedes the chemical signaling process in neural cells leading to alteration in brain’s perception,
which results in immediate abnormal behavioral acts. A significant body of evidence indicates that
alcohol consumption causes damages to central nervous system4. However, a long-term consumption of
alcohol can result in sustained structural damage to hippocampal cells/tissues potentially resulting in
permanent morphological changes in the hippocampus volume5-6. Hippocampus being an important
region of cognitive function is also a primary ethanol sensitive region in the brain7. Evidence indicates
that the impairment of hippocampal integrity caused by alcohol consumption is caused by attenuation of
neurogenesis8-11. Such structural change in hippocampus can result in neuropathological disorder,
severely affecting day-to-day life activities of a person. Therefore, there has been a significant interest in
studying the effect of alcohol on hippocampal cells.
In this work we are interested in analyzing the morphological changes (i.e. structural alterations)
in the hippocampal cells as a result of long term exposure of alcohol, in alcoholic mice model. In
particular, we are quantifying structural alterations in the nuclei of hippocampal neuron cells by
employing a recently developed powerful light localization analysis approach, which has nano-scale level
of detection sensitivity12-13. In this approach, the structural alteration is quantified in terms of the mass
density fluctuations. In particular, the quantification measurement is performed in a single parameter
called inverse participation ratio (IPR) values, which takes care all the structural heterogeneity of the
system, and is termed as degrees of structural disorder Lsd. A detailed discussion on the technique and
various terminologies associated with it are presented in the following sections.
The IPR approach is a very powerful technique, used in condensed matter physics, for
analyzing heterogeneously disordered system via statistical analysis of spatially localized eigenfunctions
of the light wave in the media14-15. The IPR value, for an eigenfunction E, is defined as IPR =∫
(in the unit of inverse area). For an optical lattice system, if E is its one of the eigenfunctions of the
Hamiltonian, then the value of the IPR measures the degree of spatial localization of that eigenfunction. It
has been shown that the strength/degree of light localization in a weakly optical disordered media
increases linearly with the increase in the strength of the disorder of the system. Therefore, IPR value
provides a measure of the degree of disorder inside an optical lattice system. The average value of the IPR
at specific length scale of a uniform lattice is a fixed universal number (~2.45), but the value linearly
increases with increase in the degree of disorder in the media. The quantum mechanical concept of IPR
and its applications in characterizing heterogeneously disordered media, in a single parameter, can be
seen elsewhere14-21. In TEM imaging, electron wave interaction with different parts of the sample
modulates the intensity at the image plane. It has been shown that the value at the spatial location of the
voxel (x, y) is ITEM (x, y) = I0 TEM + dITEM(x,y), which is related to the refractive index: n(x,y) =n0 + dn(x,y)
=ρ0 + β.ρ(x,y), where I0 and n0 are the respective mean values of TEM intensity and refractive index
medium surrounding a scattering structure, ρ(x,y) is local mass density of the biological media, and β
(≈0.185 for most biological molecules found in living cells) is the specific refraction increment.
Therefore, intensity of fluctuation in TEM imaging is reflected in refractive index fluctuations by: n(x,y)
α M(x,y) α I(x,y) 13, 22-23.
To obtain the eigenfunctions of the optical system, the Hamiltonian of the system is generated
from the optical lattice. To do this, we use Anderson’s disordered tight binding model (TBM) optical
Hamiltonian, which has been well studied in condensed matter physics and proven to be a good model to
quantify eigenfunction localization of the system of any type or geometry and disorder. A TBM
Hamiltonian with i> and j> optical wave functions for respective ith and jth lattice sites becomes:
, (1)
where t is the overlap integral between ith and jth sites. To determine the effective optical potential of ith
optical lattice, εi for the voxel around the point (x, y) can be written as εi α dn(x,y)/no = dITEM/I0 , since
dn(x,y)<< n(x, y) and dITEM(x, y) <<I0. Now, we can define average value of IPR where the average is
taken over all N eigenfunctions of a sample size LxL (i.e., <IPR(L)>sample) in an optical lattices system as
, (2)
where is the ith eigenfunction of the Hamiltonian of optical lattice of size L × L, and dx = dy = a, is the
length scale of each smallest pixel. Total number of eigenfunctions N = (L/a)2. For the average IPR, the
degree of structural disorder Lsd can be written as
. (3)
For the sake of simplicity, we will use the proportionality constant in the above equation as 1.
All work with animals was approved by the Animal Care and Use Committee of the University of
Tennessee UT, Health Science Center, Memphis, TN, USA, in accordance with institutional and U.S.
federal guidelines for animal experimentation. Adult female mice (C57BL6; 10-12 weeks) were caged
with 2 in each group and administered with or without alcohol in their diet. One group of mice were fed a
Lieber-DeCarli liquid diet (Dyets Inc., Bethlehem, PA) with f1-6% ethanol (0% for 2 days, 1% for 2 days,
2% for 2 days, 4% for one week, 5% for one week and 6% for one week), while the control group of
animals in each group were pair-fed with an isocaloric diet adjusted with maltodextrin. Animals were
iijiijjitiiH)(NiLLiSampledxdyyxENLIPR1004),(1)(sdPixelLLIPR)( euthanized and hippocampal samples were collected and fixed to suitable chemicals according to our
probing instrument.
Biopsy samples obtained from hippocampal areas of mice were fixed more than two hours in 0.1
M Na cacodylate buffer (pH 7.2 to 7.4) with 2.5 percent glutaraldehyde and 2.5 percent
paraformaldehyde. These fixed samples were then washed with several changes of 0.1 M cacodylate
buffer (pH 7.2 to 7.4) to preserve structure. We post-fixed this sample with 2 percent osmium tetroxide in
0.1 M Na cacodylate buffer for 1-2 hours and rinsed it with several changes of 0.1 M cacodylate buffer
(pH 7.2- 7.4). Following the standard protocol, samples were en bloc stained with aqueous UA and
dehydrated through ethanol series every 15 minutes after rinsing with deionized water. For further
dehydration, samples were embedded in polymer resin containing EPON to ensure sufficient stabilization
for ultrathin sectioning. Thoroughly cleaned and concentrated samples prepared above were then
sectioned with an ultra-microtome to the thickness of 70 nm so that the sample would be electron
transparent and mechanically robust. Since the resultant specimen was sufficiently thin to allow the
penetration of TEM electron beams, it was ready for imaging.
We used the classical Joel JEM-1200 TEM with its higher magnification (3000x) and working
potential of 60 kV. Acquired wavelength by incident electron beam in TEM is less than nanometer, and
we were able to get resolution down to fractions of a nanometer. TEM data were recorded in the form of
images, i.e., 2D micrographs. Several (~12-15) neuron cells from hippocampal area were selected for the
study via TEM imaging of the biopsy samples. The intra-nuclear disorder analysis was performed for the
nucleus part of the cells after cropping them out from the whole cell image. A refractive index lattice
system, ‘optical lattice’, is created from the TEM image intensity data (as described in the Method).
Subsequently, degrees of intra-nuclear disorder, <Lsd>, in terms of IPR values, were determined at
different length scales for the each lattice system. In particular, sample length L dependent average
structural disorder <Lsd(L)> values were determined at length scales ranging from ~ 33 – 200 nm. For
every cell, the IPR value was calculated at each length scales, and were averaged over the whole cell to
obtain a mean <IPR(L)>pixel value for the cell. Subsequently, an overall mean <<IPR>pixel>cell ~<Lsd>
value for each category of cell was determined by taking the average of <IPR(L)>pixel values for all the
cells of that category. Similarly, mean standard deviation in the <Lsd> values, (σ(Lsd)), was also
calculated for both, i.e., the alcoholic and non-alcoholic categories of the cells. The results are presented
in the following section.
RESULT S AND DISCUSSION
Figures 1(a) and 1(a’) show the actual TEM images of nucleus of sample neuron cells of non-alcoholic
and alcoholic mice, respectively. Conversely, figures 1(b) and 1(b’) are their respective <Lsd (L)> images
computed at length L=130 nm. The color map from blue to red represents increasing order of degree of
disorder, i.e., the <Lsd > values, inside the nucleus. As it can be seen that, the neuron cell nucleus of the
alcoholic mice has more hot spots compared to the non-alcoholic mice suggesting that there is higher
degree of disorder, i.e., mass density fluctuation inside them. A better angle view of the disorder map for
both the cells has been represented in fig 1(c) and 1(c’). The regions of higher peaks are the areas where
Control
Alcoholic
(a)
(a’)
(b)
(b’)
(c)
(c’)
TEM
Images
Lsd
Images
Figure 1. (a) and (a’) are representative TEM images of nucleus from hippocampus of normal and chronic alcoholic
mouse. (b) and (b’) are corresponding <Lsd >-sample images for pixel size LL=130 nm x 130 nm (TEM resolution
6.52 nm). (c) and (c’) are respective 3-dimensional views for the Lsd (x,y) at same length..
the fluctuation in mass density, i.e., dρ(x,y), is higher compared to average mass density of the cell. It is
to be noted that the term dρ(x,y) represents the strength of the mass density fluctuation, therefore, the
region of high peaks equally indicates about either a lower or higher mass density, in comparison to the
average mass density of the cell nucleus, in that area.
Fig. 2 shows the average disorder, <Lsd>, measured at sample length 163 nm. For the mice treated
regularly with alcohol, the <Lsd> value at 163 nm length scale was determined to be 4.05, where
the <Lsd > value for the non-alcoholic (control) mice was found to be 3.77. The student’s t test obtained a
(a)
4.1
p=0.0015
3.95
>
d
s
L
<
3.8
3.65
3.5
(b)
p=0.05
)
d
s
L
(
σ
0.5
0.45
0.4
0.35
0.3
Figure 2.(a). Average Lsd bar plot at pixel size L × L = 163 nm × 163 nm and ensemble averaging for (i)
hippocampus neural cell nuclei of control mice (control) and (ii) chronic alcoholic mice (EA treated). (b)
Corresponding standard deviation σ(Lsd) at L × L = 163 nm × 163 nm.
p-value of 0.0015, suggesting that the average degrees of disorder determined in both the cases are
significantly different. The results for standard deviation (std) in the disorder, σ(Lsd), at the same sample
length, i.e., L×L = 163 nm×163 nm, is shown in Fig. 2(b).The results clearly exhibit that the alcoholic
mice cells have considerably higher standard deviation in their σ(Lsd) values. This suggests that the
alcoholic mice have higher mass density fluctuations in their neuron cell nucleus in comparison to the
control mice. Subsequently, we also compared the <Lsd (L)> values in alcoholic and control mice at
different sample lengths L. The Fig. 3(a) shows the plot of mean <Lsd (L)> vs. L for different sample
lengths ranging from 33 nm to 195 nm. As it can be seen in Fig. 3(a), when the sample length L is
increased from L= 163 nm, the mean <Lsd (L)> values also increases, as well as the differences between
the <Lsd (L)> values of alcoholic mice and the
(a)
4.5
4
>
d
s
L
<
3.5
3
2.5
2
(b)
)
d
s
L
(
σ
0.6
0.48
0.36
0.24
0.12
Control+EA
Control
Control+EA
Control
33
65
98
130
163
195
Sample length L (nm)
33
65
98
163
Sample Length L (nm)
130
195
Figure 3(a). Graphical representation of ensemble-averaged values of <Lsd (L)> versus L (in nm) plots for (i) nuclei
of normal mice (ii) nuclei of alcoholic mice. (b) Corresponding standard deviation σ(Lsd) versus L (in nm) plots for
nuclei of normal mice and alcoholic mice.
non-alcoholic mice. Further in Fig. 3(a), it can be seen that <Lsd (L)> values decreases with the decrease
in the sample L; accordingly the difference between the <Lsd(L)> values for alcoholic mice and the control
mice decreases as well. It is worth pointing out here that the visible difference is prominent around the
length scale ~100 nm, which is the typical dimension of the building blocks of the cell nuclei, such as
DNA, RNA. etc.
It should be noted that the degree of disorder, <Lsd(L)>, which is measured in terms of IPR
numbers and calculated at a particular sample length L, basically measures the degree of spatial
inhomogeneity in the refractive index distribution (and, thus in the mass density distribution) inside the
sample area size L × L. Therefore, it is reasonable to expect the <Lsd (L)> values will increase with large
L, as the strength of inhomogeneity increases with increase in L till a saturation value of the IPR is
reached at very large L12-13. Expectedly, therefore, at higher length scales, the <Lsd(L)> values will
increase, and so will the difference between the <Lsd(L)> values of two disordered system. A similar trend
can be seen for the standard deviation (std) in <Lsd(L)> values, σ(Lsd), for both the cases, Fig. 3(b). It is
interesting to note that the present system (i.e., the biological sample) exhibits perfect example of a
weakly disordered system where the mean and the standard deviation of disorder parameter increases with
the increase in the sample size till a saturation is reached, implying that the nanoscale structural disorder
is increasing with alcoholism. Such systems are well characterized by the IPR method, based on
mesoscopic physics.
In terms of mass density fluctuations, a higher <Lsd(L)> value at a sample length L indicates a
higher mass density fluctuations, or rearrangement, in spatial mass density distribution in the sample area
L × L. Therefore the present results suggest that the alcoholic mice have higher mass density fluctuation
inside the nuclei of their hippocampal neural cells compared to the non-alcoholic (control) mice.
Similarly, a higher standard deviation in Lsd,, σ(Lsd(L)), values in alcoholic mice suggest a more spread in
mass density fluctuations in them compared to the non-alcoholic mice. It is important to point out here
that, the nucleus inside a cell is densely packed with the chromatin coils. Therefore, the results obtained in
this study indicate about a possible rearrangement in the chromatin structure. Changes in protein
expression profiles have been recorded in the hippocampus of alcoholic subjects 24. Additionally, it is also
interesting to point out here is that a typical diameter of chromatin coil is ~30 nm. Our measurement
shows not a significant difference in the <Lsd(L)> values at this length scales in the two cases, however
aroung three times of this length scale. These results invite more research in this direction in order to
elucidate the actual effects of chronic alcoholism on brain hippocampal neuron cells.
In summary, we studied quantification of nanoscale structural alteration, in terms of mass density
fluctuations, in hippocampal neuron cells as an effect of chronic alcohol consumption. Using the powerful
IPR analysis approach, the study was performed on TEM images of hippocampal neuron cells of the non-
alcoholic (control) and alcoholic mice. The structural disorders were evaluated inside the nucleus of the
hippocampal neuron cells at various length scales ranging from 33 – 195 nm. The results show higher
degree of structural alterations in alcoholic mice in comparison to the control mice, which suggests about
possible rearrangement of chromatin structure inside the nucleus.
ACKNOWLEDGEMENT:
This work was supported by NIH R01 EB003682 and UofM (PPradhan), and NIH AA12307 and
DK55532 (RKRao).
REFERENCES:
1. Andersen, P. et al., eds., The Hippocampus Book (Oxford University Press, 2006),
2. O’Keefe, J. & Nadel, L. The Hippocampus as a Cognitive Map (Oxford University Press, 1978)
3. Burgess, N., Maguire, E.A., & O’Keefe, J. The Human Hippocampus and Spatial and Episodic
Memory. Neuron 35, 625–641 (2002).
4. Walker D.W., Barnes D.E., Zornetzer S.F., Hunter B.E., & Kubanis P. Neuronal loss in
hippocampus induced by prolonged ethanol consumption in rats. Science 209, 711–713 (1980).
5. Riley, J.N., & Walker, D.W. Morphological Alterations in Hippocampus after Long-Term
Alcohol Consumption in Mice. Science 201, 646-648 (1978).
6. McMullen, P.A., Saint-Cyr, J.A., & Carlen, P.L. Morphological Alterations in Rat CA1
Hippocampal Pyramidal Cell Dendrites Resulting from Chronic Ethanol Consumption and
Withdrawal. The Journal of Comparative Neurology 225, 111–118 (1984).
7. Vilpoux, C., Warnault, V., Pierrefiche, O., Daoust, M., & Naassila, M. Ethanol-sensitive brain
regions in rat and mouse: a cartographic review, using immediate early gene expression. Alcohol
Clin Exp Res 33, 945–969 (2009).
8. Collins M.A., Corse T.D., & Neafsey E.J. Neuronal degeneration in rat cerebrocortical and
olfactory regions during subchronic “binge” intoxication with ethanol: possible explanation for
olfactory deficits in alcoholics. Alcohol. Clin. Exp. Res. 20, 284–292 (1996).
9. Bengochea O., & Gonzalo L.M., Effect of chronic alcoholism on the human hippocampus. Histol.
Histopathol. 5, 349–357 (1990).
10. Geil, C.R., Hayes, D.M., McClain, J.A., Liput, D.J., Marshall, S.A., Chen, K.Y., & Nixon, K.
Alcohol and adult hippocampal neurogenesis: Promiscuous drug, wanton effects. Prog
Neuropsychopharmacol. Biol. Psychiatry 3, 103–113 (2014).
11. Nixon K, Kim D.H., Potts E.N., He J., & Crews F.T. Distinct cell proliferation events during
abstinence after alcohol dependence: microglia proliferation precedes neurogenesis. Neurobiol
Dis. 31, 218–29 (2008).
12. Pradhan, P., Damania, D., Joshi, H.M., Turzhitsky, V., Subramanian, H., Roy, H.K., Taflove, A.,
Dravid, V.P., Backman, V. Quantification of nanoscale density fluctuations using electron
microscopy: Light-localization properties of biological cells. Appl. Phys. Lett. 97, 243704 (2010).
13. Pradhan, P., Damania, D., Joshi, H.M., Turzhitsky, V., Subramanian, H., Roy, H.K., Taflove, A.,
Dravid, V.P., Backman, V. Quantification of nanoscale density fluctuations by electron
microscopy: probing cellular alteration in early carcinogenesis. Phys. biol. 8.2, 243704 (2011).
14. Pradhan, P. & Sridhar, S. Correlations due to Localization in Quantum Eigenfunctions of
Disordered Microwave Cavities. Phys. Rev. Lett. 85, 2360–2363 (2000).
15. Pradhan, P. & Sridhar, S. From chaos to disorder: Statistics of the eigenfunctions of microwave
cavities. Pramana 58, 333–341 (2002).
16. Lee, P.A. & Ramakrishnan, T.V. Disordered electronic systems. Rev. Mod. Phys. 57, 287–337
(1985).
17. Abrahams, E., Anderson, P.W., Licciardello, D.C., & Ramakrishnan, T.V. Scaling theory of
localization—Absence of quantum diffusion in two dimensions. Phys. Rev. Lett. 42, 673–676
(1979).
18. Kramer, B & Mackinnon, A. Localization—theory and experiment. Rep. Prog. Phys. 56, 1469–
1564 (1993).
19. Prigodin, V.N. & Altshuler, B.L. Long-range spatial correlations of eigenfunctions in quantum
disordered systems. Phys. Rev. Lett. 80:9, 1944 (1998).
20. Murphy, N.C., Wortis, R. & Atkinson, W.A. Generalized inverse participation ratio as a possible
measure of localization ratio as a possible measure of localization for interacting systems. Phys.
Rev. B 83:18, 184206 (2011).
21. Jackson, S.R., Khoo, T.J., & Strauch, F.W. Quantum walks on tree with disorder: Decay,
diffusion and localization. Phys. Rev. A, 86:2, 022335 (2012).
22. Schmitt, J.M., & Kumar, G., Turbulent nature of refractive-index variations in biological tissue.
Opt. Lett. 21, 1310-1312 (1996).
23. Barer, R., Ross, K.F.A., and Tkaczyk, S. Refractometry of Living Cells. Nature 171, 720–724
(1953).
24. Matsumoto, H & Matsumoto, I., Alcoholism: Protein Expression Profiles in a Human
Hippocampal Model. Expert Review of Proteomics 5:2, 321–231, (2008).
|
1510.02943 | 1 | 1510 | 2015-10-10T14:35:24 | Defining the Free-Energy Landscape of Curvature-Inducing Proteins on Membrane Bilayers | [
"physics.bio-ph",
"cond-mat.soft",
"cond-mat.stat-mech"
] | Curvature-sensing and curvature-remodeling proteins are known to reshape cell membranes, and this remodeling event is essential for key biophysical processes such as tubulation, exocytosis, and endocytosis. Curvature-inducing proteins can act as curvature sensors as well as induce curvature in cell membranes to stabilize emergent high curvature, non-spherical, structures such as tubules, discs, and caveolae. A definitive understanding of the interplay between protein recruitment and migration, the evolution of membrane curvature, and membrane morphological transitions is emerging but remains incomplete. Here, within a continuum framework and using the machinery of Monte Carlo simulations, we introduce and compare three free-energy methods to delineate the free-energy landscape of curvature-inducing proteins on bilayer membranes. We demonstrate the utility of the Widom test-particle/field insertion methodology in computing the excess chemical potentials associated with curvature-inducing proteins on the membrane-- in particular, we use this method to track the onset of morphological transitions in the membrane at elevated protein densities. We validate this approach by comparing the results from the Widom method with those of thermodynamic integration and Bennett acceptance ratio methods. Furthermore, the predictions from the Widom method have been tested against analytical calculations of the excess chemical potential at infinite dilution. Our results are useful in precisely quantifying the free-energy landscape, and also in determining the phase boundaries associated with curvature-induction, curvature-sensing, and morphological transitions. This approach can be extended to studies exploring the role of thermal fluctuations and other external (control) variables, such as membrane excess area, in shaping curvature-mediated interactions on bilayer membranes. | physics.bio-ph | physics | Defining the Free-Energy Landscape of Curvature-Inducing
Published as Phys. Rev. E 90, 022717 (2014)
Proteins on Membrane Bilayers
Richard W. Tourdot∗
Department of Chemical and Biomolecular Engineering,
University of Pennsylvania, Philadelphia, PA, 19104
N. Ramakrishnan†
Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, 19104
Ravi Radhakrishnan‡
Corresponding Author; Department of Chemical and Biomolecular Engineering,
Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, 19104
(Dated: July 1, 2021)
5
1
0
2
t
c
O
0
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
3
4
9
2
0
.
0
1
5
1
:
v
i
X
r
a
1
Abstract
Curvature-sensing and curvature-remodeling proteins, such as Amphiphysin, Epsin, and
Exo70, are known to reshape cell membranes, and this remodeling event is essential for key
biophysical processes such as tubulation, exocytosis, and endocytosis. Curvature-inducing pro-
teins can act as curvature sensors; they aggregate to membrane regions matching their intrinsic
curvature; as well as induce curvature in cell membranes to stabilize emergent high curvature,
non-spherical, structures such as tubules, discs, and caveolae. A definitive understanding of the
interplay between protein recruitment and migration, the evolution of membrane curvature, and
membrane morphological transitions is emerging but remains incomplete. Here, within a contin-
uum framework and using the machinery of Monte Carlo simulations, we introduce and compare
three free-energy methods to delineate the free-energy landscape of curvature-inducing proteins
on bilayer membranes. We demonstrate the utility of the Widom test-particle/field insertion
methodology in computing the excess chemical potentials associated with curvature-inducing
proteins on the membrane -- in particular, we use this method to track the onset of morpho-
logical transitions in the membrane at elevated protein densities. We validate this approach
by comparing the results from the Widom method with those of thermodynamic integration
and Bennett acceptance ratio methods. Furthermore, the predictions from the Widom method
have been tested against analytical calculations of the excess chemical potential at infinite di-
lution. Our results are useful in precisely quantifying the free-energy landscape, and also in
determining the phase boundaries associated with curvature-induction, curvature-sensing, and
morphological transitions. This approach can be extended to studies exploring the role of ther-
mal fluctuations and other external (control) variables, such as membrane excess area, in shaping
curvature-mediated interactions on bilayer membranes.
PACS numbers: 87.16.-b,87.17.-d
∗ [email protected]
† [email protected]
‡ [email protected]
2
I.
INTRODUCTION
Membranes constitute the boundary of all cells and cell organelles; these structures are
primarily composed of a lipid bilayer. The curvature of a membrane (i.e., the curvature
of the lipid bilayer) is considered to play an active role in controlling the spatial inho-
mogeneity and functionality in cells. Several membrane bound proteins are thought to
be involved in generating and regulating membrane curvature, while many others sense
background membrane curvature generated through other means. The mechanisms of
membrane curvature generation and sensing have been classified into several categories
based on their distinct qualitative features [1, 2]. They include (1) protein scaffolding:
in this mechanism multiple proteins locally concentrate to a region of the membrane and
induce curvature by virtue of an intrinsic curvature in their membrane facing domains [3],
(2) hydrophobic insertion: in this mechanism, the involved proteins insert their hydropho-
bic domains into the membrane bilayer (also known as wedging) to generate curvature
[2], and (3) oligomerization: certain proteins, which cannot induce or sense membrane
curvature individually, associate into oligomeric domains and induce curvature coopera-
tively [2, 3]. Examples of curvature-inducing proteins include families of proteins with
membrane adjacent BAR (Bin/Amphiphysin/RVs) domains [4 -- 6]. BAR domains are
crescent-shaped α-helical bundles that bind to the membrane bilayer mainly through the
processes of protein scaffolding and hydrophobic insertion. Based on their detailed struc-
tures the BAR domains are further sub-classified into classical BAR, N-BAR, F-BAR, etc.
Another example is the dynamin family of proteins, which are comprised of PH domains.
A third example corresponds to proteins that employ hydrophobic insertion mechanism
to generate curvature. Typically, these proteins have an intrinsically-disordered struc-
ture. Upon binding to membrane they undergo a folding transition to form amphipathic
α-helices which are buried inside a leaflet of the bilayer -- specific examples include ENTH
and ANTH domain containing proteins [1 -- 3]. For further discussion, we refer to a recent
review article on mechanisms of curvature induction by proteins on a bilayer membrane
[7].
Most theoretical studies on protein binding are concerned with the adsorption on planar
lipid bilayers (reviewed in [1]); these studies are mainly concerned with planar membranes
and curvature effects were not discussed in these studies. Reynwar et al.
[8] performed
3
coarse-grained molecular dynamics simulations which show that once adsorbed onto lipid
bilayers, curvature-inducing proteins experience an effective curvature-mediated attrac-
tive interactions. Jiang and Powers [9] investigated lipid sorting induced by curvature for
a binary lipid mixture using a phase-field model. Das and co-workers have investigated
the effect of protein sorting on tubular membranes using theoretical techniques [10, 11].
Using Monte Carlo simulations, Sunil Kumar and coworkers [12 -- 16] studied the effects of
membrane shape transitions and protein-induced anisotropic bending elasticity and cur-
vature have on the shape of vesicles and the distribution of proteins on them. Using the
Monte Carlo method, Liu et al., and Ramanan et al. have investigated the spatial seg-
regation, curvature-sensing, and vesiculation in bilayers with curvature-inducing proteins
[17, 18]. Along with these computational studies, several experimental studies have been
carried out to investigate curvature generation.
Sorre and coworkers [19, 20] have conducted experimental investigations into the sorting
of lipids on a lipid membrane tube (tether) drawn from a giant unilamellar vesicle (GUV)
using an optical trap. Curvature sorting of lipids and its influence on the bending stiffness
of the bilayer membrane was studied by Tian et al [21, 22]. Dynamic sorting of lipids and
proteins has been studied by Heinrich and coworkers [23]. These authors observed that
the nucleation of disordered membrane domains occurs at the junction between the tether
and GUV. Several other theoretical and experimental studies have helped shed light on
the phenomena of curvature-mediated sorting [23 -- 26].
In general, curvature-inducing proteins can act as curvature sensors and aggregate on
the curved regions of the membrane [15]. In this way cells can perform protein and lipid
sorting for subsequent functions. The composition of lipids in a membrane may also mod-
ulate the curvature-sensing and curvature-generation activities of the proteins. Regulation
of membrane curvature and its sensing is also important in understanding the underlying
cellular physiology governed by trafficking, especially in the context of health conditions in
humans. A definitive understanding of the interplay between protein binding/migration
and membrane curvature evolution is emerging but remains incomplete. The mechanisms
that underpin such behavior are hugely important in intracellular assembly and stability
of organelles (which often sustain extreme curvatures). These mechanisms are also impor-
tant in intracellular transport, and sorting of proteins and cargo. Though many aspects of
these fundamental processes are well-characterized from a molecular biology perspective,
4
especially in the domain of protein-protein interactions and increasingly in the area of
protein localization, several open questions remain unanswered. These form the basis for
a complete understanding of the underlying mechanisms in these fundamental ("unit")
cellular process from a biophysical and thermodynamic perspective. The emerging picture
from a wide array of recent studies is that molecular interactions between the protein and
the lipids at the molecular scale directly determine the morphology of cellular membranes
at the micron scale primarily by setting up curvature fields [27 -- 31]. Determination and
characterization of these curvature fields is a challenging task [27, 32 -- 35].
In cell membranes, protein-induced radius of curvature ranges from a few nanometers
to a few tens of nanometers depending on the protein and lipid composition of the mem-
brane. For example, N-BAR domains stabilize curvature regions with radius of mean
curvature of 6.25 nm [36], while dynamin induced tubes have radius of 25 nm [37]. In
vitro experiments have reported epsin induced tubulation of lipsomes with average tubule
radius of 10 nm [32]. How these dimensions are related to the curvature induced by just
one functional unit (i.e., the minimal oligomer with ability to induce persistent curvature)
is not known. Multiscale modeling studies have been recently carried out to shed light
into this important question [27, 31, 32]. The membrane mediated interactions between
the different curvature induced regions can extend beyond the range of the molecular size
of these proteins. Hence in order to account for the disparate length scales, the effect of
multiple proteins, and thermal fluctuations we adopt a continuum approach in this article
which is based on the Canham-Helfrich description of membranes [38]. The subject of
this article is to define and quantify the free-energy landscape of such curvature-inducing
proteins on a fluid bilayer membrane within the context of membrane elasticity theory.
This manuscript does not model specific experimental systems but rather focusses on a
methodology to compute free energies.
It relies on the premise that for proteins that
induce curvature, when they act in dilute concentrations, the curvature induction is lo-
calized because there is only a finite amount of binding free energy available to deform
the membrane. Hence, the spontaneous curvature around the protein will be localized.
At the continuum level, just like the Canham-Helfrich description of the membrane as an
infinitesimally thin surface, irrespective of the mechanism of curvature induction, we make
the assumption that the effect of the protein is to introduce a curvature field (defined as
the function H0, see below), which is our definition for spontaneous curvature.
5
II. MODEL
A. Membrane Model
For biological membranes -- if the thickness is negligible when compared to its lateral
dimensions -- the thermodynamic behavior of the membrane can be well captured by the
elastic energy functional [38]
H =(cid:90) (cid:16)κ
2
(2H − H0)2 + ¯κK + σbare(cid:17) dA,
(1)
where the material properties are given by κ, the bending rigidity; ¯κ, the saddle splay
modulus; and σbare, the bare surface tension. The geometric properties of the surface
are given by the gauge-invariant scalars H and K, the mean and Gaussian curvatures,
respectively. H0 is a spontaneous curvature field that represents the curvature-inducing
interactions between the protein and membrane; see section II B for details. The integral
is performed over the surface area of the membrane with dA being the differential area.
This approach of treating the effect of the curvature-inducing protein as a curvature field
in the continuum field formulation has been utilized in prior studies [14 -- 17, 27, 39 -- 43].
We make the system amenable to numerical simulations by discretizing the continuous
membrane into a triangulated surface with N vertices, T triangles, and L links. Self-
avoidance is imposed by restricting the link length l to be in the range a0 ≤ l ≤ √3a0.
Here, a0 is the characteristic length scale of the membrane which is much smaller the
persistence length. We note that the length scale in the model is set by the value of a0.
We chose N = 900 and initially place them on a square planar configuration as a 30 × 30
grid. The open edges of the membrane are subjected to periodic boundary conditions
along the plane of the membrane (see Figure 1).
B. Membrane-Protein Interaction Model
For proteins that induce curvature, when they act in dilute concentrations, we expect
their curvature induction to be localized because there is only a finite amount of binding
free energy to deform the membrane. Hence, the spontaneous curvature around the pro-
tein will be localized to a finite length-scale. Irrespective of the mechanism of curvature
induction, we make the assumption that the effect of the protein is to introduce a cur-
6
FIG. 1. (Color Online) a) Confomations of an initialy planar (top panel) and an equilibrated
membrane (bottom panel); the vertices are colored based on the mean curvature which is ex-
pressed in units of a
−1
0 . b) The three Monte Carlo moves -- namely, the vertex move, link flip,
and protein move -- used to evolve the membrane are shown.
vature field H0. We justify our stance because the Canham-Helfrich formalism already
approximates the model membrane to be infinitesimally thin. So the precise mechanism
of spontaneous curvature induction, which can only be correctly modeled in an atomic
level model, has to approximated in some manner within the Helfrich framework, which
is through the choice of the H0 function. An alternative approach at the mesoscale is to
explicitly represent the protein field as particles with suitably characterized membrane
adhesion energies as done in [44 -- 46]. However we have chosen to employ the sponta-
neous curvature field model presented here since it fits well into a multi-scale framework,
where the required field parameters can be determined from an all atom or coarse grained
molecular simulation or experiment, as we describe below.
We do not know the exact nature of H0. But several functions chosen for H0 --
depending on the shape and extent of the function and depending on how many proteins
are present -- will elicit a finite number of emergent membrane morphologies (such as
vesicles, tubules, inward-tubules, caveloe etc.), a premise which is supported by a body
of work discussed in reference [7]. For example, in earlier studies, we have shown that
irrespective of whether we choose an isotropic Gaussian function or a cosine function or
a square-well function, we will get vesicular buds under certain configurations [17, 39 --
41]. Another example is that whether we choose an anisotropic (ellipse shaped) Gaussian
dimple, or an anisotropic saddle shaped function, we can induce tubules [7].
In this article, the spontaneous curvature induced in the vicinity of the membrane at
7
EquilibrationVertex MoveLink FlipProtein Moveb)a)H (units of a0-1)-1.2-0.60.00.61.2 (cid:126)rm due to a protein at (cid:126)rp is represented as,
H0((cid:126)rm, (cid:126)rp) = C0 F((cid:126)rm, (cid:126)rp).
(2)
Here C0 is the induced membrane curvature at (cid:126)rm = (cid:126)rp. As a first approximation we
choose this deformation profile F((cid:126)rm, (cid:126)rp) to be a Gaussian function. A radially symmetric
curvature profile has the form,
Fiso(r) = exp(cid:18)−
r2
2(cid:19) ,
(3)
where r = (cid:126)rm − (cid:126)rp and the 2/2 is the variance of the Gaussian.
In general, the
function F can take any arbitrary form as imposed by the protein curvature field. For
instance, proteins containing BAR domains, like Endophilin and Exo70 domain containing
Exocyst complex, are known to induce spatially anisotropic deformations [27, 32 -- 35]
that depends on the orientation of the protein θ = arccos((cid:107)(cid:126)rm · (cid:126)rp(cid:107)). Such anisotropic
curvature profiles can be modeled as,
Fani(r, θ) = exp(cid:32)−r2(cid:34)cos2 θ
2(cid:107)
sin2 θ
2⊥ (cid:35)(cid:33) .
+
(4)
2(cid:107)/2 and 2⊥/2 are the variances along the directions parallel and perpendicular to the
protein orientation, respectively.
In order perform a systematic study of the free-energy landscape associated with cur-
vature induction, we deal with curvature profiles that are analytically tractable: for this
purpose we have chosen an isotropic spontaneous curvature profile in accordance with
equation (3). Since this choice is an approximation to the exact shape of H0, we dis-
cuss the question: for a given H0 (e.g., equation (3)), how do we estimate its parameters
consistent with a given biological system?
There are three methods we employ to determine the parameters of H0 for a given
biological system, which we summarize below. Method 1 (outlined in detail in previous
work [40]) estimates the parameters in equation (3) by matching the membrane defor-
mation energy due to one spontaneous curvature field to the binding free energy of the
protein with the membrane bilayer. Method 2 (outlined in reference [17]) estimates the
parameters in equation (3) by matching the computed curvature-induced sorting proba-
bility of the proteins with those measured in experiments. In Method 3, the numerical
8
value of the field-parameters are determined based on molecular dynamics simulations at
the atomic or near-atomic (coarse-grained) scales reported in the literature [27, 31]. In
all three methods, the estimate for C0 is ∼ 0.05nm−1 and that for is ∼ 17 nm for ENTH
domain proteins on a typical cell membrane with κ = 20kBT . Later in the article, we set
a typical value of 2 = 6.3a0
2, and κ = 10kBT (typical value for a lipid bilayer in vitro),
which fixes the value of a0 ∼ 10 nm.
On a triangulated membrane, though the core of each protein is defined on a vertex it
induces curvature in the neighborhood of its core vertex in accordance with equation (3).
Each of the n proteins is associated with an unique vertex and each vertex can accom-
modate one protein at the most. The presence of multiple proteins in the vicinity of each
other leads to a superposition of the spontaneous curvature fields. The exact form of
additivity of spontaneous curvature fields is not well established and hence we employ a
simple additive rule where the multiple spontaneous curvature contributions at a given
membrane location are linearly added and truncated as,
(5)
H0((cid:126)rm) = min(cid:32)2C0,
H0((cid:126)rm, (cid:126)rp)(cid:33) .
n(cid:88)p=1
Note that H0((cid:126)rm) denotes the total spontaneous curvature at membrane location (cid:126)rm due
to all proteins in its vicinity.
By including the effect of protein-membrane interaction as a spontaneous curva-
ture field, we assume that the equilibrium behavior of the system is dominated by the
membrane-mediated protein-protein interaction. These interactions are dictated by the
strength and range of the curvature field and small-length-scale interactions (i.e., at
the atomic level) are smoothed-out. Justification for this assumption has recently been
presented by directly parameterizing such a curvature field from molecular dynamics sim-
ulations [27]. In reference [47], Aranda-Espinoza et al. employed a combination of integral
equation theory and the linearized elastic free-energy model to describe the spatial dis-
tribution of the membrane-bound proteins. Their study indicates that the interaction (in
the absence of thermal undulations) between two membrane-bound curvature-inducing
proteins is dominated by a repulsive interaction. Consistent with these published reports,
the calculated binding energy between two membrane-bound proteins interacting through
the curvature fields (again without thermal undulations) show dominant repulsive inter-
actions which is governed by the range of the curvature field [41]. Thus, purely based
9
on energetic grounds, the previous analyses have suggested that membrane-deformation-
mediated energies tend to be repulsive and should prevent, rather than promote, the
formation of protein dimers or clusters.
Kozlov has discussed how the effect of fluctuations can change the repulsive nature of
the interactions [48]. The author's discussion is based on the premise that any membrane
protein locally restrains thermal undulations of the lipid bilayer. Such undulations are
favored entropically, and so this increases the overall free-energy of the bilayer. Neigh-
boring proteins collaborate in restricting the membrane undulations and reduce the total
free-energy costs, yielding an effective (membrane-mediated) protein-protein attraction.
Indeed, for the linearized free-energy model, computing the second variation of energy,
(note that at equilibrium, the first variation is zero, while the second variation governs
the stiffness of the system against fluctuations) yields that the presence of a protein (or
equivalently a curvature-inducing function) leads to a localized suppression of membrane
fluctuations [41, 42]. This calculation has been further verified by using a free-energy
method to compute the change in Helmholtz free-energy upon the introduction of a cur-
vature field [42]. This provides for the possibility of an entropically-mediated protein-
protein attraction. The outcome of the interplay between the attractive entropic forces
and the repulsive energetic forces is context specific as both have the same dependence on
the protein-protein distance, and their absolute values differ only by coefficients with sim-
ilar values. This has been demonstrated by examining the protein-protein pair correlation
(spatial and bond-orientational) and through the effect on membrane morphology [41]. In-
deed the model predicts that the cooperative effects of membrane-mediated interactions
between multiple proteins can drive different morphological transitions in membranes
[8, 41, 47, 48]. This notion of cooperativity is also consistent with the analysis of Kim et
al. [49], who have shown using an energetic analysis that in the zero temperature limit,
clusters with size larger than five membrane-bound curvature-inducing proteins can be ar-
ranged in energetically stable configurations. It is also worth mentioning for completeness
that Chou et al. [49] have extended the energetic analysis to membrane-bound proteins
that have a noncircular cross-sectional shape and to local membrane deformations that
are saddle shaped (negative Gaussian curvature) and have shown that in such cases the
interactions can be attractive even without considering fluctuations.
10
C. Monte Carlo Moves
The accessible states of the membrane protein system are sampled using a set of three
Monte Carlo moves that mimic membrane undulations, lipid diffusion, and protein diffu-
sion. In the framework of Dynamically triangulated Monte Carlo (DTMC), a Monte Carlo
step (MCS) comprises of N attempts to randomly displace the vertices, L attempts to flip
the links and n attempts to randomly displace the protein on the membrane surface. The
various moves have been illustrated in Figure 1b and each of the attempted moves are
accepted using the Metropolis algorithm [50]. For a complete description of the Monte
Carlo moves see reference [17].
In our model, the random displacement vector is adaptively chosen to ensure that the
acceptance rate for the vertex move is 50%, while the acceptance rate for link flips and
protein diffusion are dictated by the geometry. All our simulations were equilibrated for
10 million MCS and statistics were collected over another 20 million MCS.
D. Ensemble for the Planar Membrane
A planar membrane is characterized by its extensive variables; the entropy, S; the
surface area, A; and the projected area, AP . The internal energy of the membrane with
n proteins is given by,
dU (N, n, A, AP , S) = dH = µdN + µP dn + σdA + γdAP + T dS.
(6)
Here the conjugate variables are µ, the chemical potential of the membrane, µP , the
chemical potential of a membrane protein, and γ, the tension due to the frame (also called
the frame tension).
It should be noted that for closed membranes (e.g., a cylindrical
membrane or a spherical vesicle) the volume enclosed by the membrane, V , and the
osmotic pressure difference, −P , are used in place of AP and γ, respectively; we limit
our studies in this article only to planar membranes. We can assume that l is not a
physically independent variable from the list of extensive variables defined above; rather,
l sets the length-scale or resolution of the mesh, and tuning it allows us to change A/AP
or the value of σ. Here, σ is an effective tension conjugate to A and is constituted by
11
a combination of the bare surface tension (σbare) and the area compressibility modulus.
In our simulations, we control (hold constant) N, n, σ, AP , and T . Hence, the suitable
thermodynamic potential for a planar membrane in our simulations is given by,
dF (N, n, σ, AP , T ) = dH − T dS − SdT − σdA − Adσ.
(7)
The effective surface tension σ defined in equation (6) should be distinguished from the
bare surface tension σbare defined in equation (1). We have performed all our studies
with σbare = 0. However, the effective surface tension determined from the fluctuation
spectrum can still be non-zero, (see section IV A), because the value of σ is renormalized by
an effective area compressibility modulus term; the latter arises because of the constraint
a0 ≤ l ≤ √3a0, we impose for self-avoidance, see section II A.
III. FREE-ENERGY METHODS
The free-energy landscape of the protein-membrane system drives key biophysical phe-
nomena including protein recruitment, protein membrane remodeling, curvature-sensing,
and protein clustering. Hence, in order to gain better insight into the behavior of this
system, we delineate a strategy to compute the free-energy landscape for a single protein
interacting with the membrane using the suite of free-energy methods described below.
A. Widom Test-Particle/Field Insertion Method
We determine the change in free-energy when a protein binds to the membrane by
determining the excess chemical potential using the test-particle insertion method. The
Widom particle/test-particle insertion method is a computational technique used to probe
a system's chemical potential [51]. This technique samples the excess chemical potential
by randomly inserting a virtual test (ghost) particle, and determines the change in the
system's energy due to insertion of the test particle.
Let Qn and Qn+1 be the partition functions for a membrane with n and n + 1 proteins,
respectively. The partition function is related to the configurational free-energy, (i.e., not
including the contribution from the kinetic energy or from internal degrees of freedom
such as rotation), as Fn = −kBT ln Qn for all n. Hence, the change in free-energy upon
12
insertion of a protein field (i.e., the test particle) in a membrane with n proteins is given
Qn (cid:19) .
∆F = Fn+1 − Fn = −kBT ln(cid:18) Qn+1
It can be seen from equations (6) and (7) that the above change is equal to the chemical
(8)
(9)
(10)
by:
potential,
Combining equation (9) with equation (8) we obtain:
.
∂F
µP =
∂n(cid:12)(cid:12)(cid:12)(cid:12)AP ,σ,N,T
µP = −kBT ln(cid:18) Qn+1
Qn (cid:19) ,
which can be decomposed into an ideal gas contribution and an excess contribution such
that,
µP = µid
P (ρ) + µex
P .
(11)
The configurational component of the ideal part can be calculated from the protein den-
sity ρ as kBT ln ρ; we note that the full ideal gas contribution is given by µid
P (ρ) =
kBT ln(ρΛd) not including the contributions from the internal degrees of freedom. Here,
Λ = (2πmkBT /h2)−1/2 with m the molecular mass of the protein, h the Planck's constant,
and d the dimensionality of the system. If ∆H be the energy change due to insertion of
a test curvature-inducing protein then the excess chemical potential is written as,
µex
P = −kBT ln(cid:90) (cid:104)exp(−β∆H)(cid:105)Puniform(sn+1)dsn+1.
(12)
Here, β = (kBT )−1, and ∆H = H (n + 1) − H (n) and the ensemble average (cid:104)·(cid:105) is taken
over the phase space defined by the membrane and the n protein fields. Here, sn+1 = (cid:126)rp,
with p = n + 1, is the position of the n + 1th protein field, and Puniform(sn+1) repre-
sents a uniform probability distribution from which the coordinate of the n + 1th par-
ticle/field is sampled. The integral over sn+1 amounts to the sum over all Widom test
paticle/field insertion trials, and Puniform(sn+1) equals the reciprocal of the total num-
ber of trials. For conciseness, we represent the right-hand-side term in equation (12) as
−kBT ln(cid:104)exp (−β∆H)(cid:105)n. This formulation is derived for a homogeneous membrane while
the corresponding form of equation (12) for a spatially inhomogeneous membrane, where
µex
P is a function of the phase space variables r, is given by,
µex
P (r) = −kBT ln(cid:104)exp (−β∆H(r))(cid:105)n.
(13)
13
At equilibrium the bulk chemical potential µP is a constant, hence the scaled, inhomoge-
neous, spatial density can be determined as,
ρ(r) = ρ0(cid:104)exp (−β∆H(r))(cid:105)n,
(14)
where ρ0 = exp (µP ). The Widom test-particle/field insertion method is more suitable
to probe chemical potentials in dilute systems whereas its applicability to systems with
large protein concentrations is limited; see Appendix A for a discussion. Hence, in order
to study the higher protein concentrations we also use more reliable methods based on
free-energy perturbation, which are defined in the next sections.
B. Thermodynamic Integration (TI) Method
Thermodynamic integration is a free-energy perturbation technique used to compute
the change in free-energy between two states A and B, with energies HA and HB; these
states correspond to a membrane with n and n + 1 proteins, respectively. Further, state A
is characterized by a scalar parameter λ = 0 and state B by λ = 1. The system is evolved
with a Hamiltonian (or energy function) H(λ) = (1 − λ)HA + λHB. To define a path
between A and B, the parameter λ is varied between 0 ≤ λ ≤ 1 in successive windows of
the simulation. The free-energy change along this path [50] is given by,
∆FTI = FB − FA =(cid:90) 1
0 (cid:28) ∂H(λ)
∂λ (cid:29) dλ.
(15)
TI overcomes many of the limitations of the Widom test-particle/field insertion method
(see Appendices A and B), but the results of equation (15) should match the results
from equation (11) in the dilute limit; i.e., when the concentration of protein is such that
n << N .
C. Bennett Acceptance Ratio Method (BAM)
The Bennett acceptance method is also used to approximate the free-energy difference
between two states close to each other in phase space. This method is derived from the
detailed balance equations involving two states (A and B) [52]. Namely,
M (HA − HB) exp(−βHB) = M (HB − HA) exp(−βHA),
(16)
14
where M is some function that defines the acceptance distribution for transition from state
A to state B or vice versa. In our case, we choose M to be the Metropolis function M (x) =
min(1, exp(−βx)), which defines the acceptance probability according to a Boltzmann
distribution. This yields:
exp(cid:18)−∆FBAM
kBT (cid:19)A→B
=
QB
QA
= (cid:104)M (HB − HA)(cid:105)A
(cid:104)M (HA − HB)(cid:105)B
.
(17)
Appendix C provides a brief discussion of the expected accuracy of the Bennett acceptance
methodology for the choice of the acceptance function M described above; the Bennett
acceptance method can be improved further by optimizing the function M , to decrease
the sampling error [53].
D. Analytic Approximation to the Excess Chemical Potential of Curvature-
Inducing Proteins
For some special cases the chemical potential of a curvature-inducing protein can be
derived analytically. The change in energy due to the addition of one curvature protein
can be determined from equation (1) as,
∆H =(cid:90) κ
2(cid:0)−4HH0 + H 2
0(cid:1)dA.
(18)
At infinite dilution (i.e. when n = 0) this change in energy for curvature fields given by
equation (3) can be included in the expression for the excess chemical potential, which
can be expressed as:
µex
P = −kBT ln(cid:28)exp(cid:18) −κ
kBT (cid:18)−2C0(cid:90) Hf (r)dA +
This relation can be further simplified to,
π2C 2
0
4 (cid:19)(cid:19)(cid:29)n=0
.
(19)
µex
P =
κπ2C 2
0
µT =0
4(cid:124) (cid:123)(cid:122) (cid:125)
− kBT ln(cid:28)exp(cid:18)2κC0(cid:82) Hf (r)dA
(cid:124)
(cid:123)(cid:122)
kBT
µfluc
(cid:19)(cid:29)n=0
(cid:125)
,
(20)
since the second term in the exponential depends only on constants. In the above equation,
µT =0 can be interpreted as the chemical potential to insert a protein on a flat membrane.
Cellular membranes can remain planar when the membrane is strongly bound or pinned
15
to other cellular components like the cytoskeleton and other membrane binding proteins
which can be characterized by a pinning fraction. The pinning fraction φ can range from
0 for an free membrane to 1 for a completely pinned membrane. When φ < 1 the chemical
potential has additional contributions from the undulation modes of the membrane, which
is given by µfluc.
IV. RESULTS
A. Membrane Undulations and Power Spectrum
The equilibrium properties of an undulating membrane are significantly influenced by
the choice of control variables (see section II D). Hence before delineating the protein
induced deformations we first analyze the fluctuation modes of a planar membrane in the
absence of a curvature (protein) field. The height-height correlation of a planar membrane,
described by equation (1), parameterized in the Monge Gauge [54, 55], and expressed in
Fourier space, is given by,
(cid:104)hqh−q(cid:105) =
kBT
AP [κq4 + σq2]
.
(21)
Here, the angular brackets represent the equilibrium ensemble average, and we define hq
as the 2-dimensional discrete Fourier transform of the membrane height function h((cid:126)r) =
h(x, y). Namely,
h((cid:126)r) =(cid:88)q
hq exp (i(cid:126)q·(cid:126)r) .
(22)
In equation (22), (cid:126)q = (qx, qy) = 2π(nx/L, ny/L), where AP = L2 and nx, ny are integers.
The undulation spectrum corresponding to a planar membrane with κ = 10kBT and
σbare = 0 for a range of A/AP is shown in Figure 2. The data was fit to equation (21) and
the corresponding fit parameters, κeff and σeff, are shown in the inset to Figure 2; also see
Figure S1 in the Supplementary Material.
When A/AP > 1.05 the membrane displays dominant long wavelength undulations,
represented in Figure 2 by the large intensities of the power spectrum at low q; this
results in shapes with curvatures of large magnitude.
In this regime κeff < κ -- which
corresponds to thermal (entropic) softening of the membrane -- and σeff ∼ 0, which implies
that the membrane is tensionless. Henceforth, we choose to model the membrane with
16
FIG. 2. (Color Online) Undulation spectrum (main plot) and the fit values for κeff and σeff
(inset) for different values of A/AP . Each pair of values seen in the legend of the main plot
corresponds to (A/AP , AP ) for the membrane. With increase in A/AP the small q behavior
transitions from a concave to a convex profile, which is characteristic of σeff crossing over to
negative values as shown in the inset. The effective bending rigidity is also renormalized with
change in A/AP such that κeff → 0 as A/AP → ∞. The filled symbols in the inset correspond
to values of A/AP for which σeff ∼ σbare and κeff ∼ κ.
A/AP = 1.04 (parameters corresponding to the filled symbols in the inset to Figure 2) for
which we compute κeff ∼ κ and σeff ∼ σbare + σ = 0.0; however, we note that A/AP is an
important parameter which defines the thermodynamic ensemble in section II D.
B. Membrane Conformations versus C0 and n
The equilibrium shapes of a planar membrane interacting with spontaneous curvature-
inducing proteins with fixed 2 = 6.3a2
0, and for different magnitudes of the imposed
curvature C0, is shown in Figure 3.
A comparison of the membrane conformations for C0 = 0.0, 0.4, and 0.8a0
−1, in Fig-
ure 3, shows that in the presence of a small number of the curvature-inducing proteins
(dilute limit) the membrane does not undergo any morphological changes, which is con-
sistent with previous studies [41]. This is characteristic of membranes with dilute protein
concentrations or proteins imposing weak spontaneous curvatures. In the dilute limit the
proteins localize to regions on the membrane matching their curvature field; however, the
concentration is too low to promote any spatial aggregation of proteins which can lead to
17
0.5123410-610-410-210011.21.41.60510152011.21.41.600.511.52q!2πl"!unitsofa−10"⟨hqh−q⟩!unitsofa20"A/APκeff(kBT)σeff!kBTa−20"κeff=κA/AP,AP!a20"(1.45,1010)(1.22,1198)(1.05,1405)(1.02,1631)(1.01,1873)σeffκeff a morphological transition. Hence, the proteins in this concentration regime can largely
be regarded as curvature-sensors. We note, however, that even in the dilute limit there
is significant renormalization of the bending stiffness and membrane tension (see Figures
S2 and S3 in the Supplementary Material). The effects at higher concentrations are
more drastic leading to a change in the undulation behavior; that is, for larger n, the cor-
responding governing equations are more complex than that described by equation (21)
as the undulation spectrum is two-dimensional and depends on q, q(cid:48). Specifically, when
κ = κ(x, y), σ = σ(x, y), and H0 = H0(x, y), and whose respective Fourier transforms are
given by κq, σq, h0,q, we can show that H is given by:
(cid:104)H(cid:105) =
1
2AP (cid:88)(cid:126)q (cid:88)(cid:126)q(cid:48) {[q2q(cid:48)2(cid:104)hqhq(cid:48)(cid:105)−q2(cid:104)hqh0,q(cid:48)(cid:105)−q(cid:48)2(cid:104)h0,qhq(cid:48)(cid:105)+(cid:104)h0,qh0,q(cid:48)(cid:105)]κq+q(cid:48) + qq(cid:48) [(cid:104)hqhq(cid:48)(cid:105)] σq+q(cid:48)}.
(23)
At large n, as can be seen from Figures S2 and S3, the spontaneous curvature fields
significantly influence the low q modes of the fluctuation spectrum.
Our results in Figure S4 (Supplementary Material) also quantify the increase in excess
area of the membrane (A − AP ) as a function of n for different values of C0. We find the
membrane excess area to increase with protein concentration (n), and is more pronounced
for higher values of C0. This increase is consistent with the softening of the membrane
(i.e., lowering of κ); however the effect is subtle because a positive renormalized tension
is manifested. For larger n, the membrane becomes substantially softer, however, the
undulation behavior is more complex than that described in equation (21), as discussed
above.
With increase in protein concentration, spatial aggregation is more pronounced and
cooperative effects -- due to membrane curvature-mediated interactions -- stabilize protein
clustering as well as induce morphological transitions. In this limit, the proteins collec-
tively induce stable morphological features in the membrane as seen in Figure 4; here, for
C0 = 0.8a−1
0 , protein clustering leads to tubule formation when n > 10.
Figures 3 and 4 show how the control variables such as A/AP (relative membrane area)
and n (protein concentration) govern the emergent membrane morphologies. These results
also suggest subtle competition between the translational entropy of proteins, entropy due
to membrane undulations, and membrane deformation energy due to curvature-induction
18
FIG. 3. (Color Online) Representative membrane conformations as a function of imposed cur-
vature C0 for a system with 6 proteins: (a) no protein fields; (b) six protein fields each with
C0 = 0.4a
−1
0 ; (c) six protein fields each with C0 = 0.8a
−1
0 . Color bar shows the induced curvature
field H0 in units of a
−1
0 .
FIG. 4. (Color Online) Representative membrane conformations as a function of epsin concen-
tration for C0 = 0.8a
−1
0 : (a) 2 protein fields; (b) 8 protein fields; (c) 14 protein fields. Color bar
shows the induced curvature field H0 in units of a
−1
0 ; a tubule is present in (c).
by proteins. Since the morphological changes in the membrane are associated with a large
change in the entropy of the system, they can be quantitatively tracked only by computing
the free-energy landscape of curvature induction. The following sections quantify the free-
energy landscape of protein-induced curvature deformations as a function of C0 and n.
C. Widom Test Particle/Field Insertion Method
Widom test-particle/field insertion method is used to quantify the excess chemical
potential of curvature-inducing proteins on a planar membrane. Figure 5 shows the excess
chemical potential for dilute protein concentrations (i.e., n → 0) as a function of C0
and 2, for curvature fields of the form given by equation (3). For C0 = 0.4a−1
and
0
19
0.00.40.8(a)(b)(c)1.20.40.82814(a)(b)(c)1.20.40.8n 0 , µex
0.6a−1
P is negative, and hence it is favorable to insert a protein on the membrane. In
this limit, the protein's curvature field is shallow and matches well with the equilibrium
curvature profile of the natural undulations in the membrane leading to reduced free-
energy/chemical potential. However, it should be noted that the excess chemical potential
can cross over to positive values with further increase in the value of 2 and the insertion
of a protein is no longer thermodynamically favorable. For C0 = 0.8a−1
P is observed at much lower values of 2. µex
crossover to positive µex
0 and 1.0a−1
0 , the
P increases linearly
P is a
signature of curvature induced deformation, since equilibrium membrane profiles cannot
with 2 with their respective slope depending on the value of C0. An increase in µex
accommodate such large curvatures. Hence, these results quantify both the curvature-
sensing and curvature-inducing behavior of membrane proteins.
The excess chemical potential as a function of the induced spontaneous curvature C0
is shown in Figure 6 (data from Figure 5 has been replotted). As stated before, the free-
energy for insertion of a protein is negative for small magnitudes of induced curvature and
extents (low C0 and 2). For higher values of 2 the excess chemical potential is observed
to grow quadratically with C0, as predicted by equation (20). We note that the higher
values of 2 correspond to an energy dominated regime, for which, by relative comparison,
the entropic correction (second term in RHS of equation (20)) is small enough to be
neglected.
We have shown in Figure 7 the computed chemical potential as a function of protein
concentration (n) for a planar membrane with C0 = 0.8a−1
0. For small
values of n where the concentration of proteins does not considerably affect the mem-
and 2 = 6.3a2
0
brane undulation, we observe µex
P to be positive and to increase with increasing value
of n. The excess chemical potential reaches a peak value at n ≈ 6 -- beyond which the
chemical potential drops to negative values implying that the subsequent recruitment of
proteins is favorable. In analogy, the region to the left of the peak corresponds to the
planar membrane morphology shown in Figure 4(a) and the region marked tubules to the
extreme right corresponds to the tubulated membrane conformation shown in Figure 4(c).
In the transition region we observe both tubulated and planar morphologies with equal
probabilities. This leads to large fluctuations in µex
P , which is indicated by the large error
bars in the chemical potential for protein concentrations n = 8 and n = 10 (see Figure 7).
The above example demonstrates that the Widom test-particle/field method is a pow-
20
erful approach to quantitatively map the phase boundary associated with morphological
transitions in membranes.
FIG. 5. (Color Online) Excess chemical potential as a function of 2, for an isotropic Gaussian
curvature obtained through the Widom test-particle/field insertion method. Data shown for
four values of spontaneous curvature, C0.
FIG. 6. (Color Online) Excess chemical potential as a function of C0, for an isotropic Gaussian
curvature obtained through the Widom test-particle/field insertion method. Data shown for
four values of variance, 2.
D. Comparison to Analytical Results
The values of the chemical potential at infinite dilution can also be computed an-
alytically. However, for proteins with finite curvature extent, a direct comparison with
21
-5051015202523456789ǫ2(cid:0)unitsofa20(cid:1)µexP(unitsofkBT)0.4a−100.6a−100.8a−101.0a−10-505101520250.30.40.50.60.70.80.91C0(cid:0)unitsofa−10(cid:1)µexP(unitsofkBT)2.3a204.3a206.3a208.3a20 FIG. 7. (Color Online) Excess chemical potential of an isotropic Gaussian curvature field, with
C0 = 0.8a
−1
0 and 2 = 6.3a2
0, as a function of the number of proteins (n).
analytical results is complicated by the non-trivial curvature-field dependent term in equa-
tion (20). It is possible, however, to obtain closed form analytical predictions for the excess
chemical potential when spontaneous curvature fields of the form H0 = C0δ (r − r(cid:48)) are
considered. In this section, the results obtained from the Widom test-field method are
compared against analytical predictions for such curvature fields.
In our model, proteins which do not have large extents of curvature can be approx-
imated as point sources of spontaneous curvature. A point spontaneous curvature field
can be described by,
H0((cid:126)rm, (cid:126)rp) = C0δ(r), where r = (cid:126)rm − (cid:126)rp.
(24)
Using equation (24), equation (19) can be reduced to,
Here, Avertex = √3(1.3a0)2/2 is the area per vertex in our discrete triangulated mesh, and
1.3a0 is the average link length at the value of A/AP employed here: the factor Avertex
arises because of the discrete approximation to the Dirac delta function. The ensemble
average in (25) can be evaluated in simulations through a cumulant expansion,
(cid:104)exp (tH)(cid:105) = 1 + t(cid:104)H 1(cid:105) +
t2
2!(cid:104)H 2(cid:105) +
t3
3!(cid:104)H 3(cid:105) + ...,
(26)
where, (cid:104)H i(cid:105) is the i'th moment of the mean curvature, and t = 2κC0/kBT . As demon-
strated in Appendix D, the sum of terms (cid:104)H i(cid:105) is a weakly decaying function of i, and hence
22
µex
P =
.
(25)
κC 2
0
2Avertex
(cid:124) (cid:123)(cid:122) (cid:125)
µT =0
− kBT ln(cid:28)exp(cid:18)2κC0
(cid:124)
(cid:123)(cid:122)
kBT
µfluc
H(sn+1)(cid:19)(cid:29)n
(cid:125)
-20-15-10-50510152002468101214nµexP(unitsofkBT)PlanarTubules we retain the first 15 terms in order to obtain convergence. In Figure 8 µex
P obtained from
the Widom test-field method is plotted and compared against µT =0 and (µT =0 − µfluc).
The analytical results with finite temperature corrections agree well with µex
P . The Widom
test-field method is thus validated for point spontaneous curvature fields, and hence we
are confident that the method gives reliable estimates for the excess chemical poten-
tial. It should be noted that the fluctuation corrections (µfluc) for the point spontaneous
curvature field ranges from 0 to 6kBT . This large correction is a manifestation of the
protein curvature field localizing to membrane undulations matching their profile, and
the value of µfluc depends on κ, C0, 2, and n.
In the next section, results from the
Widom test-particle/field insertion method is compared to results from both thermody-
namic integration and Bennett acceptance methods, to further validate the estimates for
the chemical potential.
FIG. 8. (Color Online) Widom test particle/field insertion results are shown in blue. Analytical
scaling with a fluctuation correction calculated from the cumulant expansion is shown in green.
Zero temperature scaling shown in red.
E. Comparison of Free-Energy Methods
The chemical potential to insert a protein field is given by,
µ =
dF
dn
=
∆F
∆n
.
(27)
Here, ∆F is the free-energy change to insert ∆n proteins. We compute ∆F using two
free-energy perturbation techniques namely thermodynamic integration and Bennett ac-
23
-1-0.500.511.522.533.500.20.40.60.81C0(unitsofa0)µ(unitsofkBT)µT=0µT=0−µflucµexP ceptance method (BAM). The techniques used to compute ∆FTI and ∆FBAM involve
growing ∆n curvature fields that have zero spontaneous curvature initially (state A), to
a desired value of C0 (state B). Since the presence of the protein is felt only through C0,
perturbing the system from state A to B is analogous to inserting ∆n proteins.
In order to make direct comparisons to results from the Widom test-field method, we
choose ∆n = 1. In this case, the values of ∆FTI and ∆FBAM are related to the chemical
potential as,
∆FTI = ∆FBAM = µ = µid
P (ρ) + µex
P ,
(28)
where, µex
P is calculated using Widom test particle/field insertion method, while the con-
P (ρ) (as discussed above in eqn.
figurational contribution to the entropic correction, µid
(11)), is given by,
µid
P (ρ) = kBT ln ρ.
(29)
Both thermodynamic integration and Bennett acceptance methods calculate the difference
in free-energy between a state with no protein field and a state with one protein field.
Results from Widom insertion method cannot be directly compared to TI or BAM to
an important difference in sampling between the methods. Widom insertion samples the
curvature field equally at all spatial locations on the membrane, whereas TI and BAM
introduce the curvature field at a specific spatial location; this difference in sampling
defines a correction of entropic origin for thermodynamic integration and Bennett methods
which needs to be accounted for before all three methods can be compared against one
another. Details of the procedure for computing µid
P (ρ) are given in Appendix E. The
values of ∆FTI and ∆FBAM are plotted and compared against Widom test-field values
for µ in Figure 9. The results show excellent agreement for small values of C0, but each
method deviates as the spontaneous curvature is increased. The estimate for the chemical
potential µ agrees very well with ∆FTI and ∆FBAM for small values of C0 < 0.6a−1
0 . For
larger values of C0 the chemical potential determined using the Widom method deviates
from the estimates derived from the perturbation techniques. The comparison between
the methods at higher protein densities is also investigated and the results are discussed
in Appendix F.
The mismatch in the values of µ between these methods at large values of C0 is well
known. In the case of Widom test-field insertion the deviation is a result of dominant
24
contributions from some rare conformations to the chemical potential. Estimates for
the chemical potential from thermodynamic integration also break down due to insuf-
ficient sampling at larger values of C0; this can be seen in the small deviations from
the corresponding values of BAM in Figure 9. Metrics to quantify the sampling error
from the three methods are discussed further in Appendices A, B, and C. The applica-
bility of each of the described free-energy methods depends on the system investigated,
desired accuracy, and the available computational resources. The Widom insertion tech-
nique gives accurate results with low computational overhead and this works very well
for dilute protein concentrations and weak curvature fields; however at higher protein
concentrations and strong curvature fields, where the energies are large, this method
becomes inaccurate and this is a know artifact of Widom insertion. On the other hand
perturbative techniques like TI and Bennett work very well for all concentrations, but
are computational expensive. For dense systems TI or Bennett methods are better suited.
FIG. 9.
(Color Online) Comparison of chemical potentials from Widom, TI, and BAM, for
different C0 with 2 = 6.3a2
0 and n = 1.
V. CONCLUSION
Three free-energy sampling methods have been used to quantify the chemical potential
of curvature-inducing proteins in a field theoretic mesoscale cell membrane model. Re-
sults show good agreement between each method for weak spontaneous curvature fields
and deviate at strong curvature field strengths due to the differences in the nature of
25
-10-8-6-4-2024681000.20.40.60.81C0(cid:0)unitsofa−10(cid:1)µ(unitsofkBT)∆FBAM∆FTIµ sampling in each method. The results from the Widom method are also in excellent
agreement with an analytical result for curvature fields approximated by a delta function,
further validating our computational approach. The analytical result also provides a basis
to explain the quadratic dependence of the excess chemical potential on the strength of
the curvature field induction in an energy dominated regime. Further, the utility of the
Widom particle/field insertion method to quantitatively track phase boundaries associ-
ated with morphological transitions of the membrane has been successfully demonstrated
in the context of a tubulation transition. Our results also indicate that the Widom test
particle/field insertion method fails to capture the correct chemical potential at high cur-
vature field strengths, as expected, due to the large perturbation in energy. In this limit,
the thermodynamic integration and the Bennett acceptance methods perform favorably
to control the statistical error. With these caveats noted, the free-energy approach to
quantify the energy landscape of protein-mediated membrane deformation is novel and
powerful in quantitatively examining protein-induced morphological transitions in bilayer
and membrane systems. Our simulations are able to recapitulate a tubulation transi-
tion above a critical density of curvature-inducing proteins. Tubulation of liposomes has
been widely observed in the literature for high concentrations of curvature-inducing pro-
teins including Epsin, Amphiphysins, and Exo70 [27, 56, 57]. Given the characteristics
of a single protein curvature field, our model would be able to predict these tubulation
thresholds for each protein species. Future work will focus on extending these methods to
study curvature-sensing in cylindrical/tether geometries, anisotropic curvature fields, sys-
tems with inhomogeneous background curvature, and also the effect of control variables
such as tension in morphological transitions of the membrane induced by spontaneous
curvature.
ACKNOWLEDGMENTS
This work was supported in part by National Science Foundation grants DMR-1120901,
CBET-1133267, and CBET-1244507, and the National Institutes of Health grant NIH
U01-EB016027. The research leading to these results has received funding from the Euro-
pean Commission grant FP7-ICT-2011-9-600841. Computational resources were provided
in part by the National Partnership for Advanced Computational Infrastructure under
26
Grant No. MCB060006 from XSEDE.
Appendix A: Widom Test Particle/Field Insertion: Quantification of Sampling
In the Widom test particle/field insertion method, the ensemble average is taken over
a Boltzmann distribution of ∆H. This means, the small or negative ∆H values will
dominate the ensemble average. The distribution of ∆H is a Gaussian, as shown in
Figure 10, with P (∆H) dependent on the strength of the curvature field. As the strength
of the curvature field increases the mean of this Gaussian distribution will shift to the
right, towards higher energies and both the precision and accuracy of the Widom method
will be impacted adversely.
FIG. 10. (Color Online) Normalized distribution of ∆H obtained using the Widom particle/field
insertion method for several C0; here 2 = 6.3a2
0.
Appendix B: Accuracy of Thermodynamic Integration
By setting up a range of simulations over the Kirkwood coupling parameter, λ, in the
interval from 0 to 1, the elastic energy of the membrane with and without H0 can be
tracked and integrated along λ. Figure 11 details the contributions of Hλ=0 and Hλ=1.
To calculate the chemical potential, which can be compared to Widom insertion, the
free-energy is computed by introducing one spontaneous curvature field (∆n = 1).
27
00.050.10.150.20.25-40-30-20-10010203040∆H(unitsofkBT)P(∆H)0.4a−100.6a−100.8a−10 FIG. 11. (Color Online) Plots of Hλ=0 and Hλ=1 as a function of λ. Data shown corresponds
to a spontaneous curvature field with C0 = 0.8a
−1
0 and 2 = 6.3a2
0.
Appendix C: Accuracy of Bennett Acceptance Method
The Bennett Acceptance method requires the two states being sampled to have a small
difference in energy. This accuracy can be quantified by plotting the distribution of ∆H
in each direction sampled (A → B and B → A). A large overlap in the distributions of
∆H describes states which have a small difference in energy. For example, in the case of
one curvature-inducing protein, the states A and B represent a membrane with curvature
0 , and state B to have C0 = 0.76a−1
fields C0 and C0 + δC0, respectively. Consider state A to have a spontaneous curvature
C0 = 0.8a−1
0. The normalized
distribution of ∆H is shown in Figure 12. As expected, the energy is normally distributed,
with the overlap between each distribution being within one standard deviation of each
0 , for a fixed 2 = 6.3a2
other. If the states are separated further apart in energy, this overlap will become minimal,
and the accuracy of Bennett will decline.
Appendix D: Convergence of the Cumulant Expansion
The number of terms to be retained in a cumulant expansion depends upon its conver-
gence behavior. Figure 13 shows µfluc computed using a cumulant expansion as a function
of the number of terms (i) retained. It can be seen that for higher C0 more terms need
to be considered in order to attain convergence. For all analysis presented in this article,
the first 15 terms were used to compute µfluc.
28
48049050051052053000.20.40.60.81λH(unitsofkBT)Rκ2(2H)2dARκ2(2H−H0)2dA FIG. 12. (Color Online) Normalized distribution of the change in energy when the membrane
transits from state A to state B and vice-versa in BAM.
FIG. 13. (Color Online) The fluctuation correction for the chemical potential (µfluc) obtained
with the cumulant expansion as a function of the number of terms (i) considered. Data shown
for three values of C0 = 0.4, 0.6, 0.8 a
−1
0 .
Appendix E: Estimation of the Entropic Correction
In order to compare TI or Bennett with Widom insertion method, the difference in
density sampling between the methods can be approximated. In TI or BAM the spon-
taneous curvature field stabilizes a bump on the membrane and this limits the lateral
diffusion of membrane protein field. This means that in the limit of a large C0 and 2,
the membrane curvature field can only sample a small region of the membrane which cuts
off entropic contributions due to diffusion.
In a Widom simulation the curvature field
probes the free-energy with equal probability across the whole membrane. This disparity
29
00.020.040.060.080.10.12-2-1012∆H(unitsofkBT)P(∆H)(HB−HA)(HA−HB)00.511.522.533.52468101214iµfluc(unitsofkBT)0.4a−100.6a−100.8a−10 in density sampling is of entropic origin and can be written as,
FTI/BAM + F (ρ) = µex
P
where
Nvert(cid:19) ,
F (ρ) = −kBT ln(cid:18) 2σψ
(E1)
(E2)
with σψ being some average number of vertices out of a total Nvert vertices that a curvature
field visits in a TI or Bennett simulation. The entropy lost in a thermodynamic integration
simulation was computed by plotting a histogram of the number of unique vertices visited
by an curvature field, ψ, and finding the standard deviation of that distribution, σψ. The
standard deviation is calculated from
σ2
ψ =(cid:88)j
j2P (j)2 −(cid:88)j
(jP (j))2 .
(E3)
Calculated values of standard deviation and their corresponding values of free-energy are
listed in Table I.
FIG. 14. (Color Online) Histogram of the number of unique vertices visited by a curvature field
(ψ) in a TI simulation as a function of C0.
Appendix F: Comparison of Free-energy Methods at Higher Densities
The Widom particle/field insertion method is known to fail at high densities due to the
nature of its sampling. Therefore a comparison of free-energy methods for higher densities
is done in order to quantify its accuracy. A comparison between the chemical potential
30
0102030405060020406080100120ψP(ψ)0.8a−100.6a−100.4a−100.2a−10 TABLE I. Estimation of the Entropic Correction
C0/(a
−1
0 )
2σψ
F (ρ) / (kBT )
0.2
0.4
0.6
0.8
1.0
242
105
74
49
37
1.31
2.15
2.50
2.90
3.17
obtained from both TI and the Widom method for several protein concentrations ranging
is calculated according to Appendix E. For C0 = 0.8a0
from n = 0 to n = 6 is shown in Figure 15. The entropic correction for the Widom method
−1 this correction is approximately
−1 its F (ρ) =
1.83kBT . The comparison in Figure 15 shows that the methods agree within statistical
−1 is systematic
and is expected due to a similar deviation seen in Figure 9 between the Widom method
−1. The deviation between the results at C0 = 0.8a0
F (ρ) = 2.85kBT , for C0 = 0.6a0
−1 its F (ρ) = 2.42kBT , and for C0 = 0.4a0
error for C0 = 0.6a0
and other free-energy methods for dilute concentrations as discussed in Figure 9.
FIG. 15.
(Color Online) Chemical potential versus n: Data from the results of the Widon
method and the TI method are shown for two C0s with κ = 10kBT and 2 = 6.3a2
0.
31
-50510152001234567nµ(unitsofkBT)µTI0.6a−10µTI0.8a−10µWidom0.6a−10µWidom0.8a−10 [1] T. Baumgart, B. R. Capraro, C. Zhu, and S. L. Das, Annu. Rev. Phys. Chem. 62, 483
(2011).
[2] H. T. McMahon and J. L. Gallop, Nature Cell Biology 438, 590 (2005).
[3] J. Zimmerberg and M. M. Kozlov, Nature 7, 9 (2005).
[4] B. J. Peter, Science 303, 495 (2004).
[5] J. Zimmerberg and S. McLaughlin, Current Biology 14, R250 (2004).
[6] J. L. Gallop, C. C. Jao, H. M. Kent, P. J. G. Butler, P. R. Evans, R. Langen, and H. T.
McMahon, The EMBO Journal 25, 2898 (2006).
[7] N. Ramakrishnan, P. B. S. Kumar, and R. Radhakrishnan, Physics Reports in press,
10.1016/j.physrep.2014.05.001 (2014).
[8] B. J. Reynwar, G. Illya, V. A. Harmandaris, M. M. Muller, K. Kremer, and M. Deserno,
Nature 447, 461 (2007).
[9] H. Jiang and T. Powers, Phys. Rev. Lett. 101, 018103 (2008).
[10] P. Singh, P. Mahata, T. Baumgart, and S. L. Das, Phys. Rev. E 85, 051906 (2012).
[11] C. Zhu, S. L. Das, and T. Baumgart, Biophys. J. 102, 1837 (2012).
[12] P. B. Sunil Kumar, G. Gompper, and R. Lipowsky, Phys. Rev. E 60, 4610 (1999).
[13] P. Sunil Kumar, G. Gompper, and R. Lipowsky, Phys. Rev. Lett. 86, 3911 (2001).
[14] N. Ramakrishnan, P. B. Sunil Kumar, and J. H. Ipsen, Phys. Rev. E 81, 041922 (2010).
[15] N. Ramakrishnan, P. B. S. Kumar, and J. H. Ipsen, Biophys. J. 104, 1018 (2013).
[16] N. Ramakrishnan, J. H. Ipsen, and P. B. S. Kumar, Soft Matter 8, 3058 (2012).
[17] J. Liu, R. Tourdot, V. Ramanan, N. J. Agrawal, and R. Radhakrishanan, Molecular Physics
110, 1127 (2012).
[18] V. Ramanan, N. J. Agrawal, J. Liu, S. Engles, R. Toy, and R. Radhakrishnan, Integr. Biol.
3, 803 (2011).
[19] B. Sorre, A. Callan-Jones, J.-B. Manneville, P. Nassoy, J.-F. Joanny, J. Prost, B. Goud,
and P. Bassereau, Proc. Natl. Acad. Sci. U.S.A. 106, 5622 (2009).
[20] B. Sorre, A. Callan-Jones, J. Manzi, B. Goud, J. Prost, P. Bassereau, and A. Roux, Proc.
Natl. Acad. Sci. U.S.A. 109, 173 (2012).
[21] A. Tian and T. Baumgart, Biophys. J. 96, 2676 (2009).
32
[22] A. Tian, B. R. Capraro, C. Esposito, and T. Baumgart, Biophys. J. 97, 1636 (2009).
[23] M. Heinrich, A. Tian, C. Esposito, and T. Baumgart, Proc. Natl. Acad. Sci. U.S.A. 107,
7208 (2010).
[24] F. Julicher and R. Lipowsky, Phys. Rev. E 53, 2670 (1996).
[25] U. Seifert, Phys. Rev. Lett. 70, 1335 (1993).
[26] B. R. Capraro, Y. Yoon, W. Cho, and T. Baumgart, J. Am. Chem. Soc. 132, 1200 (2010).
[27] Y. Zhao, J. Liu, C. Yang, B. R. Capraro, T. Baumgart, R. P. Bradley, N. Ramakrishnan,
X. Xu, R. Radhakrishnan, T. Svitkina, and W. Guo, Developmental Cell 26, 266 (2013).
[28] G. S. Ayton and G. A. Voth, Seminars in Cell and Developmental Biology 21, 357 (2010).
[29] H. Cui, G. S. Ayton, and G. A. Voth, Biophys. J. 97, 2746 (2009).
[30] H. Cui, C. Mim, F. X. V´azquez, E. Lyman, V. M. Unger, and G. A. Voth, Biophys. J.
104, 404 (2013).
[31] R. Tourdot, R. Bradley, N. Ramakrishnan, and R. Radhakrishnan, Under Review in IET
Systems Biology.
[32] C.-L. Lai, C. C. Jao, E. Lyman, J. L. Gallop, B. J. Peter, H. T. McMahon, R. Langen, and
G. A. Voth, Journal of Molecular Biology 423, 800 (2012).
[33] G. A. Voth, Biophys. J. 104, 517 (2013).
[34] M. Simunovic, C. Mim, T. C. Marlovits, G. Resch, V. M. Unger, and G. A. Voth, Biophys.
J. 105, 711 (2013).
[35] A. Arkhipov, Y. Yin, and K. Schulten, Biophys. J. 97, 2727 (2009).
[36] C. Mim, H. Cui, J. A. Gawronski-Salerno, A. Frost, E. Lyman, G. A. Voth, and V. M.
Unger, Cell 149, 137 (2012).
[37] M. Marino, K.-H. Moon, and J. E. Hinshaw, Microscopy and Microanalysis 11, 1066 (2005).
[38] W. Helfrich, Z Naturforsch C C 28, 693 (1973).
[39] J. Weinstein and R. Radhakrishnan, Molecular Physics 104, 3653 (2006).
[40] N. J. Agrawal, J. Nukpezah, and R. Radhakrishnan, PLoS Comput Biol 6, e1000926 (2010).
[41] N. J. Agrawal, J. Weinstein, and R. Radhakrishnan, Molecular Physics 106, 1913 (2008).
[42] N. Agrawal and R. Radhakrishnan, Phys. Rev. E 80, 011925 (2009).
[43] N. Ramakrishnan, P. B. S. Kumar, and J. H. Ipsen, Macromol. Theory Simul. 20, 446
(2011).
[44] A. Sari´c and A. Cacciuto, Phys. Rev. Lett. 109, 188101 (2012).
33
[45] A. H. Bahrami, R. Lipowsky, and T. R. Weikl, Phys. Rev. Lett. 109, 188102 (2012).
[46] S. Dasgupta, T. Auth, and G. Gompper, Soft Matter 9, 5473 (2013).
[47] H. Aranda-Espinoza, A. Berman, N. Dan, P. Pincus, and S. Safran, Biophys. J. 71, 648
(1996).
[48] M. M. Kozlov, Nature 447, 387 (2007).
[49] T. Chou, K. S. Kim, and G. Oster, Biophys. J. 80, 1075 (2001).
[50] D. Frenkel and B. Smit, Understanding Molecular Simulation : From Algorithms to Appli-
cations, 2nd ed. (Academic Press, 2001).
[51] B. Widom, J. Chem. Phys. 39, 2808 (1963).
[52] C. H. Bennett, Journal of Computational Physics 22, 245 (1976).
[53] A. de Ruiter, S. Boresch, and C. Oostenbrink, J. Comput. Chem. 34, 1024 (2013).
[54] U. Seifert, Advances in Physics 46, 13 (1997).
[55] F. L. H. Brown, Quart. Rev. Biophys. 44, 391 (2011).
[56] M. G. J. Ford, I. G. Mills, B. J. Peter, Y. Vallis, G. J. K. Praefcke, P. R. Evans, and H. T.
McMahon, Nature 419, 361 (2002).
[57] M. Masuda, S. Takeda, M. Sone, T. Ohki, H. Mori, Y. Kamioka, and N. Mochizuki, The
EMBO Journal 25, 2889 (2006).
34
|
1510.04778 | 1 | 1510 | 2015-10-16T05:08:23 | Mesoscale computational studies of membrane bilayer remodeling by curvature-inducing proteins | [
"physics.bio-ph",
"cond-mat.soft",
"cond-mat.stat-mech"
] | Biological membranes constitute boundaries of cells and cell organelles. Physico-chemical mechanisms at the atomic scale are dictated by protein-lipid interaction strength, lipid composition, lipid distribution in the vicinity of the protein, shape and amino acid composition of the protein, and its amino acid contents. The specificity of molecular interactions together with the cooperativity of multiple proteins induce and stabilize complex membrane shapes at the mesoscale. These shapes span a wide spectrum ranging from the spherical plasma membrane to the complex cisternae of the Golgi apparatus. Mapping the relation between the protein-induced deformations at the molecular scale and the resulting mesoscale morphologies is key to bridging cellular experiments across the various length scales. In this review, we focus on the theoretical and computational methods used to understand the phenomenology underlying protein-driven membrane remodeling. The suite of methods discussed here can be tailored to applications in specific cellular settings such as endocytosis during cargo trafficking and tubulation of filopodial structures in migrating cells, which makes these methods a powerful complement to experimental studies. | physics.bio-ph | physics | Published as Phys. Reports 543, 1–60 (2014)
Mesoscale computational studies of membrane bilayer
remodeling by curvature-inducing proteins.
N. Ramakrishnan
∗
and Ravi Radhakrishnan
†
Department of Chemical and Biomolecular Engineering,
Department of Bioengineering, Department of Biochemistry and Biophysics,
University of Pennsylvania, Philadelphia, PA-19104
Department of Physics, Indian Institute of Technology Madras, Chennai, India - 600036
‡
P. B. Sunil Kumar
5
1
0
2
t
c
O
6
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
8
7
7
4
0
.
0
1
5
1
:
v
i
X
r
a
1
Abstract
Biological membranes constitute boundaries of cells and cell organelles. These membranes
are soft fluid interfaces whose thermodynamic states are dictated by bending moduli, induced
curvature fields, and thermal fluctuations. Recently, there has been a flood of experimental
evidence highlighting active roles for these structures in many cellular processes ranging from
trafficking of cargo to cell motility.
It is believed that the local membrane curvature, which
is continuously altered due to its interactions with myriad proteins and other macromolecules
attached to its surface, holds the key to the emergent functionality in these cellular processes.
Mechanisms at the atomic scale are dictated by protein-lipid interaction strength, lipid com-
position, lipid distribution in the vicinity of the protein, shape and amino acid composition of
the protein, and its amino acid contents. The specificity of molecular interactions together with
the cooperativity of multiple proteins induce and stabilize complex membrane shapes at the
mesoscale. These shapes span a wide spectrum ranging from the spherical plasma membrane to
the complex cisternae of the Golgi apparatus. Mapping the relation between the protein-induced
deformations at the molecular scale and the resulting mesoscale morphologies is key to bridging
cellular experiments across the various length scales. In this review, we focus on the theoretical
and computational methods used to understand the phenomenology underlying protein-driven
membrane remodeling. Interactions at the molecular scale can be computationally probed by
all atom and coarse grained molecular dynamics (MD, CGMD), as well as dissipative particle
dynamics (DPD) simulations, which we only describe in passing. We choose to focus on several
continuum approaches extending the Canham - Helfrich elastic energy model for membranes to
include the effect of curvature-inducing proteins and explore the conformational phase space of
such systems. In this description, the protein is expressed in the form of a spontaneous curva-
ture field. The approaches include field theoretical methods limited to the small deformation
regime, triangulated surfaces and particle-based computational models to investigate the large-
deformation regimes observed in the natural state of many biological membranes. Applications
of these methods to understand the properties of biological membranes in homogeneous and in-
homogeneous environments of proteins, whose underlying curvature fields are either isotropic or
anisotropic, are discussed. The diversity in the curvature fields elicits a rich variety of morpho-
logical states, including tubes, discs, branched tubes, and caveola. Mapping the thermodynamic
stability of these states as a function of tuning parameters such as concentration and strength of
2
curvature induction of the proteins is discussed. The relative stabilities of these self-organized
shapes are examined through free-energy calculations. The suite of methods discussed here can
be tailored to applications in specific cellular settings such as endocytosis during cargo trafficking
and tubulation of filopodial structures in migrating cells, which makes these methods a powerful
complement to experimental studies.
PACS numbers: 87.16.-b,87.17.-d
Keywords: self-assembly, hydrophobicity, hydrophilicity, cell membrane, lipid bilayer, continuum models,
Helfrich Hamiltonian, molecular dynamics, triangulated surfaces, Monte Carlo, free energy
∗
†
‡
[email protected]
[email protected]
[email protected]
3
CONTENTS
I. Introduction to membranes
A. Physiological significance of lipid membranes
B. Molecular description of lipid membranes
C. Chemical heterogeneity and lipid organization in multicomponent GUVs
D. Morphological transitions and the role of thermodynamic variables
E. Membrane remodeling by curvature inducing factors
II. Thermodynamics-based models for membranes
A. Thin sheet approximation of membranes
B. Theory of bent plates
Uniaxial bending
Biaxial bending
C. Canham - Helfrich phenomenological theory for membranes
1. Gauss-Bonnet theorem and the Gaussian bending term:
D. Thermal softening of elastic moduli
E. Ensembles for thermodynamic description
III. Overview of analytical and computational methods
A. Energy minimization and stationary shapes
B. Membrane dynamics
C. Surface of evolution formalism
D. Direct numerical Minimization
E. Fourier space Brownian dynamics (FSBD)
F. Dynamically triangulated Monte Carlo methods.
G. Particle-based models
H. Molecular and coarse grained approach for modeling membranes
IV. Modeling membrane proteins as spontaneous curvature fields
A. Curvature induction in membranes—intrinsic and extrinsic curvatures
B. Isotropic curvature models
4
7
10
13
18
20
23
25
26
27
29
29
30
31
32
33
35
36
36
38
40
41
42
45
47
52
52
63
C. Membranes with in-plane order: Model for anisotropic curvature-inducing
proteins
1. Membrane with in-plane field
2. Monte Carlo procedure for nematic membranes
D. Anisotropic bending energy
E. Properties of nematic membranes
1. Conformational phase diagram for κ⊥ = 0
F. Pairing of defects: the role of Gaussian curvature
G. Implication of defect structures in biological membranes
H. Mapping the length scales
V. Applications of the particle-based EM2 model in the study of morphological
transitions in membranes mediated by protein fields
A. Discretization to a particle-based model
B. Isotropic and anisotropic protein fields in the EM2 model
C. Membrane remodeling by N-BAR and F-BAR proteins
VI. Free energy methods for membrane energetics
A. Thermodynamic integration (TI)
B. Free energy for a membrane in the Monge gauge subject to isotropic
curvature fields
C. Thermodynamic integration methods for the anisotropic curvature model
VII. Conclusions
VIII. Acknowledgments
IX. Appendix
A. Differential geometry
B. Planar Monge gauge
1. Elastic energy of a planar membrane
C. Surface quantifiers on a triangulated surface
5
66
67
68
70
71
72
74
77
79
80
80
82
84
88
88
89
94
97
98
99
99
100
101
102
D. Householder transformation
E. Parallel transport of a vector field on a triangulated patch
F. Properties of a membrane with in-plane order
1. Defect structure in polar and nematic fields
2. Defect core and its winding number
3. Topological charge and its minimum
4. Organization of defects of non deformable spherical surfaces
5. Non-curvature-inducing in-plane field and deformable membranes
G. Regularized delta function
References
107
107
108
109
111
112
114
115
117
117
6
I.
INTRODUCTION TO MEMBRANES
Cell membranes are biological structures involved in a wide range of biological processes
and ubiquitous in both prokaryotic and eukaryotic cells. Mouritsen, in his book “Life as a
matter of fat” [1], aptly describes a membrane as functioning as a barrier, a carrier, and a
host. Namely, membranes constitute a barrier that delineates the outside from the inside
of a cell, separates a cell from another, and in addition, encapsulates most cell organelles
in eukaryotic cells. Membranes have been long known to be involved in the trafficking of
cellular cargo. They play an integral role, as a carrier, in the processes of endocytosis and
exocytosis, which represent key inter- and intra-cellular transport mechanisms that aid
in cellular uptake of cargo ranging from nutrients to pathogens. In its role as a host, the
membrane is home to a large set of proteins, ligands, and various other macromolecules
which are involved in processes that span a wide spectrum from cell signaling to cell
replication.
A biological membrane results from the complex assembly and organization of different
kinds of fatty acid molecules called lipids. In addition to the lipid molecules, a cell mem-
brane also comprises a large number (concentration) of proteins and a relatively small
number (concentration) of carbohydrates. Depending on the type of cell membrane inves-
tigated, the protein concentration varies between 18% and 75% with the corresponding
protein to lipid ratio varying between 0.23 and 1.6 [2]. Carbohydrate molecules have
been estimated to have concentrations in the range of 3% − 10%. Molecular composition
and the complex interactions between the individual components are key factors that
determine the resulting macroscopic shapes of the biological membrane, which in their
role as a barrier also influence the shape of the cells and cell organelles they enclose.
When categorized on the basis of their complexity, membrane structures reported in the
biological literature vary from simple, symmetric, primitive spherical shapes, commonly
seen in the case of the plasma membrane, to the highly complex, convoluted structures
displayed by organelles like the endoplasmic reticulum and the golgi. The genesis of these
shapes and the other more complex shapes like those shown in Fig. I, has been extensively
investigated in cell biology. Despite the efforts over these years, a generic framework to
explain all the observed shapes does not exist. Understanding the mechanisms governing
cell membrane organization would be instructive and can help gain insight into the more
7
complex question of “how do the cell and its organelles get their shape ? ”.
8
FIG. 1. Simple and complex shapes seen in cells and cell organelles. Image from [3] (Reprinted
by permission from Macmillan Publishers Ltd: Nat. Rev. Mol. Cell Biol. 8(3), 258–264 copyright
(2007)). (a,b) Electron microscope images showing various cell organelles in a yeast cell (a)
and in a normal rat kidney cell (b). The major organelles shown—namely, the Golgi(G), the
nucleus(N), the endoplasmic reticulum(ER), the vacuole(V), and the mitochondrion(M)—have
identical shapes in both yeast and normal rat kidney cells, and the shapes of these organelles are
also preserved in other cells. (c) Fluorescence image of the nuclear envelope(NE) and ER in COS
cells, wherein the bright spots denote the localization of a particular ER associated protein. (d)
The nuclear envelope that defines the boundary of a cell nucleus is a relatively smooth, double
bilayer structure which merges with the rough and tubular endoplasmic reticulum. (e) Doubled
walled structure of the mitochondrion; its outer membrane has a smooth shape while its inner
membrane is organized into a complex network of tubular shapes called the cristae. (f ) Tubules
and flattened sacs—the primary structures that dominate the morphology of the Golgi stacks.
9
acefdbGGGNER tubuleERLumenNENVLMMMERERNECNSDGolgistackcis fingerstrans cisternaetrans ERmight be determined by the rise and pitch of the spirals formed during the oligomerization of the F1F0-ATP-synthase dimers.Bilayer-couple effect. Some proteins might use the so-called bilayer-couple effect to curve membranes and affect organelle shape. As hydrophobic interactions between the two leaflets of a membrane bilayer tend to keep them coupled together, a significant change in the surface area of either leaflet, perhaps by as much as a few percent5, causes the membrane to bend. Proteins can there-fore curve membranes by inserting into only one (or primarily one) leaflet of a bilayer.For an integral membrane protein to induce or stabilize membrane curvature throughout an organelle, it must be abun-dant enough to encompass the entire region of the organelle that is being actively shaped. Such a mechanism has been proposed recently for the reticulons and DP1/Yop1, which are abundant integral membrane proteins that are required to maintain proper ER tubular structure14. These proteins have an unusual membrane topology; they insert two pairs of α-helices partially into the ER membrane. The hydrophobic stretches that form the hairpin helices are not long enough to completely span the membrane (they are only 30–34 amino acids long). Therefore, these short hairpins probably occupy more space in the outer than the inner leaflet of the ER membrane, causing it to curve and tubulate14 (FIG. 2c).Interestingly, these proteins have a similar topology to caveolin, which also inserts an unusually short hairpin into the membrane bilayer at the plasma membrane, and this feature is at least partially responsible for generating membrane curvature at caveolae. Caveolin, the F1F0-ATP synthase, reticulons and DP1/Yop1 therefore share several proposed mechanistic features: they are all oligomerizing integral membrane proteins, they are concentrated at the region that they are shaping, and they have unusual transmembrane-domain properties that might generate membrane curvature.Tubule formation probably requires more than an increase in membrane curvature, and the abundant and oligomerizing reticulons and DP1/Yop1 probably also use a scaffold-ing mechanism to stabilize tubules (this is analogous to caveolin at caveolae and to the F1F0-ATP-synthase complex in mitochon-drial tubules). The most logical arrangement for all these proteins would be as spirals or rings that encompass the region of the mem-brane that they shape. These structures have not all been decisively demonstrated by EM.Figure 1 Organelles have complex, conserved shapes. Electron microscopy (EM) images of the yeast Pichia pastoris (a) and a mammalian normal rat kidney (NRK) cell (b) showing: the Golgi stack (G), nucleus (N), peripheral endoplasmic reticulum (ER), mitochondrion (M) and vacuole or lysosome (V or L). Note the similar organelle shapes in both cells. Scale bar in a represents 0.5 µm and in b represents 1 µm. A confocal fluorescence image of a COS cell expressing an ER-localized protein labelled with green fluorescent protein (c). Both the nuclear envelope (NE) and the ER are formed from a single, continuous membrane. Scale bar represents 5 µm. EM image of an NRK cell highlights the even spacing (~50 nm) of the two membranes of the NE (d); both are continuous with the tubular branches of the peripheral ER. Scale bar represents 0.2 µm. Single section of a three-dimensional tomogram of a chick dendrite mitochondrion (e). The boxed sections highlight the conserved ~20-nm spacing between the inner and outer mitochondrial membranes (S) and the ~30-nm diameter of cristae tubules (d). Scale bar represents 0.1 µm. EM image of an NRK cell showing the multiple cisternae of a Golgi stack (f). Note the regular intercisternal spacing (~20 nm) and the irregular intracisternal lumenal spacing. Scale bar represents 0.2 µm. Part a is reproduced with permission from REF. 54 © (1999) Rockefeller University Press. Part e is reproduced with permission from REF. 1 © (2000) Elsevier. Part c was kindly provided by J. Rist, Cambridge University, UK. Parts b, d and f were kindly provided by M. Ladinsky, University of Colorado, USA. PERSPECTIVESNATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 8 MARCH 2007 259©!2007!Nature Publishing Group Currently, in vivo experimental methods have reached a degree of sophistication high
enough to quantify the physical and chemical properties of the individual components
of a lipid membrane. Further, computer-based algorithms and supercomputers have un-
dergone a transformative change, which in turn allow one to model highly complex sys-
tems, such as membrane-bound macromolecules. These factors together have widened
the horizons of membrane science beyond single-component and multi-component lipid
membranes to also focus on the role of proteins/membrane-associated macromolecules in
curvature induction, curvature detection (or sensing), and membrane remodeling. How-
ever, the problem at hand is too complex to interpret in terms of experiments alone, and
the study of membranes at the cellular scale is not yet amenable to molecular simulations.
Mechanics- and thermodynamics-based continuum modeling is a powerful alternative to
investigate the behavior of membranes spanning length scales extending well above a few
tens of nanometers. In this approach, the large microscopic degrees of freedom associated
with the lipids and proteins in the membrane are represented in terms of a few macro-
scopic observables that obey well-defined mechanical and thermodynamic principles. In
this article, we will focus on the various theoretical and computational methods that are
widely used in the study of membranes in this continuum limit relevant to the cellular
scale.
A. Physiological significance of lipid membranes
The membrane interacts with almost all major entities of a cell—namely, proteins,
nucleic acids, and polysaccharides. In addition it is an important member of many sig-
naling pathways and hence plays a pivotal role in many crucial decisions determining
cell fate such as cell motility, metabolism, proliferation, and survival. The relationship
between cellular pathology and abnormality in the cellular membrane has been observed
in a wide variety of diseases :
for instance, sickle cell anemia is linked to enhanced
phosphatidylserine levels, Duchenne muscular dystrophy results from the breakdown of
cytoskeletal membrane anchoring, amyloid-related diseases like Alzheimer’s show signa-
tures of disrupted leaky membrane bilayers, and cancer metastasize in a tissue by breaking
the cell-cell junction, which in turn leads to a neoplasm-promoting microenvironment in
the tissue [4, 5]. As described above, the observed membrane anomalies can span length
10
scales ranging from nanometers (molecular scale) to microns and beyond (tissue scale).
For instance, at the nanoscale, one observes local perturbations in the organization of
membrane constituents, whereas at length scales comparable to a cell, the diseased cell
displays noticeable change in its structure and its organization within a tissue. In spite
of being separated over large length scales, the observations at the molecular and cellular
scales can be coupled—chemical changes precede morphological changes and vice versa.
The relation between the biochemistry and structure is inherently a multi-scale process,
which makes membranes a complex system to deal with. In this section, using the specific
examples shown in Fig. 2, we will re-iterate the multi-scale nature of membranes and their
role in maintaining the integrity of the cell and tissue.
11
FIG. 2. Membrane protein interactions at multiple length scales. (a) Cell junctions formed
by transmembrane linker proteins; the brigt regions denote the junctions formed by Integrins
and Cadherins [6](Reprinted by permission from Macmillan Publishers Ltd: Nature 453(7), 453–
456 copyright (2008)), (b) Illustration of the various inter-cellular junctions
[7](Reprinted by
permission from Macmillan Publishers Ltd: Nat. Rev. Mol. Cell Biol. 5, 542–553 copyright (2004)),
(c) The vesicular buds formed by the process of clathrin-mediated endocytosis (CME)
[8]
(Reprinted by permission from Macmillan Publishers Ltd: Nat. Cell Biol. 14(6), 634–639 copyright
(2012)), (d) A cartoon of a clathrin coated vesicle which shows the spatial localization of various
accessory-proteins involved in CME [9] (Reprinted by permission from Macmillan Publishers Ltd:
Nat. Rev. Mol. Cell Biol. 3(12), 971–977 copyright (2002)), (e) Spontaneous tubulation of a
liposome due to addition of epsins [10] (Reprinted by permission from Macmillan Publishers Ltd:
Nature 419, 361–366 copyright (2002)), (f, g) A molecular picture which shows the insertion of
the α-helix (helix-0), of an epsin, into a membrane leaflet and its specific interactions with a
negatively charged PIP2 lipid [11] (Reprinted from J. Mol. Biol., 423(5), C-L Lai et. al, Membrane
Binding and Self-Association of the Epsin N-Terminal Homology Domain, 800–817, Copyright (2012),
with permission from Elsevier).
Cadherins are a family of transmembrane proteins found in cells. The cytoplasmic
domain of a Cadherin is linked to cytoskeletal filaments while the extracellular domains
of adjacent cells interact to form adherens junctions. At the scale of a tissue, cells in the
tissue are kept together by these adherens junctions, and hence the integrity of the tissue
12
is determined by the strength of these junctions. In Fig. 2(a) the bright regions mark
the spatial location of Cadherins, and the illustration in Fig. 2(b) shows the representa-
tive position of the Cadherins with respect to the cell membrane. Though the adherens
junctions are formed in the extracellular region, its structure and strength are determined
by the local membrane environment around the Cadherins. It has been shown that the
Cadherins localize to membrane micro-domains rich in cholesterol
[12, 13] and changes in
the membrane lipid composition disrupts the structure of adherens junctions [14]. This
is an obvious case of membrane biochemistry driving tissue structure, discussed at the
start of this section.
At the cellular level, exocytosis and endocytosis are key processes in the bidirectional
transport of inbound and outbound cargo across a membrane barrier. Interestingly, these
processes also significantly influence the cadherin levels on the plasma membrane which
in turn influence the stability, polarity, and motility of the cell. Fig. 2(c) and (d) are
representative images of clathrin-mediated endocytosis (CME) that involves the formation
of vesicular buds due to the cooperative action of proteins such as clathrin, adaptor protein
2 (AP2), epsin, and dynamin on a lipid bilayer environment. However, the action of just
one of these proteins—namely, epsin—results in membrane tubulation (see Fig. 2(e)).
Epsin-membrane interactions are believed to cause deformations on the membrane at
length scales of nanometers due to the insertion of its helix-0 (α-helix) into one of the
leaflets of the bilayer (Fig. 2(f)). The nature and strength of such interactions also depend
on the lipid environment such as the presence or absence of PIP2 (Fig. 2(g)). In order
to develop an understanding of such an inherently multiscale system, there is a need to
employ a scale-dependent description of the membrane. This is achieved by the technique
of coarse graining, which allows for the flow of information between the different scales.
As a first step of coarse graining, we will develop some insights into how the molecular
picture of membranes translates into the corresponding one at the (∼ 100 nm) mesoscale.
B. Molecular description of lipid membranes
Lipids are one among the four building blocks of biology, with the other three being
amino acids, nucleic acids, and sugars [1]. These fatty acids, which are carboxyl-group-
containing hydrocarbons, are the most abundant molecules in the cell, numbering over a
13
thousand different types in both eukaryotes and prokaryotes [15]. From a biological point
of view, supramolecular organization of lipids has low functionality compared to biopoly-
mers like proteins and DNA/RNA, which are poly-amino acids and poly-nucleotides re-
spectively. A lipid molecule can either be polar or apolar, with the former being hy-
drophilic and the latter being hydrophobic, depending on the chemical moieties attached
to the carboxyl group. Apolar lipids in a polar solvent, aggregate into lipid droplets that
are known to be the energy store of a cell
[16, 17] and are interesting from a functional
point of view.
FIG. 3.
A lipid molecule and its self organized structures—shown are, (a) the chemical
composition of a DMPC lipid, with the hydrophobic and hydrophilic regions shaded differently,
(b) the representative model of a lipid, with the head and tail groups marked, (c) a cross section
of a flat membrane bilayer, and (d) a cross section of a closed bilayer (generally called a vesicle).
The structurally important polar lipid molecules are characterized by a hydrophilic
part called the head and a hydrophobic chain called the tail. This is illustrated in
Fig. 3(a) and (b) for the case of a dimyristoyl-phosphotidylcholine or DMPC lipid. When
lipid molecules are introduced into an aqueous solvent with concentration above a criti-
cal value, called the critical micelle concentration, they spontaneously partition into an
interface that shields the hydrophobic tails from the solvent. The simplest realization
of such an interface is a lipid bilayer, a cartoon of which is shown in Fig. 3(c) and (d).
Depending on the area to volume ratio of the lipid molecule, self-assembled structures
like micelles, cylindrical micelles, multi-lamellar stacks, and bi-continuous phases can also
be stabilized [18]. The phase diagrams of over 2000 well-characterized lipid mixtures has
been compiled by Koynova and Caffrey [19]. At the molecular scale, the organization and
14
(cid:1)(cid:1)(cid:1)(cid:2)(cid:3)(cid:4)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(c) flat bilayer(d) closed bilayer5 nmHydrophilic(a) 14:0 dimyristoyl-phosphodylcholineHydrophobic(b)HeadTail interaction of the lipid molecules with other biological entities is predominantly governed
by the chemistry of the lipid molecules.
Eukaryotic and prokaryotic organisms have over 1000 types of lipid molecules, and
these molecules can be broadly divided into three major classes—namely, glycerol-based
lipids, cholesterol, and ceramide based sphingolipids
[5]. Even lipids belonging to the
same class exhibit large chemical diversity due to variations in the hydrophilic head
groups and differences in the number, length and saturation of the hydrocarbon chain
(tail); see Fig. 3(a). Phospholipids for instance can have a variety of head groups, like
phosphatidylcholine (PC), phosphatidylserine (PS), phosphatidylethanolamine (PE), and
phospatidylglycerol (PG), and these groups can be uncharged, anionic, cationic, or zwitte-
rionic. The organization of lipids in a multi-component lipid membrane is well described
by the fluid mosaic model
[20], which describes cell membranes as “two-dimensional
solutions of lipids and other macromolecules”.
Membranes in mammalian cells consist primarily of phospholipids and glycerol. Other
classes of lipids that are present in smaller quantities are nevertheless essential for the
cell to perform specific biological processes. Variations in lipid composition have been
shown to impact a host of cellular properties like exocytosis, endocytosis, phagocytosis,
sensitivity of receptor molecules to extracellular signaling molecules, and cytotoxicity
[16, 21]. Systematic studies to understand the correlation between lipid composition
and membrane organization/function show that even variations in the same class of lipid
across different cells can produce different effects. Hence, in vitro experiments using re-
constituted cell membranes are hard to interpret due to the highly complex organizational
landscape of the constituent lipid molecules.
Though it is hard to understand the morphological properties and organizational pat-
terns even for membranes constituted from binary/ternary lipid mixtures, all biological
membranes display some key microscopic properties that are key to our understanding
of macroscopic models introduced later. We given a brief summary of a few of these
properties below:
15
A two-dimensional fluid: The absence of bonded interactions between the lipid molecules
in a bilayer allows the lateral diffusion of lipids in the plane of the leaflet it resides
in. Experimental observations based on fluorescent tagging and electron spin resonance
(ESR) spin tagging of lipid molecules have estimated the diffusion constant of lipids in a
bilayer membrane to be of the order of 10−12m2s−1 [1, 15]. As a result, lipid bilayers do
not resist shear stresses, like a solid, but instead sustain a flow-field when sheared, like a
liquid. Lipid molecules can also be translocated from one leaflet of the bilayer to another.
The translocation can be a result of thermal fluctuations or specialized lipid translocator
proteins, called flippases and floppases, that are normally found in the membranes of cells.
Lipid flip flop is a slow process compared to most processes associated with a membrane.
The spontaneous rate of translocation in the absence of these specialized proteins ranges
from hours to days. Although they are fluid like, lipid membranes display elastic-like be-
havior in response to normal stresses; this topic is discussed at length in section II C, and
also forms the basis for much of this article. In addition, lipid membranes are selectively
permeable to ions
[22], poly-electrolytes
[23], and many other small molecules. As a
result of this semi-permeable nature, they can maintain different chemical environments
in the interior and exterior regions. It should also be noted that every cell organelle has
a chemical environment different from the other, which allows them to perform a unique
biological function. Semi-permeability combined with flexibility makes lipid membranes
effective barriers. As will be discussed below, many biological processes can also be con-
trolled by modulating the curvature on the membrane surface.
Domain formation in multi component membranes: By the Gibbs phase rule, heterogeneity
in lipid composition can give rise to a variety of coexisting phases in the form of lipid
domains in each of the leaflets of a multi-component lipid membrane [24, 25]. These
domains are mainly formed due to mismatch in the lengths of the hydrophobic chains or
due to preferential partitioning of lipids.
16
FIG. 4. Representative phases in a lipid bilayer: shown are three distinct phases namely the
gel phase (a), the liquid-disordered, Ld phase (b), and the liquid-ordered, Lo phase (c). Solid
and liquid implies that the nature of translational correlations in the plane of the membrane
are either solid-like or liquid-like (top view of (a) shows a lattice structure, characteristic of a
solid, while (b) and (c) have no lattice structure and hence are liquid like). Hydrophobic chains
in the ordered phase and gel phase have strong orientational correlations, whereas those in the
disordered phase have random orientations and thus have weak orientation correlations. In (c)
the shaded triangles represent smaller lipid molecules like cholesterol and sphingolipids that
intercalate into the hydrophobic region.
A lipid domain can exist in three distinct phases—namely, (a) gel phase, (b) liquid-
disordered (Ld), and (c) liquid-ordered (Ld), as shown in Fig. 4 [26–28]. A lipid membrane
shows a gel phase at temperatures T < Tm and makes a transition to the Ld or Lo phase,
depending on the concentration, when the temperature exceeds the transition temper-
ature Tm. The gel phase is characterized by the presence of long-range translational
correlations in the position of the head groups, as seen in the top view of Fig. 4(a),
whereas the absence of any such correlation is a signature of the liquid phase. On the
other hand, the orientations of the lipid tails are correlated in the ordered phase and
uncorrelated in the disordered phase. The gel phase of lipid domains is highly relevant to
model bilayers and is not seen in biological membranes. In the case of the Lo phase, which
is commonly observed in multi-component lipid membranes, the hydrophobic chains ac-
quire orientational order even at T > Tm due to the intercalation of smaller lipids into
the hydrophobic region, which in turn leads to the arrest of the acyl chain degrees of
freedom. The intercalating lipids are mainly cholesterol and sphingo-lipids, shown as
17
(a)(b)(c)top viewside viewGel phaseliquid ordered ( )liquid disordered ( ) triangles in Fig. 4(c). The phase diagram of many two-component and three-component
lipid mixtures have been well studied in the literature [29–33]. Lipid domains in a bilayer
leaflet can either be correlated [34] or uncorrelated with the lipid domains in the other
leaflet of the membrane. How these microphase separations, such as lipid rafts [35–37],
seen in multi-component membranes are related to (and functionally relevant to) cellular
phenomena is still a matter of open debate.
Much of our understanding of lipid organization has been derived from the study of
an in vitro membrane system called Giant Unilamellar vesicle (GUV). These micron- to
millimeter-sized vesicles are formed from lipid mixtures through processes like sonication
of multi-lamellar vesicles and electroformation of dry lipid films; see [30] for a detailed
review of the experimental techniques. A GUV assembled from a single lipid species is
called a single-component vesicle, whereas that containing multiple lipid species is called
a multi-component vesicle. GUVs alleviate many complexities seen in in vivo membrane
systems since their chemical heterogeneity can be precisely controlled, and the size of
GUVs allows them to be observed under a microscope. These properties make GUVs the
most used experimental system to investigate lipid organization in membranes. In the
next two sections, based on experimental observations in GUVs, we will briefly describe
how change in chemical heterogeneity drives lipid organization and also how perturbations
in the thermodynamic variables drive morphological changes.
C. Chemical heterogeneity and lipid organization in multicomponent GUVs
Fig. 5 shows how the phases and organization of lipids are modulated by the compo-
sition of cholesterol in a GUV formed from a ternary lipid mixture of saturated DMPC,
unsaturated DMPC, and cholesterol. Composition of the 16 carbon chain long saturated
and unsaturated DMPC lipid were taken at 1:1. In the absence of cholesterol (Fig. 5(a)),
the saturated lipids organize into a non-circular solid phase (same as the gel phase de-
scribed in Fig. 4(a)), shown as dark regions, that coexists with a liquid phase, shown
by the bright regions in the micrograph. The solid domain is further characterized by
decrease in lipid mobility and moves and rotates as a rigid body. With increase in choles-
terol content, for concentration in the range of 10%− 50% mol, the solid phase is replaced
18
by the Lo phase (see Fig. 4(c)) that coexists with a background liquid phase, leading to a
liquid-liquid coexistence, as seen in Fig. 5(b) and (c). Phase coexistence becomes unstable
beyond a cholesterol concentration of 55% mol, and the GUV becomes uniformly bright,
which is characteristic of the Ld phase (Fig. 4(b)).
FIG. 5. Phase segregation into liquid-ordered and liquid-disordered domains in the presence
of cholesterol in multi-component vesicular membranes, constituted from saturated and unsat-
urated DMPC lipids. The fluorescence images show the coexistence of various lipid phases at
different concentrations of cholesterol; (a) gel-Ld at 0 mol%; (b) Lo-Ld at 30 mol%; (c) Lo-Ld at
40 mol%; and (d) no visible phase separation at 50 mol%. Image adopted from [29] (Reprinted
figure with permission from Sarah L Veatch and Sarah L Keller, Phys. Rev. Lett., 89 (26), 2681011
and 2002. Copyright (2002) by the American Physical Society.)
The ternary lipid mixture discussed above resembles the lipid composition called the
raft mixture. Sphingolipids in the presence of cholesterol can assemble into a special-
ized cholesterol rich structure called rafts
[38], which are lipid domains in the liquid-
ordered phase (Lo) [35]. Rafts are believed to be membrane micro-domains enriched in
glycophosphatidylinositol and GPI-anchored proteins, which play an important role in
signal transduction, membrane trafficking, cytoskeletal organization, and pathogen entry
[26]. This example clearly illustrates how lipid heterogeneity affects lipid organization in
membranes and how the organization can be used effectively by cells to perform biological
processes. It should also be noted that the stability of lipid phases described above is
a function of thermodynamic variables like temperature and pressure. The experiments
described above were performed at a temperature 5°C lower than the liquid-ordered to
liquid-disordered transition temperature. The cholesterol-dependent gel -Ld and Lo-Ld
phase coexistence, described in Fig. 5, would show a completely different behavior or may
even be unstable if the experiments are performed at temperatures above this transition
19
OrganizationinLipidMembranesContainingCholesterolSarahL.VeatchandSarahL.Keller*DepartmentsofChemistryandPhysics,UniversityofWashington,Seattle,Washington98195-1700(Received13June2002;published9December2002)Afundamentalattributeofraftformationincellmembranesislateralseparationoflipidsintocoexistingliquidphases.Usingfluorescencemicroscopy,weobservespontaneouslateralseparationinfree-floatinggiantunilamellarvesicles.Werecordcoexistingliquiddomainsoverarangeofcompo-sitionandtemperaturesignificantlywiderthanpreviouslyreported.Furthermore,weestablishcorre-lationsbetweenmiscibilityinbilayersandinmonolayers.Forexample,thesamelipidmixturesthatproduceliquiddomainsinbilayermembranesproducetwouppermiscibilitycriticalpointsinthephasediagramsofmonolayers.DOI:10.1103/PhysRevLett.89.268101PACSnumbers:87.16.Dg,64.70.Ja,64.75.+gMammaliancellsaresurroundedbyanouterwallor‘‘plasmamembrane’’ofproteinsandlipidsarrangedinopposingleafletsofabilayer.Thereisgrowingevidencethatthismembraneisnotuniform,butinsteadlaterallyorganizesintoregionscalled‘‘rafts.’’Theseareliquiddomainsinwhichcholesterol,saturatedlong-chainedlipids,andparticularproteinsareconcentrated[1].Raftsarecurrentlyofgreatinteresttocellbiologists,immu-nologists,andphysicalscientistsalike[2].Asanexample,raftshavebeenimplicatedinimportantcellfunctionssuchasendocytosis,adhesion,signaling,apoptosis,pro-teinorganization,andlipidregulation[1,3,4](andrefer-encestherein).Furthermore,pathogensmayrecruitspecificlipidstoasitetoaidinfectionofacell[5].Asimplersysteminwhichtostudyphysicalpropertiesoflipidsisinabilayermembraneofavesicle.Ex-perimentalsupportforbothraftsincellsandcoexistingliquiddomainsinvesicleshasbeengatheredbyavarietyofmethodsincludingdetergentinsolubility[6],single-moleculetracking[7],fluorescencequenching[8],fluorescenceenergytransfer[9],andaggregationoffluo-rescentlylabeledproteins[4].Recently,directvisualiza-tionofmicron-scaleliquiddomainswasaccomplishedinbilayers[10,11].Forbilayersofaparticularmixtureofphospholipids,sphingomyelin,andcholesterol,amisci-bilitytransitionwasfoundbelowthelipidmeltingtem-perature[11].Todate,fewlipidmixtureshavebeenstudiedbythismethodandthemiscibilitytransitiontemperaturehasnotbeensystematicallyexplored.Weareinterestedinhowstrictlylipidcompositionmustberegulatedforliquiddomainstoforminvesicles.There-fore,wehaveinvestigatedvesiclemembranescontainingseveraldifferentphospholipidsandawiderangeofcholesterolcompositions.Bysystematicallyvaryingthetemperatureofourvesicles,wehaveassembledex-tensivemiscibilityphasediagramsbasedonfluorescencemicroscopy.Coexistingliquidphasesalsoexistinlipidmonolayersatanair-waterinterface,asobservedbyavarietyofresearchers(e.g.,[12]).Therehasbeenapersistentques-tionofhowexperimentalresultsinmonolayersystemscanbeappliedtobilayers.Bystudyingthesamelipidmixturesinmonolayersasinourvesicles,weareabletocomparethephasebehaviorofbothsystems.Giantunilamellarvesicles.—Vesiclesweremadewithternarymixturesofasaturatedphosphatidylcholinelipid[eitherdi(14:0)PC,di(15:0)PC,di(16:0)PC,ordi(18:0)PC],anunsaturatedphospholipid[di(18:1)PC],andcholesterol.Thesemixtureswerechosentomimicraft-formingcompositionsincellmembranes[10,11].In-dividualvesicleswithinapopulationmayvaryslightlyincomposition,estimatedas2mol%cholesterol.Thisuncertaintyresultsinarangeoftransitiontemperatures[13].Phospholipids(AvantiPolarLipids,Birmingham,AL),cholesterol,anddihydrocholesterol(Sigma,St.Louis,MO)werestoredatÿ20Candusedwithoutfurtherpurification.Aminimalamount(0.8mol%)ofTexasReddi(16:0)-phosphatidylethanolamine(Mole-cularProbes,Eugene,OR)wasusedasadyeforcontrast.FIG.1.Fluorescencemicrographsofvesiclesandmonolayersof1:1di(18:1)PC/di(16:0)PCplusvaryingamountsofcholes-terol.Scalebarsare20m.Vesiclesare<5Cbelowtheirphasetransitiontemperaturesandexhibiteithersolid-liquidphasecoexistence(a),liquid-liquidphasecoexistence(b),(c),ornovisiblephaseseparation(d).Vesicledomainsarenotnecessarilyatequilibriumsizesandcoalescewithtime.Monolayersareat
270:5Candbelowtheirphasetran-sitionsurfacepressuresandexhibitliquid-liquidphasecoex-istence(e)–(g).Monolayerdomainsizesdonotsignificantlychangeonthetimescaleofexperiments(minutes).VOLUME89,NUMBER26PHYSICALREVIEWLETTERS23DECEMBER2002268101-10031-9007=02=89(26)=268101(4)$20.00©2002TheAmericanPhysicalSociety268101-1 temperature.
D. Morphological transitions and the role of thermodynamic variables
GUVs, or in general membranous structures, show noticeable morphological transitions
in response to a perturbation in the thermodynamic environmental variables, like temper-
ature and pressure. Fig. 6 shows the range of thermally undulated shapes displayed by
pure phospholipid bilayer membranes in response to temperature change. The spherical
vesicle shown in Fig. 6(1), at T = 27.2°C, transforms into the budded vesicle, shown in
Fig. 6(6), when the temperature is changed to T = 41°C. As a function of increasing
temperature, the budding process proceeds through a series of conformational changes
from spherical to oblate ellipsoid to prolate ellipsoid to pear shaped which discontinu-
ously transforms into a budded membrane. These shape changes are accompanied by an
increase in the surface area of the vesicle at nearly constant volume.
20
FIG. 6. Shape transformations in a DMPC vesicle in response to a change in temperature from
27.2࠷ - 41࠷. An initially spherical vesicle (1), at 27.2࠷, transforms into an oblate ellipsoid
(2), at T = 36.0࠷. With further increase in temperature, pear shaped vesicles (3,4,5) are
stabilized, for 36.0࠷ < T < 41.0࠷, which transforms into a budded vesicle (6), at T = 41࠷;
Image adopted from reference [39] (Reprinted from Biophys. J, 60 (4), J. Kas, E. Sackmann,
Shape transitions and shape stability of giant phospholipid vesicles in pure water induced by area to
volume changes, 825–844, Copyright (1991), with permission from Elsevier).
21
(2):T=36.00CV-I2200pm3A=2770pm2(3):T=37.50CV=10800gm3A=2750gm2(5):T=41.0CV=11900pm3A=281Ogm2BiophysicalJournal(4):T=39-1OCV-I2000pm3A=2800pm2IlOgmI(6):T=41.OOCV=12000pm3A=2820gm2Volume60October1991828(1):T=27.20CV=l2200gM3A=2570pM2 FIG. 7. Various pathways for the deformation of an initially biconcave vesicle, whose top and
side views are shown in (a), to elongated filament like stable shapes, shown in frames (x), (y),
(z), and (x
(cid:48)
), under the influence of osmotic pressure. The giant unilamellar vesicles were formed
from a binary mixture of DMPC and cholesterol and imaged using phase contrast microscopy.
Image adopted from [40] (Reprinted from J. Mol. Bio, 178 (1), H. Hotani, Transformation pathways
of liposomes, 113–120, Copyright (1984), with permission from Elsevier).
In addition to temperature, in the case of closed vesicles, morphological transformations
can also be driven by the osmotic pressure difference between the inside and outside of the
vesicle. An osmotic pressure difference may be a result of excess electrolyte concentration
in the solvent in contact with the vesicle. Pressure-induced shape changes may be quite
drastic, as seen in Fig. 7, for a two component GUV formed from a binary mixture of
DMPC and cholesterol
[40]. When the salt concentration is higher on its exterior, an
initially biconcave vesicle, whose top and side views are shown in Fig. 7(a), transforms
(cid:48)
into one of the elongated filamentous shapes, seen in frames (x), (y), (z) and (x
) of
Fig. 7. The biconcave to elongated morphological transition has multiple pathways that
are characterized by varying intermediate shapes.
Vesicular membranes support large morphological transformations in response to phys-
ical and chemical perturbations because they are soft systems. Experiments on biological
22
LETTERS TO THE EDITOR (k) (0) (r) FIG. 2. Transformation pathways of lipsomes. Circular biconcave form, of which both front and side views are shown in (a), transformed into one of the stable forms shown in (x). (y), (z) and (x’), via one of the possible pathways shown by arrows (see the text for details). Magnification of each microvideopraph is arbitrary. least four main transformation pathways. I designated them as ellipse, triangle, square and pentagon pathways, respectively. Each polygonal form continued to transform further: they transformed into either branched tube forms (Fig. 2(g) and (i)) or head-and-tail forms (Fig. 2(h), (j) and (k)). In this scheme, a peanut form (Fig. 2(f)) could be classified as a branched tube form (a straight union of two thick tubes), because it derived from the bipolar normal mode. During the transformation t’he sides of each polygon gradually collapsed inward such that the vertices became points. Each polygonal form reached a critical shape, and then one or more vertices of the polygon suddenly protruded and grew rapidly to tubular protrusions. The vesicle stopped fluctuating in shape as soon as the vertice(s) protruded. In the pentagon pathway (Figs 2 and 4) the disc-shaped head became smaller as its tubular tail grew longer (Fig. 4(d) to (h)). When the dimension of the head decreased to a critical size, it changed gradually from a disc shape into a triangular plate shape (Fig. 4(i)), and soon afterwards the triangular head transformed instantaneously into a branched short tube (Y-form) as shown in Figure 4(j). The tail still continued to grow longer such that it became very thin and flexible, while the head portion became smaller and smaller (Figs 4(k) and (1) and 2(w)). Then, a minor fraction of the filaments completed this process by assimilating their heads into their tail portion. to form filaments of homogeneous thickness throughout their length (Fig. 2(x’)). The remaining fraction of the filaments underwent a drastic t,ransformation: the head abruptly began to transform into a sphere form (Fig. 2(v)). Th e transformation progressed and synthetic membranes have estimated their bending rigidity to be of the order of
10− 100kBT [41, 42], which is a clear indication of its softness. In spite of their softness,
membranes can support large stresses and large shape fluctuations; this basically explains
the simple to complex spectrum of vesicular shapes shown in Figs. 6 and 7.
In summary, changes in the thermodynamic state variables can play a key role in
driving shape transformation in vesicular membranes. Though these experimental obser-
vations are instructive, it should be noted that the physical and chemical environments
in which these experiments have been performed are mostly non-physiological. Hence
we need to look beyond simple lipid-based systems and also include the effect of other
macromolecules to explain the occurrence and stability of complex cellular and cell or-
ganelle shapes described in section I and shown in Fig. I.
E. Membrane remodeling by curvature inducing factors
A variety of factors associated with the membrane can spontaneously deform mem-
branes and support the large curvatures associated with complex membrane shapes. A
summary of the key membrane remodeling factors is given below.
(a) Membrane inclusions:
The membrane of a cell is home to a large set of functional macromolecules that are
essential for its function. A vast set of protein machinery comprising signaling proteins,
like kinases; transport proteins,
like flippases; channel proteins,
like aquaporins;
ion
channels, like the sodium/potassium channels; pump proteins, like bacteriorhodopsins;
junction proteins, like integrins; and cytoskeletal-linking proteins, like profilin, are found
in a typical membrane that encloses a cell and its organelles. These proteins constitute
25% − 75% of the weight of the membrane and this ratio depends the type of cell and on
the organelle they are associated with (see Table 1 in [2]). Proteins can be amphipathic
(contain both polar and non-polar subunits ). The degree of amphipathicity determines
how the protein is associated with the membrane. In addition to proteins, other macro-
molecules like sugars and peptides are also found in non-negligible quantities.
In this
23
article, we will refer to all non-lipid entities as membrane inclusions, and the importance
of these molecules in controlling membrane morphologies will be discussed in detail in
section IV A.
(b) Area/compositional asymmetry:
Membranes maintain an asymmetry in the number and composition of lipids in each
monolayer for structural and functional reasons. While number asymmetry can arise due
to different sizes of lipids and/or curvature in the membrane, membranes in cells also dis-
play compositional asymmetry [43, 44] in order to orchestrate many biological processes
that are specific to a particular leaflet. For example, the inner leaflet of the membrane
is rich in phosphatidylinositol lipids, like PIP, PIP2 and PIP3, which are essential for
clathrin-mediated endocytosis in the endocytic pathway.
It should also be noted that
many of the membrane-interacting proteins reshape cellular morphologies by generating
an area asymmetry in the membrane. This mechanism will be discussed in detail in
section IV A.
The aim of this review is to explore the role of these curvature-inducing factors in
remodeling membranes into biologically relevant morphologies in order to gain insight
into the genesis of cellular shapes. As noted above, biological membranes have widely
separated length scales and time scales. They measure a few nanometers in the transverse
direction and extend to a few microns laterally. Our knowledge of biological membranes
has been derived both from top-down models developed from continuum theories of solid
and fluid mechanics as well as from bottom-up models derived from molecular structure
and packing. This review focuses on the theoretical and the associated computational
methods that are prevalently used in understanding the thermodynamics and structure of
lipid-based artificial and biological membranes under the influence of curvature-inducing
factors with characteristic lateral size in the regime 100 − 1000 nm, which includes both
the meso and continuum scales.
The article is organized as follows. In section II, we introduce the underlying theory
for the elastic model of the membrane and discuss the role of various elastic parame-
ters and relevant ensembles. Various analytical and computational methods employed
24
in the study of the Canham - Helfrich elastic energy model have been reviewed in sec-
tion III along with a brief description of the molecular methods. Starting section IV with
the biological picture relevant to protein-induced deformations, we classify the protein-
induced curvature field into two categories—namely, isotropic and anisotropic. We focus
on the theory and computational methods for the nematic membranes used as a model
for anisotropic curvature-inducing proteins interacting with the membrane. The particle-
based EM2 model is introduced in section V, and the use of the model to simulate the
remodeling behavior of BAR-domain-containing proteins is discussed.
In section VI,
we describe free-energy methods based on thermodynamic integration to delineate the
free-energy landscape of protein-induced remodeling and conclude in section VII.
II. THERMODYNAMICS-BASED MODELS FOR MEMBRANES
Fully atomistic molecular models and coarse-grained molecular models (reviewed
briefly in section III H) are useful to probe the nature of the interactions at the nanome-
ter resolution. The models closer to atomic or electronic resolution remain true to the
biochemistry of interactions and can map the sensitivity of protein-lipid interactions to
the underlying chemical heterogeneity (e.g., the effect of protonation or ion binding on the
specific interaction of proteins with lipids). On the other hand, the models closer to the
cellular length scale are adept at describing cooperative interactions and long-wavelength
deformations. Conceptually, the divisive length scale between these extremes is set by a
characteristic length (see section II B): much below this length scale, representing the
chemical heterogeneity is important (e.g., presence of cholesterol, PIP2, Ca2+) as much
of the interactions are energy dominated. For length scales much above the characteristic
length, a coarse-grained representation of the chemical interactions is often sufficient
(e.g., the effect of cholesterol, PIP2, Ca2+ on bending stiffness or intrinsic curvature)
as the emergent phenomena are often entropy dominated. Owing to the limitations
of computing power, models with molecular resolution have limited applicability in the
investigation of structures and shapes of membrane systems, whose dimensions match cel-
lular length scales. The limitation is primarily due to the system size needed to approach
the thermodynamic limit and also due to the lower cutoff on the temporal resolution,
which is essential in capturing key biochemical processes. At cellular length scales, it is
25
apparent, however, that top-down phenomenological models can be used to gain insights
into the biophysical properties of membranes. In this section, we review the Canham -
Helfrich elasticity theory for biological membranes
[45, 46].
In this phenomenological
approach, membrane response (such as shape and undulation) to external perturbations,
the associated stability, and energetics can be determined.
A. Thin sheet approximation of membranes
An elasticity-based phenomenological description of the membrane is valid if the sys-
tem under consideration obeys the following constraint: the lateral extent of the mem-
brane—for instance the diameter of a vesicle—is large (L ∼ O(µm)) compared to the
bilayer thickness (δ ∼ O(nm)).
In the above-mentioned limit, a lipid bilayer can be represented as a thin, flexible,
fluid sheet, of constant area at length scales matching its largest dimension. The sheet
represents the neutral surface, known as the unstrained plane in bent membrane, a neutral
surface is shown in Fig. 8. We will discuss the role of neutral surface in more detail in the
next section.
FIG. 8.
(a) Cross sectional view of a flat membrane, with surface area A0. (b) When bent,
the flat membrane shows stretching (A > A0) in the top monolayer and compression (A < A0)
in the bottom monolayer. The neutral surface is the plane in the bent membrane with nearly
constant surface area (A = A0). THe neutral surfaces in the flat and bent membranes are shown
as solid lines.
26
(a) Flat membrane(b) Bent membraneNeutral surface B. Theory of bent plates
The theory of membranes originates from the elastic theory of bent plates. Consider
a thin plate of thickness h and lateral dimension L, with h/L << 1. The plate is bent
along the z direction as shown in Fig. 9(a). On bending, the upper surface of the plate
stretches while the lower surface gets compressed. The extension to compression behavior
crosses over at an unstrained surface called the neutral surface of the plate. In the case of
a plate with uniform thickness, the neutral surface can be identified with the mid-plane of
the plate. The energy to bend the plate, can be written in terms of the stress and strain
tensors as [47]
(cid:90) +h/2
(cid:90) L
(cid:90) L
Hplate =
−h/2
dz
dy
dx
0
0
(cid:18)
(cid:19)
Y
2(1 + σ)
u2
ik +
σ
1 − 2σ
u2
ii
.
(1)
The indices i and k of the strain tensor take values (x, y, z), when described in a cartesian
coordinate system, and Einstein summation convention is implied. Here Y and σ are
respectively the Young’s modulus and Poisson’s ratio of the material. If ui is the i th
component of the displacement vector, the strain tensor is defined as uij = 1
).
2( ∂ui
∂xj
+ ∂uj
∂xi
FIG. 9. Bending of a thin plate of thickness h. (a) On bending the upper surface of the plate
undergoes stretching while the lower surface gets compressed. In between these two surfaces
is an unstrained surface called the neutral surface. The xy plane of the cartesian coordinate
system coincides with the mid-plane of the undeformed plate and the upper and lower surfaces
of the plate are at z = +h/2 and z = −h/2, respectively. (b) The position of a point (filled
circle) on the neutral surface before and after deformation; the point is displaced only along
the z-direction, with the z component of displacement vector given by uz = ζ(x, y). RD and
RUD are respectively the coordinates of a point on the neutral surface, with respect to a global
coordinate system, on the deformed and undeformed plate.
27
stretchedunstrainedcompressed(b) mid−plane of the plate(a) Bent platezxh2−h2ζ(x,y)~RD~RUD which, also satisfies the condition imposed on the neutral surface (z = 0). The components
of the strain tensor for a thin bent plate are
uxx = −z
uxy = −z
∂2ζ
∂y2
∂2ζ
∂x2 , uyy = −z
∂2ζ
∂x∂y
, uxz = uyz = 0
and
uzz =
σ
1 − σ
z
(cid:18) ∂2ζ
∂x2 +
(cid:19)
.
∂2ζ
∂y2
(2)
(3)
(cid:35)(cid:41)
.
A point in the undeformed plate undergoes a displacement of (cid:126)u = {ux, uy, uz} to a
new position on the deformed plate. When the point considered is on the neutral surface,
as shown in Fig. 9(b), the in-plane displacements can be neglected (ux = uy = 0) and
the point moves only in the transverse direction (uz = ζ(x, y)). But the stretching and
compressive behavior of the regions above and below the neutral surface implies that the
in-plane displacements are not negligible in these regions. Hence the displacement vector
can be shown to be
(cid:26)
(cid:27)
(cid:126)u =
−z
∂ζ
∂x
,−z
∂ζ
∂y
, ζ
Using eqn. (3) in the expression for the elastic energy (eqn. (1)), we obtain
(cid:90) L
(cid:90) L
dy
dx
0
0
(cid:40)(cid:18) ∂2ζ
∂x2 +
∂2ζ
∂y2
(cid:19)2
(cid:34)(cid:18) ∂2ζ
(cid:19)2
+ 2(1 − σ)
∂x∂y
∂2ζ
∂x2
∂2ζ
∂y2
−
Hplate =
Y h3
24(1 − σ2)
(4)
Rewriting the position vector (cid:126)RD = (cid:126)RUD + (cid:126)u, for any point on the neutral surface, it
can be shown that
∂2RD
∂x2 =
∂2RD
∂y2 =
∂x2
∂2RUD
(cid:124) (cid:123)(cid:122) (cid:125)
(cid:124) (cid:123)(cid:122) (cid:125)
∂2RUD
∂y2
0
0
+
∂2u
∂x2 =
∂2ζ
∂x2 and
+
∂2u
∂y2 =
∂2ζ
∂y2 ,
(5)
are the curvatures of the surface along the x and y directions, respectively. From eqn. (4),
it can be inferred that the energy of the bent plate is determined by the various curvature
measures on the neutral surface given by ∂2ζ
∂x2 , ∂2ζ
∂y2 and ∂2ζ
∂x∂y .
28
Uniaxial bending
The upper and lower surfaces of the plate show stretching and compression behavior
only along one direction, say for instance the x direction. The membrane has non-zero
curvature only along the x direction, given by ∂2ζ
∂x2 = 1 (cid:126)RD. A uniformly curved plate
resembles a cylinder of radius RD = R and hence the bending energy becomes
Hplate,uniaxial =
Y h3
24(1 − σ2)
Biaxial bending
(cid:90) L
(cid:90) L
(cid:18) 1
(cid:19)2
dy
dx
0
0
R
.
(6)
In the case of uniform bending along both the x and y directions, the shape of the
deformed plate is similar to the surface of a sphere. Hence ∂2ζ
For a uniform sphere of radius RD = R, the bending cost is given by,
∂x2 = ∂2ζ
∂y2 = 1 (cid:126)RD and ∂2ζ
∂x∂y = 0.
(cid:90) L
(cid:90) L
(cid:40)(cid:18) 2
(cid:19)2
dy
dx
0
0
R
(cid:41)
Hplate,biaxial =
Y h3
24(1 − σ2)
+ 2(σ − 1)
1
R2
.
(7)
Eqns. (6) and (7) reveal some general features of the bending energy:
1. The various material properties can be related to the bending rigidity of the plate
through the relation κ = Y h3
12(1−σ2). κ has the dimensions of energy.
2. The prefactor to the second term in eqn. (7), is another material property of interest
called the Gaussian rigidity, κG = 2(σ − 1)κ/2. For materials with Poisson’s ratio
σ = 0, the bending and Gaussian rigidity are related to each other as κG = −κ.
3. The mean curvatures of the cylindrical and spherical surfaces, of radius R, consid-
ered in cases of uniaxial and biaxial bending are, respectively, Hcyl = 1/2R and
Hsph = 1/R. The corresponding κ-dependent part of the bending energy, in both
eqns. (6) and (7), depends on surface curvature as Hplate,uniaxial ∝ (2Hcyl)2 and
Hplate,biaxial ∝ (2Hsph)2, respectively.
These concepts are general and are applicable to any system that can be approximated
as a thin plate. In the next section, we will apply these principles and derive the Canham-
Helfrich Hamiltonian for biological membranes in the macroscopic length scale.
29
C. Canham - Helfrich phenomenological theory for membranes
Typical problems in membrane biophysics, which are computationally expensive to
approach using molecular models, involve membranes whose lateral extension (L) exceeds
50 nm. If the average thickness (h) of a lipid membrane is taken to be 5 nm, as shown
in Fig. 3, the thickness-to-length ratio (h/L) of these bilayer structures is less than 1/10,
which is in the regime of the thin plate theory. This heuristic estimate for L can set the
lower bound for the characteristic length discussed in the introduction to section II. Later,
in various sections, we will introduce and discuss other length scales that are relevant to
the membrane system.
Further, there are two important properties of biological membranes that justify the
use of continuum theories for their description: (a) the area per lipid molecule in the
membrane is nearly constant, and (b) the fluctuation in the thickness of a lipid bilayer
has been shown to be in the range 1A − 3.5A [48], which is negligible when compared
to its average thickness of 50A.
In this regime, both the upper and lower surfaces in
the bilayer closely follow the neutral surface and guarantee the small-deformation limit
prescribed in our derivation of eqns. (6) and (7).
The elastic theory for membranes, known as the Canham - Helfrich Hamiltonian [45,
46], has the form
(cid:90) L
(cid:90) L
Helastic =
dx
dy
0
0
(cid:111)
(2H)2 + κGG
(cid:110)κ
2
.
(8)
Here G is the deviatoric curvature called the Gaussian curvature of the surface. Since
the membrane is a self-assembled system, the relevant energies are comparable to thermal
energy(O(kBT )); this also suggests that the bending modulus κ should be of the order of
kBT . Experimental measurements on a wide class of lipid membranes estimates the value
of κ to be in the range of 10 − 100kBT . Based on these results and using the definition
of κ, given in section II B, we can estimate the Young’s modulus of a 5nm thick lipid
membrane to between 107 and 108N/m2.
In addition to pure bending, the morphology of a membrane can also be affected by
other modes that alter the membrane area. The membrane area couples to the surface
tension, σ, and area elasticity modulus, KA.
pressure difference (∆p) between the inside and outside of a vesicle can also drive shape
In the case of closed vesicles, an osmotic
30
changes. Taking these contributions into account, eqn. (8) can be written in a more
general form as,
Hsur =
(cid:110)κ
2
dS
(cid:90)
S
(2H − C0)2 + κGG + σ
(cid:111)
+
1
2KA(A − A0)2 +
(cid:90)
V
dV ∆p.
(9)
Here, the equilibrium area is given by A0. The geometry of the lipids can impose a
preferred equilibrium curvature on the membrane, which is also captured by this energy
functional through the spontaneous curvature term C0. The integral in the first term is
performed over the entire surface of the membrane, and the integration in the second
term is carried out over the volume (V ) enclosed by the surface.
1. Gauss-Bonnet theorem and the Gaussian bending term:
The topology of a membrane is described by the Euler characteristic, χ, which in turn
is related to the genus of the surface g and the number of holes h as, χ = 2(1 − g) − h.
For instance, a membrane morphology with a spherical topology has g = 0, h = 0, and
χ = 2 whereas a membrane with the topology of a torus has g = 1, h = 0, and χ = 0. As
an example, a closed membrane structure with g = 2 and h = 1 is illustrated in Fig. 10.
FIG. 10. A vesicular structure with two genus (g = 2) and one hole (h = 1), for which the
Euler number χ = −3. A hole is different from a genus in that it is constituted by a tear in the
membrane surface. The membrane shape is adopted from the geometry models provided with
Javaview (URL http://www.javaview.de).
The Gauss-Bonnet theorem relates the Euler number of the surface to the total Gaus-
31
sian curvature of the surface as [50],(cid:90)
S
dS G = 2πχ.
(10)
For a membrane with fixed topology, by virtue of the Gauss-Bonnet theorem, it can be seen
that the Gaussian energy contribution in eqn. (8) is a constant, with values of 4πκG for
a spherical membrane and 0 for a toroidal membrane. For most of the studies presented
here, since the topology of the membrane remains constant, the contribution from the
Gaussian energy can be neglected during the analysis. We note, however, that when
κG is spatially inhomogeneous or when a planar patch of a membrane with a constant
projected area is subject to periodic boundaries, in general, it will have a contribution
from the Gauss curvature term.
D. Thermal softening of elastic moduli
Continuum theory has been used in the renormalization studies of membrane elastic
parameters [51–54], computations of the fluctuation spectrum, etc. The elastic moduli
(κ, κG, σ) introduced in the thermodynamic model for biological membranes (eqn. (9)) are
conceptually different from those corresponding to a solid. In contrast to a solid, regions
on a membrane away from each other by a distance r display orientational decorrelations
with increase in r. The value of r at which this decorrelation occurs, characterized by
the persistence length of the membrane (ξp), also depends on the temperature and the
bending rigidity. De Gennes has suggested that ξp ∼ a exp(4πκ/3kBT ), which implies
that for a typical bilayer, ξp >> a, where a is the characteristic molecular dimension of
an individual lipid [55]; note here that for a typical bilayer, ξp is much larger than L, the
characteristic length over which the elasticity model itself is valid. It was first suggested by
Helfrich [51] that orientational decorrelations may have their origin in a scale dependent
elastic modulus. Hence the elastic moduli employed in the thermodynamic description of
a membrane are thermal quantities in the sense that they renormalize with system size
and temperature. For example, the bending rigidity has been shown to be modulated
from its bare value (κ) to a renormalized value (κR) as,
(cid:18) L
(cid:19)
κR = κ − α
kBT
4π
log
32
a0
.
(11)
There is a general consensus on the form of this relation, but the value of the prefactor
depends on the statistical measure used in the calculations. Conflicting values for the
prefactor—α = 1 [51, 56, 57], α = 3 [52–54] and α = −1 [58, 59]—pointing to both
thermal softening (α = 1 and 3) and stiffening (α = −1) have been reported. In spite
of these contradictions, it should be remembered that we measure κR from experiments
and molecular simulations, whereas we impose κ in mesoscale/continuum simulations
based on the Helfrich energy functional. Similar renormalization behavior exists for other
elastic moduli too [54, 60, 61]. It has been shown by Cai and Lubensky that the in-
plane hydrodynamic modes arising due to the fluid nature of the membrane can lead to
renormalization of the elastic parameters [62, 63].
E. Ensembles for thermodynamic description
It has been stated earlier that the conformations of a membrane are susceptible to
thermal undulations, and hence, it is important to understand the various thermodynamic
ensembles associated with theoretical modeling of membranes. The interaction between
the membrane degrees of freedom and key thermodynamic variables like temperature, T ;
chemical potential, µ; frame tension, τ ; and osmotic pressure difference, ∆p = pout −
pin; is entirely dependent on the thermodynamic ensemble used. Here we will describe
two ensembles that are primarily used in the study of planar membranes and closed
vesicular membranes (note that here closed refers to the fact that the membrane does
not have free line boundaries). The thermodynamic formulation begins with considering
the internal energy, U , as a function of all relevant extensive variables of the ensemble.
Other free-energy functions and their independent variables can be derived from U by
defining suitable Legendre transformations as long as the free energy depends on at least
one extensive variable [64].
(a) Constant σApµT ensemble for planar membranes: A membrane patch is a model
system for supported membrane or represents a sub-region of a cell membrane. The
internal energy of the planar membrane is a function of its extensive variables S, Ap, N
and A and hence the elastic(internal) energy U = U (S, Ap, N, A). A patch enclosed in a
given frame is characterized by a fixed projected area Ap, which is conjugate to a type of
tension called the frame tension τ . We also denote the system entropy by S, the number
33
of lipids by N , and the chemical potential of the lipids by µ. In this ensemble, the surface
area of the membrane A fluctuates while we fix the surface tension σ, such that A ≥ Ap.
The appropriate free energy in this ensemble is given by dF = dU−d(σA)−d(T S)−d(µN )
(b) Constant σµV T ensemble for closed membranes: The internal energy of a closed
membrane is U = U (S, V, N, A). This ensemble closely represents a closed membrane of
arbitrary topology seen in ex-vivo and in-vivo experimental systems. The osmotic pressure
difference ∆p is determined by the solute concentration in the solvent inside and outside
the vesicle and couples to V . The appropriate free energy in this ensemble is given by
dF = dU − d(σA) − d(T S) − d(µN ).
The different ensembles for open and isolated systems are pictorially highlighted in
Fig. 11 [65, 66].
FIG. 11.
(center panel) Initial state of a membrane with surface area, A, projected area, Ap,
and number of lipids, N . (top panel) In the open state, the system exchanges lipid molecules
with a reservoir and the corresponding variables in the final state are (a) (A
(cid:48)
(cid:48)
, A
p, N
(cid:48)
) in the
(cid:48)
(σ, τ ) ensemble and (b) (A
(cid:48)
, Ap, N
) in the (σ, Ap) ensemble. (bottom panel) In the absence
of number fluctuations the state of the membrane is given by (c) (A, Ap, N ) in the (A, Ap)
ensemble, (d) (A
(cid:48)
(cid:48)
, Ap, N ) in the (σ, Ap) ensemble, and (e) (A
(cid:48)
p, N ) in the (σ, τ ) ensemble.
, A
34
IsolatedOpen(a)(b)(c)(d)(e) Multiple variants of eqn. (9) that impose constraints on the external variables associ-
ated with the membrane, like the bilayer coupling model, and the area difference elasticity
model
[67] have also been used for modeling continuum membranes. For more details,
we refer the reader to the detailed review on theoretical methods in membranes by Seifert
[42]. Any further constraints on the thermodynamic variables can be imposed in the form
of a Lagrange multiplier. Such constraints make physical sense when one accounts for
the various constituents making up the cell. A case in point is the constraint on the
volume enclosed by the membrane in the presence of cytoskeletal filaments or due to the
incompressible nature of the cytoplasmic fluid. The morphological transitions arising from
volume constraints have long been a subject of interest, and the details can be found in
an excellent review by Seifert [42].
III. OVERVIEW OF ANALYTICAL AND COMPUTATIONAL METHODS
The elastic free-energy functional can be minimized using analytical methods for mem-
branes with well-defined geometries to gain insight into the structural and statistical prop-
erties of the system. A variety of theoretical techniques can be used in the analysis, and
the choice of the method is dictated by the phenomenon investigated. The class of prob-
lems can broadly be classified into (a) determination of equilibrium properties and (b)
understanding the dynamics of membranes with and without hydrodynamics. For sake of
simplicity, we will use a planar membrane, parameterized using the planar Monge gauge,
to illustrate each of these methods.
In the planar Monge gauge, the surface of the membrane is described with respect to
the x − y plane at z = 0 by the position vector R = (x, y, h(x, y)), where h(x, y) is the
height of the membrane at position (x, y) on the reference plane. For simplicity, we will
perform our analysis on eqn. (9), with σ = 0, KA = 0, and κG = 0. The mean curvature of
a planar membrane, which has been derived in appendix B using the differential geometry
methods described in appendix A, is given by,
2H =
∇2h(cid:113)
1 + (∇h)2
.
(12)
Using this form of mean curvature in eqn. (9), for the parameters given above, the
35
elastic energy of the membrane in the Monge Gauge has the form
(cid:90)
(cid:113)
1 + (∇h)2
Hsur =
κ
2
dx
∇2h(cid:113)
1 + (∇h)2 − C0
2
.
(13)
A. Energy minimization and stationary shapes
The planar membrane will take the shape that minimizes its bending energy. Hence
the shape of the planar membrane that exhibits pure bending can be determined by
minimizing the linearized form of eqn. (13) with respect to height h,
2 ∇2(cid:0)
κ
δHsur
δh
=
∇2h − C0
(cid:1) = 0.
(14)
Hence the minimum energy conformations of the membrane are those satisfying the dif-
ferential equation ∇2(∇2h− C0) = 0 for the given boundary conditions. This implies that
all height profiles for the membrane with curvature ∇2h = C0 are possible solutions.
B. Membrane dynamics
Equilibrium shape analysis introduced in section III A, is useful in determining the
long-time-scale behavior of a membrane in response to a perturbation. The short-time-
scale relaxation of the membrane can be studied by analyzing the equations of motion for
the membrane, which for the Monge gauge is given by the dynamical equation [68],
∂h(x)
∂t
= −
Γ(x, x
(cid:48)
)
+ η(x, t)
(cid:48)
.
dx
(15)
(cid:26) δHsur
(cid:48)
δh(x
)
(cid:27)
(cid:90)
(cid:90)
The hydrodynamic kernel Γ(x, x
(cid:48)
) captures the long-range interactions between different
regions of the membrane, mediated by the surrounding fluid, and η(t) is the noise term
with (cid:104)η(x, t)(cid:105) = 0 and (cid:104)η(x, t)η(x
[69]. For a
membrane in the Monge gauge (with C0 = 0 and assuming √g = 1) and using the energy
given by eqn. (13), the dynamical equation becomes,
)(cid:105) = 2kBT ApΓ(x, x
)δ(x − x
)δ(t − t
(cid:48)
(cid:48)
, t
(cid:48)
(cid:48)
)
(cid:48)
∂h(x, t)
∂t
= −
Γ(x, x
(cid:48)
)
(cid:48)
, t) + η(x, t)
(cid:48)
.
dx
(16)
The above integro-differential equation becomes amenable to theoretical analysis when
represented in Fourier space. Defining h(k, t) =(cid:82) dx exp(−i k · x)h(x, t), eqn. (16) can
(cid:111)
(cid:110)
κ∇4h(x
36
be written in Fourier space as,
∂h(k, t)
∂t
= Γ(k)(cid:8)
−κk4h(k, t) + η(k, t)(cid:9) .
(17)
For an almost planar membrane, Γ(k) can be derived from the Stokes equation and is
given by Γ(k) = 1/(4ηk), where η is the fluid viscosity. See reference [42] for a complete
discussion of the method and its applications. This form of the elastic energy forms the
basis for Fourier space Brownian dynamics [70, 71]; see section III E.
The static and dynamical properties of the membrane are well represented by dynam-
ical height correlation function given by [42],
(cid:68)
(cid:69)
(cid:48)
(cid:48) (t, t
) =
Skk
h(k, t)h(k
(cid:48)
(cid:48)
, t
)
=
kBT
κk4 exp(−γk(t − t
(cid:48)
))δ(k − k
(cid:48)
),
(18)
where k = k and γk = κk4Γ(k). In the limit t − t
correlation function,
(cid:48)
→ 0, Skk
(cid:48)
(cid:48) (t, t
) reduces to the static
Ckk
(cid:48) =
kBT
κk4 δ(k − k
(cid:48)
),
(19)
(cid:112)
(cid:112)
which can also be derived from eqn. (13). In this case, the amplitude of each undulation
mode scales as k−4, and it can be seen that the intensities of the long-wavelength modes
(small k) dominate the fluctuation spectrum. In our description, we have considered a
tensionless membrane, whereas non-zero tension gives rise to undulation modes whose
spectrum scales as k−2 for k ≤
σ/κ. The crossover from the
tension-dominated regime (k−2 scaling) to the bending-dominated regime (k−4 scaling)
κ/σ. For deformations with radius of
defines a tension-dependent length scale ltension =
σ/κ and as k−4 for k >
(cid:112)
curvature smaller that ltension, the bending energy term dominates the Hamiltonian and for
those with radius of curvature much larger than ltension, the interfacial tension dominates.
The typical value of this length scale in the cellular context (assuming κ = 20kBT and
σ = 30µN/m) is ltension ∼ 50 nm. At this length scale, the interfacial tension can compete
with the bending energy to influence the mean shape of the membrane as well as the nature
of the undulations. Additionally, we note that other length scales can also be relevant
depending on the ensemble. For example, for the closed membrane ensemble discussed in
section II E, one can define lpressure = (κ/∆P )1/3, which in the cellular context assumes a
typical value of 10 nm (assuming a typical value of ∆P ∼ 106Pa).
37
C. Surface of evolution formalism
The analytical techniques discussed above are extremely useful but can only handle
membrane geometries that can be readily parametrized in the small-slope limit, (i.e.
∇h << 1). In addition to the studies in the small-slope approximation, analytical ap-
proaches have also been used to examine the behavior of axisymmetric vesicular structures
with large curvatures (i.e. curvatures beyond the small slope limit). A notable example is
the study of a membrane tether pulled out from a spherical vesicle [72–76]. The analyti-
cal approach, introduced in secs. III A and III B, breaks down when the membrane shape
becomes nonaxi-symmetric and also when thermal fluctuations are accounted for. Here,
we describe one such method in detail. The surface-of-evolution approach [77–80] to
model the membrane at equilibrium is useful in considering large (albeit axi-symmetric)
deformations, including those in which h is a multivalued function of r. We consider a
generating curve γ parametrized by arc length s lying in the x − z plane. The curve γ is
expressed as (see Fig. 12)
FIG. 12. Schematic of a membrane profile that shows the different variables used in the surface
of evolution formalism.
γ(0, s1) → R3γ(s) = (R(s), 0, z(s))
(20)
where s1 is the total arc length, which is not known a priori. This generating curve leads
to a global parameterization of the membrane expressed as
X : (0, s1) × (0, 2π) → R3
(21)
38
s=s1s=0Rzψ X(s, u) = (R(s) cos(u), R(s) sin(u), z(s))
(22)
where u is the angle of rotation about the z-axis. With this parameterization, the mean
curvature H and the Gaussian curvature G are given as follows;
− z(cid:48)(cid:48)R(cid:48))
and
z(cid:48) + R(z(cid:48)R(cid:48)(cid:48)
R
R(cid:48)(cid:48)
R
,
2H = −
G = −
(23)
(24)
where the prime indicates differential with respect to arc-length s. The expressions ob-
tained above for the mean curvature and the Gaussian curvature are cumbersome. To
simplify them, an extra variable ψ, where ψ(s) is the angle between the tangent to the
curve and the horizontal direction, is introduced which leads to the following two geomet-
ric constraints:
(cid:48)
(cid:48)
R
z
= cos(ψ(s))
= − sin(ψ(s))
(25)
(26)
These two constraints lead to the following simplified expressions for the mean curvature
and the Gaussian curvature:
2H = ψ
G = ψ
sin(ψ(s))
(cid:48)
+
R(s)
(cid:48) sin(ψ(s))
R(s)
Starting with the membrane energy Hsur defined by
(cid:90) 2π
(cid:90) s1
(cid:110)κ
0
0
2
Hsur =
(2H − C0)2 + κGG + σ
(27)
(28)
(29)
(cid:111)
dA
where dA is the area element given by Rdsdu, and substituting for H, G, we obtain the
following expression for Hsur:
(cid:90) 2π
(cid:90) s1
(cid:40)
(cid:18)
Hsur =
0
0
κ
2
ψ
(cid:48)
+
sin(ψ(s))
R(s) − C0
(cid:19)2
+ κGψ
(cid:48) sin(ψ(s))
R(s)
(cid:41)
+ σ
R ds du.
(30)
We now proceed to determine the minimum-energy shape of the membrane. The condition
that specifies the minimum-energy profile is that the first variation of the energy should
be zero. That is:
δHsur = 0,
39
(31)
subject to the geometric constraints R(cid:48) = cos(ψ(s)), z(cid:48) = − sin(ψ(s)). These constraints
can be re-expressed in an integral form as follows:
(cid:90) s1
(cid:90) s1
0 {R
0 {z
(cid:48)
− cos(ψ(s))} ds = 0
(cid:48)
+ sin(ψ(s))} ds = 0
(cid:40)
(cid:18)
0
0
(cid:90) s1
(cid:90) 2π
(cid:90) s1
(cid:90) s1
+ ν
0
(cid:48)
R
(cid:40)
(cid:18)
(cid:19)2
(cid:90) s1
(cid:19)2
Introducing Lagrange multipliers, we solve our constrained optimization problem as fol-
lows. We introduce the Lagrange function ν, η and minimize the quantity F :
(cid:41)
F =
κ
2
(cid:48)
+
ψ
sin(ψ(s))
R(s) − C0
+ κGψ
(cid:48) sin(ψ(s))
R(s)
σ
R ds du
− cos(ψ(s))ds + η
0
(cid:48)
z
+ sin(ψ(s))ds.
Since the integrand of the double integral is independent of u, F simplifies to:
F =2π
0
κR
2
ψ
(cid:48)
+
sin(ψ(s))
R(s) − C0
sin(ψ(s)) + σR
(cid:48)
+ κGψ
(cid:41)
(cid:48)
+ ν (R
− cos(ψ(s))) + η (z
(cid:48)
+ sin(ψ(s)))
ds
(32)
(33)
(34)
(35)
(36)
The minimization problem is then expressed as:
δF = 0.
The resulting Euler-Lagrange equations and applications of this approach in studying
membrane tethers in GUV experiments and nucleation of vesicles mediated by curvature-
inducing and force-mediating proteins on cell membranes can be found in the published
literature [74, 81–84].
D. Direct numerical Minimization
Complex shapes that are both axisymmetric and non-axisymmetric can be studied
using direct numerical simulations that use well-known numerical methods to compute
the elastic energy and curvature forces in the membrane.
Here the continuous membrane is discretised into a computational mesh, and the energy
and forces are computed from the conformation of the mesh, which evolves with time. A
key requirement for the applicability of this method is that the membrane shape should
40
be continuous and derivatives can be computed everywhere. Using the computed values of
Hsur and −∇Hsur, the state of the membrane can be evolved using a suitable minimization
technique or by integrating the equation of motion given in eqn. (15).
Surface evolver [85] is a powerful numerical minimization software package that can
be used to determine the zero temperature shapes of symmetric and non-axisymmetric
vesicular membranes. This tool can simultaneously minimize energy contributions from
surface tension, curvature energy, and gravitational energy. Further, in this approach,
the total energy can be minimized subject to constraint on the external variables A, the
surface area, and V , the enclosed volume, and in effect, one determines the membrane
shape minimizing the energy given in eqn. (9).
The applications of this approach in studying the interaction of the membrane with
nanoparticles of different shapes can be found in the published literature [86, 87]. Extend-
ing these analyses of non-axi-symmetric shapes to thermally undulating systems requires
the use of other numerical methods, which are discussed next. We do this in two stages:
first, we discuss one numerical method to incorporate thermal undulations in small slopes,
and then, we proceed to describe methods handling undulations in the non-small-slope
limit.
E. Fourier space Brownian dynamics (FSBD)
For a planar membrane with periodic boundaries, the starting point for FSBD is the
discretization and forward integration of eqn. (17) as:
(cid:126)h(k, t + ∆t) = (cid:126)h(k, t) +
(cid:110)
−Λ(k) · (cid:126)h(k) + Γ(k) · (cid:126)η(k, t)
(cid:111)
∆t,
(37)
with Λ(k) = κk4Γ(k) being a diagonal matrix. The shape profile of the periodic membrane
at every time can be obtained by an inverse Fourier transform of h(k, t). In the Fourier
space, the noise η(k), which is a complex number, obeys
(cid:104)η(k, t)(cid:105) = 0 and (cid:104)η(k, t)η(k
(cid:48)
(cid:48)
, t
)(cid:105) = 2kBT ApΓ
−1(k)δ(k − k
(cid:48)
(cid:48)
)δ(t − t
),
(38)
and is drawn from a Gaussian distribution with zero mean and variances—2kBT ApΓ(k)
and kBT ApΓ(k) for the real and imaginary parts, respectively. At every time step, the
forces are calculated in real space, and the shape profile is computed in the Fourier space
41
and inverted to get the final shape of the membrane. In addition to planar membranes,
this model has also been used to study protein mobility on fluctuating surfaces, and the
effect of cytoskeletal pinning on membrane dynamics [88–92].
F. Dynamically triangulated Monte Carlo methods.
The numerical approach described above deals with smooth geometries and is hence
a zeroth-order approximation to the natural state of bilayer configurations. Membrane
shapes are naturally irregular since they are susceptible to thermal fluctuations, and a
rigorous analysis of these shapes should also involve higher order terms. Monte Carlo
based methods account for the effect of thermal undulations and hence are extremely
useful in studying membranes in their canonical ensemble. Triangulated random surface
models have been extensively used in high-energy physics, mainly in Euclidean string
theory [93–96]. In the context of membranes, these techniques were first used to study
the crumpling of self-avoiding tethered membranes [97, 98]. Fluid membranes were first
studied by Ho and Baumgartner [99, 100] using the method of dynamically triangulated
Monte Carlo (DTMC), wherein the triangulation map of the membrane is dynamic. In
this representation, the membrane surface is discretized into a set of N vertices that
form T triangles and L links, with surface topology defined by the Euler number χ =
N + T − L. The elastic energy of the membrane Hsur can estimated from the orientation
of the triangles as,
42
FIG. 13. A patch of a triangulated membrane depicting the one-ring neighbourhood around a
vertex v. The edge e connects vertices v and 1, and is shared by the faces f1(e) and f2(e). The
normal to the surface at vertex v is given by Nv.
(cid:88)
(cid:88)
(cid:110)
{e(v)}
v
Hsur = λb
(cid:111)
1 − Nf (1, e) · Nf (2, e)
− ∆p V.
(39)
As illustrated in Fig. 13, {e(v)} denotes the set of links vertex v makes in its one-ring
neighbourhood. In general, Nf denotes the normal to any face f , whereas Nf (1, e) and
Nf (2, e) explicitly represent the normals to faces f1 and f2 sharing an edge e. V is the
volume enclosed by the surface. λb in eqn. (39) and κ in (9) are related to each other in
a geometry-dependent manner. For example, λb = √3κ for a sphere and λb = 2κ/√3 for
a cylinder [66].
In this discretization technique, the squared mean curvature at a vertex v is approxi-
mated in an implicit manner, as given in eqn. (39). This calculation does not involve the
computation of the principal curvatures c1(v) and c2(v) . Recently, an alternate technique
to compute principal curvatures on triangulated surfaces was introduced by Ramakrish-
[101] (see Appendix C for details), which allows one to represent the elastic
nan et al.
energy as,
N(cid:88)
v=1
Hsur =
κ
2
{c1(v) + c2(v)}2 Av − ∆pV.
(40)
The equilibrium properties of a self-avoiding membrane are determined by analyzing
43
Nf(1,e)Nf(2,e)Nvv123456 the total partition function,
Z(N, κ, ∆p) =
1
N !
{T }
v=1
(cid:88)
(cid:90)
N(cid:89)
(cid:104)
(cid:110)
−β
Hsur
(cid:16)
{ (cid:126)X},{T }
(cid:17)
(cid:105)(cid:111)
+ VSA
d(cid:126)x(v) exp
(41)
VSA is the self-avoidance potential, usually chosen to be the hard sphere potential. (cid:126)X is
the position vector of all vertices in a triangulated surface, and T is the corresponding
triangulation map. The temperature of the system is expressed in units of β = 1/kBT , and
the integral is carried over all vertex positions and summed over all possible triangulations.
A tuple, η = [{ (cid:126)X},{T }], represents a state of the membrane in its phase space. In the
case of Monte Carlo (MC) studies, a change in state, η → η
, is effected by means of
Monte Carlo moves, the rules of which correspond to importance sampling [102]. The
(cid:48)
time in MC simulations is expressed in units of Monte Carlo steps(MCS).
FIG. 14. Monte Carlo moves involved in the equilibrium simulations of a fluid random surface.
(a) A vertex move emulates thermal fluctuations in the membrane, and (b) a link flip simulates
fluidity in the membrane.
A membrane quenched to a particular microstate (η) relaxes to its equilibrium confor-
mation mainly through thermal fluctuations and in-plane diffusion. A Monte Carlo step
captures these modes by performing N attempts to displace a randomly chosen vertex
and L attempts to flip a randomly chosen link, as in Fig. 14. The rules of importance
sampling and details of each move are as follows:
44
a) Vertex moveb) Link flip{~X}→{~X′}{T}→{T′}Pacc=min[1,exp(−β∆Hsur)]Pacc=min[1,exp(−β∆Hsur)] (a) Vertex move : The vertex positions of the surface are updated, { (cid:126)X} → { (cid:126)X
},
by displacing a randomly chosen vertex within a cube of side 2σ around it, with fixed
triangulation {T }. As a result, the old configuration of the membrane η = [{ (cid:126)X},{T }] is
},{T }]. The total probability of this MC move
updated to a new configuration η
= [{ (cid:126)X
(cid:48)
(cid:48)
(cid:48)
obeys the detailed balance condition given by,
(cid:48)
ω(η → η
)Pacc(η → η
(cid:48)
(cid:48)
) = ω(η
→ η)Pacc(η
(cid:48)
→ η).
Choosing the attempt probability ω for the forward transition (η → η
transition (η
) = ω(η
→ η) to be equal, ω(η → η
→ η) = (8σ3N )−1, we get the
(cid:48)
(cid:48)
(cid:48)
(42)
(cid:48)
) and backward
probability of acceptance as
Pacc(η → η
(cid:48)
) = min
(cid:110)
(cid:104)
(cid:105)(cid:111)
(cid:48)
)
1, exp
−β∆Hsur(η → η
,
(43)
which is the well-known Metropolis scheme [103]. σ defines the maximum displacement
of the vertex and is chosen appropriately, so that the acceptance of vertex moves is close
to 50%.
(b) Link flips : An edge shared between two triangles is flipped to link the previously
(cid:48)
unconnected vertices of the triangles. Such a move changes the triangulation map from
{T } → {T
}, in the process of which it changes the neighborhood of some vertices, which
is effectively a diffusion. With fixed vertex positions, the old and new configurations in this
case are η = [{ (cid:126)X},{T }] and η
}], respectively. The attempt probability for
→ η) = (L)−1, and the acceptance probability
flipping a link is given by ω(η → η
is as in eqn. (43).
= [{ (cid:126)X},{T
) = ω(η
(cid:48)
(cid:48)
(cid:48)
(cid:48)
The thermodynamic properties of the fluid membrane computed using the DTMC
approach can be found in published studies [101, 104–106].
G. Particle-based models
Field-based continuum models, described above, assume the presence of a self-assembled
membrane, and hence are extremely useful in understanding the membrane response to
an imposed curvature field or fluctuations in its environment. This model becomes too
complicated to handle when one needs to investigate the role of many other physical
parameters like topological fluctuations and hydrodynamics. Meshless, particle-based
45
mesoscopic models are powerful alternatives in this regime. In this approach, the micro-
scopic structure of the membrane is coarse grained into particles, whose lengths are in
the mesoscale, that interact via well-defined interparticle potentials. In this framework
the membrane structure is formed due to the self-assembly of the coarse grained particles
and alleviates many of the limitations seen in the mesh representation of the membrane.
However, the formulation of interaction potentials that can reproduce both the elastic
and transport coefficients of the membrane at the mesoscale is still less understood.
The earliest particle-based mesoscopic model was proposed by Drouffe et. al.
[107],
in which the spherical coarse grained particle has a hydrophobic region surrounded by
two hydrophilic regions. The repulsive hard sphere potential between the particles en-
sures incompressibility, whereas an orientation-dependent potential ensures flexibility and
bending. A variety of mesoscale particle models—namely, the EM-DPD model
[108],
where the bending modulus is related to the imposed bulk strain modulus; the rigid and
flexible rod approximation of lipids [109–112]; the spherocylinder representation [113];
EM2 model based on both the bending rigidity and stretching modules [114]; and vari-
ants of the Drouffe model
[115–117]—have been proposed. In this section, we will focus
on the EM2 model based on eqn. (8) introduced by Ayton et al.
[114], which has been
extended to include the presence of curvature-inducing proteins in reshaping membranes
[118–122].
In the EM2 model, a bilayer membrane is coarse grained into a set of quasi particles
of length LEM 2, as shown in Fig. 15, with area density ρA = hρ0; here h and ρ0 are,
respectively, the thickness and volume density of the membrane.
FIG. 15.
Illustration of the EM2 model where each quasiparticle is of size LEM 2. Two particles
i and j, with respective positions (cid:126)ri and (cid:126)rj, interact via pair potentials that depend on their
separation, (cid:126)rij, and normal orientations, ni and nj. (Bilayer image is courtesy of Ryan Bradley.)
46
EM2 quasi particle The effective interaction of two EM2 particles
[114], positioned at (cid:126)ri and (cid:126)rj and
separated by (cid:126)rij, has contributions from a bending potential (Hbend) and a stretching
potential (Hstretch). These potentials have the form,
N(cid:88)
N(cid:88)
i=1
i=1
Nc,i(cid:88)
Nc,i(cid:88)
j∈i
j∈i
1
2
1
2
Hbend =
Hstretch =
8κ
ρANc,i
{(ni.rij)2 + (nj.rij)2}
(cid:26) rij
(cid:126)rij2
(cid:27)2
ij − 1
r0
ij)2hλ
8π(r0
N 2
c,i
.
(44)
(45)
The unit normal vectors corresponding to particles i and j are ni and nj, respectively.
Summation over j runs over all the Nc,i particles found with a cutoff distance rc around
particle i. λ is the stretch modulus of the membrane, computed from all atom non-
equilibrium MD simulations, and r0
ij is the distance between the particles in the unde-
formed state of the membrane. The properties of the EM2 model and its coupling to
mesoscopic solvents like WCA and BLOBS [123] can be found in the original article by
Ayton et al.
[114]. Details of the model and its use in studying phenomena related to
protein-induced membrane remodeling will be discussed in detail in Sec. V.
H. Molecular and coarse grained approach for modeling membranes
The roles of lipids and protein-lipid interactions can be described/quantified by the-
oretical models at multiple resolutions of length and timescales, as depicted in Fig. 16.
In the previous sections, we have focused on the continuum scale, which is the major
emphasis in this article. However, we note that models closer to atomic or electronic
resolution remain true to the biochemistry of interactions and can map the sensitivity
of protein-lipid interactions to the underlying chemical heterogeneity (e.g., the effect of
protonation or ion binding on the specificity interaction of proteins with lipids). Hence,
we conclude this section by providing a brief overview of molecular simulation methods.
47
FIG. 16. DMPC lipid molecules described at multiple resolutions, ranging from the nanoscale
to the macroscopic scale. All atoms in a lipid are explicitly represented at the nanoscale, whereas
in the mesoscale, where coarse grained simulations are used, only a reduced number of atomic
coordinates are used to represent the lipid molecule. At length and time scales that fall into
the continuum regime the relevant parameters are determined from the structural and physical
properties of the membrane.
Molecular dynamics (MD), where a molecule can be resolved to the level of an atom,
is the most popular simulation technique and is widely used in the study of material
systems. In this approach, the atoms in a molecule interact among themselves and other
atoms through a set of bonded and non-bonded potentials. These potentials are in turn
derived from extensive quantum calculations of the molecule involved. MD simulations are
extremely useful in understanding various physical and biological processes in membranes
that span over nanoscopic length and time scales, and they can also be used to understand
many chemical events with average life times of the order of femto seconds. Technical
48
details of MD simulations of lipid membranes can be found in several reviews on this
topic [124–127]. The spectrum of membrane-related problems studied using molecular
simulations ranges from properties of self-assembled, single- and multi-component lipid
bilayers [128–140], fusion of vesicles [141], and interactions of membranes with sugars
[142], proteins [143–147], and peptides [148].
The advent of faster computer processors and efficient algorithms has scaled membrane
system sizes amenable to molecular simulations, with explicit atomistic representation, as
shown in Fig. 16 for a DMPC molecule, from 128 lipids [128] to hundreds of thousands
at present
[127]. However, a simple estimate shows that the total number of atoms,
inclusive of both lipids and water, involved in a MD simulation of a micron-sized vesicular
membrane is of the order of 1011. The number of degrees of freedom involved is an over
representation of the membrane if one is only interested in studying its physical properties
at length and time scales far separated from the atomistic scales. As an alternative, coarse
graining techniques can be used to retain only the degrees of freedom that are relevant to
the time and length scales under investigation, which in turn can reduce the number of
atomic units used in the simulations.
The concept of coarse graining has been used extensively used in constructing a macro-
scopic theory of material systems from their microscopic degrees of freedom [68]. An
example of coarse graining in soft-matter systems is the blob representation of polymers,
where a segment of the polymer is represented as an independent entity of a given size
[149]. Using a similar strategy, the multiple microscopic variables associated with a set of
atoms in a lipid molecule can be replaced by fewer macroscopic variables that capture the
essential physics of these atoms: for instance, the spatial position of the atoms in the head
group of a lipid molecule can be substituted with a coarse grained (CG) particle that has
an average position and a characteristic size. The various coarse graining strategies differ
in how these macroscopic variables are handled: is the CG particle placed at the center of
mass or around the position of the largest atom? What macroscopic properties does the
choice of interaction potential between the CG particles intend to capture? Coarse grained
lipid models have evolved rapidly from simple bead-spring representations [150, 151] to
more sophisticated models with the choice of interaction potentials representing functions
that bridge certain membrane properties between the atomistic and coarse grained picture
[127, 152].
49
There are three well established coarse graining strategies for simulations of soft-matter
systems—namely, structure based, force based and potential-energy based. The key differ-
ences between these models lies in the choice of the interaction potential between the CG
beads, which are chosen to reproduce some specific physical properties observed in experi-
ments or computed through detailed molecular simulations. In the case of structure-based
models, the CG potentials are designed to minimize the difference between the coarse- and
fine-grained radial distribution functions. The resulting model retains the chemical struc-
ture of the atomistic system and can be an excellent choice if the study aims to investigate
the structural properties of the molecule. Such a model has also been used in scale hopping
and adaptive resolution simulations [153–155]. Structure-based CG methods have been
employed in simulating lipid membranes [156, 157] and their interactions with proteins
[158–160]. On the other hand, the CG potentials for the force-based [161, 162] method
are chosen to preserve the total force as one moves up from the atomistic description to
a coarse grained description of the membrane.
The conservation of the underlying structure and forces in the CG methods described
above, do not necessarily imply conservation of thermodynamic averages of various ob-
servables. Alternatively, the free-energy-based MARTINI force field by Marrink and co-
workers for lipids [163, 164] and proteins [165] is another widely used parameterization to
semiquantitatively reproduce a wide distribution of thermodynamic data. The resulting
force field is thus maximally transferable to novel systems with different temperatures,
pressures, and compositions. It has found wide application in modeling vesiculation [137],
membrane fusion [141] and pore formation [140, 146], peptide-lipid interactions, protein-
gated ion channels, and lipid raft formation. Several mechanisms of curvature interactions
on membranes have been carried out using particle-based simulation methods [166–169].
CGMD models can also be used in combination with MD models in bottom-up ap-
proaches of systematically coarse-graining the atomistic description. Bridging techniques
that seamlessly integrate two distinct length scales in this category include the inte-
grated molecular mechanics/coarse-grained or MM/CG approaches [170]. Another goal
of molecular/coarse-grained modeling can also be to rigorously bridge the molecular scale
with the continuum scale, so that information is effectively passed between scales in a
self-consistent manner.
The Canham - Helfrich model in eqn. (8) is only an idealization of the free-energy of
50
the interface. The true bilayer system can only be accurately described by invoking the
statistical mechanics of inhomogeneous phases
[68]. The MD and CGMD models can
in principle capture the nature of such inhomogeneities when a computational approach
to modeling is sought. However, a theoretical approach for handling the density inho-
mogenieties can also be pursued within the purview of classical density functional theory,
wherein the free-energy can be described as a unique functional of the spatially varying
density [171]. In this context, the membrane thickness is finite with an inhomogeneous
density profile, and the conjugate variables to the strain or deformation fields are defined
through the pressure tensor. For an inhomogeneous system, the pressure tensor P at
position r can be expressed in a tensorial form of the virial equation, and it can be split
into a kinetic part, PK, derived from the kinetic energy and a potential part, PU, derived
from the potential energy U ({ri}) [172, 173],
P(r) = PK(r) + PU(r), where PK(r) = kBT ρ(r)I.
(46)
Here, U is the potential energy function, {ri} represents the vectorial coordinate of atom
i, ρ is the density, kB is the Boltzmann constant, and I is the identity matrix. For pair-
j, i(cid:54)=u(ij), where u(ij) = u(rij) with rij = ri − rj,
(cid:80)
wise additive potentials U = (1/2)(cid:80)
(cid:42)(cid:88)
(cid:88)
PU(r) is given by:
(cid:43)
(cid:90)
i
dl δ(r − l)
.
Cij
(47)
PU(r) = −
1
2
∂u(ij)
∂rij
j, i(cid:54)=j
i
Here, Cij is a contour from ri to rj. Different conventions to choose the contour yield differ-
ent (non-equivalent, see however [174]) expressions for the pressure tensor [172, 173, 175].
For a review comparing the different methods, see references [176, 177]. Applications of
these methods for the characterization of interfacial stresses in lipid bilayers are available
[150, 178].
brane as an infinitesimally thin interface, and the Hamiltonian H =(cid:80)
The correspondence between equation for Hsur (eqn. (9)), which describes the mem-
i + U ({ri})
governing P(r), which provides a molecular view of the membrane interface of finite width,
2mir2
1
i
can be derived by equating the corresponding free-energy changes or work done in de-
forming the interface in the two descriptions [182, 183]. The expression for the surface
tension is given by (here, we choose the direction normal to the bilayer as the z−axes and
51
calculate the pressure averages over the lateral plane [172, 179, 184]):
γ =
0
dz[PN (z) − PL(z)] ,
(48)
where PN (z) = (cid:104)Pzz(z)(cid:105) and PL = 1
2((cid:104)Pxx(z) + Pyy(cid:105)) are the average normal and lateral
pressure components of the tensor P, respectively. The expressions for bending stiffness
constants are given by [182, 183]:
(cid:90) Lz
(cid:90) Lz
(cid:90) Lz
0
κC0 =
dz(z − zref)[PN (z) − PL(z)] ,
κG =
0
dz(z − zref)2[PN (z) − PL(z)] .
(49)
(50)
These expressions were derived assuming that deforming the interface conserves the total
volume. In general, the expressions relating the elastic constants and the pressure tensor
are ensemble dependent; for example, analogous expressions in the constant projected area
ensemble are derived by Farago [185]. Alternatively, the equivalence between Hsur and
H can be established by requiring that the height-height fluctuation spectrum in both
models are in agreement. This equivalence provides a powerful technique to estimate
membrane elastic constants in both microscopic and macroscopic simulations. Such an
approach has been used recently to determine the macroscopic bending modulus κ and
Gaussian modulus κG from molecular simulations [186–188].
IV. MODELING MEMBRANE PROTEINS AS SPONTANEOUS CURVATURE
FIELDS
A. Curvature induction in membranes—intrinsic and extrinsic curvatures
Based on eqn. (19), the natural thermal undulations are pronounced at large wave-
lengths and supressed at small wavelengths due to equipartition of energy. Thus, it is
highly unlikely that curvature at the nanoscale can arise spontaneously due to bending-
mediated thermal undulations. The regular but complex membrane morphologies ob-
served in cellular systems is expected to be influenced by additional interactions defining
an energy landscape associated with curvature. Experimental observations suggest that
52
the process of systematic deformation of cellular membranes may be local, and the macro-
scopic reorganization of membranes is likely driven by cooperative interactions.
In a bilayer membrane, both the number and composition of lipid molecules in each
monolayer can vary. This difference is an inherent source of spontaneous curvature in
membranes. Consider a bilayer membrane, with its monolayers uniformly separated by
a distance δ and an equilibrium area difference equal to ∆A0 ( in the case of a planar
membrane ∆A0 = 0). When the area difference between the monolayers deviates from its
equilibrium value to ∆A, the membrane develops a spontaneous curvature, which can be
expressed in terms of their material properties, using the area difference elasticity (ADE)
model, as [67],
C0 = δ
∂G
∂(∆A)
Amidδ
(∆A − ∆A0) .
π
+
κnl
κl
(51)
Here κnl and κl are the non-local and local bending rigidities, with κnl ≈ κl for most
phospholipid bilayers [67]. G =(cid:82) dA H 2 is a dimensionless measure of the squared curva-
ture of the surface, and Amid =(cid:82) dA is the area of the reference surface. In the presence
of an area asymmetry between the monolayers, which implies non-zero spontaneous cur-
vature (C0 (cid:54)= 0), the equilibrium morphology of the membrane would assume shapes with
2H = C0.
A variety of mechanisms can induce an area asymmetry in the monolayers of the
membrane. Common examples include (a) flip-flop of lipids between the monolayers, (b)
mixing of lipids with the solvent, (c) presence of lipids whose geometry differs from that
of the bulk membrane, and (d) insertion of a non-lipid molecule into the bilayer. Such
an induced spontaneous curvature can either be local or be felt over the entire membrane
domain. In case of (c) and (d) the strength and magnitude of C0 is dependent on the
geometry and composition of the lipids and on the macromolecules in the membrane. Un-
catalyzed inter leaflet lipid transduction is an extremely slow process, and lipid molecules
have diminishingly small solubility in aqueous solvents and hence can be considered to be
insoluble for all practical purposes. For these reasons, the spontaneous curvature contri-
butions (a) and (b) are negligible in our time scales of observation. It should also be noted
that in experiments on model membranes, large deformations in membrane shapes are
observed by changing environmental conditions like temperature and salt concentration.
53
However, the role of these variables as drivers of membrane shape change can be neglected
since they fluctuate weakly in physiological conditions.
Under physiological conditions, the spontaneous curvature induced by area asymmetry
is considered too weak [189, 190] to induce large deformations in the membrane shapes.
Zimmerberg and Kozlov [191] have shown that it takes large area asymmetry to generate
vesicular shapes observed in cellular systems; e.g. their calculations show that the area
asymmetry required to bud off vesicles of size 100 − 200nm, which matches transport
vesicles in biological cells, from a cellular size GUV is of the order of ∆A = 0.1Amid.
Known lipid compositions of biological membranes do not support such a large differ-
ence in the area of the monolayers, which in turns calls for other external mechanisms of
curvature induction. In the past decade, a vast body of experimental evidence has high-
lighted the role of membrane-associated macromolecules, mainly proteins, in remodeling
the curvature of membranes. Proteins induce membrane curvature through the two key
mechanisms discussed below.
(a)
(b)
FIG. 17.
(a) States of a symmetric transmembrane ion channel embedded inside a membrane
bilayer. The membrane is flat and has no induced-curvature when the channel is in the closed
state (i) and when the channel transitions to the open state the membrane is spontaneously
curved under the influence of the induced-curvature (ii). (b) Curvature induction due to the
insertion of an asymmetric transmembrane protein (i) and due to the partial insertion of an
amphipathic entity as shown in (ii).
1. Curvature generation by wedging: The simplest example for protein induced sponta-
neous curvature can be seen in the case of membrane-spanning proteins generally called
54
(i) closed stateIONCHANNELflat membrane(ii) open stateGenerates curvatureNa+ ions(i) Asymmetric wedging(ii) Amphipathic insertion transmembrane proteins. This family of proteins represents a large class of macro-
molecules like pore proteins aquaporin, ion channels, ion pumps, enzymes, and light
harvesting complexes. The membrane-deforming properties of the transmembrane pro-
teins is demonstrated in Fig. 17(a) using a sodium ion channel. The functionality of the
ion channel is dependent on its conformation, which switches between a closed and open
state, as shown in Fig. 17(a), (i) and (ii), respectively. Mismatch in the size and shape of
the proteins and lipids tends to distort the membrane around a transmembrane protein
to induce an area asymmetry between the upper and lower leaflet of the bilayer. This
distortion is minimal when the channel is in the inactive state (closed) resulting in a flat
membrane. When the channel opens to allow the diffusion of sodium ions, the surrounding
membrane is more distorted, and as a result, the vicinity of the channel becomes curved,
as shown in Fig.
17(a)(b). In the event of cell signaling, when all the channels open
up in unison, the induced area asymmetry can be large enough to induce spontaneous
curvatures required for systemic membrane deformations
[192]. A recent experimental
report on curvature sensing and partitioning ability of potassium channels reinforces the
above notion [193].
2. Curvature induction by scaffolding:
Proteins that localize and interact with the
surface of the membrane are called peripheral proteins. Many of these proteins, in ad-
dition to their definite functional role, also interact with the membrane. These inter-
actions result in membrane curvature induction as observed in BAR-domain-containing
proteins, dynamin superfamily proteins [194], nexins, ENTH-domain-containing epsins,
reticulon-containing Dp1/Yop1 [195, 196], C2-domain-containing synaptotagmin, and
clathrin complexes
[197]. The underlying cytoskeletons are also known to tether to the
membrane and pull out long tethers, whose effects will not be considered here. Some of the
peripheral proteins can also induce curvature through the wedging mechanism described
earlier.
55
FIG. 18. Classes of the BAR domains. Shown are the crystal structures for a classical BAR
domain (a), a F-BAR domain (b), and an I-BAR domain (c): the monomeric units that
constitute the dimeric molecule are shown in blue and red, and the positive charges on each
protein are localized to the curved surface; this surface being concave for the classical- and F-
BAR and convex for an I-BAR. (d) The N-terminal amphipathic helices in an N-BAR domain,
whose dimeric unit is shown in red.
The study of BAR-domain-containing proteins and their role in membrane reshaping is
currently an active area of research. The BAR (Bin/Amphiphysin/Rvs) domain, a dimeric
molecule made from 260−280 amino acids arranged primarily as α-helices, is conserved in
many proteins of the endocytic pathway [198] and conserved in most organisms ranging
from yeast to humans [199–201]. These domains are known to generate high curvature
due to their interactions with the membrane [202] and also by recruiting other proteins
like Dynamin, Arp 2/3 [203]. The monomeric units are arranged such that the resulting
BAR domain has a curved shape, as in Fig. 18, that interacts with the underlying lipid
membrane. Based on the motif of the interacting regions, the BAR domains are further
classified into classical-BAR, F-BAR, and I-BAR [202]. Both the classical and F-BAR
domains are crescent (or) banana shaped, with the former being more curved than the
latter. The shallowness of F-BAR proteins, which is ∼ 3 fold smaller than the classical
[204–206].
BAR, arises due to a slightly different arrangement of the monomeric units
In the conventional sense, the membrane bending into the extracellular space is said to
56
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++a) Classical BAR domainb) F- BAR domainc) I-BAR domaind) Endophilin N-BARN-terminal amphipathic helix be positively curved. I-BAR domains, on the other hand, are negatively curved [207].
A special class of the positively curved dimers is the N-BAR domains, characterized by
the presence of an amphipathic helix at the N-terminal of each monomer unit
[208],
which in shown in Fig. 18(d). The presence of the helix enhances the domain membrane
interaction. For example, the N-BAR containing Endophilin tubulates liposomes even
at lower concentrations where the classical BAR fails to [209]. The extra curvature is
induced by the amphipathic helix that embeds into a monolayer, to the depth of the
glycerol groups, serving as a wedge that generates a spontaneous curvature [203, 210]
due to area asymmetry in the bilayer.
Interaction of the BAR domains with the underlying membrane is primarily electro-
static in nature. The positive charges of the protein are concentrated on the membrane
proximal surface of the BAR domains (see Fig. 18), which is in turn strongly attracted by
the negatively charged lipid heads. As the protein is fairly stiff, the membrane responds
to the above-mentioned attractive interactions by curving itself to match the curved sur-
face of the BAR domain, but at the cost of the bending energy. The strong electrostatic
interactions dominate over the soft elastic energy resulting in a spontaneous deformation
of the membrane [211]. As shown in Fig. 19, the presence of a crescent like BAR domain
induces a spontaneous positive curvature on the membrane. The degree of deformation
depends on the type of BAR domain, the distribution of the cationic residues on the
domain, and also on the nature of the lipid molecules making up the membrane.
FIG. 19.
An illustration of the membrane reshaping behavior of positively and negatively
curving BAR domains. A BAR domain inducing positive curvature, like the classical-, F- and N-
BAR, induces a positive curvature along its orientation (left panel), whereas an I-BAR stabilizes
negative curvature in the membrane (right panel).
It has further been observed that the membrane deforms preferentially in particular
57
Stabilizes positive curvatureGenerates negative curvature Classical BAR (or) F-BAR (or) N-BARI-BAR (or) MIM directions. This is a key feature that differentiates a scaffolding protein from a transmem-
brane protein, whose locally induced membrane deformations tend to be isotropic. The
directional bending anisotropy of the scaffolding proteins can be mapped to a n-atic tan-
gent plane order, which is invariant under rotations by 2π/n. The deformations induced
by the amphipathic helices of the N-BAR domain are also directional in nature, hence
fitting naturally into this scheme. Computational investigations into the BAR domain-
membrane interactions have also established the directional nature of the BAR-domain
containing proteins [212–214]. More details on the various classes of membrane remod-
eling proteins, their mechanisms of membrane bending, and the associated biochemical
network can be found in various reviews on this topic [3, 189, 191, 211, 215–219].
The spatial pattern of the spontaneous curvature induced in the membrane through
the mechanisms listed above can have entirely different forms. For the purpose of iden-
tifying the suitable continuum model for a given system, we classify the spontaneous
curvature(C0) induced by membrane-curving macromolecules (henceforth we refer to them
as curvactants) into two major classes based on the principal radii of curvature R1 and
R2.
Isotropic curvactants: Local membrane deformations induced by this class of pro-
teins/protein complexes and the associated curvature profile have a radial symmetry
(R1 = R2 = R). The curvature induced by symmetric transmembrane proteins/inclusions,
illustrated in Fig. 17(a) and (b,i), falls into this category. As will be seen later, the pres-
ence of these proteins preserves the rotational symmetry of the membrane and hence does
not involves additional energy contributions due to symmetry breaking [68] (i.e.
the
energy of the membrane in a given conformation remains invariant under change in orien-
tation of the proteins). Adhesive functionalized nanoparticles, like those used in targeted
drug delivery applications, can strongly bind to membranes, which in turn indents the
membrane in their vicinity leading to an isotropic spontaneous curvature [220–222]
Anisotropic curvactants: The curvature profile induced by BAR-domain-containing pro-
teins, dynamin, and exo70-domain-containing proteins are inherently anisotropic (R1 (cid:54)=
R2). This anisotropy arises due to a variety of reasons like the secondary structure of the
protein, the charge distribution in the membrane-facing domain of the protein, and dis-
tribution of lipids around the proteins. Direct experimental/computational evidence, for
58
the anisotropic nature of the curvature profile is hard to obtain since these deformations
at the single molecule level are not large enough to be distinguished from those resulting
from thermal undulations in the membrane. Alternately, the anisotropic form of the cur-
vature field can be observed in the macroscopic structures stabilized when these proteins
interact with a liposome. For instance, liposomes tubulate in the presence of different
F-BARs, syndapin, and dynamin proteins, as shown in Fig. 20(a). It is hypothesized that
the spontaneous curvature induced by a protein is then related to the radius of the tube
it stabilizes. The macroscopic organization of membrane-remodeling proteins on tubular
liposomes are known only for a few proteins like dynamin [223]—Fig. 20(b) shows the
helical arrangement of dynamin on a tube of radius 25nm, with the pitch of the helix
measuring 15 nm [224, 225]. Computational studies have also been supportive of the
anisotropic nature of many curvature-remodeling proteins. All atom simulations of N-
BAR domains [212] and BAR domains [160], shown respectively in Figs. 20(c) and 20(d),
confirm the presence of directional curvatures in these systems. The latter study [160]
and others
[159, 226] also point to the role of cooperativity amongst proteins in re-
shaping membranes. For a given protein density, it has been shown that the radius of
curvature of the deformed membrane depends on the relative orientation between the
proteins. Furthermore, the proteins were required to be on a lattice to facilitate extreme
membrane deformations, as seen when a planar membrane folds to a tube
[11]. We
characterize the class of anisotropic curvature-inducing proteins by two values of sponta-
neous curvature—C
(cid:107)
0 = 1/R1 and C⊥
0 = 1/R2—which are respectively the spontaneous
curvatures induced along and normal to the orientation of the long axis of the protein.
59
(a)
(b)
(c)
(d)
FIG. 20.
(a) Liposome tubulation by F-BAR proteins, Synapdin, and Dynamin (Reprinted
from Dev. Cell, 9 (6), Itoh et. al., Dynamin and the Actin Cytoskeleton Cooperatively Regulate
Plasma Membrane Invagination by BAR and F-BAR Proteins, 791–804, Copyright (2005), with
permission from Elsevier), (b) A cartoon of the helical organization of Dynamin proteins on
a tube they constrict. (c) Snapshot from a 27 ns molecular dynamics simulation of an N-
BAR domain interacting with a DOPC/DOPS membrane (Image from P. Blood and G. Voth,
Direct observation of Bin-Amphiphysin-Rvs(BAR) domain-induced membrane curvature by means
of molecular dynamics simulations, Proc. Nat. Acad. Sci. 103, 15068-15072 [212] - Copyright
(2006) National Academy of Sciences, U.S.A) (d) All atom simulations of Amphiphysin N-BAR
domains—in its dimeric form and arranged in a staggered conformation—shows the generation
of spontaneous directional curvature (Reprinted from Structure, 17 (6), Ying Yin, Anton Arkhipov,
Klaus Schulten, Simulations of Membrane Tubulation by Lattices of Amphiphysin N-BAR Domains,
882–892, Copyright (2009), with permission from Elsevier) .
Intrinsic anisotropy: Anisotropic behavior in membranes can also be intrinsic, originating
60
isalsosharedbyF-BARdomainproteins.RecombinantFBP17,CIP4,syndapin,aswellastheF-BARdomainsofFBP17(aminoacids1–284)orCIP4(aminoacids1–284)wereincubatedwithliposomescomposedofacrudebrainlipidextract(FolchfractionI).Negativestainingelectronmicroscopyofthesepreparationsrevealedthatalloftheseproteinshadapowerfultubulatingactiv-ity(Figure2A,upperpanels).ThediameterofF-BAR-inducedtubuleswassomewhatvariableanddependentontheexperimentalconditions(outerdiameterbetween40and200nm).Thepolymerizationpropertywassug-gestedtoplayaroleinthelipid-tubulatingactivityofBARdomains(FarsadandDeCamilli,2003;Peteretal.,2004).Interestingly,theF-BARdomainofFBP17,whenincubatedalonewithoutliposomes,polymerizedintolongfilaments(FigureS2).Inthepresenceofliposomesanddynamin,amphiphy-sinandendophilingeneratenarrowtubulesofahighlyuniformsmalldiameterwithacharacteristicstripedhy-bridcoat(thickstripeswithapitchofw20.261.5mm)differentfromthestripedcoatsgeneratedbyBARpro-teinsalone(verythin,tightlyapposedstripes)ordynaminalone(thinstripeswithapitchofw13.260.5mm[Farsadetal.,2001;Takeietal.,1999]).Inasimilarfashion,whenincubationsofFBP17withliposomesweresupple-mentedwithdynamin,narrowtubulesofauniformdiam-eterandwithastripedcoatresemblingtheamphiphysin/dynaminorendophilin/dynaminhybridcoat(thickstripes;pitchof23.760.88mm)wereobserved(Figure2A,lowerpanels).ThepowerfulliposometubulationactivityofF-BARproteinswasconfirmedbydynamiclightmicroscopy-basedassays.Inafirstassay,lipiddropletswerefirstdriedinasmallchamberonaglassslide,andthenhydratedtogeneratelargebilayersheets(Figure2B).Asshownbydifferentialinterferencecontrast(DIC)mi-croscopy,theadditionofallF-BARdomainproteinstestedinthisassay,namelyFBP17,CIP4,andsyndapin,Figure2.F-BARDomainsTubulateLiposomes(A)Negativestainingelectronmicroscopy.Upperpanels:LiposomeswereincubatedwithrecombinantF-BAR-containingproteinsortheiriso-latedF-BARdomains.Lowerpanels:LiposomeswereincubatedeitherwithFBP17orwithdynaminalone,orwithbothproteinstogether,asindicated.ThepresenceofbothproteinsinducedthereorganizationoftheFBP17-only-anddynamin-only-coatedtubulesintonarrowtubulesdecoratedbyastripedcoat(withthickerstripesthanthedynamin-onlycoat)reminiscentofthehybridamphiphysin/dynaminorendophilin/dynamincoat(Farsadetal.,2001;Takeietal.,1999).Thescalebarsare100nm.(B)Left:Schematicdrawingoftheexperimentalsystemusedtoanalyzethetubulationoflipidbilayersbydifferentialinterferencecontrast(DIC)microscopyasshownintherightfield.Lipidsarespottedinasmallchamberbetweentwoglasssurfaces,dried,andthenrehydratedtogeneratebilayersheets,typicallycomposedofafewsuperimposedlayers.Protein-containingsolutionsareaddedtothechamberduringmicroscopicobservation.Thediagramofalipidspotbeforeandafterapplicationofatubulatingprotein.Right:Timecourseanalysisofedgesofmembranesheets(MS)duringtheincubationwithacontrolprotein(top)orwithFBP17(bottom).(C)EdgesoflipidsheetsafterincubationwithCIP4andsyndapin1asin(B),demonstratingthepresenceoflipidtubules.(D)Liposomesweregeneratedinsuspension,thenlabeledwiththefluorescencedyeFM2-10(top)andincubatedwithFBP17(middle)orCIP4(bottom).Thescalebarsare5mm.DevelopmentalCell794Nature Reviews Molecular Cell BiologyNeckNeckNeck1301520627751864GTPaseMiddlePHGEDPRDG domain‘Stalk’‘Stalk’‘Foot’PRDa Dynamin 1bdcePolymerization ofdynamin helix(cid:47)(cid:71)(cid:79)(cid:68)(cid:84)(cid:67)(cid:80)(cid:71)(cid:2)(cid:558)(cid:85)(cid:85)(cid:75)(cid:81)(cid:80)GTP-dependent dimerization ofG domains, GTP hydrolysis andresulting membrane constriction G domainGTPNeck‘Stalk’‘Foot’132290°The G domain and BSE. The G domain sits on a helical bundle, known as the bundle signalling element (BSE)30 or neck40, which is formed by three helices derived from sequences at the N-terminal and C-terminal sides of the G domain and from the C-terminal region of the GED, respectively (FIG. 2). Consistent with the idea that this region of the GED is in close physical proximity with, and is functionally linked to, the G domain, a screen for suppressor mutations of a mutation in the G domain of D. melanogaster dynamin identified a mutation within the GED C terminus41. The BSE is followed by a stalk, which is composed of helices from the middle domain and the N-terminal region of the GED30–33, and a PH domain42, which forms the vertex or ‘foot’ of the stalk hairpin and binds membranes. The PRD, which is expected to be unfolded, emerges at the boundary between the BSE and the G domain, most likely project-ing away from the membrane, where it might interact with other proteins.Dimerization through the stalk. The stalk of dynamin dimerizes in a cross-like fashion (FIG. 2a,b) to yield a dynamin dimer in which the two G domains are oriented in opposite directions30–33 (FIG. 2b). This dimer is the basic dynamin unit and is different from the additional dimer-ization interface that is generated by the interaction of two G domains (FIG. 2b,c), which is discussed below.Phospholipid association through the PH domain. The PH domain binds acidic phospholipids in the cytosolic leaflet of the plasma membrane, and phosphatidyl-inositol-4,5-bisphosphate (PtdIns(4,5)P2) in particular, via a positively charged surface at the foot of the dynamin hairpin42,43. PH-domain mutants that impair phospho-inositide binding exert dominant-negative effects on clathrin-mediated endocytosis44,45. The binding between the isolated dynamin PH domain and phosphoinositides is of very modest affinity (>1 mM), but this membrane interaction is strengthened by charge-dependent asso-ciation with other negatively charged phospholipids and by avidity afforded by dynamin polymerization43–46. A hydrophobic loop emerging from the PH domain may promote membrane interactions and may also have curvatur e-generatin g or -sensing properties28,47.Coordinating dynamin function through the PRD. The PRD contains an array of PXXP amino acid motifs, which interact with many SH3 domain-containing proteins (see Supplementary information S1 (table)) to localize dynamin at endocytic sites and coordinate dynamin’s function with these other factors during endocytosis48–51. Accordingly, dynamin lacking the PRD cannot rescue endocytic defects in dynamin-knockout fibroblasts19. The PRD–SH3 interactions are typically of moderate affinity, but the presence of multiple SH3-binding motifs in the PRD and multiple SH3 domain-containing pro-teins at endocytic sites, as well as the polymeric state of these proteins, results in a significant avidity effect, which enhances the ability of such interactions to concentrate dynamin. At least some interactions of the dynamin PRD are regulate d by phosphorylation48.(cid:40)(cid:75)(cid:73)(cid:87)(cid:84)(cid:71)(cid:2)(cid:20)(cid:2)(cid:94)(cid:2)Structure of dynamin and putative mechanism of dynamin-mediated membrane fission. a(cid:124)(cid:94)(cid:124)(cid:46)(cid:75)(cid:80)(cid:71)(cid:67)(cid:84)(cid:2)(cid:84)(cid:71)(cid:82)(cid:84)(cid:71)(cid:85)(cid:71)(cid:80)(cid:86)(cid:67)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:70)(cid:81)(cid:79)(cid:67)(cid:75)(cid:80)(cid:2)(cid:81)(cid:84)(cid:73)(cid:67)(cid:80)(cid:75)(cid:92)(cid:67)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:81)(cid:72)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)based on its three-dimensional structure, as revealed by crystallographic studies (cid:10)(cid:80)(cid:87)(cid:79)(cid:68)(cid:71)(cid:84)(cid:85)(cid:124)(cid:75)(cid:80)(cid:70)(cid:75)(cid:69)(cid:67)(cid:86)(cid:71)(cid:2)(cid:67)(cid:79)(cid:75)(cid:80)(cid:81)(cid:2)(cid:67)(cid:69)(cid:75)(cid:70)(cid:2)(cid:82)(cid:81)(cid:85)(cid:75)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:89)(cid:75)(cid:86)(cid:74)(cid:75)(cid:80)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:82)(cid:84)(cid:75)(cid:79)(cid:67)(cid:84)(cid:91)(cid:2)(cid:85)(cid:71)(cid:83)(cid:87)(cid:71)(cid:80)(cid:69)(cid:71)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:74)(cid:87)(cid:79)(cid:67)(cid:80)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:124)(cid:19)(cid:2)(cid:90)(cid:67)(cid:2)(cid:85)(cid:82)(cid:78)(cid:75)(cid:69)(cid:71)(cid:2)(cid:88)(cid:67)(cid:84)(cid:75)(cid:67)(cid:80)(cid:86)(cid:11)(cid:16)(cid:2)(cid:52)(cid:71)(cid:73)(cid:75)(cid:81)(cid:80)(cid:85)(cid:2)(cid:86)(cid:74)(cid:67)(cid:86)(cid:2)(cid:68)(cid:71)(cid:78)(cid:81)(cid:80)(cid:73)(cid:2)(cid:86)(cid:81)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:85)(cid:67)(cid:79)(cid:71)(cid:2)(cid:72)(cid:81)(cid:78)(cid:70)(cid:71)(cid:70)(cid:2)(cid:79)(cid:81)(cid:70)(cid:87)(cid:78)(cid:71)(cid:2)(cid:67)(cid:84)(cid:71)(cid:2)(cid:85)(cid:74)(cid:81)(cid:89)(cid:80)(cid:2)(cid:75)(cid:80)(cid:124)(cid:86)(cid:74)(cid:71)(cid:2)(cid:85)(cid:67)(cid:79)(cid:71)(cid:2)(cid:69)(cid:81)(cid:78)(cid:81)(cid:87)(cid:84)(cid:16)(cid:2)(cid:54)(cid:74)(cid:71)(cid:2)(cid:69)(cid:84)(cid:91)(cid:85)(cid:86)(cid:67)(cid:78)(cid:2)(cid:85)(cid:86)(cid:84)(cid:87)(cid:69)(cid:86)(cid:87)(cid:84)(cid:71)(cid:2)(cid:81)(cid:72)(cid:2)(cid:67)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)(cid:70)(cid:75)(cid:79)(cid:71)(cid:84)(cid:2)(cid:75)(cid:85)(cid:2)(cid:85)(cid:74)(cid:81)(cid:89)(cid:80)(cid:2)(cid:68)(cid:71)(cid:78)(cid:81)(cid:89)(cid:2)(cid:10)(cid:79)(cid:75)(cid:80)(cid:87)(cid:85)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:50)(cid:52)(cid:38)(cid:85)(cid:14)(cid:124)(cid:89)(cid:74)(cid:75)(cid:69)(cid:74)(cid:2)(cid:67)(cid:84)(cid:71)(cid:2)(cid:86)(cid:74)(cid:81)(cid:87)(cid:73)(cid:74)(cid:86)(cid:2)(cid:86)(cid:81)(cid:2)(cid:68)(cid:71)(cid:2)(cid:87)(cid:80)(cid:72)(cid:81)(cid:78)(cid:70)(cid:71)(cid:70)(cid:11)(cid:14)(cid:2)(cid:67)(cid:80)(cid:70)(cid:2)(cid:75)(cid:85)(cid:2)(cid:69)(cid:81)(cid:78)(cid:81)(cid:87)(cid:84)(cid:15)(cid:69)(cid:81)(cid:70)(cid:71)(cid:70)(cid:2)(cid:86)(cid:81)(cid:2)(cid:79)(cid:67)(cid:86)(cid:69)(cid:74)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:78)(cid:75)(cid:80)(cid:71)(cid:67)(cid:84)(cid:2)representation (created with PyMOL (Schrödinger); Protein Data Bank code (cid:21)(cid:53)(cid:48)(cid:42)(cid:21)(cid:19)). b(cid:124)(cid:94)(cid:124)(cid:53)(cid:69)(cid:74)(cid:71)(cid:79)(cid:67)(cid:86)(cid:75)(cid:69)(cid:2)(cid:84)(cid:71)(cid:82)(cid:84)(cid:71)(cid:85)(cid:71)(cid:80)(cid:86)(cid:67)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:81)(cid:72)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)(cid:70)(cid:75)(cid:79)(cid:71)(cid:84)(cid:85)(cid:2)(cid:67)(cid:80)(cid:70)(cid:2)(cid:81)(cid:72)(cid:124)(cid:74)(cid:71)(cid:78)(cid:75)(cid:69)(cid:67)(cid:78)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)(cid:82)(cid:81)(cid:78)(cid:91)(cid:79)(cid:71)(cid:84)(cid:85)(cid:2)(cid:67)(cid:84)(cid:81)(cid:87)(cid:80)(cid:70)(cid:2)a tubular template in two different orientations (with 90° rotation between them). The colour-coding of the domains matches the colours in part a (minus the PRDs, which are (cid:86)(cid:74)(cid:81)(cid:87)(cid:73)(cid:74)(cid:86)(cid:2)(cid:86)(cid:81)(cid:2)(cid:82)(cid:84)(cid:81)(cid:76)(cid:71)(cid:69)(cid:86)(cid:2)(cid:81)(cid:87)(cid:86)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:82)(cid:81)(cid:78)(cid:91)(cid:79)(cid:71)(cid:84)(cid:75)(cid:92)(cid:71)(cid:70)(cid:2)(cid:74)(cid:71)(cid:78)(cid:75)(cid:90)(cid:11)(cid:16)(cid:2)(cid:54)(cid:74)(cid:71)(cid:2)(cid:67)(cid:82)(cid:82)(cid:84)(cid:81)(cid:90)(cid:75)(cid:79)(cid:67)(cid:86)(cid:71)(cid:2)(cid:78)(cid:81)(cid:69)(cid:67)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:68)(cid:81)(cid:87)(cid:80)(cid:70)(cid:2)nucleotide is highlighted in yellow (small circles). Dynamin polymerization occurs as a (cid:84)(cid:71)(cid:85)(cid:87)(cid:78)(cid:86)(cid:2)(cid:81)(cid:72)(cid:2)(cid:75)(cid:80)(cid:86)(cid:71)(cid:84)(cid:67)(cid:69)(cid:86)(cid:75)(cid:81)(cid:80)(cid:85)(cid:2)(cid:68)(cid:71)(cid:86)(cid:89)(cid:71)(cid:71)(cid:80)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:110)(cid:85)(cid:86)(cid:67)(cid:78)(cid:77)(cid:85)(cid:111)(cid:2)(cid:81)(cid:72)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)(cid:79)(cid:81)(cid:80)(cid:81)(cid:79)(cid:71)(cid:84)(cid:85)(cid:2)(cid:10)(cid:75)(cid:80)(cid:86)(cid:71)(cid:84)(cid:72)(cid:67)(cid:69)(cid:71)(cid:124)(cid:20)(cid:11)(cid:2)(cid:67)(cid:80)(cid:70)(cid:2)(cid:68)(cid:71)(cid:86)(cid:89)(cid:71)(cid:71)(cid:80)(cid:2)(cid:85)(cid:86)(cid:67)(cid:78)(cid:77)(cid:2)(cid:70)(cid:75)(cid:79)(cid:71)(cid:84)(cid:85)(cid:2)(cid:10)(cid:75)(cid:80)(cid:86)(cid:71)(cid:84)(cid:72)(cid:67)(cid:69)(cid:71)(cid:85)(cid:124)(cid:19)(cid:2)(cid:67)(cid:80)(cid:70)(cid:124)(cid:21)(cid:11)(cid:16)(cid:2)(cid:54)(cid:74)(cid:71)(cid:124)(cid:41)(cid:54)(cid:50)(cid:15)(cid:70)(cid:71)(cid:82)(cid:71)(cid:80)(cid:70)(cid:71)(cid:80)(cid:86)(cid:2)(cid:70)(cid:75)(cid:79)(cid:71)(cid:84)(cid:75)(cid:92)(cid:67)(cid:86)(cid:75)(cid:81)(cid:80)(cid:2)(cid:81)(cid:72)(cid:2)(cid:41)(cid:124)(cid:70)(cid:81)(cid:79)(cid:67)(cid:75)(cid:80)(cid:85)(cid:2)(cid:68)(cid:71)(cid:86)(cid:89)(cid:71)(cid:71)(cid:80)(cid:2)(cid:67)(cid:70)(cid:76)(cid:67)(cid:69)(cid:71)(cid:80)(cid:86)(cid:2)(cid:84)(cid:87)(cid:80)(cid:73)(cid:85)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:124)(cid:74)(cid:71)(cid:78)(cid:75)(cid:90)(cid:2)(cid:10)(cid:74)(cid:75)(cid:73)(cid:74)(cid:78)(cid:75)(cid:73)(cid:74)(cid:86)(cid:71)(cid:70)(cid:2)(cid:68)(cid:91)(cid:2)(cid:91)(cid:71)(cid:78)(cid:78)(cid:81)(cid:89)(cid:2)(cid:85)(cid:86)(cid:67)(cid:84)(cid:85)(cid:2)(cid:75)(cid:80)(cid:2)(cid:86)(cid:74)(cid:71)(cid:2)(cid:78)(cid:81)(cid:80)(cid:73)(cid:75)(cid:86)(cid:87)(cid:70)(cid:75)(cid:80)(cid:67)(cid:78)(cid:2)(cid:88)(cid:75)(cid:71)(cid:89)(cid:2)(cid:81)(cid:72)(cid:124)(cid:86)(cid:74)(cid:71)(cid:2)(cid:74)(cid:71)(cid:78)(cid:75)(cid:90)(cid:11)(cid:14)(cid:2)(cid:75)(cid:85)(cid:2)(cid:86)(cid:74)(cid:81)(cid:87)(cid:73)(cid:74)(cid:86)(cid:2)(cid:86)(cid:81)(cid:2)(cid:82)(cid:84)(cid:81)(cid:79)(cid:81)(cid:86)(cid:71)(cid:2)(cid:67)(cid:85)(cid:85)(cid:71)(cid:79)(cid:68)(cid:78)(cid:91)(cid:15)(cid:85)(cid:86)(cid:75)(cid:79)(cid:87)(cid:78)(cid:67)(cid:86)(cid:71)(cid:70)(cid:2)(cid:41)(cid:54)(cid:50)(cid:67)(cid:85)(cid:71)(cid:2)(cid:67)(cid:69)(cid:86)(cid:75)(cid:88)(cid:75)(cid:86)(cid:91)(cid:14)(cid:2)(cid:84)(cid:71)(cid:85)(cid:87)(cid:78)(cid:86)(cid:75)(cid:80)(cid:73)(cid:2)(cid:75)(cid:80)(cid:2)membrane constriction and ultimately fission. c(cid:124)(cid:94)(cid:124)(cid:50)(cid:84)(cid:81)(cid:82)(cid:81)(cid:85)(cid:71)(cid:70)(cid:2)(cid:41)(cid:54)(cid:50)(cid:2)(cid:74)(cid:91)(cid:70)(cid:84)(cid:81)(cid:78)(cid:91)(cid:85)(cid:75)(cid:85)(cid:15)(cid:70)(cid:71)(cid:82)(cid:71)(cid:80)(cid:70)(cid:71)(cid:80)(cid:86)(cid:2)lever-like movement of dynamin’s neck (the bundle signalling element), relative to the (cid:41)(cid:124)(cid:70)(cid:81)(cid:79)(cid:67)(cid:75)(cid:80)(cid:16)(cid:2)d(cid:124)(cid:94)(cid:124)(cid:53)(cid:69)(cid:74)(cid:71)(cid:79)(cid:67)(cid:86)(cid:75)(cid:69)(cid:2)(cid:88)(cid:75)(cid:71)(cid:89)(cid:2)(cid:81)(cid:72)(cid:2)(cid:86)(cid:74)(cid:71)(cid:85)(cid:71)(cid:2)(cid:77)(cid:71)(cid:91)(cid:2)(cid:85)(cid:86)(cid:71)(cid:82)(cid:85)(cid:2)(cid:78)(cid:71)(cid:67)(cid:70)(cid:75)(cid:80)(cid:73)(cid:2)(cid:86)(cid:81)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:15)(cid:79)(cid:71)(cid:70)(cid:75)(cid:67)(cid:86)(cid:71)(cid:70)(cid:2)(cid:79)(cid:71)(cid:79)(cid:68)(cid:84)(cid:67)(cid:80)(cid:71)(cid:2)fission. e(cid:124)(cid:94)(cid:124)(cid:37)(cid:84)(cid:91)(cid:81)(cid:15)(cid:71)(cid:78)(cid:71)(cid:69)(cid:86)(cid:84)(cid:81)(cid:80)(cid:2)(cid:79)(cid:75)(cid:69)(cid:84)(cid:81)(cid:85)(cid:69)(cid:81)(cid:82)(cid:91)(cid:2)(cid:75)(cid:79)(cid:67)(cid:73)(cid:71)(cid:2)(cid:85)(cid:74)(cid:81)(cid:89)(cid:75)(cid:80)(cid:73)(cid:2)(cid:67)(cid:2)(cid:74)(cid:71)(cid:78)(cid:75)(cid:69)(cid:67)(cid:78)(cid:2)(cid:82)(cid:81)(cid:78)(cid:91)(cid:79)(cid:71)(cid:84)(cid:2)(cid:81)(cid:72)(cid:2)(cid:82)(cid:87)(cid:84)(cid:75)(cid:72)(cid:75)(cid:71)(cid:70)(cid:2)(cid:70)(cid:91)(cid:80)(cid:67)(cid:79)(cid:75)(cid:80)(cid:2)that has driven the formation of a tubule from a liposome. GED, GTPase effector domain; (cid:50)(cid:42)(cid:14)(cid:124)(cid:82)(cid:78)(cid:71)(cid:69)(cid:77)(cid:85)(cid:86)(cid:84)(cid:75)(cid:80)(cid:2)(cid:74)(cid:81)(cid:79)(cid:81)(cid:78)(cid:81)(cid:73)(cid:91)(cid:29)(cid:2)(cid:50)(cid:52)(cid:38)(cid:14)(cid:2)(cid:50)(cid:84)(cid:81)(cid:15)(cid:84)(cid:75)(cid:69)(cid:74)(cid:2)(cid:70)(cid:81)(cid:79)(cid:67)(cid:75)(cid:80)(cid:16)(cid:2)(cid:43)(cid:79)(cid:67)(cid:73)(cid:71)(cid:2)(cid:75)(cid:80)(cid:2)(cid:82)(cid:67)(cid:84)(cid:86)(cid:2)e(cid:2)(cid:69)(cid:81)(cid:87)(cid:84)(cid:86)(cid:71)(cid:85)(cid:91)(cid:2)(cid:81)(cid:72)(cid:2)(cid:35)(cid:16)(cid:124)(cid:40)(cid:84)(cid:81)(cid:85)(cid:86)(cid:14)(cid:2)(cid:55)(cid:80)(cid:75)(cid:88)(cid:71)(cid:84)(cid:85)(cid:75)(cid:86)(cid:91)(cid:2)(cid:81)(cid:72)(cid:2)(cid:55)(cid:86)(cid:67)(cid:74)(cid:14)(cid:2)(cid:55)(cid:53)(cid:35)(cid:14)(cid:2)(cid:67)(cid:80)(cid:70)(cid:2)(cid:56)(cid:16)(cid:124)(cid:55)(cid:80)(cid:73)(cid:71)(cid:84)(cid:14)(cid:2)(cid:48)(cid:81)(cid:84)(cid:86)(cid:74)(cid:89)(cid:71)(cid:85)(cid:86)(cid:71)(cid:84)(cid:80)(cid:2)(cid:55)(cid:80)(cid:75)(cid:88)(cid:71)(cid:84)(cid:85)(cid:75)(cid:86)(cid:91)(cid:14)(cid:2)(cid:55)(cid:53)(cid:35)(cid:16)REVIEWSNATURE REVIEWS MOLECULAR CELL BIOLOGY VOLUME 13 FEBRUARY 2012 77© 2012 Macmillan Publishers Limited. All rights reservedbindsdirectlytothemaximumcurvaturesurfaceoftheBARdomain,resultinginaveryhighdegreeofmembranecurvature.Intheothercase(NBR2),theN-BARdomaintiltssothatthemaximumcurvaturesurfaceoftheBARdomainisnolongerparalleltothesurfaceofthebilayer.Inthisorientation,theBARdomainstillpresentsaconcavesurfacetothebilayer,butthissurfacehasasmallerdegreeofcurvature.Thislowercurvaturesurface,inturn,resultsinalowerdegreeofmembranecurva-ture.Fig.1cshowsthebindingofArgandLysresiduestothenegativelychargedoxygenatomsonthelipidheadgroupsduringsimulationsNBR1andNBR2.Interestingly,althoughthepro-cessofbindingdiffersinthetwocases,thefinalnumberofboundresiduesisquitesimilar.Inaddition,althoughtheybindinadifferentorientation,theN-BARdomainsinbothsimulationsuseessentiallythesamechargedresiduestobindthelipidbilayer.Asexpected,theinitialjumpinthenumberofboundresiduescoincideswith,andisslightlyprecededby,anabruptincreaseinthecurvatureofthemembrane.Thisobservationisconsistentwiththecurvaturebeingdrivenbyanelectrostaticattraction.Fig.2containssnapshotsofthemembranebindingandcurvaturedevelopmentduringthecourseofthesimulations.Thetwosystemsevolvequitedifferently.InNBR1(Fig.2a),abendingmodedevelopsdirectlyunderneaththeBARdomain,whereasinNBR2(Fig.2c),aglobalbendingmodeformsinitially.ThisdifferenceintheundulationdevelopmentinthetwomembranesmayhavecontributedtothedifferenceinBARdomainorientationduringthesimulation.Afteraninitialde-velopmentphase,bothmembranesbindcompletelytothesurfacethatispresentedtothemandlocallyadopttheintrinsiccurvatureofthatsurface(Fig.2bandd).OncetheBARdomainsareboundintheseconfigurations,theinducedmem-branecurvatureappearstobequitestableonthetimescaleofthesesimulations.Fig.3showstheaveragecurvatureandshapeforthemem-branesandtheBARdomainoverthefinal7nsofthesimula-tions,duringwhichthemembranecurvatureappearstohavepeaked.Thebindingofthemembranetothemaximumcurva-turesurfaceoftheN-BARdomaininNBR1correspondstothegenerallyspeculatedmannerofBARdomainbinding;however,itresultsinahigherdegreeoflocalmembranecurvature(peakradiusofcurvatureof⬇6.7nm;Fig.3a)thanissuggestedbydrawinganarcalongthemaximumcurvaturebindingsurfaceFig.1.TimeevolutionofmembranecurvaturewithN-BARdomainbindingandorientation.(a)MembranecurvaturedevelopmentforNBR1(solidline),NBR2(dashedline),andplainlipidbilayer(dottedline)intheregionoccupiedbytheBARdomain.Curvatureiscalculatedfromthediscretizedmembraneshape(seeFig.3b)ateachtimepoint.(b)AnglebetweentheBARdomainconcavebindingsurfaceandthemembranesurfaceforNBR1(solidline)andNBR2(dashedline).TheangleBARistheanglebetweentheycoordinateaxis(whichisparalleltothemembranesurface)andtheBARdomainprincipalaxisthatis(initially)mostnearlyparalleltotheycoordinateaxis(seeFig.2).(c)Numberoflipidheadgroupoxygenatomswithin4.2ÅofanArgorLysnitrogen(minimumassociationtimeof1ns;50-pssamplinginterval)forNBR1(solidline)andNBR2(dashedline).Fig.2.SimulationsnapshotsofN-BARdomainsinducinglocalmembranecurvature.(a)SnapshotofsimulationNBR1att⫽10ns.(b)SnapshotofsimulationNBR1att⫽27ns.(c)SnapshotofsimulationNBR2att⫽10ns.(d)SnapshotofsimulationNBR2att⫽27ns.InsimulationNBR1,theN-BARdomainmaintainsitsconcavesurfacefacingthemembraneandinducesahigherdegreeofcurvature.InsimulationNBR2,theN-BARdomaintiltsalongtheycoordinatedirection(intotheplaneoftheimage)andpresentsalessconcavesurfacetothemembrane.Inbothcases,themembranebendstomatchthecurvatureoftheN-BARdomainsurfacefacingthemembrane.Chargedphosphatidylserineheadgroupsareshowninpurple,andpolarphosphatidylcholineheadgroupsaregreen.(Solventisnotshown.)BloodandVothPNAS兩October10,2006兩vol.103兩no.41兩15069BIOPHYSICSofSandqisalsodescribed).ThevaluesofSandqinFigure1aregivenfortheinitialstate,ascharacteristicsofthepreassembledlattices.Duringsimulations,theseparametersdeviateslightlyfromtheinitialvaluesduetomembranebendingandN-BARdomaindisplacement,deviationsbeingwithin5forqand0.5nm(rarelyupto1nm)forS.ThehighestmembranecurvaturesobservedinallsimulationscorrespondtoradiiofR=13–20nm,whereasthecurvatureradiusoftheproteinitselfis11nm.Suchstronglybentmembranesareproducedbylatticeswithapproximately10–20dimersper1000nm2,S=3–6nm,andq=0–5(althoughonelatticewithq=20producedhighcurvatureaswell,seeleftpanelinthemiddlerowinFigure1B).Theseobservationsareinqual-itativeagreementwithcryo-EMimagingofmembranetubessculptedbyamphiphysinN-BARdomains(Takeietal.,1999;Peteretal.,2004),showingstriationswithS5–10nm.High-resolutioncryo-EMreconstructionsforF-BARdomains(Frostetal.,2008),whicharelargerandfeatureshallowerintrinsiccurvature,demonstrateS6–8nmandq10–15.Figure2illustratessimulationsofoneN-BARlatticeinbothSBCGandall-atomrepresentations(simulations8BARs-AAand8BARs-CG;seealsoMoviesS1andS2).TheN-BARlatticeinthesesimulationsistheclosesttothe16BARslatticeprobedinSBCGsimulationsofthe64316nm2membranepatch(Figure1B).SimilartoSBCGsimulations(Figure2A),theall-atomsimulationshowsthattheassumedN-BARlatticeremainsstableandinducesglobalmembranecurvature(Figure2B).Theglobalcurvaturecontinuestodevelopthroughoutthe300nssimulation.ThemembranebeneathsevenofeightN-BARdomainsestablishesaclosecontactwiththeconcavesurfaceoftheN-BARdomains.TheeighthN-BARdomaindoesnotyetestablishaclosecontacttothemembranesurface.DifferencesinmembranecontactsbetweendifferentN-BARdomainswithinthelatticearealsoseenintheSBCGsimulations.Membranebendingintheall-atomsimulationdevelopsmoreslowlythanthatinSBCGsimulations:at300ns,theradiusofcurvatureisR=54nm,whereasfortheSBCGsimulationofthesamelattice,thisvalueofRisreachedwithin80ns,andat300nsonefindsintheSBCGcaseR=27±7nm(averagedoverallfivesimulations8BARs-CG).Alikelyreasonforthespeeddiscrepancyisthattheall-atomrepresentationcontainsmanymoredegreesoffreedomthantheSBCGmodel,whichresultsinstrongerfrictionmanifestedthroughslowerbending.Anotherreasonisthatatthebeginningoftheall-atomsimulation,moretimethanintheSBCGsimulationisrequiredfortheN-BARdomainstoformpropercontactswiththemembraneandwitheachother.Despitethedifferenceinbendingspeed,theN-BARdomainlatticesgenerateglobalmembranecurvatureinasimilarfashion.CompleteMembraneTubulationbyLatticesofN-BARDomainsInsimulations43BARs-iand24BARs-i,i=1,2,3,4(seeTable1),weconsiderarelativelybroad(200nm)membranepatch.Thesesimulationsareaimedatobtainingcompletetubulationfromaninitiallyflatmembrane,asshowninFigure3.Allsimulationsi=1,2,3,4areindependentandstartfromthesameinitialstruc-ture,butusedifferentinitialvelocities,whicharerandomlygeneratedforeachsimulationaccordingtoaMaxwelldistribu-tionatT=310K.Anothersourceofdifferencebetweentrajecto-riesi=1,2,3,4arerandomforcesactingoneachCGbeadateachtimestep,theforcesbeingintroducedthroughtheLange-vinthermostat(asreportedinArkhipovetal.[2008]andPhillipsetal.[2005])tomaintainconstanttemperature.AsshowninFigures3Aand3B(seealsoMoviesS3andS4),membranebendingisinitiatedattheedgesofthemembranewithinafewhundrednanoseconds.Thebendingpropagatesthentowardthecenter,roundingtheentiremembraneonatimescaleof30–200ms.Eventuallythemembraneisdrivenintoanear-tubularstate,inwhichtheedgescancomeintocontactandfuse,producingacomplete,stabletube.Figures3Aand3Bshowsnapshotsfromtwosimulationsinwhichtubu-lationiscompletedwithin35and200ms(simulations43BARs-3and24BARs-3;seeFigure4).Theradiiofthetubesformedbyeither43or24N-BARdomainsareapproximatelythesame(i.e.,25nm).Theseradiiaremainlydeterminedbytheoriginallengthofthemembrane,whichwaschosentoallowforaformationofatubewithasizethatistypicallyobservedinexperimentsforamphiphysinN-BARFigure2.All-AtomandSBCGSimulationsofanN-BARDomainLattice(AandB)SnapshotsofthesimulationsinSBCGandall-atomrepresentations,respectively.StructureMembraneTubulationbyLatticesofN-BARDomainsStructure17,882–892,June10,2009ª2009ElsevierLtdAllrightsreserved885 mainly from the geometry of the lipid molecules. The hydrophobic part of the lipid is a
long molecule and hence can have a multitude of orientations. The effect of chain orien-
tations on the conformation and physical properties of membranes have been known for
some time. Of interest is the tilted bilayer phase, where the hydrocarbon chain orients at
an angle with respect to the normal defining the vesicle surface, as shown in Fig. 21(b).
The lipid tilt normally seen in the liquid ordered Lβ
tially selects a direction in the membrane that constitutes an in-plane vector field on its
[228, 229] and Pβ
(cid:48)
(cid:48) phases preferen-
surface [230]. Membranes with tilted lipids have been observed to stabilize highly curved
tubular shapes.
(a)Flat bilayer
(b)Tilted phase of bilayer
(c)Dimeric head Gemini surfac-
tant
FIG. 21. Shown in (a) is a flat bilayer with negatively charged head group and its hydrophobic
tails oriented along the surface normal, in (b) is the tilted phase of a membrane, where the
orientation of its tail molecules subtends an angle with the surface normal, and in (c) is a
Gemini surfactant molecule, with its dimeric heads linked by a rigid spacer.
Surface vector order is also seen in membranes constituted from lipid molecules with a
chiral center, like 1,2-bis-(tricosa-10,12-diynoyl)-sn-glycero-3-phosphocholine [231], which
is a diacteylene-containing lipid. The presence of chirality induces a spontaneous twist
[232, 233] in the self-assembled structures—chiral membranes are also commonly associ-
ated with tubular shapes. The role of lipid tilt and chirality has been previously recog-
nized, and mean field models that include their contributions have predicted the stability
of tubular membranes and helical ribbons [234–237]. Signatures of in-plane order have
also been observed in membrane systems such as POPC + water [238]. It has been ob-
served that the membrane self-assembles into long tubes, and the authors have theorized
that this arises from the inherent anisotropy of the lipid molecules with respect to the
surface normal. In addition to the hydrophobic chains, the polar head of the lipid can
also induce an in-plane order. For instance, Gemini surfactants with dimeric heads group
61
normalTilt normalTilt Rigid spacer are connected by a rigid spacer, as in Fig. 21(c), whose orientations constitute an in-plane
order. The morphologies of the vesicles with random spacer orientations are different from
those with spacers in the ordered phase. In contrast to other pure lipid systems, bilayer
vesicles of Gemini surfactants have been observed to stabilize tubular structures [239].
At this point, it should be noted that the mentioned intrinsic mechanisms can only
promote weakly curved regions and hence their role in stabilizing highly curved organelles,
like the golgi, and endoplasmic reticulum or in the formation of endocytic vesicles is limited
[189]; the latter would require specialized proteins to perform the job as discussed above.
Different modes by which proteins modulate the curvature of membranes have also
been discussed by McMahon et al.
[216]. On the experimental front, direct measurements
of bending-mediated force transduction and molecular organization in lipid membranes
based on interferometry and fluorescence measurements have been reviewed [240, 241].
The dynamics of molecules and the mosaic organization of the plasma membrane and
their implications in cellular physiology, primarily focused on studies involving fluores-
cent labeling and imaging, have also been recently reviewed [242]. Several experimental
studies have been carried out to investigate curvature generation and sensing. Sorre and
coworkers
[243, 244] conducted experimental investigation of the sorting of lipids on a
lipid membrane tube (tether) drawn from a giant unilamellar vesicle (GUV) using an
optical trap. Curvature sorting of lipids and its influence on the bending stiffness of the
bilayer membrane was studied by Tian et al
[245, 246]. Dynamic sorting of lipids and
proteins has been studied by Heinrich and coworkers [247]. These authors observed that
nucleation of disordered membrane domains occurs at the junction between the tether
and giant unilamellar vesicles. Several theoretical and experimental studies have helped
shed light on the curvature-mediated sorting phenomenon [79, 193, 247–249].
In cell membranes, protein-induced radius of curvature ranges from few a nanometers to
a few tens of nanometers depending on the protein and lipid composition of the membrane.
For example N-BAR domains stabilize tubular membranes with radius in the range 25 −
32 nm [250], whereas dynamin-induced tubes have a radius of 25 nm [251]. In vitro
experiments have reported epsin-induced tubulation of liposomes with a tubule radius
of 10nm [11].
In the following sections, we will discuss how the triangulated surface
method and particle-based EM2 method can be readily extended to include the effect of
curvature-inducing proteins. The effect of protein-induced membrane remodeling can be
62
captured in field theoretic models, based on eqn. (9), by substituting the protein with a
suitable spontaneous curvature field whose magnitude and extent matches the deformation
profile of the membrane in the vicinity of the protein. In this section, we will describe
two curvature field based models that can be used to study the effect of proteins on the
properties of membranes.
B.
Isotropic curvature models
Most theoretical studies on protein binding focus on protein adsorption to planar lipid
bilayers (reviewed in [252]); these studies were mainly concerned with planar mem-
branes, and curvature effects were not discussed. Reynwar et al.
[166] performed coarse-
grained molecular dynamics simulations, which show that once adsorbed onto lipid bi-
layers, curvature-inducing proteins experience effective curvature-mediated attractive in-
teractions. Jiang and Powers [253] investigated lipid sorting induced by curvature for a
binary lipid mixture using a phase-field model. Das and co-workers have investigated the
effect of protein sorting on tubular membranes using theoretical techniques
[254, 255].
Using Monte Carlo simulations, Sunil Kumar and coworkers [101, 256–259] have studied
the effects protein-induced isotropic and anisotropic curvatures have on the shape of vesi-
cles. Using the Monte Carlo method, Liu et al. and Ramanan et al. have investigated
spatial segregation, curvature sensing, and vesiculation in bilayers with curvature-inducing
proteins [260, 261].
Models for protein diffusion in ruffled surfaces
[262] and the simultaneous diffusion
of protein and membrane dynamics [263–266] have also been reported. In such models,
there is only a one-way coupling between the membrane dynamics and the protein dy-
namics—i.e., a change in membrane morphology affects the diffusion of the proteins and
not the reverse. These studies have been extended to simultaneous protein diffusion and
membrane motion models to treat the case of curvature-inducing proteins diffusing on
the membrane [92, 267, 268]. The new aspect introduced in these latter models is the
two-way coupling between the protein and membrane motion. In this case, the membrane
morphology not only influences the protein diffusion by presenting a curvilinear manifold,
but also presents an energy landscape for protein diffusion. The protein diffusion in turn
affects membrane dynamics because the spatial location of the proteins determine the
63
intrinsic curvature functions and hence the elastic energy of the membrane [267]. This
methodology has been utilized in exploring the equilibrium behavior of bilayer membranes
under the influence of cooperative effects induced by the diffusion of curvature-inducing
proteins [268].
Isotropic curvature models have been utilized to study the energetics of curvature-
inducing protein interactions on a membrane [269–279]. Experimental methods to probe
such interactions have also been discussed [280]. By including the effect of protein-
membrane interaction as a curvature field, the assumption is that the equilibrium be-
havior of the system is dominated by the membrane-mediated protein-protein interaction
dictated by the strength and range of the curvature field and that small-length-scale inter-
actions (i.e. at the atomic level) are smoothed out. Justification for this assumption has
recently been presented by directly parameterizing such a curvature field from molecular
dynamics simulations [281]. In [274], Aranda-Espinoza et al. employed a combination
of integral equation theory to describe the spatial distribution of the membrane-bound
proteins and the linearized elastic free-energy model and reported that the interaction (in
the absence of thermal undulations) between two membrane-bound curvature-inducing
proteins is dominated by a repulsive interaction. Consistent with these published re-
ports, the calculated binding energy (again without thermal undulations) between two
membrane-bound proteins interacting through the curvature fields show dominant repul-
sive interactions governed by the range of the curvature field [268]. Thus, purely based
on energetic grounds, the previous analyses have suggested that membrane-deformation-
mediated energies tend to be repulsive and should prevent, rather than promote, the
formation of protein dimers or clusters.
Kozlov has discussed how the effect of fluctuations can change the repulsive nature of
the interactions [282]. The author’s discussion is based on the premise that any mem-
brane protein locally restrains thermal undulations of the lipid bilayer. Such undulations
are favored entropically, and so this increases the overall free energy of the bilayer. Neigh-
boring proteins collaborate in restricting the membrane undulations and reduce the total
free-energy costs, yielding an effective (membrane-mediated) protein-protein attraction.
Indeed, for the linearized free-energy model, computing the second variation of energy
(note that at equilibrium, the first variation is zero, whereas the second variation governs
the stiffness of the system against fluctuations), we see that the presence of a protein (or
64
equivalently a curvature inducing function) leads to a localized suppression of membrane
fluctuations [268, 283]. As will be discussed later in section VI, this calculation has been
further verified by using a free-energy method to compute the change in Helmholtz free
energy upon the introduction of a curvature field [283]. This provides for the possibility
of an entropically mediated protein-protein attraction. The outcome of the interplay be-
tween the attractive entropic forces and the repulsive energetic forces is context specific
because both have the same dependence on the protein-protein distance and their abso-
lute values differ only by coefficients with similar values. This has been demonstrated
by examining the protein-protein pair correlation (spatial and bond-orientational) and
through the effect on membrane morphology [268].
Indeed, the model predicts that
the cooperative effects of membrane-mediated interactions between multiple proteins can
drive different morphological transitions in membranes
[166, 268, 274, 282]. This no-
tion of cooperativity is also consistent with the analysis of Kim et al.
[269], who have
shown using an energetic analysis that in the zero-temperature limit, clusters of larger
than five membrane-bound curvature-inducing proteins can be arranged in energetically
stable configurations. It is also worth mentioning for completeness that Chou et al.
[269]
have extended the energetic analysis to membrane-bound proteins that have a noncir-
cular cross-sectional shape and to local membrane deformations that are saddle shaped
(negative Gaussian curvature) and have shown that, in such cases, the interactions can
be attractive even without considering fluctuations. Hence, based on the afore-mentioned
simulations and analysis of the continuum models, it is hypothesized that attractive in-
teractions between curvature-inducing proteins can result from entropy of membrane un-
dulations
[261, 282–284] (the same phenomenon was investigated using particle-based
simulations by Reynwar et al.
[166]). Using related continuum methods, preserving bi-
directional coupling of protein-induced curvature migration and membrane undulations,
the role of adhesive forces as well as anisotropy of curvature fields on membrane-mediated
protein interactions have also been reported [260, 285–288]. The isotropic curvactants,
described in section IV A, can be represented by the spontaneous curvature field C0,
defined in the continuum model described by eqn. (9).
The role of the spontaneous curvature C0 in determining the conformational and ther-
modynamic properties of the membrane has also been studied in a variety of contexts—curvature-
induced instabilities [289, 290], intramembrane- and crystalline-domain-induced budding
65
in multi-component lipid membranes [79, 257, 291–294], effects of lipid packing [295],
and effects of trans-bilayer sugar asymmetry [296]; see chapter by Gompper and Kroll
[66] for further details.
C. Membranes with in-plane order: Model for anisotropic curvature-inducing
proteins
The organelles of a biological cell have membranes with highly curved edges and tubes,
as seen in the endoplasmic reticulum, the golgi, and the inner membrane of mitochondria.
Tubulation has also been observed, in vitro, in self-assembled systems of pure lipids [297].
It has been shown that macromolecules, which constitute and decorate the membrane
surface, strongly influence the morphology of membranes. For instance, proteins from
the dynamin superfamily are known to pull out membrane tubes while oligomerizing
themselves into a helical coat along the tube [194]. The BAR domain containing proteins
in general can induce a wide spectrum of membrane shapes, ranging from protrusions to
invaginations, depending on the geometry and interaction strength of the BAR domain [3,
189, 191].
The models for membranes with isotropic curvature fields (such as those discussed in
section IV B) cannot explain the emergence and stability of such highly curved structures.
We will show in the remainder of this section (and also in section V) that a source for
anisotropic bending energy will be the minimal requirement to explain the emergence and
stability of tubular shapes, which could arise from an in-plane orientational field on the
membrane [298, 299]. In general, the different types of intrinsic and extrinsic in-plane
order, discussed in section IV A, can be represented as a p-atic in-plane field. This in-plane
order, tangential to the surface of the membrane, can capture curvature modulations in the
membrane arising from anisotropic membrane inclusions. The p-atic field has a rotation
symmetry of 2π/p ; for instance, a nematic field (p = 2) has a π rotational symmetry,
and a hexatic field (p = 6) has a π/3 rotational symmetry. In this section, we will only
be dealing with nematic in-plane order since this closely resembles the curvature profile
induced by a protein interacting with a membrane. As we will discuss below, a nematic
field is sufficient to demonstrate the applicability of this model to explain a variety of
vesicular morphologies observed in in vitro experiments involving membrane remodeling
66
proteins like Dynamin [300, 301], Epsin [302], BAR domains [303], and Exo70 [281],
which are all believed to induce an anisotropic curvature field on the membrane. The
anisotropic contributions to the energy can be accommodated by extending the isotropic
elastic energy with additional anisotropic terms, obeying all relevant symmetries, as shown
below.
1. Membrane with in-plane field
Ramakrishnan et al.
[101, 305] considered the vertices of the triangulated surface
(introduced in section III F) by additionally decorating them with a nematic in-plane
field m, of unit length, defined on the tangent plane at each vertex. Fig. 22(a) shows
the neighbourhood of a vertex with an in-plane field and fig. 22(b) shows a vesicular
membrane with in-plane order defined at all vertices; the surface coverage can be set to
the desired concentration. The field m is defined on the tangent plane of vertex v (the
Darboux frame to be precise, see Appendix C), as m(v) = a t1(v) + b t2(v), and this
representation is illustrated in Fig. 22(a). If m subtends an angle ϕ with the maximum
principal direction t1, then a = cos ϕ and b = sin ϕ.
(a)In-plane field at vertex v
(b)Membrane surface with decorated in-plane field.
FIG. 22. (a) An in-plane polar field, m(v) = a t1 + b t2, defined on the tangent plane of vertex v.
(b) A discretized membrane, of spherical topology, decorated with an in-plane nematic director
field.
The self-interaction between the in-plane field is given by Hp−atic. The form of this
interaction potential differs with the value of p. For example, the self-interaction between
the in-plane field with polar symmetry (p = 1) can be represented by the standard XY -like
67
planetangentt1t2ϕmv interactions [68],
(cid:88)
(cid:104)v,v
(cid:48)(cid:105)
H1−atic = −J1
cos(θvv
(cid:48) ),
(52)
For a nematic field, the Lebwohl-Lasher interaction potential
[306],
H2−atic = −
J2
2
(cid:88)
(cid:104)v,v
(cid:48)(cid:105)
(cid:8)3 cos2(θvv
(cid:48) ) − 1(cid:9) ,
(53)
has been used by Ramakrishnan et al.
[101, 305] to model the self-interaction of the
nematic in-plane field defined on the vertices of the membrane. J1 and J2 are the in-
teraction strengths of the polar and nematic fields, respectively. Note that θvv
curvature-dependent angle between the orientations of the in-plane field at vertices v and
is the
(cid:48)
(cid:48)
v
, computed using the parallel transport technique defined in appendix E. This depen-
dence implicitly couples the in-plane field to the membrane, which in turn considerably
affects the morphology of the fluid membrane.
2. Monte Carlo procedure for nematic membranes
As described by Ramakrishnan et al. [101, 305], the Monte Carlo techniques for a fluid
random surface can be extended to a field-decorated membrane when the contributions
arising from field orientations are accounted for in the phase space integral. The total
partition function of the membrane with the in-plane field has the form,
Z(N, κ, ∆p, Jp) =
(cid:90)
(cid:88)
N(cid:89)
{T }
v=1
1
N !
(cid:90)
dϕ(v)
68
d(cid:126)x(v) exp{−βHtot} ,
(54)
FIG. 23. Monte Carlo moves for a membrane with an in-plane field.
where Htot = Hsur + H2−atic + VSA, and the integral is performed over all possible in-
plane orientations, in addition to the phase space of the random surface in eqn. (41).
The membrane explores the accessible states in its phase space, now represented by η =
[{ (cid:126)X},{T },{ϕ}], through the set of Monte Carlo moves shown in Fig. 23. A vertex
move and a link flip, shown, respectively, in Fig. 23(a) and (b), are exactly performed as
described in section III F, except that in these moves, the field orientation {ϕ} remains
unchanged before and after the move. The third class of move aims at changing the
orientations of the in-plane nematic field, {ϕ} → {ϕ
}. As shown in Fig. 23(c) the in-
(v).
plane field m(v), at a randomly chosen vertex v, is updated to a new orientation m
(cid:48)
(cid:48)
The move is accepted using the Metropolis scheme with a probability given by,
(cid:48)
(cid:48)
Pacc({ϕ} → {ϕ
with η = [{ (cid:126)X},{T },{ϕ}] and η
is ω(η → η
nematic field is dependent on the membrane curvature, which in turn implicitly couples
}]. Attempt probability for the move
→ η) = (2∆θN )−1. As noted before, the self-interaction between the
−β∆H (η → η
1, exp
}) = min
= [{ (cid:126)X},{T },{ϕ
) = ω(η
(55)
,
(cid:48)
(cid:48)
the membrane geometry to the texture of the nematic field. In appendix F, we describe
how the equilibrium shape of an otherwise spherical membrane is changed in the presence
of a polar and nematic field. The explicit interaction of the protein with the membrane
was introduced by Ramakrishnan et al.
[101, 305] through an additional energy term
69
(cid:110)
(cid:16)
(cid:48)
(cid:17)(cid:111)
(cid:48)
)
(a)Vertex move(c) In−plane field flip(b) Link Flipϕϕϕϕϕϕ′t1t1t1t1t1t1mm{~X}→{~X′}{T}→{T′}{ϕ}→{ϕ′} (Hanis) that depends on the directional spontaneous curvature induced by the protein
and the curvature of the membrane surface; this form of Hanis is described in the section
below.
D. Anisotropic bending energy
The elastic behavior of the membrane becomes anisotropic when the in-plane field m
couples to its curvature tensor K. As comprehensively discussed in [66, 234, 235], such
an interaction is described by an energy functional containing all possible gauge-invariant
scalars constructed out of m and K. The full form of the energy will have contribu-
tions from terms like ( mK m), ( m⊥K m⊥), ( mK m⊥), ( m⊥K m), ( mKK m), ( m⊥KK m⊥),
( mK m)2, ( m⊥K m⊥)2, ( mK m⊥)2, and ( m⊥K m)2; in addition, one may also consider
gradients of m.
m⊥ is the in-plane field orientation perpendicular to m, defined as
ma⊥ = gacγcbmb, with γab being the antisymmetric tensor. For computational purposes,
following the definition of m in section IV C 1, we approximate the perpendicular nematic
orientation as m⊥(v) = bt1(v) − at2(v), and it can be verified that m · m⊥ = 0.
To keep the complexity of the problem tractable, we choose the in-plane field to be an
achiral nematic director of unit magnitude. The terms of the interaction energy are chosen
such that the membrane is invariant under m → − m and K → −K. To lowest order in
m, the explicit interaction of the nematic field with the membrane has the form [307]:
(cid:90)
Hanis =
1
2
(cid:124)
dS κ(cid:107)
S
(cid:104)
(cid:123)(cid:122)
maKabmb − C
(cid:107)
0
(cid:105)2
(cid:125)
(cid:107)
anis
H
(cid:90)
S
+
1
2
(cid:124)
dS κ⊥(cid:2)ma⊥Kabmb⊥ − C
⊥
0
(cid:123)(cid:122)
H ⊥
anis
(cid:3)2
(cid:125)
.
(56)
Here, the Einstein summation convention over repeated indices is implied. It has to be
noted that higher order terms like ( mKK m), which possess all the symmetries listed
above, have been neglected, and contributions from second order gradients are zero due
to the constraint mimi = 1 [307].
The directional rigidities κ(cid:107) ≥ 0 and κ⊥ ≥ 0 are, respectively, the additional stiffness
along and perpendicular to the nematic field orientation m. If the nematic field represents
the curvature due to BAR domain proteins, κ(cid:107) and κ⊥ can be thought of as a measure
of the modified rigidities due to the protein structure and the molecular-level interaction
strength with the lipids on the bilayer membrane. Similarly, the ensuing curvatures
70
resulting from the interaction of the protein with the membrane are approximated by C
and C⊥
on the membrane in its vicinity. Their values can be positive, negative, or zero, depending
0 , which are the directional spontaneous curvatures imposed by the in-plane field
(cid:107)
0
on the type of proteins they represent. For example, C
domain, and C
(cid:107)
0 < 0 when it contains an I-BAR domain [202].
(cid:107)
0 > 0 for a protein with a F-BAR
On a triangulated surface, the anisotropic bending energy can be computed as,
N(cid:88)
v=1
1
2
Av
Hanis =
(cid:26)
(cid:104)
(cid:105)2
+ κ⊥(cid:2)c
⊥
(cid:3)2
⊥
0
(v) − C
(cid:27)
κ(cid:107)
(cid:107)
c
(cid:107)
0
(v) − C
.
(57)
Here, c(cid:107)(v) = c1 cos2 ϕ(v)+c2 sin2 ϕ(v) and c⊥(v) = c2 cos2 ϕ(v)+c1 sin2 ϕ(v) are the direc-
tional curvatures on the membrane parallel and perpendicular to the nematic orientation
calculated using Euler’s theorem [50].
E. Properties of nematic membranes
The equilibrium shapes of the nematic membrane, with total energy Htot = Hsur +
H2−atic + Hanis, were determined by Ramakrishnan et al.
[259] using Monte Carlo
techniques for surfaces with in-plane field, defined in section IV C 2. As discussed in
appendix F, the nematic in-plane field can affect membrane shapes only when they are in
the nematic phase; this can be achieved by setting J2 = 3kBT , at which the 3-dimensional
system is above the critical value for the isotropic-nematic transition. In this case, the
thermal fluctuations in the orientational field do not change the qualitative behavior,
and the nematic texture only allows the presence of minimum number of defects; see
appendix F for a detailed discussion on defects in field texture and surface topology. Any
additional defect generation is due to the large deformations of the underlying membrane.
The predictions of the mean field model are reproduced when the bending stiffness κ is
set to large values. The thermally excited shapes of a membrane with a 1-dimensional
nematic field (κ⊥ = 0) have been studied by Ramakrishnan et al.
[259], and we discuss
this below.
71
1. Conformational phase diagram for κ⊥ = 0
The equilibrium shapes of the nematic membrane as a function of C
(cid:107)
0 are given in Fig.
24 for κ(cid:107) = 5kBT [259].
FIG. 24. Classes of shapes of a nematic membrane for C
(cid:107)
0 = 0.0 (a), −0.3 (b), −0.4 (c), −0.6
(d), 0.2 (e), 0.4 (f ), 0.5 (g) and 0.6 (h), with κ = 10kBT , κ(cid:107) = 5kBT , κ⊥ = 0 and J2 = 3kBT
(Reprinted figure with permission from N. Ramakrishnan, John H. Ipsen, P. B. Sunil Kumar, Soft
Matter, 8(11), 3058 and 2012. Copyright (2012) by the Royal Society of Chemistry).
The directional deformation induced by the in-plane nematic field augments the equi-
librium shapes of the model membrane. Non-zero values of the directional spontaneous
(cid:107)
0 ) stabilize many shapes ubiquitous in the biological cell: the shapes include
curvature (C
tubes (−0.3 to 0.3), corkscrews (0.35 to 0.5), branched shapes (> 0.5), discs (−0.35 to
−0.55), and inner tubes or caveola like shapes (< −0.55). Recall that a fluid surface
containing a nematic field with no anisotropic interactions deforms in a manner such that
the four +1/2 disclinations are localized to the vertices of a tetrahedron. The presence
72
of anisotropic interactions alters the free-energy landscape and allows for the presence
of other defect textures. The prominent features distinguishing a class of shapes from
another, for a membrane of constant surface area, are as follows:
1. Tubes are cylindrical structures with exactly four +1/2 disclinations, with the tube
(cid:107)
0 , κ(cid:107), and κ, as in Fig. 24(a,b,e). Two individual
disclinations pair up at the end cap of the tube, resulting in two defect pairs each
radius dependent on the value of C
of net charge +1. The average nematic orientation also responds to change in the
(cid:107)
0 ≤ 0 and (cid:104)ϕ(cid:105) ∼ 0 for
directional spontaneous curvature, with (cid:104)ϕ(cid:105) ∼ π/2 for C
(cid:107)
C
0 > 0.
2. The canal surface, a snapshot of which is shown in Fig. 24(f), closely resembles a
tube spiralling around its long axis.
3. Branched membranes are seen for large C
(cid:107)
0 , where multiple tubes originate from a
common region called an intersection or a neck. Due to these large deformations,
the number of +1/2 disclinations in a branched membrane exceeds four but still
on the branched shape shown in Fig.
preserves the Euler characteristic of the surface. For example, the nematic field
24(g), contains six +1/2 and two −1/2
disclinations. A pair of negative disclinations, each of charge Qneck, are localized to
the region of intersection. This neck is finite sized (fig. 24(g)) with Qneck = −1/2
and has exactly three tubular protrusions. On the other hand, the neck is said to be
narrow (Fig. 24(h)) when Qneck = −1 and bridges two membrane tubes. A similar
organization has also been seen in the simulation of N-BAR domains on tubulated
membranes [122].
4. Discs: The highly curved rim along with the low curvature planar regions charac-
terize a disc. The four +1/2 defects are far separated from each other and localize
to regions of negative principal curvatures as in Fig. 24(c).
5. Inner tubes: In the context of a biological cell, inner tubes are the invaginations in
the plasma membrane into the cytoplasmic side. These caveola like shapes, named
after the membrane deformations caused by the caveolin proteins, are common in
neuron cells and T-tubules. Inner tubes, shown in Fig.
24(d) and Fig.
25, are
analogous to branched membranes in terms of the additional topological defects
73
they posses. Unlike in the latter, the lengths of the tubes are dependent on the
volume and self avoidance constraints of the membrane.
FIG. 25. Shown are two different aspects of the inner tubes. On the left, the transparent view
reveals the tubes grown into the interior of the membrane. The nematic field orientation on
the exterior surface and also on the inner tubes are shown in the right panel; for C
(cid:107)
0 < 0, the
nematic field always orients along the minimal principal curvature direction, with this direction
corresponding to c2 = 0 on the outer surface and c2 < 0 on the inner tubes.
It is helpful to consider the predominant sign of the principal curvatures, c1 and c2, for
these shapes. The possible values of c1 and c2, for ideal geometries that belong to the
above mentioned classes, are listed in Table I.
All shapes stabilized by scanning the value of the directional bending stiffness, κ(cid:107), fall
value of C
into one of the five classes listed above. It should also be observed that the characteristic
(cid:107)
0 , for which different classes of shapes are stabilized, is also a function of κ(cid:107). The
precise state boundaries delineating the different states can be identified by computing the
relative free energies, which we discuss in section VI. The distribution of directional and
principal curvatures as well as the effect of changing the bending rigidity on the emergent
morphological states of the membrane are discussed further in Ramakrishnan et. al.
[259].
F. Pairing of defects: the role of Gaussian curvature
Charge of a topological defect is analogous to an electric charge. Two like charged
defects repel each other with an interaction energy logarithmically dependent on their
separation, Hdef ∝ − ln(Rdef); this has been shown to be true for defects on both pla-
74
c2=0c2<0 Class of shape
c1
c2
1. Tubular
2. Canal surface
3. Branched
4. Disc like
a. On the rim
b. Region enclosed by rim
5. Inner tubes
a. Exterior side
b. Interior side
> 0
> 0
> 0
> 0
= 0
> 0
= 0
= 0
= 0
= 0
= 0
= 0
= 0
< 0
TABLE I. Predominant values of the principal curvatures for various classes of shapes.
nar
[308] and curved surfaces
[309].
In order to understand the spatial organization
of topological defects, Ramakrishnan et al.
[259] computed the geodesic distance (ξ)
between these defects on the triangulated surface, using the Dijkstra algorithm [310].
75
FIG. 26. P (ξ) is the distribution of the geodesic distance between a pair of defect cores, for
shapes corresponding to different values of C
(cid:107)
0 and κ(cid:107) (Reprinted figure with permission from N.
Ramakrishnan, John H. Ipsen, P. B. Sunil Kumar, Soft Matter, 8(11), 3058 and 2012. Copyright
(2012) by the Royal Society of Chemistry).
The distribution of ξ on a nematic membrane, for various set of parameters, is shown
in Fig. 26. The analysis has been performed on tubular membrane shapes with exactly
four +1/2 disclinations, but the results hold for other shapes too. When κ(cid:107) (cid:54)= 0, two
+1/2 disclinations come close to each other on a region of positive Gaussian curvature,
resulting in defect pairs, each of strength +1, at either ends of the tube. In Fig. 26, this
pairing is reflected as two distinct peaks in P (ξ) for κ(cid:107) (cid:54)= 0, which is expressly different
from the broad distribution seen when κ(cid:107) = 0. The peak at small ξ represents the geodesic
connecting defects within a pair and shifts to the left with increasing C
(cid:107)
0 . The curvature
dependence of defect localization is in good agreement with theoretical predictions of this
phenomenon [311, 312].
Coupling of nematic defects to the curvature of the membrane and the resulting dy-
namics play an important role in the tubulation mechanisms of nematic membranes. At
temperatures where the nematic order is susceptible to thermal fluctuations, proliferation
and annihilation of additional defect pairs control the resulting tubular morphologies. The
76
025507510012500.010.020.030.040.05separation between defects pairs.Defects at equal distancereplacemenGeodesicdistance,ξP(ξ)κ!=0;C!0=0κ!=5;C!0=0.2κ!=5;C!0=0.3κ!=5;C!0=0.4κ!=5;C!0=0 different mechanisms leading to stabilization of tubular and branched membranes have
been discussed in [259]. Further, nematic membranes display long-wavelength thermal
undulations in the form of canal-surface-like structures that are intermediate to the tubu-
lar and branched morphologies [259]. The qualitative behavior of a nematic membrane
with both its anisotropic stiffnesses being non-zero (κ(cid:107) (cid:54)= 0 and κ⊥ (cid:54)= 0) is the same as
described above, and hence is not discussed here (see [313] for details).
G.
Implication of defect structures in biological membranes
Even though the nematic ordering and the presence of defect structures in biological
membranes has not been directly verified experimentally, Ramakrishnan et al.
[258] ex-
tended the anisotropic curvature model to investigate the behavior of partially decorated,
single-, and multi-component nematic membranes. In the case of a partially decorated
membrane, it has been shown that the presence of an anisotropic ordering interaction
promotes the aggregation of proteins with similar curvature properties into spatial do-
mains [258]. Such a response leads to many interesting behaviors and also enriches the
conformational phase space of the vesicular membrane with the protein field.
FIG. 27. Equilibrium shapes of a nematic membrane with varying concentration of the nematic
field (NF = 0.1N−0.7N ), for a prescribed value of the directional curvature, C
from Biophysical Journal, 104(5), N. Ramakrishnan, P. B. Sunil Kumar, John H. Ipsen, Membrane-
(cid:107)
0 = 0.5 (Reprinted
Mediated Aggregation of Curvature-Inducing Nematogens and Membrane Tubulation, 1018–1028,
Copyright (2013), with permission from Elsevier).
77
With NF being the relative concentration of the nematic field, Fig.
27 shows the
equilibrium shapes of a nematic membrane with NF in the range 0.1N − 0.7N , βκ =
(cid:107)
10, βκ(cid:107) = 5, and C
0 = 0.5. At low concentrations, the field localizes to the rim of a
disc that becomes unstable at large concentrations (NF > 0.6N ), resulting in membrane
tubulation. The equilibrium shapes of a partly decorated membrane as a function of
the field concentration show striking similarities to the arrangement of reticulons and
Dp1/Yop1 proteins in the peripheral ER [218]; the sheet like pattern (Fig.
27(c)) of
the nematic membrane and the coexisting sheet and tubes (Fig. 27(g)) have both been
observed in experiments on the ER. Green fluorescent regions in Fig.
28(a), reprinted
from [218], denote the spatial locations of the reticulon proteins on a sheet like ER
and confirms that these proteins are confined to the rim of the sheet. Furthermore,
at high concentrations of the nematic field, the observed conformations of the nematic
membrane and the orientational pattern of the in-plane field are in good agreement with
the predictions of reference [218], shown in Fig. 28(b).
Localization of calreticulons to ER edges.
α-Calreticulon proteins have been colored green.
An illustration of reticulon proteins on ER sheets.
(a)
(b)
FIG. 28.
(a) Immunofluorescence micrographs of a ER sheet with α - Calreticulon pro-
teins (marked green with fluorescent markers) and (b) an illustration showing the pattern of
Dp1/Yop1p and reticulon proteins on an ER sheet and tubules (Reprinted from Cell , 143 /5, Y.
Shibata et. al., Mechanisms Determining the Morphology of the Peripheral ER, 774–788, Copyright
(2010), with permission from Elsevier).
In this context, the conditions for tubulation; estimates for the tube radius, orientation
angle, and their fluctuations; and application of the two-component nematic model to the
78
sheetformation.Indeed,whenthereticulonRtn4bwasoverex-pressedinCOS7cells,peripheralERsheetsbecamediminishedwithincreasingexpressionlevels(Figures5Eand5F;quantifica-tioninFigures5Gand5H).Concomitantwiththedecreaseinsheetstructures,thenormaltubularnetworkwasgraduallyre-placedwithlong,unbranchedtubules(quantificationinFigure5I).Figure2.MembraneProteinsEnrichedinERSheets(A)TheendogenouslocalizationofthemembraneproteinClimp63iscomparedwiththatoftheluminalERproteincalreticulininCOS7cells,usingindirectimmunofluorescencewithspecificantibodies.Thefar-rightpanelshowsamergedimage.JunctionsbetweenperipheralERsheetsandtubulesarehighlightedinthemagnifiedviewoftheboxedarea(inset).Scalebar,10mm.(B)Asin(A)butcomparingthelocalizationofkinectin(KTN)andcalreticulin.(C)Asin(A)butcomparingthelocalizationofp180andcalreticulin.(D)Climp63,p180,andkinectinweredepletedinCOS7cellsbyRNAi(C/P/KsiRNA),andClimp63,TRAPa,andcalreticulinwerevisualizedusingindirectimmunofluorescencewithspecificantibodies.Scalebar,10mm.(E)Asin(D)butwithcellstransfectedwithcontrolsiRNAoligonucleotides.SeealsoFigureS2andFigureS5.778Cell143,774–788,November24,2010ª2010ElsevierInc.transferasecomplex,andp180(GorlichandRapoport,1993).Theseproteinsstayboundtoribosomesupondetergentsolu-bilizationofroughERmembranes,buttheycanbereleasedfromtheribosomesbypuromycin/highsalttreatment.Climp63andkinectinarenotboundtodetergent-solubilizedtranslocons(datanotshown),sohowtheyarerecruitedremainstobeclarified.OurresultsindicatethatERsheetscorrespondtoroughERandtubulestosmoothER.Weproposethattheassemblyoftranslatingmembrane-boundribosomesintopolysomesconcentratestheassociatedmembrane-proteins,includingClimp63,p180,andkinectin.Theirconcentrationmightfacilitatetheirhigher-orderoligomerization,whichmayberequiredfortheirexclusionfromhigh-curvatureareasandthusfortheirFigure7.ModelingoftheEffectofCurvature-StabilizingandSheet-PromotingProteinsonERMorphology(A)ThereticulonsandDP1/Yop1p(yellowarcs)areassumedtolocalizeexclusivelytotubulesandsheetedges,generatingandstabilizingthesehigh-curvaturemembranes.Stabilizationofsheetedgesenablestheupperandlowermembranesofthesheettoadoptplanarshapes.(B)TopviewofmembraneshapescomputedbythetheoreticalmodelforincreasingGvalues.Thecomputationwasperformedforatotalmembraneareacor-respondingto1mmradiusoftheinitialdisc-likeshape,a15nmcross-sectionradiusofthetubulesandedges,anda40nmoptimaldistancebetweenthearc-likeproteinsattheedge(seeSupplementalInformation).ChangeofGfrom1to2.1(bluetored)correspondstoincreasingthenumberofcurvature-stabilizingproteinsNcfrom140to290.(C)Gvaluesandmembraneshapeswerecalculatedfordifferentnumbersofcurvature-stabilizingandsheet-promotingproteins,NcandNs.Thecolorscorre-spondtothemembraneshapesshowninFigure7B.ThecoloredlinesonthebottomplaneofthediagramrepresenttherelationshipbetweenNcandNsforagivenshapeofthesystem.(D)GvaluesandmembraneshapeswerecomputedfordifferentNsvaluesatNc=290.TheshapesrefertoNs=0,500,and1000.Cell143,774–788,November24,2010ª2010ElsevierInc.785 problem of tube constriction by ATP activated dynamin have been discussed in detail in
[258].
H. Mapping the length scales
How do the model parameters like the spontaneous curvature and the density of
nematic inclusions compare with experiments ? The size of the tubes remodeled by
curvature-inducing protein can range from a few nanometers to a few hundred nanome-
ters. Hence the length scale that can be associated with the triangulated model is not
generic but is specific only to a class of proteins. For comparison, one can use the well-
studied system of dynamin-driven tubulation of membranes. Dynamin proteins tubulate
a spherical liposome, due to the curvature-inducing properties of their γ-GTPase and PH
domains, and the resulting shape resembles the branched membrane discussed in Fig. 24.
Electron microscopy studies have shown that the radius of these tubes to be in the range
of 10 − 40 nm [194, 314–316], depending on the state and concentration of dynamin
proteins. These proteins are observed to form a helical coat, each ring of the helix being
formed from approximately 20 units, and the pitch of the helix, which is the separation
between successive rings, has been observed to be ∼15nm.
membrane shown in Fig. 24 for C
These experimentally observed network of tubes can be compared with the branched
(cid:107)
0 ∼ 1.0. In this limit, five nematics (five membrane
vertices) make up the circumference of the tube, which, when compared with the experi-
mental values, yield the length of the tether to be ≈ 25nm. Hence, C
a curvature of ≈ (25nm)−1, in real units, which is not far from the suggested value of the
(cid:107)
0 = 1.0 translates to
intrinsic curvature of dynamin. The five in-plane director fields when mapped to the 20
dynamin units, making up a ring, leads us to the estimate that the in-plane nematic field
at each vertex is the average orientation of approximately 4 dynamin proteins. A map
comparing the coarse grained lengths, used in the simulation, with the biological scales
is given in Fig. 29. As mentioned earlier this mapping changes with the choice of the
curvactant protein.
79
FIG. 29. A comparison of the coarse grained lengths to biological length scales for Dynamin.
V. APPLICATIONS OF THE PARTICLE-BASED EM2 MODEL IN THE STUDY
OF MORPHOLOGICAL TRANSITIONS IN MEMBRANES MEDIATED BY
PROTEIN FIELDS
The EM2 model can be employed as a competing approach to investigate the curvature-
mediated morphological transitions induced by isotropic as well as anisotropic curvature-
inducing proteins. In particular, it has been used to understand the membrane remodeling
behavior of N-BAR-domain and F-BAR-domain containing proteins [119, 317]. The EM2
model
[114], introduced in section III G, is a coarse grained simulation technique based
on the elastic energy description of a lipid bilayer (eqn.
(8)). In this model, the field
theoretic continuum membrane is discretized into a set of quasiparticles that interact
with each other through a bending potential, which is derived from eqn.
(9). In spite
of being a particle-based model, the EM2 approach allows one to access length and time
scales corresponding to the macroscopic limit (µm, ms) and hence can be used to study
the dynamics of protein-induced self-assembly of membranes. Formulation of this method
involves a suite of techniques adopted from non-equilibrium molecular dynamics (NEMD),
smoothed particle hydrodynamics (SPH) and smooth-particle applied mechanics (SPAM)
and recasting the continuum elastic free energy in terms of pair potentials. Being able
to express the curvature energy in terms of a binary interaction potential gives one a
template for a general binary interaction potential.
A. Discretization to a particle-based model
In order to model the membrane interactions in terms of pair potentials, which repro-
duces the key thermodynamic properties, the continuum membrane should be discretized
80
C0!1(50nm)(25nm)!1!1(12nm)(6nm)!1(3nm)!1340.502Biological length scales(Directional spontaneous curvature)1 (cid:88)
j
into a set of quasiparticles. The local number density of the quasiparticles (ρ(r)) can be
expressed using a regularized delta function (see Appendix G for details) as,
ρ(r) =
δh(r − rj).
(58)
For a quasiparticle j, located at position rj, the regularized delta function δh(r− rj) =
D)−1 for all values of r satisfying r − rj ≤ rc and δh(r − rj) = 0 if otherwise. Here,
(hL2
the quasiparticle corresponds to a cuboid of height h and sides LD, which defines the
length scale of coarse graining.
For any field variable a(r), we have a(r)ρ(r) = (cid:80)
(cid:90)
dra(r)δh(r − rj).
ajδh(r − rj), with aj given as
If the initially flat membrane patch, with surface area A, is dis-
j
cretized into N quasiparticles, then the average density of the membrane is given by
ρA = N/A, and the lipid density per quasiparticle is given by ρ∗ = ρAL2
D.
A quasiparticle i with position ri and outward unit normal ni interacts with another
quasiparticle j with position rj and outward unit normal nj through a discretized bending
potential given by,
N(cid:88)
i=1
1
2
(cid:124)
Heff =
Nc,i(cid:88)
(cid:123)(cid:122)
(cid:125)
∆U stretch
j(cid:54)=i
Hσ,eff
ij
.
(59)
Nc,i is the number of quasiparticles around particle i, within a cutoff distance of rc,
and the bending and stretching potentials are given by,
ij
+
i=1
1
2
(cid:124)
j(cid:54)=i
Hκ,eff
N(cid:88)
Nc,i(cid:88)
(cid:123)(cid:122)
(cid:125)
∆U bend
8κ
(cid:18) ni · rij
(cid:1)(cid:3)2
(cid:2)2(cid:0)rij − r0
ρANc,i
Φij
0
ij
2πλh
N 2
c,i
0
81
∆U bend
ij =
∆U stretch
ij
=
The curvature function is given by Φij =
rij
bulk modulus, the energy contribution due to stretching is given by,
rij
+
for rij ≤ rc,i
if otherwise.
(cid:19)2
(cid:18) nj · rij
(cid:19)2
(60)
, and if λ is the
for rij ≤ rc,i
if otherwise.
(61)
rij and r0
ij are the vectors connecting particles i and j in the deformed and undeformed
states, respectively, with rij = rij and r0
ij. The bending and stretching poten-
tials have been chosen to ensure that the membrane has zero energy in the undeformed
ij = r0
state (here ni and nj are perpendicular to rij). The EM2 particles self-assemble into
a membrane both in the absence and presence of an explicit solvent; however the mem-
brane displays slightly altered dynamics in the presence of a solvent. Two explicit solvent
models—namely, the WCA solvent, and BLOBS solvent—have been used to model the
interaction of the EM2 particle with the surrounding fluid. Details of the membrane and
solvent models have been described in reference [114].
The topology of the self-assembled structures formed by the quasi-particles is a dy-
namic variable in the EM2 model. Since topological changes involve energy changes, the
topology-dependent Gaussian curvature term, which was not accounted for, but could
be easily incorporated, in the triangulated surface model (section III F), should also be
accounted for. The contribution from the Gaussian curvature term can be expressed as,
(cid:88)
i=1
HκG,eff =
κG
ρA
N c2
1,i.
(62)
Here κG is the Gaussian modulus introduced in eqn. (9). The total elastic contribution
to the EM2 model has three contributions and is given by,
Heff = Hκ,eff + Hσ,eff + HκG,eff.
(63)
B.
Isotropic and anisotropic protein fields in the EM2 model
The quasiparticle discretization of the elastic energy function, given in eqn. (59), can
be extended to accommodate the effect of both isotropic and anisotropic curvature fields
that were discussed earlier in section IV A. The presence of membrane curving proteins
enters into the model through the curvature function Φij, as in reference [317],
(cid:18) ni · rij
(cid:19)2
(cid:18) nj · rij
rij
(cid:19)2
Φij =
rij − γ
+
+ γ
.
(64)
Here γ is the protein-dependent spontaneous curvature function which for isotropically
curving proteins, with spontaneous curvature C0, has the form γ = C0/2. Alternately, for
anisotropically curving proteins, γ is a more complex function that depends both on the
82
particle position and orientation. Anisotropy is introduced into the EM2 model through
a unit vector field m whose orientation along the tangent plane defines the direction of
anisotropy. For an anisotropic protein, with maximum spontaneous curvature C0,
(cid:20)(cid:16)
(cid:17)2
(cid:16)
(cid:17)2(cid:21)
γ(rij, mi, mj) =
C0
2
(cid:107)
m
i .rij
(cid:107)
j .rij
m
+
.
(65)
If Pi be the tangent plane projection operator at quasiparticle i, then m
(cid:107)
i = Pimi.
A more sophisticated version of the EM2 model has been proposed by Ayton et al.
[119], where coupling between the density of the protein field ( given by φα for quasiparticle
α ) and membrane composition has been explicitly considered. In the proposed model,
for a pair of quasiparticles i and j, the following quantities are defined:
1. Membrane coupling to protein density
(a) The bending modulus depends on protein field density as κij = κ − ηκ(φi +
φj)/2.
(b) The spontaneous curvature is given by C0,ij = C0f (φi, φj), with f (φi, φj) =
−(φi + φj)/2 if (φi + φj) < 0 and zero otherwise.
2. Protein density coupling to membrane composition and geometry : Curvature-
inducing proteins diffuse on the membrane, and hence, their density can respond to
heterogeneities on the membrane surface. Two key factors driving change in protein
density—curvature sensing and sensitivity to lipid charge distribution—have been
considered. The compositional coupling is represented by an additional energy
contribution, HS,M .
(a) Intrinsic coupling (IC): In this formulation, the interaction between the mem-
brane and the protein field is sensitive to the background membrane curvature.
(b) Compositional coupling (CC): The protein field density is dependent on the
density of negatively charged lipids.
3. Free energy contributions arising from spatially varying densities of the protein field
and membrane lipid composition have been accounted for in this model
4. The effect of protein oligomerization has been included through an explicit term
that captures the energy costs due to protein oligomerization.
83
Technical details of the model along with the expression for various energy contribu-
tions can be found in [119]. In the following section we will highlight the key findings
on membrane remodeling studied using the EM2 model for two BAR domain containing
proteins namely N-BAR and F-BAR. The atomistic to coarse grained mapping for the
protein field and membrane quasiparticles has been described in detail in [321].
C. Membrane remodeling by N-BAR and F-BAR proteins
In the studies of Ayton et al.
[119, 317], the protein is represented as a curvature
field with curvature profiles given by eqns. (64) and (65).
In this multi-scale model,
the value of the spontaneous curvature C0 is determined from molecular simulations of a
single N-BAR domain interacting with the membrane of interest [212]. If ρ is the density
of the N-BAR proteins, δA a small area patch, and HN−BAR the spontaneous curvature
induced by a single protein, then
C0 = ρδAHN−BAR.
(66)
C0 has a maximum value of HN−BAR ∼ 0.15nm−1, which corresponds to the maximum
packing of N-BAR proteins on the membrane.
In this framework, the density of the
protein field is set by choosing an appropriate value of C0. For example C0 = 0.15nm−1
and C0 = 0.05nm−1 correspond to protein densities of ρ = 1 and ρ = 0.3333, respectively
.
Figs. 30(a) and (b) show the membrane remodeling behavior of N-BAR domains us-
ing the isotropic curvature model (eqn. (64)). At low N-BAR densities (small values of
C0), the initially vesicular membrane remains nearly spherical; a representative shape
obtained for C0 = 0.0nm−1 has been shown in Fig. 30(a). However, with increase in the
spontaneous curvature (increasing protein density), the spherical shape becomes unstable
and breaks up into an ensemble of smaller vesicles; this phenomenon has been shown in
Fig. 30(b) for C0 = 0.14nm−1. The radius of the smaller vesicles is set by the imposed
spontaneous curvature and can be approximated to be (C0)−1. These results obtained
from the EM2 model are consistent with those obtained from theoretical analysis and
also from computations using dynamical triangulation Monte Carlo techniques.
84
FIG. 30. Membrane remodeling behavior studied using the EM2 model with the N-BAR domain
modelled as an isotropic homogeneous curvature field. (a) An initially vesicular membrane
remains spherical when C0 = 0.0 nm−1 , and (b) the vesicular structure breaks up into a
collection of smaller vesicles at higher values of C0; data has been shown for C0 = 0.14 nm−1
(Reprinted from Biophys. J, 92 (10), G. Ayton, P. D. Blood, and G. A Voth, Membrane Remodeling
from N-BAR Domain Interactions:
Insights from Multi-Scale Simulation, 3595–3602, Copyright
(2007), with permission from Elsevier).
The membrane starts to show interesting remodeling dynamics when the curvature
field due to the BAR domains is treated as an anisotropic curvature field, whose curvature
profile follows eqn. (65). Fig. 31 shows the steady-state self-assembled shapes of the EM2
particles for four different spontaneous curvatures. When C0 <= 0.06 nm−1, the protein-
induced curvature field is too weak to promote membrane remodeling, and hence, the
quasi particles assemble into vesicular shapes as seen in Fig. 31(a).
85
experimentalobservations(5).Assuch,beforediscussingthecurrentsimulationresults,abriefsummaryofpreviousex-perimentalworkwillbegiven.Afterthat,theEM2results,atvariousincreasingN-BARdomaindensitystrengths,willbepresented.SummaryofexperimentalresultsInPeteretal.(5),liposomeswereexaminedforN-BAR(BARdomainplustheamphipathichelixattheN-terminus,seeFig.1),aswellasBARdomainsatvariousconcentra-tions(5 40mm)usingelectronmicroscopy.FortheN-BARcaserelevanttothepresentmulti-scalesimulations,itwasobservedthatliposometubulationoccurredatintermediateconcentrations(20mM),whereasliposomevesiculationoccurredathighconcentration(40mM).Tubulationresultedintubuleswithanouterdiameterof;46nm,whereasvesic-ulationoftheliposomeresultedinanarrayofsmallerlipo-someswitharangeofshapesandsizes.LowtomediumN-BARconcentrationsFortheisotropicspontaneouscurvaturefieldscenario(n¼2inEq.1),andforthelowandmediumvaluesofC0,theliposomeexhibitedadistortedstructurewithirregulardints.However,averydifferentpicturewasfoundwhentheanisotropiccurvaturefieldwasused(n¼1inEq.1).Atlowconcentrations,theliposomeremainedintact(asshowninFig.3a),whereasatintermediateconcentrations(C0¼0:08nm 1)theliposomewastubulatedintoacomplicatedstructureasshowninFigs.2cand3c.NotethatthecolorsoftheEM2particlesinFig.3arerepresentedbytheircurvaturefieldvectors,nTi,sothattheorientationalcorrelationsinthelocalcurvaturefieldscanbeidentified.Inthiscase,acloseinspectionofFig.3arevealsanalmostisotropicdistributionofcurvaturefields.Thecross-sectionaldiameterofthetubulatedstructurewasintherangeof;40–50nm.Itispossible,however,thatthisstructurecould,oververylongsimulations,annealintoasingletubule.Thelocalspontaneouscurvaturefieldsinteractto‘‘wrap’’aroundtheemergenttubulestructures.ThiseffectisshowninFig.3,b–d,wherethenTivectorslieroughlyperpendiculartothevectordescribingthelocalsymmetryaxisofthetubulatedstructures.Todeterminewhethertheoriginalstartingstructureoftheliposomehadanypersistenteffectontheresultingstructures,amacro-tubulesystemwasalsoexamined.Theinitialradiusofthemacro-tubulewas44nm;thissystemmimicsatubulated‘‘neck’’betweentwolargevesicles,forexample.InFig.4,themacro-tubuleisshownunderconditionssimilartothatusedinthepreviousliposomesimulation,exceptthatitwasfoundthatthemacro-tubulewasnotstablewhennoFIGURE2(a)AnEM2liposomewithaninitialradiusasinTable1andnoN-BARspontaneouscurvature,i.e.,C0¼0nm 1.(b)Theliposomeasinasubjectedtoanisotropic(n¼2inEq.1)spontaneouscurvaturefieldwithavalueofC0¼0.14nm 1.(c)Theliposomeasinasubjectedtoananisotropic(n¼1inEq.1)spontaneouscurvaturefieldwithavalueofC0¼0.08nm 1.Theyellowscalebarcorrespondsto100nm.TABLE1KeyparametersParameterSymbolValueLiposomeradiusrL13667nmTemperatureT308KEM2lengthscales6.8nmTimestepdt0.02psEM2energye5.5–6.5amu(nm/ps)2Bendingmoduluskc27amu(nm/ps)2SpontaneouscurvatureC00.05–0.2nm 1NumberofEM2particlesN4000Bilayerthicknessh3.4nmNumberoftimestepst23106TubuleradiusrT44.660.2nm3598Aytonetal.BiophysicalJournal92(10)3595–3602 FIG. 31.
Steady state shapes of a vesicular membrane evolved using the EM2 model with
anisotropic curvature field. Spherical shapes which are stable for C0 ≤ 0.06 nm−1 (a) evolve
into a network of tubules when C0 = 0.07 nm−1 (b) (Reprinted from Biophys. J, 92 (10), G.
Ayton, P. D. Blood, and G. A Voth, Membrane Remodeling from N-BAR Domain Interactions:
Insights from Multi-Scale Simulation, 3595-3602, Copyright (2007), with permission from Elsevier).
With further increase in C0, the membrane is driven into a network of tubular shapes.
The topology of this tubular network depends on the values of C0 and also on the strength
of κG. The transition from spherical to tubular shapes, with increasing C0, is consistent
with that seen in Fig. 24.
86
spontaneouscurvaturefieldwaspresent.Rather,themacro-tubule‘‘split’’overtimeintotwoseparatemembranesheets.Incontrast,themacro-tubulestructureremainedstablewhenaweakanisotropiccurvaturefieldwasused,asshowninFig.4a.Thisresultmirrorsthatobservedintheliposomecase;theinsetshowsthattheorientationaldistributionofthespontaneouscurvaturefieldisveryclosetoisotropicoverthemacro-tubulesurface.Atmoderateanisotropiccurvaturefieldstrengths,apronouncedtubulationresults(Fig.4c);thetubulatedshapeis‘‘heldintact’’bytheanisotropiccurvaturefielddensity,asshownintheinset.Withtheaboveresultsinhand,itispossibletorelatethemulti-scaleEM2simulationsbacktothemediumconcen-trationN-BARexperiments,whereliposomeswereobservedtoundergotubulation.Fromthesimulationresults,tubulatedliposomeswereonlyobservedwithananisotropicN-BARcurvaturefield.ThisleadsustoconcludethatsomedegreeofcollectiveorientationalorderoftheN-BARdomainsisrequiredexperimentallytofacilitatetubulation,andthatsomedegreeofspatialorderingoftheN-BARdomainsontherealliposomesurfaceexists.HighN-BARconcentrationWiththeisotropiccurvaturefield(n¼2inEq.1)ataroundthemaximumpossiblecurvatureC0¼HN-BAR,apro-nouncedvesiculationwasobserved.InFig.2b,asnapshotofanisotropicEM2simulationwithC0¼0:14nm 1isshown.Anarrayofdifferentsizedandshapedvesiclesemergedfromtheoriginalsingleliposome.Someofthevesicleswereelongated,somewerequitesmall(withdiametersaround30nm),whereasotherswerelargerwithdiametersaround100nm.Interestingly,theelongatedstructureshadcross-sec-tionaldiameterssimilartothoseobservedinexperiment(5)at;40nm.Theseresultsindicatethatauniformandisotro-picN-BARdomainspontaneouscurvaturefieldcanindeedresultinvesiculationathighN-BARdensities.Incontrast,simulationswiththehighdensityanisotropicN-BARcurvaturefieldresultedintubulatedstructuresasshowninFig.3d.Asthefieldstrengthisincreased,thecross-sectionaldiameterofthetubulesdecreasedtothepointthatbyC0¼0:10nm 1,thecross-sectionaldiameterwas;30nm.TheanisotropicN-BARcurvaturefieldathighconcentrationsimulationsneverresultedinvesiculatedstructuressimilartothoseobservedexperimentally.Inthecaseoftheinitialmacro-tubule,thestrongisotropiccurvaturefieldwithC0¼0:14nm 1(Fig.4b)againyieldedvesiculationintoavarietyofvesicleshapesandsizes.Thestronganisotropicfield,however,tubulatedthemacro-tubuleinasimilarmanneraswasobservedwiththeoriginalliposome(imagenotshown).Asinthelow/mediumN-BARconcentrationcase,thepresentEM2simulationresultscanbecomparedwiththeexperimentalobservationswherevesiculationwasobservedathighN-BARconcentrations.FromthehighdensityN-BARdomaindensityEM2simulations,vesiculatedstructureswereonlyobservedwiththeisotropicspontaneouscurvaturefield.HighdensityanisotropicN-BARcurvaturefieldsgeneratedtubulatedstructures.Combiningtheseresults,itissuggestedherethattheexperimentalhighconcentrationN-BARdomainsystemlikelyhasanisotropicspatialdistributionofN-BARdomainsontheliposomesurface.Therearethreepossibleexplanationsforwhyhighdensityliposome-boundN-BARdomainscouldhaveanisotropic(asopposedtoanisotropic)spatial/orientationaldistributionandthereforeresultinliposomevesiculation.ThefirstexplanationisthattheembeddedN-terminalhelicescouldresultinanadditionalradiusofcurvatureinadirectionnotFIGURE3LiposomeasinFig.1asubjectedtovariousanisotropicspontaneouscurvaturefields.Thespecificfieldvalueisshowninthefigure.Theyellowscalebariscorrespondsto100nm.Theorientationalstructureofthesingleradiusofspontaneouscurvatureisshownviathestick-representationoftheEM2quasi-particles.Thedifferentcolorsindicatecorrelationsinthelocalorientation:blueregionshavearadiusofcurvaturedirectionorthogonaltoredregions.Orangeregionsareintermediate.MembraneRemodelingSimulation3599BiophysicalJournal92(10)3595–3602 FIG. 32. (a) An atomistic representation of N-BAR proteins oligomerized on the membrane
surface. Solid arrows show the local spontaneous curvature direction on the membrane. (b)
Evolution (1-6) of an initially flat 250 nm2 membrane patch (1) into a tubule configuration (6)
(Reprinted from Biophys. J, 97 (6), G. Ayton, E. Lyman, V. Krishna, R. D. Swenson, C. Mim, V. M.
Unger and G. A Voth, New Insights into BAR Domain-Induced Membrane Remodeling, 1616–1625,
Copyright (2009), with permission from Elsevier).
The dynamics of the EM2 particles and of the in-plane orientation field has been
shown in Fig. 32(b), where an initially planar membrane remodels into a tubule. Locally
averaged in-plane orientations are shown as solid lines while the orientations of the local
membrane curvature are represented as diffuse lines in the background. For positive
values of C0, the orientation of the in-plane field is highly correlated with the maximum
curvature direction on the membrane, and the steady state is reached when the tubule
radius approaches (C0)−1. This finding is consistent with that observed in the case of the
nematic membrane model.
The EM2 model provides an excellent simulation technique to study the dynamics of
the membrane-protein system with explicit hydrodynamics in the meso- and continuum
87
solventismodeledwithsmoothparticleappliedmechanics(SPAM)(32–35),aspreviouslydonetomodelflagellarfilaments(36).SuperimposedonboththemesoscopicsolventandEM2membraneisanadditionalN-BARdensityvariablefB,whichdescribeslocalenhancementsordepletionsofN-BARdensityinboththesurroundingsolventandontheliposomesurface.ThetotalfreeenergyofthediscreteN-BAR/membranesystemisgivenbyH¼HSþHEM2þHMþHS;MþHO;(1)wherethediscreteenergy,H,originatesfromacontinuumlevelfreeenergydifferencemodel,andthuseachterminEq.1canberelatedtoaspecificfreeenergycomponentofthesystem.ThephysicalsignificanceofeachterminEq.1issummarizedinTable1anddiscussedintheAppendices.SpecificvaluesofvariouscoefficientsthatappearineachofthetermsaregiveninTableS1intheSupportingMaterial.BothfBandfMevolveasthemembranestructureremodels,usingaSPAM(32–35)discretizedversionofLandau-Ginzburgdynamics(25,26,37–39).Assuch,bothHSandHMinEq.1arediscreteversionsofLandaumodelsfortheN-BARdensity(27,28)andmembranecomposition(25,26).Anumberofnewfeaturesarealsoincludedinthismodel,andtheseare:1.ThemembranebendingmodulusincreaseswithincreasingBARdensityontheliposomesurfaceasonewouldexpectphysically.2.ThemagnitudeofthespontaneouscurvatureincreaseswithBARdensity.3.BARbindingonthemembranecanoccurviatwodifferentmechanisms,denotedasintrinsiccurvaturecoupling(IC),andcompositioncoupling(CC).ICcausesN-BARdensitytoaccumulateonthemembraneinregionswherethelocalcurvatureofthemembranematchesthelocalanisotropicspontaneouscurvaturethatisgeneratedbytheBARdensity.Thistypeofcurvaturecouplingisrelatedtodifferinginteractionpropen-sitiesfortheN-BARamphipathicheliceswiththemembranesurface.Indeed,thiscurvature-sensingpropertycanalsobeinterpretedasdensitycoupling(14)withintheEM2model.ThebendingenergyoftheEM2membranearisesfromtheinteractionbetweenmembranenormalvectors,U;however,thisenergycanalsobegeneratedviatwoeffectivemem-branesheetsseparatedbyamembranethickness,h(40,41),where,uponbending,theouterleafletdilates,whereastheinneronecompresses.Furthermore,itcanbeshownthatchangesinthelocalmeancurvatureintheEM2membraneareproportionaltoinversedensitychangesintheeffectiveouterleafletofthemembrane.ThispointiselaboratedfurtherinAppendixA.CCcausesN-BARdensitytoaccumulateinregionswithahighnegativelipidchargedensity,andismotivatedbytheconceptofprotein-inducedelectrostaticlipidsequestration(4,29–31).Combina-tionsofICandCCarealsoobviouslypossible.Thetypeofcoupling,eitherICorCC,willdeterminethefunctionalformusedforHS,M;thisisfurtherdiscussedinAppendixB.4.ThelocalspontaneouscurvatureisrelatedtothebulkdensityofN-BARsthroughthespontaneouscurvaturemagnitude,denotedC0.IthasamaximumvalueofC0,a-BAR,a¼N,F,whichisthemaximumpossiblemembranecurvatureforaspecific(NorF)BAR/membranesystematthemolecularscale.Inthecasewherea¼N,C0,N-BARisfoundfromMDsimulationsofsingleN-BARremodeling(9,10).Whena¼F,C0,F-BARisobtainedinthisarticlefromexperimentalobservations(5,7)butcouldalsobeobtainedfromMDsimulations.Thegeneralbehaviorofthepresentmesoscopicmodelcanbedemon-stratedbyexaminingtheremodelingofasquarepatchofEM2membrane.Fig.1adepictsasnapshotofanatomisticallydetailed,putativeN-BARolig-omerstructurethatissimilartooneproposedrecently(18)butisofalargerlengthscale.TheyellowarrowdesignatesthedirectionoftheanisotropicspontaneouscurvatureinducedbytheoligomerizedN-BARproteincoat.Fig.1bdemonstrateshowtheeffectoftheoligomerizedproteinisincorpo-ratedintotheEM2model.AninitiallyflatsquarepatchofEM2membrane(withdimensionsroughly250nm2)ispreparedwithaconstantN-BARdensityandmembranecomposition.Acorrespondinganisotropicsponta-neouscurvaturefieldasinFig.1aissuperimposedandisalsoshownbyTABLE1Physicaldescriptionofthedifferenttermsinthemesoscopicmodel,Eq.1TermFreeenergycomponentcontributionHSFreeenergycostarisingfromaspatiallyvaryingBARdensity,givenbyfB.HEM2Elastic(i.e.,quadratic)membranebendingfreeenergy.HMFreeenergycostofvariationsinthemembranelipidcomposition,givenbyfM.HS,MFreeenergycontributionarisingfromthecouplingoftheBARdensityandmembranelipidcomposition(i.e.,fBandfM,respectively).HOExplicitoligomerizationfreeenergy.AccountsforthepossibilitythatBARsmayoligomerizeintheprocessofmembranetubulationandismotivatedbytheexperimentalobservationofstriationsonamphiphysinN-BAR(8)andF-BARtubules(6).FIGURE1(a)SnapshotofaputativeN-BARoligomer-izationstructureonanatomisticmembrane.(b)SquareslabofEM2membrane,250nm2inarea,andispreparedwithaconstantN-BARdensityandmembranecomposition.Panelsb-1–6followthemembranethroughtheremodelingprocess.Thearrowsdesignatethelocalspontaneouscurvaturedirections;theintersectinglinesshowthelocalcurvaturedirectionsontheEM2membrane.BiophysicalJournal97(6)1616–16251618Aytonetal. scales. The parameters for the protein field are determined from an all atom or coarse
grained molecular simulation, but the validity of the linear relation between the protein
density and spontaneous curvature employed here warrants further investigation. In the
EM2 model, the membrane is formed as a result of self-assembly of the quasiparticles and
hence the model can accommodate morphologies with varying topologies. The discretiza-
tion of the elastic energy Hamiltonian assumes small curvature gradients locally, and as
a result, the application of the model to systems with extreme curvature requires further
sensitivity analysis.
VI. FREE ENERGY METHODS FOR MEMBRANE ENERGETICS
In the previous sections, we studied the conformational phase of a biological membrane,
with and without externally induced curvature fields, using analytical and computational
techniques. We have shown that the evolution of suitable order parameters, in parametric
space, in finite-temperature simulations can provide valuable insight into the nature of the
morphological transitions sustained in specific model systems. However, in order to estab-
lish the stability of the different morphological states/phases, the free-energy landscape
has to be delineated. Free-energy methods complement finite-temperature simulations
by providing insight into how entropic contributions modulate the energy landscape as-
sociated with various morphologies. In this section, we describe methods to numerically
delineate the free-energy landscape for the membrane with isotropic and anisotropic cur-
vature fields, using the method of thermodynamic integration.
A. Thermodynamic integration (TI)
In this approach, the free-energy difference between two membrane states A and B
with energies EA and EB, respectively, is computed by constructing a reversible path to
transition from A to B. The intermediate configurations of the membrane are controlled
by the Kirkwood coupling parameter 0 ≤ λ ≤ 1, such that the membrane is in state A
when λ = 0 and state B when λ = 1. The energy of any given intermediate state can
be written as E(λ) = (1 − λ)EA + λEB. Hence, the change in free-energy involved in a
88
transition from state A to B, calculated along the path S, is given by [102],
∆FA→B =
dλ.
(67)
(cid:90) 1
(cid:18) ∂F
(cid:19)
0
∂λ
For a system whose energy depends on a coupling parameter, λ, the partition function
can be written as,
[102]:
Q(λ) = c
(cid:90)
where c is a constant. Since the Helmholtz free-energy F (λ) = −kBT ln Q(λ), the deriva-
tive of the free-energy with respect to λ can be written as:
exp [−βE(λ)] drN ,
(68)
(cid:18) ∂F
(cid:19)
(cid:19)
(cid:18) ∂F
∂λ
= −
N,V,T
=
∂λ
N,V,T
1
β
∂
∂λ
(cid:28) ∂E
∂λ
ln Q,
(cid:29)
.
λ
(69)
(70)
yielding
Using eqn. (70) in eqn. (67) we can evaluate the change in free-energy without com-
puting the absolute free energies for various membrane states along the path S as,
∆FA→B =
dλ.
(71)
(cid:90) 1
(cid:28)∂E
(cid:29)
0
∂λ
This method can be used to compute the free energies associated with the membrane
shapes stabilized by isotropic and anisotropic curvactant fields.
B. Free energy for a membrane in the Monge gauge subject to isotropic curva-
ture fields
To investigate this phenomenon from a free-energy perspective, we employ thermo-
dynamic integration (TI). The elastic energy of a planar membrane with an external
curvature field C0, given by eqn. (B11), has the form,
(cid:90) (cid:110)κ
(cid:0)
Hsur =
∇2h − C0
2
+
(cid:1)2
(cid:16) σ
2
(cid:17)
(∇h)2(cid:111)
dxdy.
(72)
+
κ
4
C 2
0
Agrawal and Radhakrishnan [283] demonstrated the applicability of TI to a model
system of membrane deformations caused by a static (i.e. non-diffusing) heterogeneous
curvature field. The spontaneous curvature field C0 has a radially symmetric profile on
89
the membrane over a localized region characterized by a linear extent r0. The value of
C0 is taken to be zero in the membrane regions falling outside the localized region. Thus,
the induced curvature field is described by:
C0 = c0Γ(r0),
(73)
where Γ(r0) is a function that is unity within a circular domain (centered at zero) of
radius r0 and zero otherwise, and r0 is the linear extent (radius) of the curvature field
projected on the x-y plane. For the sake of illustration, we choose c0 = 0.04 (nm)−1.
The free-energy change of the membrane is calculated as a function of the extent of the
curvature field (r0) as well as the magnitude of the curvature field (c0).
In eqn. (72) and eqn. (73), when c0 is set to zero, the planar state of the membrane is
recovered, whereas for non-zero values of c0, the desired state of the curvilinear membrane
is obtained. We also note that the energy functional (eqn.
(72)) is differentiable with
respect to c0 but not differentiable with respect to r0. Hence, to compute the free-energy
changes, we choose c0 as the thermodynamic integration variable (i.e. as the coupling
parameter λ in eqn.
(70)) to obtain:
.
c0
=
∂c0
∂F
∂c0
(cid:28) ∂E
(cid:29)
(cid:0)
∇2h − c0Γ(r0)(cid:1) +
(cid:90) (cid:90) (cid:104)
(72), we obtain:
(cid:0)
(cid:17)
(cid:16) c0
(∇h)2(cid:105)
(cid:17)
(cid:16) c0
∇2h − c0Γ(r0)(cid:1) +
2
2
dxdy
(cid:29)
(∇h)2(cid:105)
c0
(74)
,
(75)
(cid:29)
dxdy
dc0.
c0
Using the expression for E from eqn.
∂F
∂c0
=
Γ(r0)κ
(cid:28)
(cid:90) (cid:90) (cid:104)
(cid:28)
−
(cid:90) c0
Upon integration along c0, this yields:
F (c0, r0) − F (0, r0) =
Γ(r0)κ
0
−
(76)
Here, F (c0, r0) − F (0, r0) is the free-energy change as derived from the partition function
(72). However, if we are interested
in eqn.
(68), where the energy is defined in Eq.
in deformation free-energy, F0, with reference to a state where C0 = 0, we employ the
relationship:
F0 = F + (cid:104)E0(cid:105) − (cid:104)E(cid:105),
(77)
where E0 is eqn. (72) with C0 = 0. Thus, ∆F0 = F0(C0, r0) − F0(0, r0) gives the defor-
mation free-energy change for a given extent of the localized region r0 (such as size of the
90
FIG. 33. Thermodynamic cycle to calculate ∆F0; ∆F0 = −∆F0,1 + ∆F0,0 + ∆F0,2. a and b
are the extent of the curvature-induced regions (r0 values) while c0 is the magnitude of the
curvature. ∆F0,1 and ∆F0,2 are computed using eqn.
(76).
clathrin coat) when C0 is varied. To calculate the free-energy as a function of r0 for a
fixed C0, we employ a thermodynamic cycle defined in Fig. 33. In this cycle, ∆F0,1 and
∆F0,2 required to deform a planar membrane to C0 = c0Γ(r0 = a) and C0 = c0Γ(r0 = b),
respectively, are calculated through eqn.
(76) and eqn.
(77).
The numerical results for the free-energy changes obtained by Agrawal and Radhakr-
ishnan [283] using thermodynamic integration are shown in Fig. 34. ∂F
∂c0
increases with
increasing value of c0 implying that the free-energy of the membrane, F , increases with
increasing magnitude of c0. Furthermore, for a larger extent r0, the increase in free-energy
(cid:16) ∂(cid:104)E(cid:105)
(cid:17)
for
∂C0
r0
is larger for the same change in c0. Fig. 34 shows the calculated values of
different values of c0 and r0. The quantity ∂(cid:104)E(cid:105)
the entropic contributions T ∂S
∂c0
, which are plotted in the inset of Fig.
∂c0 − ∂F
∂c0
derived from these two plots yields
34. As evident
from these figures, the entropic contribution to the membrane free-energy decreases with
C0, with the decrease being more prominent for larger values of r0. This provides sup-
port for the mechanism of entropy-mediated attraction between curvature-inducing bodies
discussed in section IV B.
91
D(a,C0)(b,0)(b,C0)F0,0=0F0,1F0,2F0DDD(a,0)∆F0,1(a,c0)(b,c0)(b,0)(a,0)∆F0,2∆F0∆F0,0=0 FIG. 34.
∂F
∂C0
(solid lines) and
30 nm. The inset shows T
∂S
∂C0
(dotted lines) plotted for two values of r0 = 20nm and
∂(cid:104)E(cid:105)
∂C0
as a function of c0 (Reprinted figure with permission from N. J.
Agrawal, Radhakrishnan, Phys. Rev. E, 80, 011925 and 2009. Copyright (2009) by the American
Physical Society).
Using the thermodynamic cycle shown in Fig. 33, Agrawal and Radhakrishnan [283]
calculated the membrane deformation free-energy change as a function of the extent of
r0 (data not shown). Computing the deformation free-energy change with respect to a
planar membrane (i.e. C0 = 0) gives the mean energy (cid:104)E0(cid:105) with respect to the planar
membrane. The change in the deformation free-energy (F0) with respect to a planar
membrane is plotted in Fig. 35, and the corresponding plots for change in deformation
energy (E0) and entropic energy (T S) can be found in reference [283]. Data are shown
for four different systems varying in length (L), bending rigidity (κ), and surface tension
(σ).
92
00.010.020.030.04051015202520nm30nm00.010.020.030.04−5−4−3−2−101Tκ∂S∂c0c0(nm)−1c0(nm)−11κ∂F∂c0,1κ∂"E#∂c0 FIG. 35. Membrane energy change as a function of r0 (Reprinted figure with permission from N.
J. Agrawal, Radhakrishnan, Phys. Rev. E, 80, 011925 and 2009. Copyright (2009) by the American
Physical Society).
The deformation free-energy of the membrane increases as the extent of the curvature
field r0 increases. Furthermore, changes in the non-dimensional deformation free-energy,
F0/κ, show similar trends for different systems of equal size, L = 250nm. Thus, ∆F0/κ
depends only weakly on membrane bending rigidity κ and membrane frame tension σ.
The inset to Fig. 35 depicts the variation of ∆F0/κ with area of the localized region
subject to the curvature field, Ac, defined as:
(cid:90) (cid:90)
AC =
Γ(r0)
1 +
(cid:18)
(cid:19)
1
2
(∇h)2
dxdy.
(78)
In this region, ∆F0/κ scales almost linearly demonstrating that membrane free-energy is a
linear function of AC for small deformations considered here. Interestingly, the increase in
∆F0/κ is smaller for the larger membrane size. Noting that the difference in the entropy
change for different sizes of membranes is small, the changes in ∆F0/κ are a reflection of
the changes in ∆E0/κ.
93
0510152025303500.511.5205001000150020002500300000.511.52r0(nm)σ(µN/m)κ[L,κ,σ(µN/m)]250nm,20kBT,0250nm,50kBT,0250nm,50kBT,3500nm,50kBT,3∆F0κ∆F0κAC(nm2) C. Thermodynamic integration methods for the anisotropic curvature model
In the anisotropic curvature model, the membrane-curving nature of curvactant in-
clusions is captured in the anisotropic elastic term given by eqn. (56). Thermodynamic
integration can also be used to compute the change in free-energy when a spherical ho-
mogeneous membrane undergoes a morphological change to the various phases listed in
Figs. 24 and 27 due to its interaction with these membrane inclusions. In this model, the
anisotropic bending rigidities can be chosen to be the Kirkwood coupling parameters, so
that eqn. (56) can now be written as,
Hanis(λ(cid:107), λ⊥) =
1
2
dS
λ(cid:107)κ(cid:107)
maKabmb − C
(cid:26)
(cid:104)
(cid:90)
S
(cid:105)2
(cid:107)
0
+ λ⊥κ⊥(cid:2)ma⊥Kabmb⊥ − C
⊥
0
(cid:3)2
(cid:27)
.
(79)
Here, two coupling parameters, λ(cid:107) and λ⊥, that couple respectively to κ(cid:107) and κ⊥ have been
used to construct the two dimensional free-energy landscape associated with a nematic
field coupled to the membrane. For simplicity, we will focus only on the effect of a nematic
field with curvature profile given by κ(cid:107) (cid:54)= 0 and κ⊥ = 0. The total energy of the nematic
field has three contributions given by Htot = Hsur + H2−atic + Hanis(λ(cid:107)). In the above-
mentioned framework, the free-energy can be computed from thermodynamic integration
using the relation,
∆F (κ, κ(cid:107), C
(cid:107)
0 ) =
(cid:90) 1
(cid:28) ∂Htot
(cid:29)
0
∂λ(cid:107)
dλ(cid:107).
(80)
The dynamically triangulated Monte Carlo technique, introduced in section IV C 2,
has been used to sample the phase space of the membrane. So far, all our dicussions
related to nematic membranes explored the effect of directional spontaneous curvatures
with fixed isotropic and directional bending rigidities, κ = 20kBT and κ(cid:107) = 5kBT . To
(cid:107)
0
complement those results, we will construct the free-energy landscape as a function of C
with the set of bending rigidities mentioned above.
Before proceeding to the discussion of the free-energy, we will recall the behavior
of a membrane when subjected to a directional spontaneous curvature that promotes
the formation of branched tubular shapes. The thermodynamic integration has been
performed for C
(cid:107)
0 = 0.6, for which branched tubes were shown to be the equilibrium
shapes in Fig. 24. The isotropic and anisotropic energy contributions, Hsur and Hanis, for
a triangulated membrane with N = 2030 vertices, are shown along with the corresponding
94
membrane conformations as a function of λ(cid:107) in Fig. 36.
FIG. 36. The elastic energy of a membrane (Hsur) and its interaction energy with the nematic
field (Hanis(λ(cid:107))) as a function of λ(cid:107) for a nematic membrane, with κ = 20kBT , C
(cid:107)
0 = 0.6
and κ(cid:107) = 5kBT . Shapes transitions are seen when the nematic coupling strength goes from
κ(cid:107) = 0 → 5kBT . Hanis starts to dictate the shape of the membrane when λ(cid:107) (cid:39) 0.7 leading to
tubular shapes.
Whereas the elastic energy Hsur prefers minimum curvature on the membrane surface,
the anisotropic elastic term Hanis(λ(cid:107)) would prefer membrane conformations with maxi-
mum principal curvature at every vertex, equal to C
(cid:107)
0 . As can be seen from Fig. 36, when
a spherical membrane transforms to a tube, Hanis(λ(cid:107)) starts to minimize its contribution
at the cost of Hsur when λ(cid:107) (cid:39) 0.7, which is a measure of the critical nematic membrane
interaction strength required to tubulate a membrane. In the example above, this critical
value can be shown to be equal to κ∗
(cid:107) = 0.7κ(cid:107) = 3.5kBT , since we have fixed κ(cid:107) = 5kBT
throughout our simulations. Such an analysis would be extremely helpful in protein de-
sign; say, in predicting the possible set of amino acids in a protein interacting with a
membrane, provided a mapping scheme from the mesoscale to the atomistic scale exists.
The free-energy change, ∆F (C
in the range −1.0 ≤ C
(cid:107)
0 ), for a nematic membrane with directional curvatures
(cid:107)
0 < 1.0, computed using eqn. (80), is shown in Fig. 37. Plotted
95
00.20.40.60.8100.51λ∥Energy(kBT)Hsur(λ∥)Hanis(λ∥) alongside are the average energy difference (cid:104)∆Htot(cid:105) and the entropic contribution to
(cid:107)
the free-energy, calculated as −T ∆S = ∆F (C
0 ) − (cid:104)∆Htot(cid:105). The energy difference in
the thermodynamic state of the membrane for a given value of directional spontaneous
curvature is defined as (cid:104)∆Htot(cid:105) = Htot(λ(cid:107) = 1) − Htot(λ(cid:107) = 0).
FIG. 37. The free-energy landscape connecting two membrane states with directional spon-
taneous curvatures 0 and C
(cid:107)
0 , computed from the total energy Htot and free-energy ∆F (C
(cid:107)
0 ).
∆F (C
(cid:107)
0 ) approaches (cid:104)∆Htot(cid:105) for smaller values of C
(cid:107)
0 , whereas at large values of C
(cid:107)
0 the free-
energy has contributions from the entropy, to the order of 600kBT ; the sizeable entropic contri-
bution stabilizes caveolae like morphologies (inward growing tubules) at large negative C
tubular shapes at large positive C
(cid:107)
0 .
(cid:107)
0 and
As expected, the energy barrier given by the difference in the internal energies of the
membrane is always greater than or equal to the energy barrier computed through free-
energy methods. (cid:104)∆Htot(cid:105) ∼ ∆F when the membrane remains quasi-spherical at small
(cid:107)
magnitudes of C
0. When the membrane transforms into tubes and inward tubes, for
(cid:107)
0 > 0.6, the computed free-energy is lower than the (cid:104)∆Htot(cid:105) by around 600 kBT due
C
to the large conformational entropy associated with these morphologies. The estimated
value of −T ∆S is ∼ 10% of the total energy of the membrane, which agrees well with
96
-1-0.500.5105e+031e+04Ck0Energy,(kBT)h∆Htoti∆F(Ck0)−T∆S similar estimates for this value [283].
Other methods for computing free energies can be readily adapted within the contin-
uum framework discussed above and using the machinery of Monte Carlo simulations.
Recently, Tourdot et al.
[322] introduced and compared three free-energy methods to
delineate the free-energy landscape of curvature-inducing proteins on bilayer membranes.
Specifically, they showed the utility of the Widom-test-particle/field insertion method-
ology in computing the excess chemical potential associated with curvature-inducing
proteins on the membrane and in tracking the onset of morphological transitions on the
membrane at elevated protein density. The authors validated their approach by compar-
ing the results with those of thermodynamic integration and Bennett acceptance ratio
methods.
VII. CONCLUSIONS
In summary, we have discussed various theoretical and modeling strategies along with
specific applications utilized in the study of protein-induced curvature in membranes from
equilibrium and dynamic/hydrodynamic perspectives. The future of such studies is very
bright and ripe as we are beginning to unravel the true complexity of biological systems
through highly quantitative experiments at the mesoscale. Given the central importance
of protein-induced curvature mechanisms in cellular transport, an integration of these bio-
physical methods along with models of signal transduction is expected to provide unprece-
dented valuable insight into cell biology. With the advent of advances in algorithms for
multiscale modeling as well as high-performance computing, future models are expected
to mimic the underlying biological complexity, such as the signaling microenvironment,
mechanotransduction machinery, and tissue-specific boundary conditions. Future exten-
sions of such models to the non-equilibrium regime (which we have completely ignored in
this article) are also expected to define as well as elucidate crucial cellular mechanisms
that fall outside the purview of equilibrium or linear-response regimes.
97
VIII. ACKNOWLEDGMENTS
The authors thank Dr. Neeraj Agrawal and Dr. John H. Ipsen for many insightful
discussions. They also thank numerous colleagues at the University of Pennsylvania and
the Indian Institute of Technology Madras for their various inputs. The authors are
also very grateful to the anonymous referees who helped transform this article to its
current form through their extensive and insightful feedback. The authors acknowledge
funding from the US National Science Foundation (CBET-113267, CBET-1236514, DMR-
1120901), the US National Institutes of Health (1R01EB006818 and U01-EB016027), and
the European Commission (7th programme for research VPH-600841). Supercomputing
resources were made available through XSEDE (extreme science and engineering discovery
environment, MCB060006). Source files and example files for some of the codes discussed
in this review are available online as supplementary material to this article.
98
IX. APPENDIX
Appendix A: Differential geometry
In this appendix, we will focus on the general mathematical techniques used to compute
the mean and Gaussian curvature associated with the continuum elastic energy given in
eqn. (8). This material can be found elsewhere [50, 63, 66], but we present it here for
completeness.
In the continuum approximation, a membrane is described by a two-dimensional sur-
face, S, embedded in a three-dimensional space, R3. The membrane is parameterized by
a position vector (cid:126)R(x1, x2, x3), with the gauge defined by the orthonormal coordinates
x1, x2, and x3. If x1 and x2 are orthonormal vectors in the tangent plane of the mem-
brane, the problem is simplified when one choses a suitable gauge for the surface such
that x3 = f (x1, x2), where f is a function of x1 and x2. The tangent vector at any point
on the membrane surface is defined as,
(cid:126)tα = ∂α (cid:126)R.
(A1)
Here, ∂α = ∂/∂xα, with α = 1, 2. The unit surface normal can be determined from the
tangents as,
n =
(cid:126)t1 × (cid:126)t2
(cid:126)t1 × (cid:126)t2
.
(A2)
The metric of the gauge, defined by the metric tensor, is given by the first fundamental
form,
The metric tensor is a fundamental measure of length on the surface: the distance between
gα,β = (cid:126)tα · (cid:126)tβ = ∂α (cid:126)R · ∂β (cid:126)R,
(A3)
any two points seperated by d(cid:126)x can be written in terms of the metric tensor as ds2 =
dxαdxβgαβ, and the area of a surface patch is given by dA = √det g dx1dx2. The raising
operator gαβ, which is the inverse of the metric tensor, is defined as,
gαβ =(cid:0)g
(cid:1)
−1
γν
αβ
,
(A4)
(A5)
and satisfies the following relation,
gαγgγβ = δα
β .
99
The curvature profile of a membrane is determined by the curvature tensor, given by the
second fundamental form, as,
A surface is characterized by the two curvature invariants determined from the curvature
Kαβ = ∂αβ (cid:126)R · n.
(A6)
tensor—namely,
Mean Curvature : H =
1
2
K α
α =
1
2
Tr(Kαβ) and
Gaussian Curvature : G =
det (Kαβ)
det(gab)
.
(A7)
(A8)
Appendix B: Planar Monge gauge
Small deformations of a planar membrane about the reference plane can be formulated
and studied in the planar Monge gauge. The position vector in this gauge is a functional
(cid:126)R(x, y, h(x, y)), where h(x, y) is the normal displacement of the membrane region (x, y).
If the position vector is (cid:126)R = {x, y, h(x, y)}, various geometrical quantities, defined in A,
in this gauge are as follows:
Tangent vector : With α = x, y, the unit tangent vectors at any point on the membrane
surface are given by,
and
(cid:126)tx = { 1 , 0 , ∂xh} ,
Normal vector : The unit normal vector to the surface is given by,
Covariant metric tensor, gαβ : The metric tensor in the Monge gauge takes the form,
(B1)
(B2)
(B3)
n =
.
(cid:126)tx × (cid:126)ty
(cid:126)ty = { 0, 1, ∂yh} .
(cid:12)(cid:12) = {−∂xh, −∂yh, 1}
(cid:12)(cid:12)(cid:126)tx × (cid:126)ty
(cid:113)
1 + (∇h)2
.
1 + (∂xh)2
∂xh ∂yh
gαβ =
∂xh ∂yh 1 + (∂yh)2
100
Contravariant metric tensor: The contravariant metric tensor gαβ, the inverse of gαβ, in
The metric g in the Monge gauge is,
the Monge gauge is,
gαβ =
1
g = det(gαβ) =(cid:0)1 + (∇h)2(cid:1) .
1 + (∂yh)2 −∂xh ∂yh
∂x∂xh ∂x∂yh
.
−∂xh ∂yh 1 + (∂xh)2
1(cid:113)
1 + (∇h)2
∂x∂yh ∂y∂yh
1 + (∇h)2
.
K =
(B4)
(B5)
(B6)
Curvature tensor : The components of the curvature tensor are given by Kαβ = ∂α∂β (cid:126)R · n
. In the Monge gauge,
Curvature invariants : The gauge-invariant quantities that can be constructed from the
curvature tensor are
• Mean curvature
H =
1
2
Tr(K) =
1
2
• Gaussian curvature
G =
1
2
αβγδKαγKβδ = det(K) =
is the anti-symmetric Levi-Cevita tensor.
1. Elastic energy of a planar membrane
.
∇2h(cid:113)
1 + (∇h)2
(cid:113)
1 + (∇h)2
(∂x∂xh)(∂y∂yh) − (∂x∂yh)2
(B7)
.
(B8)
Using the curvature invariants derived in appendix B, we obtain the expression for the
elastic energy in the planar Monge gauge. In the Monge gauge, eqn. (8) (and by including
the contributions from surface tension) can be written as,
(cid:90) (cid:112)
Hsur =
1 + (∇h)2
∇2h(cid:112)
1 + (∇h)2 − C0
(cid:33)2
dxdy.
(cid:32)
κ
2
101
+ σ
(B9)
In the limit of small deformations around the planar reference plane, where ∇h (cid:28) 1,
expanding and retaining terms to fourth order gives the linearized form of the elastic
energy truncated to quadratic order in curvature terms; it has the form:
(cid:90) (cid:110)κ
2
(cid:0)
∇2h − C0
(cid:1)2
(cid:111)
+ σ
Hsur =
dxdy.
(B10)
An alternate form, to quadratic order, is given by,
(cid:90) (cid:110)κ
(cid:0)
2
(cid:1)2
+
(cid:16) σ
2
(cid:17)
(∇h)2(cid:111)
+
κ
4
C 2
0
∇2h − C0
Hsur =
dxdy.
(B11)
Appendix C: Surface quantifiers on a triangulated surface
The principal curvatures and their corresponding directions are calculated at a vertex
using its one-ring neighbourhood shown in Fig. 38. The approach is based on the con-
struction of the discretized “shape operator” given by the differential form −d N in the
plane of the surface, which contains all information about the local surface topography.
Consider a local neighbourhood around a vertex v in Fig. 38. (cid:126)re is the edge vector that
links v to a neighbouring vertex. The set of edges linked to v is {e}v, and the oriented
triangles or faces with v as one of their vertices is {f}v. The one-ring neighbourhood
around vertex v is well defined by {e}v and {f}v.
Edge and vertex normals: An edge e, tethering two vertices, is characterized by its length
(cid:126)re and orientation re. As shown in Fig. 38, re along with the normal to the edge, Ne,
and binormal, be = Ne × re, define the Frenet frame on e. The edge normal is entirely
dependent on the orientation of the set of faces, {f}e = [f1(e), f2(e)], sharing it. It is
determined as,
102
FIG. 38. (cid:126)re = (cid:126)x(v) − (cid:126)x(1) is the vector along the edge e, connecting vertices v and 1. Ne and
be are respectively its normal and binormal.
Nf (1, e) + Nf (2, e)
(cid:13)(cid:13)(cid:13) Nf (1, e) + Nf (2, e)
(cid:13)(cid:13)(cid:13) .
Ne =
(C1)
Nf (1, e) and Nf (2, e) are the unit normal vectors to faces f1(e) and f2(e), respectively,
sharing the edge e. The term in the denominator is the normalization factor, and hence-
forth, the edge normal is a unit vector. Unlike in the case of an edge, there are ambiguities
in the calculation of the normal at a vertex.
For instance, let S denote a continuous surface around vertex v and let C be a closed
contour on S, enclosing v. The normal to the vertex v is calculated by integrating the
surface normal along the contour,
(cid:90)
(cid:126)Nv =
NS (C) dC
(C2)
C
with NS(C) being the unit surface normal on the contour. When S is replaced by a trian-
gulated patch, the value of NS(C) changes only at the interface between the faces—that
is when the contour crosses an edge. Hence, the contributions from each face, in {f}v,
should be appropriately weighed when approximating the vertex normal using the discrete
form of eqn. (C2). The above-mentioned ambiguity arises in the choice of this weight.
We have approximated the normal at a vertex v as,
(cid:80){f}v Ω[Af ] Nf
(cid:13)(cid:13)(cid:13)(cid:80){f}v Ω[Af ] Nf
Nv =
103
(cid:13)(cid:13)(cid:13) ,
(C3)
reNebeNf(1,e)Nf(2,e)Nvv123456 with Af denoting the surface area of the face f . We have chosen the weight factor Ω[Af ]
to be proportional to the area of the face, as in [323–325].
Shape Operator at an edge: The topographic details of the triangulated surface around a
vertex are contained in the faces and edges. The shape operator, the discrete form of the
curvature tensor, is constructed at a chosen vertex by superposing various measures that
quantify the curvedness of the surface. One such measure that can be constructed from
the geometry of the faces and edges is the shape operator at an edge, SE(e).
As noted above, an object traversing a triangulated surface feels its curvature only
when it crosses an edge. At any point p on such an edge e, the curvedness can be
quantified by the edge curvature, which is the gradient of the area vector of the triangles
sharing e [326],
h(e) = ∇p(area) ≈ 2(cid:107)(cid:126)re(cid:107) cos
.
(C4)
(cid:18)Φ(e)
(cid:19)
2
h(e) takes a non-zero value when the faces sharing an edge e are non-planar. Note
that h(e) has the dimensions of length, and dividing it by an area makes it a curvature,
an operation performed later.
FIG. 39. An illustration of the signed dihedral angle between two faces sharing an edge e.
Φ(e) is the signed dihedral angle between the faces f1(e) and f2(e) (see Fig. 39) sharing
edge e, calculated as
Φ(e) = sign
(cid:104)(cid:110)
(cid:111)
Nf (1, e) × Nf (2, e)
(cid:105)
(cid:104)
(cid:105)
Nf (1, e) · Nf (2, e)
arccos
· (cid:126)re
+ π.
(C5)
The edge curvature contribution at each edge can be used to construct the discretized
“edge shape operator”, which quantifies both the curvature and the orientation of e. This
tensor has the form
SE(e) = h(e)
.
(C6)
(cid:105)
(cid:104)be ⊗ be
Shape operator at a vertex: The complete description of a triangulated surface should
contain details of its geometry and orientation. When defining the geometry of the surface
104
replacemenΦ(e)126Nf(1,e)Nf(2,e)v the zero-dimensional vertices are primary since the higher-dimensional entities—namely,
the edges and faces—are constructed from it. Hence it is natural to also define the
orientational details of the surface on the vertices itself. This is done by constructing the
shape operator at the tangent plane of a vertex, as was done for an edge. The shape
operator on every edge originating from a vertex contributes to the shape operator at
the vertex. Hence at a vertex v, the vertex operator SV(v) is a superposition of all edge
operators {SE(e)}v around it. Further, it is instructive to recall that SE(e) was constructed
in the direction of be, which need not be along the tangent plane at v. The component
of the edge shape operator along the tangent plane, at vertex v, is the relevant part in
the construction of the SV(v). This contribution can be obtained by diagonalizing SE(e)
using the projection operator at v [324, 325],
(cid:105)
(cid:104)
Nv ⊗ Nv
Pv = 1 −
,
(C7)
where 1 is the unit diagonal matrix. The shape operator at v is then a weighted sum
W (e) Pv
† SE(e) Pv.
(C8)
of these contributions, given by
(cid:88)
{e}v
SV(v) =
1
Av
Av =(cid:80){f}v
Af /3 is the average surface area around v, whereas the weight factor for an
edge is calculated as W (e) = Nv · Ne. The shape operator, eqn. (C8), at the vertex v is
a rank 2 tensor expressed in the coordinates of the global reference system [x, y, z]. The
curvature tensor, in conventional differential geometry, is defined on the tangent plane of
the vertex [50, 66], with the principal directions t1(v) and t2(v) being the basis vectors.
It should be observed that, the vertex normal Nv is orthogonal to the principal directions.
These three directions define the local frame of reference, [ t1(v), t2(v), Nv], also called
the Darboux frame. If the curvature tensor is constructed in the Darboux frame, then
the vertex normal Nv would be the eigendirection corresponding to an eigenvalue of zero.
Since the eigenvalues of the curvature tensor are gauge invariant, we expect one eigen-
value of SV(v) to be zero and the other two, namely c1 and c2, to be the principal
curvatures. A numerical solution for the eigenspectrum of SV(v) in the native cartesian
frame of reference, is computationally expensive. Using a series of unitary transforma-
tions we can extract not only the principal curvatures but also the transformation matrix
TGL, which allows us to switch forth and back between the global and Darboux frames.
105
Knowledge of TGL is crucial when studying membranes with an in-plane vector field. In
the next part, techniques for the transformations mentioned are outlined.
Transformation to the Darboux frame: Computation of the eigenvalues of SV(v), the
vertex shape operator constructed in the cartesian frame with basis vectors [x, y, z], is
complicated by its non-zero off-diagonal elements. On the other hand, SV(v) has a di-
agonal form when represented in the Darboux frame, [ t1, t2, N ], with diagonal elements
c1, c2, and zero, a representation more suitable for extracting the eigenspectrum of the
shape operator. However, a direct transformation into the local frame is not feasible,
since the principal directions t1 and t2 are not known a priori. In our approach, we use
the Householder transformation (see Appendix D) technique to overcome this problem,
as shown in Fig. 40.
FIG. 40. Transformation from a global to local coordinate frame, using the Householder trans-
formation.
The Householder matrix TH uses the vertex normal Nv to compute the tangent plane
at vertex v, which need not be the principal directions themselves. Technical details per-
taining to the construction of TH are outlined in Appendix D. We choose to rotate the
global z direction into Nv, whereas x and y are rotated into vectors x(cid:48), y(cid:48) in the tangent
†
H(v) SV(v) TH(v)
plane at the vertex v. The shape operator, at v, in this frame C(v) = T
is a 2x2 minor, with the two principal curvatures c1(v) and c2(v) as its eigenvalues. The
principal curvatures and directions are computed from C(v), which is fairly simple. The
, N (v)] into the Darboux frame at v. Any
eigenvector matrix, TE(v), transforms [x
(cid:48)
(cid:48)
, y
tensor in the global frame can now be transformed to this local frame by the transfor-
mation matrix TGL = TE(v) TH(v). The curvature invariants—namely, the mean and
Gaussian curvature—are H(v) = (c1(v) + c2(v))/2 and R(v) = c1(v)c2(v), respectively.
106
DarbouxGlobalTGL=TETHTETHt1t2x′y′x′′=t1y′′=t2z′′=Nz′=NxyzN There are also other methods that can be used for computing the principal surface
properties on a triangulated surface
[323, 327]. However, the results of the current
algorithm for construction of the vertex shape operator are more accurate, especially
when it comes to the prediction of principal directions.
Appendix D: Householder transformation
Consider two orthonormal frames of reference given by the coordinates (x, y, z) and
(a, b, c). The Householder matrix, TH, can be used to rotate z in frame 1 to c in frame
2, such that (x, y) now are some arbitrary vectors in the plane formed by (a, b). Define a
vector,
W =
x ± c
x ± c
(D1)
with a minus sign if x − c > x + c and a plus if otherwise. The Householder matrix
is then defined as,
TH = 1 − 2W W
†
(D2)
Appendix E: Parallel transport of a vector field on a triangulated patch
In order to compare the orientations of two distant in-plane vectors on a curved surface,
it is necessary to perform a parallel transport of the vectors on the discretized surface [50,
66].
In practice, we need only to define the parallel transport between neighbouring
vertices, i.e. a transformation m(v
the tangent plane of the vertex v
(cid:48)
(cid:48)
v and v
is preserved.
(cid:48)
(cid:48)
) → Γ(v, v
) m(v), which brings m(v) correctly into
, so that its angle with respect to the geodesic connecting
As shown in Fig. 41, if r(v, v
(cid:48)
and (cid:126)ζ(v)=Pv r(v, v
) and (cid:126)ζ(v
(cid:48)
(cid:48)
v
(cid:48)
) is the unit vector connecting a vertex v to its neighbor
)= Pv
(cid:48)
(cid:48) r(v
, v) are its projections on to the tangent planes
, then our best estimate for the directions of the geodesic connecting them are
(cid:48)
at v and v
the unit vectors ζ(v) and ζ(v
(cid:48)
).
107
FIG. 41.
Illustration of the parallel transport of a vector between two neighbouring vertices v
(cid:48)
and v
in a triangulated surface.
The decomposition of m(v) along the orientation of the geodesic and its perpendicular
in the tangent plane of v, is thus,
(cid:104)
m(v) · ζ(v)
(cid:105) ζ(v) +
(cid:104)
(cid:105)(cid:16)
(cid:17)
Nv × ζ(v)
m(v) · ( Nv × ζ(v))
(E1)
m(v) =
Parallelism now demands that these coordinates, with respect to the geodesic orientation,
are the same in the tangent plane of v
; therefore,
(cid:104)
(cid:48)
(cid:105) ζ(v
(cid:48)
) +
(cid:110)
(cid:111)(cid:104)
m(v) · ( Nv × ζ(v))
Nv
(cid:105)
)
(cid:48)
(cid:48) × ζ(v
(E2)
(cid:48)
Γ(v, v
) m(v) =
m(v) · ζ(v)
This parallel transport operation allows us to define the angle θvv
tangent plane at neighbouring vertices and, in turn, their cosine and sine as:
(cid:48) between vectors in the
(cid:48)
(cid:104)
(cid:48)
(cid:105)
cos(θvv
(cid:48) ) = m(v
sin(θvv
(cid:48) ) =
Nv
) · Γ(v, v
(cid:48) × m(v
)
(cid:48)
) m(v)
(cid:48)
) m(v)
· Γ(v, v
(E3)
Appendix F: Properties of a membrane with in-plane order
The term implicit coupling indicates that the anisotropic constituents do not have any
specific interaction with the membrane. The membrane exhibits anisotropic behavior,
when it prefers a particular direction over others, which can be due to the presence
of an ordering interaction, leading to the formation of spatially ordered patterns of its
anisotropic constituents. The presence of such an order can limit the number of micro
states accessible to the membrane, and hence, the membrane feels the presence of the
anisotropic components through a reduction in the entropic contribution, which is implicit.
108
vv′ϑ(v)ϑ(v′)t1(v)t1(v′)m(v)m(v′)~ζ(v)~ζ(v′)r(v,v′)ϑvv′=ϑ(v)−ϑ(v′) The equilibrium properties of the membrane significantly change when the surface vector
field orients into an ordered state. This section describes how the morphology of the fluid
membrane is remodeled due to the presence of an ordered phase of the in-plane field, using
the Monte Carlo techniques described above. For this purpose, a nematic field (p = 2)
has been chosen to decorate the vertices of the triangulated surface, with 100% surface
coverage. The nematic field is ordered by choosing a non zero value for the exchange
interaction J2. In order to establish the importance of anisotropy the conformations of
the field decorated membrane, with κ = 0, are compared against the branched polymer
phase of a pure fluid membrane of similar rigidity.
1. Defect structure in polar and nematic fields
Textures of a vector field on curved surfaces significantly differ from those on a plane
surface. Unlike on flat surfaces, spatial orientations of the field are frustrated by the
topology of the curved surface. Geometric frustration is defined as the inability to impose
the desired local order throughout the system, due to constraints that are purely geometric
in nature. A well-known example is a system of Ising spins arranged on a triangular lattice
with antiferromagnetic bonds; here, the ground state prefers anti-parallel spins at the ends
of each bond. The triangular nature of the lattice, as illustrated in Fig. 42(a), disallows
such a state leaving one unsatisfied bond per plaquette, with parallel spin orientations,
as a result of which the minimal energy state differs from the ground state. Frustration is
also observed in a wide range of physical and biological systems like water, spin ice, high
-transition superconductors, buckled colloidal monolayers, etc.
[328]. Further, the 5-fold
and 7-fold disclinations in the tessellation of a sphere, giving rise to a scarred pattern, is
another case of frustration in nature and is shown in Fig. 42(b).
109
(a)Frustrated Ising magnet
(b)Disclinations on a tesellated sphere
FIG. 42. Two examples of geometrically frustrated systems in nature. (a) An Ising magnet
with antiferromagnetic bonds on a triangular lattice; a bond with unfavorable spins orientations
is marked with a wiggly arrow. (b) Five fold and seven disclinations are marked with blue and
red shades on a tesellated sphere. Euler theorem requires the presence of at least 12 five fold
disclinations to account for the non zero Gaussian curvature of the sphere.
Similarly, an in-plane surface vector field is also frustrated by the topology of the un-
derlying surface that gives rise to orientational singularities, also called topological defects.
In the case of an in-plane field on planar surfaces, the defects are not part of the zero-
temperature state and are seen only at high temperatures due to the thermally induced
vortex unbinding. However, for non-planar topologies, with non-zero total Gaussian cur-
vature, a minimal number of defects, each of strength 1/p is part of the field texture at
zero temperature. The number of defects is related to the surface topology through the
Euler number χ, introduced earlier in section II C 1 as,
ND(cid:88)
D=1
qD,
χ =
(F1)
where qD is the strength or the topological charge of defect D. The number of defects
ND and the charge of the defect qD depend on the symmetry of the in-plane field and
topology of the membrane surface. The following section discusses the calculation of the
defect charge from the winding number of the in-plane field.
110
IsingspinsonatriangularlatticeAnunsatisfiedbondforJ>0 2. Defect core and its winding number
In the continuum description of the tangential field, a defect core represents a math-
ematical singularity, i.e. spatial locations at which the gradient of the vector field di-
verges. [68, 308]. This singular behavior is normally taken care of by setting the order
parameter at the core, of radius Rcore, to zero and the texture of a field around a vortex
is characterized by its vorticity or winding number.
(a)Continuum : W=(cid:82)
dθ (C) = 2π
C
(b)Discrete : W(v)=(cid:80)
vi∈v ϑvivi+1 = 2π
FIG. 43. (a) A curve C enclosing a defect core is shown, along with the vector field m. (b)
Calculation of winding number around a vertex v on a triangulated surface : θ61 is the angle
the vector field at vertex 6 subtends with the field at vertex 1.
As in Fig. 43(a), if C is a curve enclosing a defect core and dϑ(C) the change in field
C dϑ(C).
Unlike in a continuum, the notion of a defect core has a slightly different meaning on a
orientation along a small segment dC, then the winding number is given by W =(cid:82)
number around each vertex, calculated as W (v) =(cid:80)
discrete surface since the in-plane field is well defined at all vertices of the triangulated
membrane. Hence, the singularities in the field orientation are identified using the winding
vi∈v ϑvivi+1. The relative orientation
of the vector field at vertices vi and vi+1, both in the one ring neighbourhood of vertex v,
is calculated using the parallel transport defined in section E.
111
coreCdϑm345261ϑ61v 3. Topological charge and its minimum
The topological or defect charge is a non-dimensional measure of the vorticity of the
defect defined from the winding number as
qD =
W
2π
.
(F2)
FIG. 44. Polar field on a planar surface with its texture described by m(x, y) = cos φ x+sin φ y,
with φ(r, θ) = qDθ. Shown are patterns corresponding to (a) qD = +1, (b) qD = −1, (c)
qD = +2, (d) qD = −2, (e) qD = +3, and (f ) qD = −3.
Patterns of a planar vector field corresponding to varying defect strengths are shown
in Fig. 44 and Fig. 45 for polar and nematic fields, respectively.
112
111111(c)(b)(a)(d)(e)(f) FIG. 45.
Nematic field textures illustrating the various defect structures. Shown are the
patterns with (a) qD = +1, (b) qD = −1, (c) qD = +1/2, and (d) qD = −1/2
In general, all possible topological defects of a p-atic vector field can be represented by
n/p, with n = ±1,±2, . . . .
The energy of a defect core is a function of its topological charge (qD) and core ra-
dius (Rcore).
In the case of a p-atic field, it scales quadratically with the charge and
logarithmically with the core radius and has the form [308, 311],
Hcore = πJpq2
D ln (Rcore/a) ,
(F3)
with a being the short distance cutoff. As a result of such a dependence, only defects of
strength 1/p, which minimize Hcore, are accommodated in the low-temperature equilib-
rium texture of the p-atic field. From eqn. (F1), the minimal number of defects in the
equilibrium texture of an in-plane field on a surface with Euler number χ can be shown
to be equal to χp. The textures corresponding to polar (p = 1) and nematic (p = 2)
fields on a surface of spherical topology (χ = 2) are shown in Fig. 46(a) and Fig. 46(b),
respectively.
113
1111(a)(b)(c)(d) (a)Polar field : χp = 2, qD = +1
(b)Nematic field : χp = 4, qD = +1/2
FIG. 46.
Patterns of an in-plane field on a rigid spherical surface, with the shaded regions
marking the defect core. (a) A polar field with two +1 defects at the poles of the sphere. (b)
Four +1/2 disclinations of the nematic field arrange themselves on the vertices of a tetrahedron,
resulting in the baseball texture.
4. Organization of defects of non deformable spherical surfaces
When the membrane surface is rigid, as expected, the polar field on a spherical sur-
face organizes such that the two +1 defects, arising due to topological frustration, are
positioned at the poles of the sphere, giving rise to the familiar hairy ball pattern. The
nematic field texture, on the other hand, stabilizes four disclinations, each of strength
+1/2, as shown in Fig. 46(b). Earlier computational studies on packing nematic like ob-
jects on spherical surfaces have shown that the four +1/2 disclinations are located on a
great circle inscribed on the surface of the sphere [329]. The pattern of the nematic field,
in our system, instead displays the baseball texture [312] with the defects positioned at
the vertices of a tetrahedron, as predicted in [309]. The observed difference has it origin
in the one constant approximation of the Frank’s free energy used here.
114
(a)qD = +1
(b)qD = −1
(c)qD = +1/2
(d)qD = +1/2
FIG. 47. Mugshot of four different defect cores; +1 (a) and -1 (b) defects for a polar field;
and +1/2 (c) and -1/2 (d) defects for a nematic field.
The ground state defects of a polar and nematic field, as seen in the simulations, are
presented in Fig. 47. So far, the behavior of the orientational field was investigated in
the context of non-deformable surfaces. These properties hold even when the underlying
surface deforms into complicated morphologies. The role of an in-plane field in deforming
membrane surfaces and the host of other interesting phenomena arising due to it are
investigated in the next section.
5. Non-curvature-inducing in-plane field and deformable membranes
Though the interaction of the surface vector field with the membrane is implicit, it
tends to deform the membrane into a variety of shapes. Simple arguments based on
energy minimization could shed some light on this interesting behavior. The total energy
of a field-decorated membrane Htot, previously defined in eqn. (54), contains contributions
from the elastic energy (Hsur), due to surface deformations, and the self-interaction energy
(Hp−atic), due to the field. In addition to these is the energy of the defect core, Hcore. In
the case of membrane conformations with minimal topological defects (χp) and smooth
field texture, Hcore and Hp−atic are nearly constants. The minimum of Htot would then
be seen for membrane conformations with minimum elastic energy, which corresponds to
a sphere. This indicates that, changes in the membrane shape are driven by a competing
interaction that is much stronger than Hsur.
A known source of such an implicit term is the interaction between the topological de-
fects, which are now a part of the membrane surface. Topological charges are analogous to
electric charges and hence attract when unlike and repel when alike. This interaction, for
115
defect charges on a planar surface, is known to depend logarithmically on the separation
between the defects. The arrangement of the nematic disclination on the vertices of a
tetrahedron, in Fig. 46(b), indicates that such a behavior holds for defects on a curved
surface too. Hence, the defects of the vector field maximize their spatial separation by
deforming the membrane at the cost of the much softer elastic energy.
An alternate reasoning for the deformation of a membrane by its in-plane polar field
is due to Mackintosh and Lubensky [330]. They argue that, as the field couples to the
Gaussian curvature (G) of the surface, it tends to minimize G everywhere by deforming
the surface into a cylinder, except at the defect core located at the pole.
(a) Polar field : Ellipsoidal membrane
(b) Nematic field : Tetrahedron
FIG. 48. Deformations of a stiff membrane (κ = 10) due to an ordered in-plane field; (a)
A polar field stabilizes cylindrical shapes; and (b) a tetrahedron is stabilized when the field is
nematic, with J2 = 5kBT .
Low-temperature equilibrium conformations of the membrane, in our simulations, are
in excellent agreement with earlier theoretical predictions [309, 330, 331]. Figure 48(a)
shows the predicted ellipsoidal membrane for a polar field(p = 1), whereas the tetrahedral
arrangement of defects, characteristic of a nematic field, is shown in Figure 48(b). In the
case of polar fields, the two +1 antipodal vortices are positioned at the ellipsoidal caps,
whereas the four +1/2 disclinations are at the vertices of the tetrahedron for the nematic
field. Mean field membrane shapes associated with a p-atic vector field, with p=1 through
6, are given in [331].
116
Appendix G: Regularized delta function
The discretization of a continuum membrane to quasiparticles is obtained through the
use of a regularized delta function [332]. These functions with compact support over a
given interval satisfy some (most) of the properties of the Dirac delta in the distributional
sense and are represented by δh(r), where h would denote the spread of the function’s
support.
For r ∈ Rn, δh(r) =
, ∆V is the elemental volume (discrete) and
{ri} are the components of the position vector. The function φ(x) satisfies the following
properties:
i=1..n
φ
h
1
∆V
(cid:81)
(cid:16) ri
(cid:17)
(cid:88)
(cid:88)
j
j
φ(x) = φ(−x)
φ(x − xj) = 1
xjφ(x − xj) = x
∀x
∀x
(G1)
To uniquely fix the form of φ(x), more conditions on its derivatives, support, and other
tion. One common condition is(cid:80)
features, will have to be imposed, the choice of which depends on the particular applica-
(φ(x − j))2 = C ∀x, for a C which is independent of
j
x.
[1] O. G. Mouritsen, Life - As a matter of fat : The emerging science of lipidomics, The
Frontiers collection, Springer, Germany, 2005.
[2] G. Guidotti, Membrane proteins, Ann. Rev. Biochem. 41 (1972) 731–752.
[3] G. K. Voeltz, W. A. Prinz, Sheets, ribbons and tubules - how organelles get their shape,
Nat. Rev. Mol. Cell Biol. 8 (2007) 258–264.
[4] J. A. Joyce, J. W. Pollard, Microenvironmental regulation of metastasis, Nat Rev Cancer
9 (2008) 239–252.
[5] P. V. Escrib´a, J. M. Gonz´alez-Ros, F. M. Goni, P. K. J. Kinnunen, L. Vigh, L. S´anchez-
Magraner, A. M. Fern´andez, X. Busquets, I. Horv´ath, G. Barcel´o-Coblijn, Membranes: a
meeting point for lipids, proteins and therapies, J Cellular Mol Med 12 (2008) 829–875.
117
[6] M. Cavey, M. Rauzi, P.-F. Lenne, T. Lecuit, A two-tiered mechanism for stabilization and
immobilization of e-cadherin, Nature 453 (2008) 751–756.
[7] J. J. Jefferson, C. L. Leung, R. K. H. Liem, Plakins: Goliaths that link cell junctions and
the cytoskeleton, Nature 5 (2004) 542–553.
[8] P. N. Dannhauser, E. J. Ungewickell, Reconstitution of clathrin-coated bud and vesicle
formation with minimal components, Nature Cell Biology 14 (2012) 634–639.
[9] B. Wendland, Epsins: adaptors in endocytosis?, Nat. Rev. Mol. Cell Biol. 3 (2002)
971–977.
[10] M. G. J. Ford, I. G. Mills, B. J. Peter, Y. Vallis, G. J. K. Praefcke, P. R. Evans, H. T.
McMahon, Curvature of clathrin-coated pits driven by epsin, Nature 419 (2002) 361–366.
[11] C.-L. Lai, C. C. Jao, E. Lyman, J. L. Gallop, B. J. Peter, H. T. McMahon, R. Langen,
G. A. Voth, Membrane binding and self-association of the epsin n-terminal homology
domain, Journal of Molecular Biology 423 (2012) 800–817.
[12] B. D. Angst, C. Marcozzi, A. I. Magee, The cadherin superfamily: diversity in form and
function, Journal of Cell Science 114 (2001) 629–641.
[13] M. Causeret, N. Taulet, F. Comunale, C. Favard, C. Gauthier-Rouvi`ere, N-cadherin
association with lipid rafts regulates its dynamic assembly at cell-cell junctions in c2c12
myoblasts, Mol. Biol. Cell 16 (2005) 2168–2180.
[14] M. G. M´arquez, N. O. Favale, F. L. Nieto, L. G. Pescio, N. Sterin-Speziale, Changes in
membrane lipid composition cause alterations in epithelial cell-cell adhesion structures in
renal papillary collecting duct cells, BBA - Biomembranes 1818 (2012) 491–501.
[15] B. Alberts, A. Johnson, J. Lewis, M. Raff, K. Roberts, P. Walter, Molecular Biology of
the cell, Garland Publishing, Singapore, third edition, 1994.
[16] G. van Meer, D. R. Voelker, G. W. Feigenson, Membrane lipids: where they are and how
they behave, Nat. Rev. Mol. Cell Biol. 9 (2008) 112–124.
[17] R. V. Farese, T. C. Walther, Lipid droplets finally get a little r-e-s-p-e-c-t, Cell 139 (2009)
855–860.
[18] J. N. Israelachvili, D. J. Mitchell, B. W. Ninham, Theory of self-assembly of lipid bilayers
and vesicles, Biochimica Biophysics Acta (BBA) - Biomembranes 470 (1977) 185–201.
[19] R. Koynova, M. Caffrey, An index of lipid phase diagrams, Chem. Phys. Lipids 115 (2002)
107–219.
118
[20] S. J. Singer, G. L. Nicolson, The fluid mosaic model of the structure of cell membranes,
Science 175 (1972) 720–731.
[21] A. A. Spector, M. A. Yorek, Membrane lipid-composition and cellular function, The
Journal of Lipid Research 26 (1985) 1015–1035.
[22] S. Paula, A. G. Volkov, A. N. Van Hoek, T. H. Haines, D. W. Deamer, Permeation of
protons, potassium ions, and small polar molecules through phospholipid bilayers as a
function of membrane thickness, Biophys. J. 70 (1996) 339–348.
[23] A. Finkelstein, Water and nonelectrolyte permeability of lipid bilayer membranes, J. Gen.
Physiol. 68 (1976) 127.
[24] A. R. Honerkamp-Smith, S. L. Veatch, S. L. Keller, An introduction to critical points for
biophysicists; observations of compositional heterogeneity in lipid membranes, Biochimica
et Biophysica Acta (BBA) - Biomembranes 1788 (2009) 53–63.
[25] S. Semrau, T. Schmidt, Membrane heterogeneity - from lipid domains to curvature effects,
Soft Matter 5 (2009) 3174.
[26] S. Munro, Lipid rafts: Elusive or illusive?, Cell 115 (2003) 377–388.
[27] M. Edidin, Lipids on the frontier: a century of cell-membrane bilayers, Nat. Rev. Mol.
Cell Biol. 4 (2003) 414–418.
[28] H.-J. Kaiser, D. Lingwood, I. Levental, J. L. Sampaio, L. Kalvodova, L. Rajendran, K. Si-
mons, Order of lipid phases in model and plasma membranes, Proc. Natl. Acad. Sci. USA.
106 (2009) 16645–16650.
[29] S. L. Veatch, S. L. Keller, Organization in lipid membranes containing cholesterol, Phys.
Rev. Lett. 89 (2002) 268101.
[30] L. Bagatolli, P. B. Sunil Kumar, Phase behavior of multicomponent membranes: Experi-
mental and computational techniques, Soft Matter 5 (2009) 3234.
[31] S. L. Veatch, S. L. Keller, Separation of liquid phases in giant vesicles of ternary mixtures
of phospholipids and cholesterol, Biophys. J. 85 (2003) 3074–3083.
[32] F. M. Goni, A. Alonso, L. A. Bagatolli, R. E. Brown, D. Marsh, M. Prieto, J. L. Thewalt,
Phase diagrams of lipid mixtures relevant to the study of membrane rafts, Biochimica et
Biophysica Acta (BBA) - Molecular and Cell Biology of Lipids 1781 (2008) 665–684.
[33] T. Hamada, Y. Kishimoto, T. Nagasaki, M. Takagi, Lateral phase separation in tense
membranes, Soft Matter 7 (2011) 9061.
119
[34] T. S. Ursell, W. S. Klug, R. Phillips, Morphology and interaction between lipid domains,
Proc. Natl. Acad. Sci. USA. 106 (2009) 13301–13306.
[35] K. Simons, D. Toomre, Lipid rafts and signal transduction, Nat. Rev. Mol. Cell Biol. 1
(2000) 31–39.
[36] K. Simons, J. L. Sampaio, Membrane organization and lipid rafts, Cold Spring Harb
Perspect Biol (2011).
[37] K. Simons, W. L. C. Vaz, Model systems, lipid rafts, and cell membranes, Ann. Rev.
Biophys. Biomol. Struct. 33 (2004) 269–295.
[38] K. Simons, E. Ikonen, Functional rafts in cell membranes, Nature 387 (1997) 569–572.
[39] J. Kas, E. Sackmann, Shape transitions and shape stability of giant phospholipid vesicles
in pure water induced by area-to-volume changes, Biophys. J. 60 (1991) 825–844.
[40] H. Hotani, Transformation pathways of liposomes, Journal of Molecular Biology 178
(1984) 113–120.
[41] R. Lipowsky, The conformation of membranes, Nature 349 (1991) 475.
[42] U. Seifert, Configurations of fluid membranes and vesicles, Adv. Phys. 46 (1997) 13.
[43] P. F. Devaux, Static and dynamic lipid asymmetry in cell membranes, Biochemistry 30
(1991) 1163–1173.
[44] D. R. Slochower, Y.-H. Wang, R. W. Tourdot, R. Radhakrishnan, P. A. Janmey,
Counterion-mediated pattern formation in membranes containing anionic lipids, Advances
in Colloid and Interface Science (2014) in press. DOI 10.1016/j.cis.2014.01.016.
[45] P. B. Canham, The minimum energy of bending as a possible explanation of the biconcave
shape of the human red blood cell, J. Theor. Biol. 26 (1970) 61–81.
[46] W. Helfrich, Elastic properties of lipid bilayers: theory and possible experiments, Z.
Naturforsch. C 28 (1973) 1–12.
[47] L. D. Landau, E. M. Lifshitz, Theory of Elasticity, Pergamon Press, Oxford, 1970.
[48] A. C. Woodka, P. D. Butler, L. Porcar, B. Farago, M. Nagao, Lipid bilayers and membrane
dynamics: Insight into thickness fluctuations, Phys. Rev. Lett. 109 (2012) 058102.
[49] F. Jahnig, What is the surface tension of a lipid bilayer membrane?, Biophys. J. 71 (1996)
1348–1349.
[50] M. P. do Carmo, Differential geometry of curves and surfaces, Prentice Hall, Engelwood
Cliffs, New Jersey, 1976.
120
[51] W. Helfrich, Effect of thermal undulations on the rigidity of fluid membranes and inter-
faces, J. Phys. France 46 (1985) 1263–1268.
[52] L. Peliti, S. Leibler, Effects of thermal fluctuations on systems with small surface tension,
Phys. Rev. Lett. 54 (1985) 1690–1693.
[53] D. Foster, On the scale dependence, due to thermal fluctuations, of the elastic properties
of membranes, Phys. Lett. A 114 (1986) 115–120.
[54] H. Kleinert, Thermal softening of curvature elasticity in membranes, Phys. Lett. 114A
(1986) 263–268.
[55] P. G. De Gennes, C. Taupin, Microemulsions and the flexibility of oil/water interfaces, J.
Phys. Chem. 86 (1982) 2294–2304.
[56] W. Helfrich, Size distributions of vesicles : the role of the effective rigidity of membranes,
J. Phys. France 47 (1986) 321–329.
[57] W. Helfrich, Measures of integration in calculating the effective rigidity of fluid surfaces,
J. Phys. France 48 (1987) 285–289.
[58] W. Helfrich, Stiffening of fluid membranes and entropy loss of membrane closure: Two
effects of thermal undulations, Eur. Phys. J. B 1 (1998) 481.
[59] H. A. Pinnow, W. Helfrich, Effect of thermal undulations on the bending elasticity and
spontaneous curvature of fluid membranes, Eur. Phys. J. E 3 (2000) 149–157.
[60] D. Marsh, Renormalization of the tension and area expansion modulus in fluid membranes,
Biophys. J. 73 (1997) 865–869.
[61] D. Marsh, Elastic curvature constants of lipid monolayers and bilayers, Chemistry and
Physics of Lipids 144 (2006) 146–159.
[62] W. Cai, T. Lubensky, Covariant hydrodynamics of fluid membranes, Phys. Rev. Lett. 73
(1994) 1186–1189.
[63] W. Cai, T. Lubensky, Hydrodynamics and dynamic fluctuations of fluid membranes, Phys.
Rev. E 52 (1995) 4251–4266.
[64] K. A. Dill, S. Bromberg, Molecular driving forces, Garland Science, New York, 2003.
[65] F. David, S. Leibler, Vanishing tension of fluctuating membranes, J. Phys. II France 1
(1991) 959–976.
[66] D. Nelson, T. Piran, S. Weinberg (Eds.), Statistical Mechanics of Membranes and Surfaces,
World Scientific, Singapore, second edition, 2003.
121
[67] L. Miao, U. Seifert, M. Wortis, H.-G. Dobereiner, Budding transitions of fluid-bilayer
vesicles - the effect of area-difference elasticity, Phys. Rev. E 49 (1994) 5389–5407.
[68] P. M. Chaikin, T. C. Lubensky, Principles of condensed matter physics, Cambridge Uni-
versity Press, Cambridge ; New York, 1995.
[69] E. Reister-Gottfried, S. Leitenberger, U. Seifert, Hybrid simulations of lateral diffusion in
fluctuating membranes, Phys. Rev. E 75 (2007) 011908.
[70] L. C.-L. Lin, F. L. Brown, Brownian dynamics in fourier space: Membrane simulations
over long length and time scales, Phys. Rev. Lett. 93 (2004) 256001.
[71] F. L. H. Brown, Elastic modeling of biomembranes and lipid bilayers, Annu. Rev. Phys.
Chem. 59 (2008) 685–712.
[72] E. Evans, W. Rawicz, Entropy-driven tension and bending elasticity in condensed-fluid
membranes, Phys. Rev. Lett. 64 (1990) 2094–2097.
[73] E. Evans, H. Bowman, A. Leung, D. Needham, D. Tirrell, Biomembrane templates for
nanoscale conduits and networks, Science 273 (1996) 933–935.
[74] I. Der´enyi, F. Julicher, J. Prost, Formation and interaction of membrane tubes, Phys.
Rev. Lett. 88 (2002) 238101.
[75] T. Powers, G. Huber, R. Goldstein, Fluid-membrane tethers: Minimal surfaces and elastic
boundary layers, Phys. Rev. E 65 (2002) 041901.
[76] J. M. Allain, C. Storm, A. Roux, M. Amar, J. F. Joanny, Fission of a multiphase membrane
tube, Phys. Rev. Lett. 93 (2004) 158104.
[77] O. Zhong-can, W. Helfrich, Bending energy of vesicle membranes: General expressions for
the first, second, and third variation of the shape energy and applications to spheres and
cylinders, Phys. Rev. A 39 (1989) 5280–5288.
[78] F. Julicher, U. Seifert, Shape equations for axisymmetric vesicles: A clarification, Phys.
Rev. E 49 (1994) 4728–4731.
[79] F. Julicher, R. Lipowsky, Shape transformations of vesicles with intramembrane domains,
Phys. Rev. E 53 (1996) 2670–2683.
[80] U. Seifert, K. Berndl, R. Lipowsky, Shape transformations of vesicles: Phase diagram for
spontaneous- curvature and bilayer-coupling models, Phys. Rev. A 44 (1991) 1182.
[81] H. Gao, W. Shi, L. B. Freund, Mechanics of receptor-mediated endocytosis, Proc. Natl.
Acad. Sci. USA. 102 (2005) 9469–9474.
122
[82] Z.-L. Liu, K.-L. Yao, X.-B. Jing, X.-A. Li, X.-Z. Sun, Endocytic vesicle scission by lipid
phase boundary forces, Proc. Natl. Acad. Sci. USA. 103 (2006) 10277–10282.
[83] J. Liu, M. Kaksonen, D. G. Drubin, G. Oster, Endocytic vesicle scission by lipid phase
boundary forces, Proc. Natl. Acad. Sci. USA. 103 (2006) 10277–10282.
[84] N. J. Agrawal, J. Nukpezah, R. Radhakrishnan, Minimal mesoscale model for protein-
mediated vesiculation in clathrin-dependent endocytosis, PLoS Comput. Biol. 6 (2010)
e1000926.
[85] K. A. Brakke, The surface evolver, Experimental mathematics 1 (1992) 141–165.
[86] S. Dasgupta, T. Auth, G. Gompper, Wrapping of ellipsoidal nano-particles by fluid mem-
branes, Soft Matter 9 (2013) 5473.
[87] S. Dasgupta, T. Auth, G. Gompper, Shape and orientation matter for the cellular uptake
of nonspherical particles, Nano Lett. 14 (2014) 687–693.
[88] L. C.-L. Lin, F. Brown, Brownian dynamics in fourier space: Membrane simulations over
long length and time scales, Phys. Rev. Lett. 93 (2004) 256001.
[89] L. C.-L. Lin, F. Brown, Dynamic simulations of membranes with cytoskeletal interactions,
Phys. Rev. E 72 (2005) 011910.
[90] L. C.-L. Lin, F. L. H. Brown, Simulating membrane dynamics in nonhomogeneous hydro-
dynamic environments, J. Chem. Theory Comput. 2 (2006) 472–483.
[91] F. L. H. Brown, Simple models for biomembrane structure and dynamics, Computer
Physics Communications 177 (2007) 172–175.
[92] J. K. Sigurdsson, F. L. H. Brown, P. J. Atzberger, Hybrid continuum-particle method for
fluctuating lipid bilayer membranes with diffusing protein inclusions, J. Comp. Phys. 252
(2013) 65–85.
[93] A. M. Polyakov, Quantum geometry of fermionic strings, Phys. Lett. B 103 (1981) 207 –
210.
[94] F. David, Randomly triangulated surfaces in - 2 dimensions, Phys. Lett. B 159 (1985)
303 – 306.
[95] V. A. Kazakov, I. K. Kostov, A. A. Migdahl, Critical properties of randomly triangulated
planar random surfaces, Phys. Lett. B 157 (1985) 295–300.
[96] A. Maritan, A. L. Stella, Some exact results from self-avoiding random surfaces, Nucl.
Phys. B 280 (1987) 561 – 575.
123
[97] Y. Kantor, M. Kardar, D. R. Nelson, Statistical mechanics of tethered surfaces, Phys.
Rev. Lett. 57 (1986) 791.
[98] Y. Kantor, D. R. Nelson, Phase transitions in flexible polymeric surfaces, Phys. Rev. A
36 (1987) 4020.
[99] J.-S. Ho, A. Baumgartner, Simulations of fluid self-avoiding membranes, Europhys. Lett.
12 (1990) 295.
[100] J.-S. Ho, A. Baumgartner, Crumpling of fluid vesicles, Phys. Rev. A 41 (1990) 5747 –
5750.
[101] N. Ramakrishnan, P. B. Sunil Kumar, J. H. Ipsen, Monte carlo simulations of fluid vesicles
with in-plane orientational ordering, Phys. Rev. E 81 (2010) 041922.
[102] D. Frenkel, B. Smit, Understanding Molecular Simulation : From Algorithms to Applica-
tions, Academic Press, 2 edition, 2001.
[103] N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, E. Teller, Equation of
state calculations by fast computing machines, J. Chem. Phys. 21 (1953) 1087–1092.
[104] D. Kroll, G. Gompper, Scaling behavior of randomly triangulated self-avoiding surfaces,
Phys. Rev. A 46 (1992) 3119–3122.
[105] G. Gompper, D. Kroll, Phase diagram and scaling behavior of fluid vesicles, Phys. Rev.
E 51 (1995) 514–525.
[106] J. Paulose, G. A. Vliegenthart, G. Gompper, D. R. Nelson, Fluctuating shells under
pressure, Proc. Natl. Acad. Sci. USA. 109 (2012) 19551–19556.
[107] J. M. Drouffe, A. C. Maggs, S. Leibler, Computer simulations of self-assembled mem-
branes, Science 254 (1991) 1353–1356.
[108] G. Ayton, G. Voth, Bridging microscopic and mesoscopic simulations of lipid bilayers,
Biophys. J. 83 (2002) 3357–3370.
[109] H. Noguchi, M. Takasu, Self-assembly of amphiphiles into vesicles: A brownian dynamics
simulation, Phys. Rev. E 64 (2001) 041913.
[110] H. Noguchi, M. Takasu, Adhesion of nanoparticles to vesicles: a brownian dynamics
simulation, Biophys. J. 83 (2002) 299–308.
[111] O. Farago, “water-free” computer model for fluid bilayer membranes, J. Chem. Phys. 119
(2003) 596.
[112] I. R. Cooke, M. Deserno, Solvent-free model for self-assembling fluid bilayer membranes:
124
Stabilization of the fluid phase based on broad attractive tail potentials, J. Chem. Phys.
123 (2005) 4710.
[113] G. Brannigan, F. L. H. Brown, Solvent-free simulations of fluid membrane bilayers, J.
Chem. Phys. 120 (2004) 1059.
[114] G. S. Ayton, J. L. McWhirter, G. A. Voth, A second generation mesoscopic lipid bilayer
model: Connections to field-theory descriptions of membranes and nonlocal hydrodynam-
ics, J. Chem. Phys. 124 (2006) 064906.
[115] H. Noguchi, G. Gompper, Meshless membrane model based on the moving least-squares
method, Phys. Rev. E 73 (2006) 21903.
[116] T. Kohyama, Simulations of flexible membranes using a coarse-grained particle-based
model with spontaneous curvature variables, Physica A 388 (2009) 3334–3344.
[117] H. Yuan, C. Huang, J. Li, G. Lykotrafitis, S. Zhang, One-particle-thick, solvent-free,
coarse-grained model for biological and biomimetic fluid membranes, Phys. Rev. E 82
(2010) 011905.
[118] G. S. Ayton, P. D. Blood, G. A. Voth, Membrane remodeling from n-bar domain interac-
tions: Insights from multi-scale simulation, Biophys. J. 92 (2007) 3595–3602.
[119] G. S. Ayton, E. Lyman, V. Krishna, R. D. Swenson, C. Mim, V. M. Unger, G. A. Voth,
New insights into bar domain-induced membrane remodeling, Biophys. J. 97 (2009) 1616–
1625.
[120] H. Cui, G. S. Ayton, G. A. Voth, Membrane binding by the endophilin n-bar domain,
Biophys. J. 97 (2009) 2746–2753.
[121] H. Cui, E. Lyman, G. A. Voth, Mechanism of membrane curvature sensing by amphipathic
helix containing proteins, Biophys. J. 100 (2011) 1271–1279.
[122] E. Lyman, H. Cui, G. A. Voth, Reconstructing protein remodeled membranes in molecular
detail from mesoscopic models, Phys. Chem. Chem. Phys. 13 (2011) 10430–10436.
[123] G. S. Ayton, H. L. Tepper, D. T. Mirijanian, G. A. Voth, A new perspective on the
coarse-grained dynamics of fluids, J. Chem. Phys. 120 (2004) 4074.
[124] D. J. Tobias, K. Tu, M. L. Klein, Atomic-scale molecular dynamics simulations of lipid
membranes, Current Opinion in Colloid & Interface Science 2 (1997) 15–26.
[125] D. P. Tieleman, S.-J. Marrink, H. J. Berendsen, A computer perspective of membranes:
molecular dynamics studies of lipid bilayer systems, Biochimica Biophysics Acta (BBA) -
125
Reviews on Biomembranes 1331 (1997) 235–270.
[126] S. E. Feller, Molecular dynamics simulations of lipid bilayers, Curr Opin Colloid in 5
(2000) 217–223.
[127] S.-J. Marrink, A. H. de Vries, D. P. Tieleman, Lipids on the move: simulations of mem-
brane pores, domains, stalks and curves, Biochimica Biophysics Acta (BBA) - Biomem-
branes 1788 (2009) 149–168.
[128] P. van der Ploeg, Molecular dynamics simulation of a bilayer membrane, J. Chem. Phys.
76 (1982) 3271–3276.
[129] H. Heller, M. Schaefer, K. Schulten, Molecular dynamics simulation of a bilayer of 200
lipids in the gel and in the liquid crystal phase, J. Phys. Chem. 97 (1993) 8343–8360.
[130] K. Tu, D. J. Tobias, J. K. Blasie, M. L. Klein, Molecular dynamics investigation of the
structure of a fully hydrated gel-phase dipalmitoylphosphatidylcholine bilayer, Biophys.
J. (1996).
[131] D. P. Tieleman, H. J. C. Berendsen, Molecular dynamics simulations of a fully hydrated
dipalmitoylphosphatidylcholine bilayer with different macroscopic boundary conditions
and parameters, J. Chem. Phys. 105 (1996) 4871.
[132] O. Berger, O. Edholm, F. Jahnig, Molecular dynamics simulations of a fluid bilayer of
dipalmitoylphosphatidylcholine at full hydration, constant pressure, and constant temper-
ature, Biophys. J. (1997).
[133] S.-J. Marrink, O. Berger, P. Tieleman, F. Jahnig, Adhesion forces of lipids in a phos-
pholipid membrane studied by molecular dynamics simulations, Biophys. J. 74 (1998)
931–943.
[134] M. Pasenkiewicz-Gierula, T. R´og, K. Kitamura, A. Kusumi, Cholesterol effects on the
phosphatidylcholine bilayer polar region: A molecular simulation study, Biophys. J. 78
(2000) 1376–1389.
[135] R. J. Mashl, H. L. Scott, S. Subramaniam, E. Jakobsson, Molecular simulation of di-
oleoylphosphatidylcholine lipid bilayers at differing levels of hydration, Biophys. J. 81
(2001) 3005–3015.
[136] K. Murzyn, T. Rog, G. Jezierski, Y. Takaoka, M. Pasenkiewicz-Gierula, Effects of phos-
pholipid unsaturation on the membrane/water interface: a molecular simulation study,
Biophys. J. 81 (2001) 170.
126
[137] S.-J. Marrink, A. E. Mark, Molecular dynamics simulation of the formation, structure, and
dynamics of small phospholipid vesicles, J. Amer. Chem. Soc. 125 (2003) 15233–15242.
[138] C. Hofsass, E. Lindahl, O. Edholm, Molecular dynamics simulations of phospholipid bi-
layers with cholesterol, Biophys. J. 84 (2003) 2192.
[139] A. H. de Vries, A. E. Mark, S.-J. Marrink, Molecular dynamics simulation of the sponta-
neous formation of a small dppc vesicle in water in atomistic detail, J. Amer. Chem. Soc.
(2004).
[140] H. Leontiadou, A. E. Mark, S.-J. Marrink, Molecular dynamics simulations of hydrophilic
pores in lipid bilayers, Biophys. J. 86 (2004) 2156–2164.
[141] S.-J. Marrink, A. E. Mark, The mechanism of vesicle fusion as revealed by molecular
dynamics simulations, J. Amer. Chem. Soc. (2003).
[142] A. K. Sum, R. Faller, J. J. de Pablo, Molecular simulation study of phospholipid bilayers
and insights of the interactions with disaccharides, Biophys. J. 85 (2003) 2830–2844.
[143] O. Edholm, O. Berger, F. Jahnig, Structure and fluctuations of bacteriorhodopsin in the
purple membrane: a molecular dynamics study, J. Mol. Biol. (1995).
[144] M. C. Pitman, A. Grossfield, F. Suits, S. E. Feller, Role of cholesterol and polyunsatu-
rated chains in lipidprotein interactions: Molecular dynamics simulation of rhodopsin in
a realistic membrane environment, J. Amer. Chem. Soc. 127 (2005) 4576–4577.
[145] M. S. P. Sansom, P. J. Bond, S. S. Deol, A. Grottesi, S. Haider, Z. A. Sands, Molecu-
lar simulations and lipid–protein interactions: potassium channels and other membrane
proteins, Biochem Soc Trans 33 (2005) 916.
[146] E. Lindahl, M. Sansom, Membrane proteins: molecular dynamics simulations, Current
Opinion in Structural Biology 18 (2008) 425–431.
[147] F. J.-M. de Meyer, M. Venturoli, B. Smit, Molecular simulations of lipid-mediated protein-
protein interactions, Biophys. J. 95 (2008) 1851–1865.
[148] P. Lague, B. Roux, R. W. Pastor, Molecular dynamics simulations of the influenza hemag-
glutinin fusion peptide in micelles and bilayers: conformational analysis of peptide and
lipids, J. Mol. Biol. (2005).
[149] M. Doi, S. F. Edwards, The theory of polymer dynamics, Oxford University Press, Oxford,
UK, 1988.
[150] R. Goetz, R. Lipowsky, Computer simulations of bilayer membranes: Self-assembly and
127
interfacial tension, J. Chem. Phys. 108 (1998) 7397–7409.
[151] M. J. Stevens, Coarse-grained simulations of lipid bilayers, J. Chem. Phys. 121 (2004)
11942–11948.
[152] A. P. Lyubartsev, Multiscale modeling of lipids and lipid bilayers, Eur. Biophys. J. 35
(2005) 53–61.
[153] M. Praprotnik, L. Delle Site, K. Kremer, Multiscale simulation of soft matter: From scale
bridging to adaptive resolution, Annu. Rev. Phys. Chem. 59 (2008) 545–571.
[154] C. Peter, K. Kremer, Multiscale simulation of soft matter systems–from the atomistic to
the coarse-grained level and back, Soft Matter (2009).
[155] C. Peter, K. Kremer, Multiscale simulation of soft matter systems, Faraday Discuss. 144
(2009) 9–24.
[156] T. Murtola, E. Falck, M. Patra, M. Karttunen, I. Vattulainen, Coarse-grained model for
phospholipid/cholesterol bilayer, J. Chem. Phys. 121 (2004) 9156.
[157] T. Murtola, E. Falck, M. Karttunen, I. Vattulainen, Coarse-grained model for phospho-
lipid/cholesterol bilayer employing inverse monte carlo with thermodynamic constraints,
J. Chem. Phys. 126 (2007) 075101.
[158] A. Y. Shih, A. Arkhipov, P. L. Freddolino, K. Schulten, Coarse grained proteinlipid model
with application to lipoprotein particles , J. Phys. Chem. B 110 (2006) 3674–3684.
[159] A. Arkhipov, Y. Yin, K. Schulten, Four-scale description of membrane sculpting by bar
domains, Biophys. J. 95 (2008) 2806–2821.
[160] Y. Yin, A. Arkhipov, K. Schulten, Simulations of membrane tubulation by lattices of
amphiphysin n-bar domains, Structure 17 (2009) 882–892.
[161] S. Izvekov, G. A. Voth, Multiscale coarse graining of liquid-state systems, J. Chem. Phys.
123 (2005) 134105.
[162] W. G. Noid, J.-W. Chu, G. S. Ayton, V. Krishna, S. Izvekov, G. A. Voth, A. Das, H. C.
Andersen, The multiscale coarse-graining method. i. a rigorous bridge between atomistic
and coarse-grained models, J. Chem. Phys. 128 (2008) 244114.
[163] S.-J. Marrink, A. H. de Vries, A. E. Mark, Coarse grained model for semiquantitative
lipid simulations, J. Phys. Chem. B 108 (2004) 750–760.
[164] S.-J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tieleman, A. H. de Vries, The martini
force field: Coarse grained model for biomolecular simulations, J. Phys. Chem. B 111
128
(2007) 7812–7824.
[165] L. Monticelli, S. K. Kandasamy, X. Periole, R. G. Larson, D. P. Tieleman, S.-J. Marrink,
The martini coarse-grained force field: Extension to proteins, J. Chem. Theory Comput.
4 (2008) 819–834.
[166] B. J. Reynwar, G. Illya, V. A. Harmandaris, M. M. Muller, K. Kremer, M. Deserno,
Aggregation and vesiculation of membrane proteins by curvature-mediated interactions,
Nature 447 (2007) 461–464.
[167] J. C. Shillcock,
Insight or illusion? seeing inside the cell with mesoscopic simulations,
HFSP journal 2 (2008) 1–6.
[168] J. C. Shillcock, Vesicles and vesicle fusion: coarse-grained simulations, Methods Mol.
Biol. 924 (2013) 659–697.
[169] R. V´acha, F. J. Martinez-Veracoechea, D. Frenkel, Intracellular release of endocytosed
nanoparticles upon a change of ligand-receptor interaction, ACS Nano 6 (2012) 10598–
10605.
[170] M. Klein, W. Shinoda, Large-scale molecular dynamics simulations of self-assembling
systems, Science 321 (2008) 798–800.
[171] D. Henderson, Fundamentals of Inhomogeneous Fluids, CRC Press, 1992.
[172] P. Schofield, J. R. Henderson, Statistical mechanics of inhomogeneous fluids, Proc. R.
Soc. Lon. A 379 (1982) 231.
[173] J. Walton, K. E. Gubbins, The pressure tensor in an inhomogeneous fluid of non-spherical
molecules, Molecular Physics (1985).
[174] G. Rossi, M. Testa, The stress tensor in thermodynamics and statistical mechanics, J.
Chem. Phys. (2009).
[175] J. Irving, J. Kirkwood, The statistical mechanical theory of transport processes. the
equations of hydrodynamics, J. Chem. Phys. 18 (1950) 817.
[176] F. Varnik, J. Baschnagel, K. Binder, Molecular dynamics results on the pressure tensor
of polymer films, J. Chem. Phys. 113 (2000) 4444.
[177] M. Venturoli, M. Maddalena Sperotto, M. Kranenburg, B. Smit, Mesoscopic models of
biological membranes, Phys. Reports 437 (2006) 1–54.
[178] O. H. S. Ollila, H. J. Risselada, M. Louhivuori, E. Lindahl, I. Vattulainen, S.-J. Marrink,
3d pressure field in lipid membranes and membrane-protein complexes, Phys. Rev. Lett.
129
102 (2009) 078101.
[179] A. F. Jakobsen, O. G. Mouritsen, G. Besold, Artifacts in dynamical simulations of coarse-
grained model lipid bilayers, J. Chem. Phys. 122 (2005) 204901.
[180] M. Venturoli, M. M. Sperotto, M. Kranenburg, Mesoscopic models of biological mem-
branes, Phys. Reports (2006).
[181] A. Grafmuller, J. Shillcock, R. Lipowsky, Pathway of membrane fusion with two tension
dependent energy barriers, Phys. Rev. Lett. 98 (2007) 218107.
[182] S. A. Safran, Curvature elasticity of thin films, Advances in Physics 48 (1999) 395–448.
[183] I. Szleifer, D. Kramer, A. Benshaul, W. M. Gelbart, S. A. Safran, Molecular theory of
curvature elasticity in surfactant films, J. Chem. Phys. 92 (1990) 6800–6817.
[184] R. Goetz, R. Lipowsky, Computer simulations of bilayer membranes: Self-assembly and
interfacial tension, J. Chem. Phys. 108 (1998) 7397.
[185] O. Farago, P. Pincus, Statistical mechanics of bilayer membrane with a fixed projected
area, J. Chem. Phys. 120 (2004) 2934–2950.
[186] M. Hu, D. H. de Jong, S.-J. Marrink, M. Deserno, Gaussian curvature elasticity determined
from global shape transformations and local stress distributions: a comparative study
using the martini model, Faraday Discuss. 161 (2012) 365–382.
[187] M. Hu, J. J. Briguglio, M. Deserno, Determining the gaussian curvature modulus of lipid
membranes in simulations, Biophys. J. 102 (2012) 1403–1410.
[188] M. Hu, P. Diggins, IV, M. Deserno, Determining the bending modulus of a lipid membrane
by simulating buckling, J. Chem. Phys. 138 (2013) 214110.
[189] Y. Shibata, J. Hu, M. M. Kozlov, T. A. Rapoport, Mechanisms shaping the membranes
of cellular organelles, Ann. Rev. Cell Dev. Biol. 25 (2009) 329–354.
[190] M. M. Kozlov, Biophysics: Joint effort bends membrane, Nature 463 (2010) 439–440.
[191] J. Zimmerberg, M. M. Kozlov, How proteins produce cellular membrane curvature, Nat.
Rev. Mol. Cell Biol. 7 (2006) 9–19.
[192] P. Wiggins, R. Phillips, Analytic models for mechanotransduction: gating a mechanosen-
sitive channel, Proc. Natl. Acad. Sci. USA. 101 (2004) 4071–4076.
[193] S. Aimon, A. Callan-Jones, A. Berthaud, M. Pinot, G. E. S. Toombes, P. Bassereau,
Membrane shape modulates transmembrane protein distribution, Developmental Cell 28
(2014) 212–218.
130
[194] G. J. K. Praefcke, H. T. McMahon, The dynamin superfamily: universal membrane
tubulation and fission molecules?, Nat. Rev. Mol. Cell Biol. 5 (2004) 133–147.
[195] Y. Shibata, C. Voss, J. M. Rist, J. Hu, T. A. Rapoport, W. A. Prinz, G. K. Voeltz, The
reticulon and dp1/yop1p proteins form immobile oligomers in the tubular endoplasmic
reticulum, J. Biol. Chem. 283 (2008) 18892–18904.
[196] J. Hu, Y. Shibata, C. Voss, T. Shemesh, Z. Li, M. Coughlin, M. M. Kozlov, T. A. Rapoport,
W. A. Prinz, Membrane proteins of the endoplasmic reticulum induce high-curvature
tubules, Science 319 (2008) 1247–1250.
[197] F. Campelo, H. McMahon, M. Kozlov, The hydrophobic insertion mechanism of membrane
curvature generation by proteins, Biophys. J. 95 (2008) 2325.
[198] J. C. Dawson, J. A. Legg, L. M. Machesky, Bar domain proteins: a role in tubulation,
scission and actin assembly in clathrin-mediated endocytosis, Trends Cell Biol. 16 (2006)
493–498.
[199] K. Farsad, N. Ringstad, K. Takei, S. R. Floyd, K. Rose, P. D. Camilli, Generation of high
curvature membranes mediated by direct endophilin bilayer interactions, J. Cell Biol. 155
(2001) 193–200.
[200] B. J. Peter, H. M. Kent, I. G. Mills, Y. Vallis, P. J. G. Butler, P. R. Evans, H. T. McMahon,
Bar domains as sensors of membrane curvature: the amphiphysin bar structure, Science
303 (2004) 495–499.
[201] B. Habermann, The bar-domain family of proteins: a case of bending and binding?,
EMBO Rep. 5 (2004) 250–255.
[202] A. Frost, V. M. Unger, P. de Camilli, The bar domain superfamily: membrane-molding
macromolecules, Cell 137 (2009) 191–196.
[203] J. L. Gallop, C. C. Jao, H. M. Kent, P. J. G. Butler, P. R. Evans, R. Langen, H. T.
McMahon, Mechanism of endophilin n-bar domain-mediated membrane curvature, EMBO
J. 25 (2006) 2898–2910.
[204] W. M. Henne, H. M. Kent, M. G. J. Ford, B. G. Hegde, O. Daumke, P. J. G. Butler,
R. Mittal, R. Langen, P. R. Evans, H. T. McMahon, Structure and analysis of fcho2
f-bar domain: a dimerizing and membrane recruitment module that effects membrane
curvature, Structure 15 (2007) 839–852.
[205] A. Frost, P. de Camilli, V. M. Unger, F-bar proteins join the bar family fold, Structure
131
15 (2007) 751–753.
[206] A. Frost, R. Perera, A. Roux, K. Spasov, O. Destaing, E. H. Egelman, P. de Camilli,
V. M. Unger, Structural basis of membrane invagination by f-bar domains, Cell 132
(2008) 807–817.
[207] S. Ahmed, W. I. Goh, W. Bu, I-bar domains, irsp53 and filopodium formation, Semin.
Cell Dev. Biol. 21 (2010) 350–356.
[208] W. Weissenhorn, Crystal structure of the endophilin-a1 bar domain, J. Mol. Biol. 351
(2005) 653–661.
[209] W. Huttner, A. Schmidt, Membrane curvature: a case of endofeelin’, Trends Cell Biol. 12
(2002) 155–158.
[210] M. Masuda, S. Takeda, M. Sone, T. Ohki, H. Mori, Y. Kamioka, N. Mochizuki, En-
dophilin bar domain drives membrane curvature by two newly identified structure-based
mechanisms, EMBO J. 25 (2006) 2889–2897.
[211] J. Zimmerberg, S. McLaughlin, Membrane curvature: How bar domains bend bilayers,
Current Biology 14 (2004) R250–R252.
[212] P. D. Blood, G. A. Voth, Direct observation of bin/amphiphysin/rvs (bar) domain-induced
membrane curvature by means of molecular dynamics simulations, Proc. Natl. Acad. Sci.
USA. 103 (2006) 15068–15072.
[213] A. Arkhipov, Y. Yin, K. Schulten, Four-scale description of membrane sculpting by bar
domains, Biophys. J. 95 (2008) 2806–21.
[214] A. Arkhipov, Y. Yin, K. Schulten, Membrane-bending mechanism of amphiphysin n-bar
domains, Biophys. J. 97 (2009) 2727–2735.
[215] R. N. Collins, How the er stays in shape, Cell 124 (2006) 464–466.
[216] H. T. McMahon, J. L. Gallop, Membrane curvature and mechanisms of dynamic cell
membrane remodelling, Nature 438 (2005) 590–6.
[217] Y. Shibata, G. K. Voeltz, T. A. Rapoport, Rough sheets and smooth tubules, Cell 126
(2006) 435–439.
[218] Y. Shibata, T. Shemesh, W. A. Prinz, M. M. Kozlov, T. A. Rapoport, Mechanisms
determining the morphology of the peripheral er, Cell 143 (2010) 774–788.
[219] T. R. Graham, M. M. Kozlov, Interplay of proteins and lipids in generating membrane
curvature, Curr. Opin. Cell Biol. 22 (2010) 430–436.
132
[220] A. H. Bahrami, R. Lipowsky, T. R. Weikl, Tubulation and aggregation of spherical
nanoparticles adsorbed on vesicles, Phys. Rev. Lett. 109 (2012) 188102.
[221] A. Sari´c, A. Cacciuto, Mechanism of membrane tube formation induced by adhesive
nanocomponents, Phys. Rev. Lett. 109 (2012) 188101.
[222] S. Zhang, A. Nelson, P. A. Beales, Freezing or wrapping: The role of particle size in the
mechanism of nanoparticle–biomembrane interaction, Langmuir 28 (2012) 12831–12837.
[223] P. Zhang, J. E. Hinshaw, Three-dimensional reconstruction of dynamin in the constricted
state, Nat. Cell Biol. 3 (2001) 922–926.
[224] A. Roux, G. Koster, M. Lenz, B. Sorre, J.-B. Manneville, P. Nassoy, P. Bassereau, Mem-
brane curvature controls dynamin polymerization, Proc. Natl. Acad. Sci. USA. 107 (2010)
4141–4146.
[225] S. Morlot, M. Lenz, J. Prost, J.-F. Joanny, A. Roux, Deformation of dynamin helices
damped by membrane friction, Biophys. J. 99 (2010) 3580–3588.
[226] P. Blood, R. Swenson, G. Voth, Factors influencing local membrane curvature induction
by n-bar domains as revealed by molecular dynamics simulations, Biophys. J. 95 (2008)
1866–1876.
[227] M. Kranenburg, B. Smit, Phase behavior of model lipid bilayers, J. Phys. Chem. B 109
(2005) 6553–6563.
[228] G. S. Smith, E. B. Sirota, C. R. Safinya, N. A. Clark, Structure of the lβ phases in a
hydrated phosphatidylcholine multimembrane, Phys. Rev. Lett. 60 (1988) 813.
[229] G. S. Smith, E. B. Sirota, C. R. Safinya, R. J. Plano, N. A. Clark, X-ray structural
studies of freely suspended ordered hydrated dmpc multimembrane films, J. Chem. Phys.
92 (1990) 4519–4529.
[230] A. Tardieu, V. Luzzati, F. C. Reman, Structure and polymorphism of the hydrocarbon
chains of lipids: A study of lecithin-water phases, J. Mol. Bio. 75 (1973) 711 – 718.
[231] J. Selinger, M. Spector, J. Schnur, Theory of self-assembled tubules and helical ribbons,
J. Phys. Chem. B 105 (2001) 7157–7169.
[232] R. Sarasij, S. Mayor, M. Rao, Chirality-induced budding: A raft-mediated mechanism for
endocytosis and morphology of caveolae?, Biophys. J. 92 (2007) 3140.
[233] A. B. Harris, R. D. Kamien, T. C. Lubensky, Molecular chirality and chiral parameters,
Rev. Mod. Phys. 71 (1999) 1745.
133
[234] W. Helfrich, J. Prost, Intrinsic bending force in anisotropic membranes made of chiral
molecules, Phys. Rev. A 38 (1988) 3065.
[235] P. Nelson, T. Powers, Rigid chiral membranes, Phys. Rev. Lett. 69 (1992) 3409–3412.
[236] J. V. Selinger, J. M. Schnur, Theory of chiral lipid tubules, Phys. Rev. Lett. 71 (1993)
4091.
[237] J. M. Schnur, Lipid tubules: a paradigm for molecularly engineered structures, Science
262 (1993) 1669–1676.
[238] V. Kralj-Iglic, A. Iglic, G. Gomiscek, F. Sevsek, V. Arrigler, H. Hagerstrand, Microtubes
and nanotubes of a phospholipid bilayer membrane, J. Phys. A 35 (2002) 1533–1549.
[239] R. Oda, I. Huc, S. Candau, Gemini surfactants, the effect of hydrophobic chain length
and dissymmetry, Chem Commun (1997) 2105–2106.
[240] J. T. Groves, Bending mechanics and molecular organization in biological membranes,
Ann. Rev. Phys. Chem. 58 (2007) 697–717.
[241] J. C. Neto, U. Agero, R. T. Gazzinelli, O. N. Mesquita, Measuring optical and mechanical
properties of a living cell with defocusing microscopy, Biophys. J. 91 (2006) 1108–1115.
[242] D. Marguet, P. F. Lenne, H. Rigneault, H. T. He, Dynamics in the plasma membrane:
how to combine fluidity and order, Embo J. 25 (2006) 3446–3457.
[243] B. Sorre, A. Callan-Jones, J.-B. Manneville, P. Nassoy, J.-F. Joanny, J. Prost, B. Goud,
P. Bassereau, Curvature-driven lipid sorting needs proximity to a demixing point and is
aided by proteins, Proc. Natl. Acad. Sci. USA. 106 (2009) 5622–5626.
[244] B. Sorre, A. Callan-Jones, J. Manzi, B. Goud, J. Prost, P. Bassereau, A. Roux, Nature of
curvature coupling of amphiphysin with membranes depends on its bound density, Proc.
Natl. Acad. Sci. USA. 109 (2012) 173–178.
[245] A. Tian, T. Baumgart, Sorting of lipids and proteins in membrane curvature gradients,
Biophys. J. 96 (2009) 2676–2688.
[246] A. Tian, B. R. Capraro, C. Esposito, T. Baumgart, Bending stiffness depends on curvature
of ternary lipid mixture tubular membranes, Biophys. J. 97 (2009) 1636–1646.
[247] M. Heinrich, A. Tian, C. Esposito, T. Baumgart, Dynamic sorting of lipids and proteins in
membrane tubes with a moving phase boundary, Proc. Natl. Acad. Sci. USA. 107 (2010)
7208–7213.
[248] U. Seifert, Curvature-induced lateral phase segregation in two-component vesicles, Phys.
134
Rev. Lett. 70 (1993) 1335–1338.
[249] B. R. Capraro, Y. Yoon, W. Cho, T. Baumgart, Curvature sensing by the epsin n-terminal
homology domain measured on cylindrical lipid membrane tethers, J. Amer. Chem. Soc.
132 (2010) 1200–1201.
[250] C. Mim, H. Cui, J. A. Gawronski-Salerno, A. Frost, E. Lyman, G. A. Voth, V. M. Unger,
Structural basis of membrane bending by the n-bar protein endophilin, Cell 149 (2012)
137–145.
[251] M. Marino, K.-H. Moon, J. E. Hinshaw, The role of dynamin in membrane constriction
revealed by cryo-em, Microscopy and Microanalysis 11 (2005).
[252] T. Baumgart, B. R. Capraro, C. Zhu, S. L. Das, Thermodynamics and mechanics of
membrane curvature generation and sensing by proteins and lipids, Annu. Rev. Phys.
Chem. 62 (2011) 483–506.
[253] H. Jiang, T. Powers, Curvature-driven lipid sorting in a membrane tubule, Phys. Rev.
Lett. 101 (2008) 018103.
[254] P. Singh, P. Mahata, T. Baumgart, S. L. Das, Curvature sorting of proteins on a cylindrical
lipid membrane tether connected to a reservoir, Phys. Rev. E 85 (2012) 051906.
[255] C. Zhu, S. L. Das, T. Baumgart, Nonlinear sorting, curvature generation, and crowding
of endophilin n-bar on tubular membranes, Biophys. J. 102 (2012) 1837–1845.
[256] P. B. Sunil Kumar, G. Gompper, R. Lipowsky, Modulated phases in multicomponent fluid
membranes, Phys. Rev. E 60 (1999) 4610–4618.
[257] P. B. Sunil Kumar, G. Gompper, R. Lipowsky, Budding dynamics of multicomponent
membranes, Phys. Rev. Lett. 86 (2001) 3911–3914.
[258] N. Ramakrishnan, P. B. Sunil Kumar, J. H. Ipsen, Membrane-mediated aggregation of
curvature-inducing nematogens and membrane tubulation, Biophys. J. 104 (2013) 1018–
1028.
[259] N. Ramakrishnan, J. H. Ipsen, P. B. Sunil Kumar, Role of disclinations in determining
the morphology of deformable fluid interfaces, Soft Matter 8 (2012) 3058.
[260] J. Liu, R. Tourdot, V. Ramanan, N. J. Agrawal, R. Radhakrishanan, Mesoscale simula-
tions of curvature-inducing protein partitioning on lipid bilayer membranes in the presence
of mean curvature fields, Molecular Physics 110 (2012) 1127–1137.
[261] V. Ramanan, N. J. Agrawal, J. Liu, S. Engles, R. Toy, R. Radhakrishnan, Systems biology
135
and physical biology of clathrin-mediated endocytosis, Integr. Biol. 3 (2011) 803.
[262] N. Gov, Diffusion in curved fluid membranes, Phys. Rev. E 73 (2006) 041918.
[263] F. Divet, G. Danker, C. Misbah, Fluctuations and instability of a biological membrane
induced by interaction with macromolecules, Phys. Rev. E 72 (2005) 041901.
[264] A. Naji, F. Brown, Diffusion on ruffled membrane surfaces, J. Chem. Phys. 126 (2007)
235103–16.
[265] E. Reister-Gottfried, S. M. Leitenberger, U. Seifert, Hybrid simulations of lateral diffusion
in fluctuating membranes, Phys. Rev. E 75 (2007) 011908–11.
[266] E. Atilgan, S. X. Sun, Shape transitions in lipid membranes and protein mediated vesicle
fusion and fission, J. Chem. Phys. 126 (2007) 095102.
[267] J. Weinstein, R. Radhakrishnan, A coarse-grained methodology for simulating interfacial
dynamics in complex fluids: application to protein mediated membrane processes, Mol.
Phys. 104 (2006) 3653–3666.
[268] N. J. Agrawal, J. Weinstein, R. Radhakrishnan, Landscape of finite-temperature equilib-
rium behaviour of curvature-inducing proteins on a bilayer membrane explored using a
linearized elastic free energy model, Molecular Physics 106 (2008) 1913–1923.
[269] T. Chou, K. S. Kim, G. Oster, Statistical thermodynamics of membrane bending-mediated
protein–protein attractions, Biophys. J. 80 (2001) 1075–1087.
[270] M. Grabe, J. Neu, G. Oster, P. Nollert, Protein interactions and membrane geometry,
Biophys. J. 84 (2003) 854–868.
[271] K. S. Kim, J. Neu, G. Oster, Curvature-mediated interactions between membrane proteins,
Biophys J 75 (1998) 2274–2291.
[272] N. Dan, A. Derman, P. Pincus, S. Safran, Membrane-induced interactions between inclu-
sions, J. Phys. II 4 (1994) 1713–1725.
[273] E. J. Wallace, N. M. Hooper, P. D. Olmsted, The kinetics of phase separation in asym-
metric membranes, Biophys. J. 88 (2005) 4072–4083.
[274] H. Aranda-Espinoza, A. Berman, N. Dan, P. Pincus, S. Safran,
Interaction between
inclusions embedded in membranes, Biophys. J. 71 (1996) 648–656.
[275] A. Berman, P. Pincus, S. A. Safran, Membrane-induced interactions between inclusions,
J. de Physique II 4 (1994) 1713–1725.
[276] N. Dan, P. Pincus, S. A. Safran, Membrane-induced interactions between inclusions,
136
Langmuir 9 (1993) 2768–2771.
[277] M. Goulian, R. Bruinsma, P. Pincus, Long-range forces in heterogeneous fluid membranes,
Europhys. Lett. 22 (2007) 145–150.
[278] R. Golestanian, M. Goulian, M. Kardar, Fluctuation-induced interactions between rods
on a membrane, Phys. Rev. E 54 (1996) 6725–6734.
[279] M. Goulian, Inclusions in membranes, Curr. Opinion in Coll.& Interface Sci. 1 (1996) 358.
[280] M. Goulian, A. Libchaber, A new technique for probing inter-membrane interactions, The
J. of General Physiology. 107 (1996) 311–312.
[281] Y. Zhao, J. Liu, C. Yang, B. R. Capraro, T. Baumgart, R. P. Bradley, N. Ramakrishnan,
X. Xu, R. Radhakrishnan, T. Svitkina, W. Guo, Exo70 generates membrane curvature
for morphogenesis and cell migration, Developmental Cell 26 (2013) 266–278.
[282] M. M. Kozlov, Biophysics: Bending over to attract, Nature 447 (2007) 387–389.
[283] N. Agrawal, R. Radhakrishnan, Calculation of free energies in fluid membranes subject
to heterogeneous curvature fields, Phys. Rev. E 80 (2009) 011925.
[284] T. R. Weikl, Fluctuation-induced aggregation of rigid membrane inclusions, Europhys.
Lett. 54 (2001) 547.
[285] S.-J. E. Lee, Y. Hori, J. T. Groves, M. L. Dustin, A. K. Chakraborty, The synapse
assembly model, Trends Immunol. 23 (2002) 500–502.
[286] J. Liu, N. Agrawal, D. Eckmann, P. Ayyaswamy, R. Radhakrishnan, Top-down mesocale
models and free energy calculations of multivalent protein-protein and protein-membrane
interactions in nanocarrier adhesion and receptor trafficking,
in: T. Schlick (Ed.), In-
novations in Biomolecular Modeling and Simulations, Royal Society of Chemistry, Cam-
bridge UK, 2012, pp. 272–287.
[287] A. Iglic, B. Babnik, K. Bohinc, M. Fosnaric, H. Hagerstrand, V. Kralj-Iglic, On the role
of anisotropy of membrane constituents in formation of a membrane neck during budding
of a multicomponent membrane, J. Biomech. 40 (2007) 579–585.
[288] K. Bohinc, D. Lombardo, V. Kraljiglic, M. Fosnaric, S. May, F. Pernus, H. Hagerstrand,
A. Iglic, Shape variation of bilayer membrane daughter vesicles induced by anisotropic
membrane inclusions, Cell Mol. Biol. Lett. 11 (2006) 90–101.
[289] S. Leibler, D. Andelman, Ordered and curved meso-structures in membranes and am-
phiphilic films, J. Phys. France 48 (1987) 2013–2018.
137
[290] S. Leibler, Curvature instability in membranes, J. Phys. France 47 (1986) 507–516.
[291] R. Lipowsky, Budding of membranes induced by intramembrane domains, J. de Physique
II 2 (1992) 1825–1840.
[292] F. Julicher, R. Lipowsky, Domain-induced budding of vesicles, Phys. Rev. Lett. 70 (1993)
2964–2967.
[293] P. B. Sunil Kumar, M. Rao, Shape instabilities in the dynamics of a two-component fluid
membrane, Phys. Rev. Lett. 80 (1998) 2489–2492.
[294] T. Kohyama, D. M. Kroll, G. Gompper, Budding of crystalline domains in fluid mem-
branes, Phys. Rev. E 68 (2003) 061905.
[295] S. W. Hui, A. Sen, Effects of lipid packing on polymorphic phase behavior and membrane
properties, Proc. Natl. Acad. Sci. USA. 86 (1989) 5825–5829.
[296] H.-G. Dobereiner, O. Selchow, R. Lipowsky, Spontaneous curvature of fluid vesicles in-
duced by trans-bilayer sugar asymmetry, Eur. Biophys. J. 28 (1999) 174–178.
[297] M. Markowitz, A. Singh,
Self-assembling properties of 1,2-diacyl-sn-glycero-3-
phosphohyroxyethanol - a headgroup modified diacetylenic phospholipd, Langmuir 7
(1991) 16–18.
[298] J.-B. Fournier, Nontopological saddle-splay and curvature instabilities from anisotropic
membrane inclusions, Phys. Rev. Lett. 76 (1996) 4436–4439.
[299] J.-B. Fournier, P. Galatola, Bilayer membranes with 2d-nematic order of the surfactant
polar heads, Braz. J. Phys. 28 (1998) 329.
[300] J. E. Hinshaw, Dynamin and its role in membrane fission, Annu. Rev. Cell Dev. Biol. 16
(2000) 483–519.
[301] G. J. K. Praefcke, H. T. McMahon, The dynamin superfamily: universal membrane
tubulation and fission molecules?, Nature 5 (2004) 133–147.
[302] G. K. Voeltz, W. A. Prinz, Y. Shibata, J. M. Rist, T. A. Rapoport, A class of membrane
proteins shaping the tubular endoplasmic reticulum, Cell 124 (2006) 573–586.
[303] J. C. Dawson, J. A. Legg, L. M. Machesky, Bar domain proteins: a role in tubulation,
scission and actin assembly in clathrin-mediated endocytosis, Trends in Cell Biology 16
(2006) 493–498.
[304] T. Lopez-Leon, V. Koning, K. B. S. Devaiah, V. Vitelli, A. Fernandez-Nieves, Frustrated
nematic order in spherical geometries, Nat. Phys. 7 (2011) 391.
138
[305] N. Ramakrishnan, P. B. Sunil Kumar, J. H. Ipsen, Modeling anisotropic elasticity of fluid
membranes, Macromol. Theory Simul. 20 (2011) 446–450.
[306] P. A. Lebwohl, G. Lasher, Nematic-liquid-crystal order - monte-carlo calculation, Phys.
Rev. A 6 (1972) 426–429.
[307] J. R. Frank, M. Kardar, Defects in nematic membranes can buckle into pseudospheres,
Phys. Rev. E 77 (2008) 41705.
[308] P. G. De Gennes, J. Prost, The Physics of Liquid Crystals, The International Series of
Monographs on Physics, Clarendon Press, Oxford, 1993.
[309] T. Lubensky, J. Prost, Orientational order and vesicle shape, J. Phys. II France 2 (1992)
371–382.
[310] E. W. Dijkstra, A note on two problems in connexion with graphs, Numerische Mathe-
matik 1 (1959) 269–271.
[311] M. Bowick, L. Giomi, Two-dimensional matter: order, curvature and defects, Adv. Phys.
58 (2009) 449.
[312] V. Vitelli, D. R. Nelson, Nematic textures in spherical shells, Phys. Rev. E 74 (2006)
21711.
[313] N. Ramakrishnan, Effect of in-plane order and activity on vesicular morphology, Ph.D.
thesis, Thesis submitted to Indian Institute of Technology Madras, Chennai, 2012.
[314] S. M. Sweitzer, J. E. Hinshaw, Dynamin undergoes a gtp-dependent conformational change
causing vesiculation, Cell 93 (1998) 1021–1029.
[315] J. E. Hinshaw, Dynamin and its role in membrane fission, Ann. Rev. Cell Dev. Biol. 16
(2000) 483–519.
[316] A. Roux, G. Koster, M. Lenz, B. Sorre, J.-B. Manneville, P. Nassoy, P. Bassereau, Mem-
brane curvature controls dynamin polymerization, Proc. Natl. Acad. Sci. USA 107 (2010)
4141–4146.
[317] G. S. Ayton, P. D. Blood, G. A. Voth, Membrane remodeling from n-bar domain interac-
tions: Insights from multi-scale simulation, Biophys. J. 92 (2007) 3595–3602.
[318] H. Noguchi, G. Gompper, Fluid vesicles with viscous membranes in shear flow, Phys.
Rev. Lett. 93 (2004) 258102.
[319] H. Noguchi, G. Gompper, Vesicle dynamics in shear and capillary flows, J. Phys.-Condens.
Mat. 17 (2005) 3439.
139
[320] H. Noguchi, G. Gompper, Shape transitions of fluid vesicles and red blood cells in capillary
flows, Proc. Natl. Acad. Sci. USA. 102 (2005) 14159–14164.
[321] E. Lyman, H. Cui, G. A. Voth, Reconstructing protein remodeled membranes in molecular
detail from mesoscopic models, Phys. Chem. Chem. Phys. 13 (2011) 10430.
[322] R. W. Tourdot, N. Ramakrishnan, R. Radhakrishnan, Defining the free energy landscape
of curvature inducing proteins on membrane bilayers, Phys. Rev. E (2014) at review.
[323] G.Taubin, Estimating the tensor of curvature of a surface from a polyhedral approxima-
tion, Proc. Int. Conf. Comp. Vision (1995).
[324] K. Hildebrandt, K. Polthier, Anisotropic filtering of non-linear surface features, Euro-
graphics 23 (2004) 1–10.
[325] K. Hildebrandt, K. Polthier, M. Wardetzky, Smooth feature lines on surface meshes,
Eurographics Symposium on Geometry Processing (2005) 1–6.
[326] K. Polthier, Polyhedral Surfaces of Constant Mean Curvature, Ph.D. thesis, TU-Berlin,
2002.
[327] E. Hameiri, I. Shimsoni, Estimating the principal curvatures and the darboux frame from
real 3-d range data, IEEE transactions on systems man and cybernetics : Cybernetics 33
(2003) 626–633.
[328] Y. Han, Y. Shokef, A. M. Alsayed, P. Yunker, T. C. Lubensky, A. G. Yodh, Geometric
frustration in buckled colloidal monolayers, Nature 456 (2008) 898–903.
[329] H. Shin, M. Bowick, X. Xing, Topological defects in spherical nematics, Phys. Rev. Lett.
101 (2008) 037802.
[330] F. MacKintosh, T. Lubensky, Orientational order, topology, and vesicle shapes, Phys.
Rev. Lett. 67 (1991) 1169–1172.
[331] J. Park, T. C. Lubensky, F. C. Mackintosh, n-atic order and continuous shape changes of
deformable surfaces of genus zero, Europhys. Lett. 20 (1992) 279.
[332] C. S. Peskin, The immersed boundary method, ANU 11 (2002) 479–517.
140
|
1712.04565 | 1 | 1712 | 2017-12-12T23:05:38 | Intra-axonal Diffusivity in Brain White Matter | [
"physics.bio-ph",
"physics.med-ph"
] | Biophysical modeling is the mediator of evaluating the cellular structure of biological tissues using diffusion-weighted MRI. It is however the bottleneck of microstructural MRI. Beyond the complexity of diffusion, the current development is hindered by the fact that biophysical models heavily rely on diffusion-specific properties of diverse cellular compartments that are still unknown and must be measured in vivo. Obtaining such parameters by straightforward fitting is hindered by the degenerated landscape of the likelihood functions, in particular, the signal obtained for multiple diffusion directions and moderate diffusion weighting strength is not enough to estimate these parameters: different parameter constellations explain the signal equally well. The aim of this study is to measure the central parameter of white matter models, namely the intra-axonal water diffusivity in the normal human brain. Proper estimation of this parameter is complicated due to (i) the presence of both intra- and extra-axonal water compartments and (ii) the orientation dispersion of axons. Our measurement involves an efficient suppression of extra-axonal space and all cellular processes oriented outside a narrow cone around the principal fiber direction. This is achieved using a planar water mobility filter -- a strong diffusion weighting that suppresses signal from all molecules that are mobile in the plane transverse to the fiber bundle. Following the planar filter, the diffusivity in the remaining compartment is measured using linear and isotropic weighting. We find the specifically averaged intra-axonal diffusivity $D_0 = 2.25\pm 0.03{\,\rm \mu m^2/ms}$ for the timing of the applied gradients. Extrapolation to the infinite diffusion time gives $D_\infty \approx 2.0{\,\rm \mu m^2/ms}$. This result imposes a strong limitation on the parameter selection for biophysical modeling of diffusion-weighted MRI. | physics.bio-ph | physics | Intra-axonal Diffusivity in Brain White Matter
Bibek Dhital,1, ∗ Marco Reisert,1 Elias Kellner,1 and Valerij G. Kiselev1, †
1Department of Diagnostic Radiology, Medical Physics, University Medical Center Freiburg, Germany
(Dated: December 14, 2017)
Biophysical modeling is the mediator of evaluating the cellular structure of biological tissues
using diffusion-weighted MRI. It is however the bottleneck of microstructural MRI. Beyond the
complexity of diffusion, the current development is hindered by the fact that biophysical models
heavily rely on diffusion-specific properties of diverse cellular compartments that are still unknown
and must be measured in vivo. Obtaining such parameters by straightforward fitting is hindered
by the degenerated landscape of the likelihood functions, in particular, the signal obtained for
multiple diffusion directions and moderate diffusion weighting strength is not enough to estimate
these parameters: different parameter constellations explain the signal equally well. The aim of this
study is to measure the central parameter of white matter models, namely the intra-axonal water
diffusivity in the normal human brain. Proper estimation of this parameter is complicated due to (i)
the presence of both intra- and extra-axonal water compartments and (ii) the orientation dispersion
of axons. Our measurement involves an efficient suppression of extra-axonal space and all cellular
processes oriented outside a narrow cone around the principal fiber direction. This is achieved
using a planar water mobility filter – a strong diffusion weighting that suppresses signal from all
molecules that are mobile in the plane transverse to the fiber bundle. Following the planar filter,
the diffusivity in the remaining compartment is measured using linear and isotropic weighting. We
find the specifically averaged intra-axonal diffusivity D0 = 2.25 ± 0.03 µm2/ms for the timing of the
applied gradients. Extrapolation to the infinite diffusion time gives D∞ ≈ 2.0 µm2/ms. This result
imposes a strong limitation on the parameter selection for biophysical modeling of diffusion-weighted
MRI.
resonance
Diffusion-weighted magnetic
imaging
(dMRI) in brain white matter (WM) has been used to
detect tissue anomalies [1–3] and reconstruct axonal
tracts in-vivo [4–7]. A currently booming research area
is evaluation of the microstructure of living tissue at the
cellular scale much below the nominal MRI resolution.
While the role of the light illuminating biological cells is
taken by water diffusion as measured by MRI, the role
of the microscope is played by biophysical modeling that
enables interpretation of diffusion measures in terms of
the underlying tissue microstructure. Although much is
known about the tissue microstructure from histology, it
cannot access parameters that are central to dMRI such
as diffusivities inside different cell species. Furthermore,
as diffusivities change dramatically upon cell death,
they must be measured in vivo – their physical meaning
does not leave much room for alternative (non-MRI)
measurement techniques. In this way, the development
gets in a vicious circle: Biophysical models need dMRI-
specific parameters that should be found using dMRI
supplied with biophysical models.
The problem is exacerbated by the typically feature-
less shape of commonly acquired dMRI signal. This ren-
ders the problem of parameter determination from fitting
model to data extremely ill-posed: Essentially different
parameter sets can explain the measured data equally
well [8].
∗ [email protected]
† [email protected]
This study is an attempt to break the vicious circle of
parameter determination by measuring the intra-axonal
water diffusivity in the normal human brain. Neuronal
axons are considered as the main contributor to dMRI
signal from brain white matter at strong diffusion sen-
sitization [9]. While it is hardly possible to completely
abandon modeling of dMRI signal, we rely on mini-
mal model assumptions in this study. We developed a
dedicated dMRI measurement technique to suppress the
signal from extra-axonal space and measure water dif-
fusivity, D0, of the remaining intra-axonal water. We
find D0 = 2.25 ± 0.03 µm2/ms, which is rather close
to the free water diffusivity at the body temperature
(3 µm2/ms). This result rules out a large domain of the
parameter space available for dMRI signal interpretation.
I. MEASUREMENT TECHNIQUE MATCHES
THE TARGET GEOMETRY
dMRI measures the loss of coherence between indi-
vidual spins, which causes signal attenuation. The loss
of coherence is due to molecular motion in the pres-
ence of magnetic field gradients. In the commonly em-
ployed Stejskal-Tanner method [10], the gradient direc-
tion is constant during the measurement so that the re-
sulting signal is sensitized to diffusion in that direction.
Varying gradient direction during measurement can sen-
sitize the signal to motion in all three spatial direction,
which is referred to as the isotropic weighting or to mo-
tion in a selected plane, the planar weighting.
7
1
0
2
c
e
D
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
5
6
5
4
0
.
2
1
7
1
:
v
i
X
r
a
2
FIG. 1. The principle of the measurement. A: Neuronal axons
are represented by cylinders positioned at a common origin
to show their orientation distribution (which results in the
massive overlapping as a side effect). The in-plane cylinders
represent axons incoherent with the principal direction and
glial processes in extra-axonal space. The second row shows
the top views. B: Application of a linear water mobility filter
suppresses extra-axonal space and a part of axons, that are
shown with the darker color. Diffusivity measured along the
fiber bundle underestimates the true value due to the contri-
bution of tilted axons. C: The planar water mobility filter,
which is used in this study, suppresses the signal from all cells,
but those that are close to the principal fiber direction. Diffu-
sivity in this direction approaches the ground truth. Isotropic
diffusion measurement is applied to eliminate the residual ef-
fect of the orientation dispersion.
For the present measurement, we select the geometry of
diffusion weighting to match that of neuronal fiber bun-
dles in white matter, Fig. 1. We apply a strong planar
weighting that acts as a planar water mobility filter by
suppressing the signal from water molecules that are mo-
bile in the plane orthogonal to a fiber bundle. Since the
axonal diameter is of the order of a micrometer [11], ra-
dial water mobility inside the axons is negligible. There-
fore, the planar filter performs a robust suppression of
signal from extra-axonal space, independently of whether
water is mobile in the plane or confined in glial processes
or axons in the bundle's transverse plane. The suppres-
sion would not act on water confined in small compact
cells in which water motion is limited in all three direc-
tions. However, the presence of such cells in the normal
white matter is limited to maximum 2% as indicated by
measurements with strong isotropic diffusion weighting
[9, 12]. Following the planar water mobility filter, the
remaining signal is primarily contributed by the axons
that are close the the axis of the fiber bundle, Fig. 1.
The strength of the planar filter is characterized by its
b-value, bP, which can be understood as the signal sup-
pression by the factor e−bPD in a homogeneous medium
with the isotropic diffusivity D (in more detail, the filter
b-matrix has the eigenvalues bP/2 , bP/2 , 0). The planar
weighting reshapes the native axonal orientation distri-
bution by narrowing it around the principal fiber direc-
FIG. 2. Axial diffusion coefficient, D(cid:107), and trace of diffusion
tensor, Tr D, in single-fiber voxels in all subjects as functions
of the strength of the planar water mobility filter, bP. With in-
creasing bP, Tr D (circles) rapidly approaches the intra-axonal
Tr Da, which is dominated by the intra-axonal diffusivity, D0.
D(cid:107) (triangles) is measured in the principal fiber bundle direc-
tion and underestimates D0 due to the residual axonal orien-
tation dispersion (Fig. 1). The slow approach to the asymp-
tote contains information about the width of the distribution
as discussed in the text. The black lines and the text show
the results of fitting the corresponding models to the pooled
data for all subjects.
tion. Water diffusivity measured along this direction,
D(cid:107), approaches the intrinsic intra-axonal diffusivity, D0,
but still underestimates it due to the residual dispersion
of axons, Figures 1 and 2. It is straightforward (section
Methods) to quantify this effect as
D(cid:107) = D0(cid:104)cos2 θ(cid:105)F ,
(1)
where θ is the angle an axon makes with the principal
fiber direction and the averaging is taken over the filter-
reshaped orientation distribution. The mean cos2 θ ap-
proaches unity for very strong planar filter, which is ob-
viously limited by the associated decrease in the signal
magnitude and hardware limitations.
The issue of underestimation can be circumvented
by following the planar water mobility filter with the
isotropic encoding [12–14]. In principle, isotropic encod-
ing measures the trace of diffusion tensor inside axons,
but for long diffusion times, the eigenvalues of this tensor
can be well approximated by 0 , 0 , D0, so that Tr D = D0.
Isotropic encoding gives an estimate of the intra-axonal
diffusivity, which is insensitive to the residual axonal ori-
entation distribution.
The described measurement was performed in the
brains of healthy volunteers. Diffusion-weighted signal
was measured in many pre-defined directions after appli-
cation of the planar filter with variable strength in the
orthogonal plane (18 directions in subjects 1 and 2 and 30
0123456bP(ms/µm2)00.511.522.53D∥,Diso(µm2/ms)D0=2.38±0.02D0=2.25±0.03θ0=24◦±2◦DisoD∥S1 isoS1 parallelS2 isoS2 parallelS3 isoS3 parallelS4 isoS4 parallel directions in subjects 3 and 4). Data with the weakest fil-
ter (0.1 ms/µm2) were used to estimate the unsuppressed
local diffusion tensor [15]. The tensor was used to select
voxels containing predominantly single fiber bundles (see
Methods). Among these single-fiber voxels, those were
further selected in which the principal fiber orientation
formed a small angle (less than 15◦) with one of the mea-
surement directions. Both D(cid:107) and Tr D were calculated
in these voxels as functions of the planar filter strength
(Fig. 2) and interpreted in the context of the model pre-
sented in Fig. 1 to obtain the intra-axonal diffusivity, D0,
and the width of the axonal orientation distribution, σ0,
(further details in section Methods).
II. METHOD
Experimental Design
Our measurement technique depends on two ansatze
• Diffusion in the
one-
dimensional.
Since the axon diameter is very
small [11, 16], the radial diffusivity is close to zero
for the diffusion time of the order of 100 ms used
in this study.
effectively
axons
is
• Diffusion in each compartment is Gaussian: The
correlation time of water motion through the cellu-
lar environment is much smaller than the diffusion
time [17].
In contrast to the majority of the present models, we
do not assume the three-dimensional water mobility in
the extra-axonal compartment. Our approach copes with
possible complex composition of this compartment in-
cluding one-dimensional cellular processes, restrictions to
planes and connected three-dimensional space.
Diffusion-weighted signal for Gaussian diffusion takes
the form
S = e− Tr(bD)
(2)
where b is the b-matrix as defined by applied diffusion-
weighting gradients and D is the diffusion tensor. Both
b and D that are real, symmetric matrices. The signal
form is applied to all individual signal-contributing tissue
compartments.
Each b-matrix can be decomposed into linear bL, pla-
nar bP and spherical bS components [18] in its eigenvector
basis,
+
0 0 0
0 0 0
0 0 1
b = bL
1 0 0
0 1 0
0 0 0
+
bP
2
.
1 0 0
0 1 0
0 0 1
bS
3
(3)
The diffusion weighting gradient waveforms can be de-
signed to obtain the desired b-matrix. We employed a
3
planar filter to suppress signal from the extra-axonal
compartment. A planar filter is designed by ensuring
bL = bS = 0.
The planar filter is designed to take advantage of the
difference in the radial diffusivities of the two compart-
ments. For any diffusion tensor D, the signal due to the
planar filter can be written as
ln S = − bP
2
− bP
2
− bP
2
d1 sin2 θ
(cid:0)1 − sin2 θ sin2 φ(cid:1)
(cid:0)1 − sin2 θ cos2 φ(cid:1)
d2
d3
(4)
where d1,2,3 are the largest, middle and the smallest
eigenvalues of D, φ is the rotation axis in the d23 plane
and θ is the rotation angle. When the planar filter is
normal to the axonal direction, it suppresses the signal
only from the extra-axonal compartment. The filter is
therefore appropriate for small θ. As we increase bP, the
planar filter increasingly suppresses signals from all com-
partment that are mobile in the plane leaving only signal
from axons that are close to being normal to the plane,
Fig. 1.
In the absence of signal from the extra-axonal com-
partment, the orientation dispersion of axons impedes di-
rect measurement of intra-axonal diffusivity and the mea-
sured apparent diffusion coefficient (ADC) in the parallel
direction results in a downward biased estimate. We cir-
cumvent this issue by applying, in addition to the parallel
diffusivity, the measurement with the isotropic weighting.
In the absence of any contribution from the extra-axonal
compartments, the difference between trace and ADC is
only solely due to dispersion.
As shown in Fig. 3, the planar filtering was obtained
by applying two orthogonal gradient waveforms G1 and
G2 on either side of the refocusing pulse. The resulting
b-matrix contains two degenerate eigenvalues that span
the plane we intend to filter. The linear diffusion weight-
ing is obtained by applying a third gradient G3 that is
orthogonal to both G1 and G2 (the mutual orthogonality
of all three gradients insures the absence of any cross-
terms). Measurements with and without G3 allow us to
measure diffusivity of the remaining signal after the pla-
nar filter. To measure the trace of the remaining signal,
small portions of G1 and G2 with the b-values equal to
that of G3 are re-attributed from the planar filter to the
measurement of Tr D.
dMRI measurements.
In-vivo measurements were performed on four in-
formed volunteers in a 3 T human scanner (Siemens
PRISMA, max gradient strength 80 mT/m, 32 channel
receive coil). The procedure was approved by the ethics
4
F
R
1
G
2
G
3
G
FIG. 3. Schematic of the sequence used for our measurements. All the three gradient directions are orthogonal to each other
and the gradient waveforms do not have any cross terms. The top two gradients provide planar diffusion filtering, the bottom
gradient provides an additional linear weighting. Each of these gradient lobes has ramp-up time of 1.5 ms and flat-top time of
4.3 ms. Slice selection gradients, spoilers and gradients for imaging are not shown.
board of University Medical Center, Freiburg. Written
consents were obtained from all volunteers. For all volun-
teers, 14 slices were acquired with field of view (FOV) =
25.6 cm, 4 mm isotropic resolution, echo time (TE) = 140
ms, repetition time (TR) = 2500 ms, bandwidth = 2005
Hz/pixel and partial Fourier factor of 0.75. All imaging
parameters, FOV, TE, bandwidth, resolution and num-
ber of slices, were kept the same for all measurements.
A simple dMRI sequence comprising of 30 single direc-
tion measurements was also measured for two b-values of
0.05 and 1 ms/µm2 to estimate the diffusion tensor [15].
In two subjects (S1 and S2), diffusion measurements
were performed for 18 directions of G3, with strength of
the planar filter 0.1, 1.0, 1.8, 2.7, 3.6 and 4.5 ms/µm2.
To account for signal-to-noise ratio (SNR) loss, data at
higher b-values were acquired with more (up to six) rep-
etitions. In two other subjects, the measurements were
performed for 30 directions of G3, and, instead of the rep-
etitions, the bP values were uniformly distributed in the
same interval. Each planar filter was applied once with-
out the linear gradient, G3, and once with linear gradient
with a b-value of 0.45 ms/µm2.
Data Analysis.
The data was analyzed using in-house written code
in Matlab R(cid:13). All
images with corrected for Gibbs-
ringing by interpolating the image based on subvoxel-
shifts that samples the ringing pattern at the zero-
crossings of the oscillating sinc-function [19]. Diffusion
tensor [15] estimation was performed using the measure-
ment with the weakest planar filter. Single bundle voxels
were selected using the fractional anisotropy (FA) along
with measures of linearity (cl = (λ1 − λ2)/λ1), planarity
(cp = (λ2 − λ3)/λ1) and sphericity (cs = λ3/λ1), where
λ1, λ2, and λ3 are the first, second, and third eigenvalues
of the diffusion tensor estimated at that voxel. These co-
efficients describe proximity of the tensor to a line, plane
and sphere [20]. The single bundle voxels had to ful-
fill the limit on the fractional anisotropy, FA > 0.5 and
cl ≥ 0.4, cp ≤ 0.2, cs ≤ 0.35 [21]. The rest of the analysis
focused on single bundle voxels in which the direction of
the primary eigenvector reflects the fiber orientation.
The DTI data was used to obtained the relative an-
gle between the primary eigenvector and the measured
direction of the gradient waveform G3. Only those sin-
gle bundle voxels were selected for further analysis in
which the principal fiber direction formed an angle less
than 15◦ with one of measured directions. For each
planar filter, the parallel ADC was estimated by ordi-
nary least squares method where the natural logarithm
of the signal were fitted against the two b-values of 0
and 0.45 ms/µm2. The trace was obtained by a sim-
ilar fitting, but in this case signal from each planar
filter with no linear weighting was fitted with signal
from an increased planar filter and a linear weighting
of 0.45 ms/µm2. For example, estimating trace after
a planar filter of 1.9 ms/µm2, required fitting data ob-
tained from planar filter of 1.9 ms/µm2and linear weight-
ing of 0 ms/µm2and planar filter of 2.8 ms/µm2and lin-
ear weighting of 0.45 ms/µm2. In other words, a planar
filter of 2.8 ms/µm2and a linear filter of 0.45 ms/µm2can
also be considered as a planar filter of 1.9 ms/µm2and
an isotropic weighting of 1.35 ms/µm2. Trace, therefore,
could only be calculated up to the second strongest pla-
nar filter.
III. RESULTS
D(cid:107) = D0
r
e
t
l
fi
r
a
n
a
l
p
r
e
t
f
a
l
a
n
g
i
s
d
e
z
i
l
a
m
r
o
N
1
0.8
0.6
0.4
0.2
0
0
1
3
2
bP ( ms/µm2)
4
5
6
FIG. 4. Normalized signal after planar weighting in WM vox-
els where the relative angle between the fibers and normal to
the plane is within 15◦.
An example of the filter effect on the signal is shown in
Fig. 4. Figure 2 shows the resulted diffusivities obtained
from the signal averaged over all selected voxels for both
the isotropic and the linear diffusion weighting as a func-
tion of planar water mobility filter strength. With the in-
creasing filter strength, both results asymptotically level
off leaving a small gap in between. In what follows, we
discuss the limits in more detail.
The isotropic measurement estimates the trace of the
compartment-averaged diffusion tensor, D,
Tr D = vF Tr Da + (1 − vF ) Tr De ,
(5)
where Da and De are the intra- and extra-axonal diffu-
sion tensors, respectively, Tr Da ≈ D0 as discussed above
and vF is the relative fraction of axonal signal after the
application of the planar water mobility filter,
vF =
va
va + (1 − va)e−bPD⊥ ,
(6)
where va is the genuine water fraction of axons (more
precisely, of all effectively one-dimensional processes) and
D⊥ is the diffusivity in extra-axonal space in the trans-
verse direction. The value of vF approaches unity for
very strong filter, bPD⊥ (cid:29) 1.
The value of Tr D for zero filter (Fig. 2) is only slightly
above its asymptotic value. This implies that the Tr De
is only slightly larger than Tr Da, in agreement with the
recent conclusion about a small difference between them
[12]. We also observe that Tr D starts leveling off at
moderate filter strength. This agrees with the expected
5
exponential suppression of extra-axonal space where wa-
ter can move in the plane normal to the principal fiber
direction.
The axial diffusivity, D(cid:107) (Fig. 2) provides two main in-
sights, one from weak filter strengths and the other from
strong filter strengths. For weak filter, D(cid:107) is the weighted
mean of intra- and extra-axonal compartments. While
the traces of the two compartments are close, the extra-
axonal diffusion tensor has appreciable transverse com-
ponent. Therefore, its longitudinal component is smaller
than the intra-axonal diffusivity.
Increasing the filter
strength reduces the weight of the extra-axonal contribu-
tion and results in some increase in D(cid:107). For strong filter,
the mechanism of the increase in D(cid:107) is different. It agrees
with the argument of narrowing the axonal orientation
distribution (Fig. 1) for the increasing filter strength (as
discussed after Eq. (1)).
We estimated the right-hand side of Eq. (1) by approx-
imating the axonal orientation distribution with a Gaus-
sian distribution of sin θ with the variance σ2. This ap-
proximation is justified for distributions effectively nar-
rowed by the planar filter, Fig. 1. This gives
(cid:35)
(cid:18) 1
(cid:19)1/2
,
ez2√
π erfi (z)
− 1
2z2
, z =
2σ2 +
bPD0
2
(7)
where the error function of imaginary argument is defined
as
(cid:34)
(cid:90) z
0
erfi (z) =
2√
π
dt et2
.
(8)
For increasing bP, D(cid:107) slowly approaches its asymptotic
value, D0, according to
(cid:34)
(cid:35)
D(cid:107) ≈ D0
1 −
1
(1/σ2 + bPD0)2
.
(9)
We fitted a constant, D0, which is the asymptotic form
of Eq. (5), to isotropically measured data and Eq. (7) to
corresponding data for the linear weighting. This posed
a question about the selection of the data fitting inter-
val, bP > bmin
P . Since both fitting procedures do not
take into account the extra-axonal signal (with a num-
ber of associated unknown parameters), a too low bmin
results in a bias in the estimated parameter. On the
other hand, a too large bmin
reduces the precision, since
fitting is applied within too short intervals, Fig. 5. The
choice was made by visual inspection of Fig. 5, some ar-
bitrariness of this action is alike the selection of the com-
monly used significance level. With this choice made,
we obtain the values D0 = 2.38 ± 0.02 µm2/ms and
D0 = 2.25 ± 0.03 µm2/ms for the isotropic measurement
(bmin
P = 3 ms/µm2) and the single-direction measurement
P = 1.2 ms/µm2), respectively.
(bmin
The value of σ = 0.41 ± 0.03 and the corresponding
angle θ = arcsin σ = 27 ± 2◦ overestimate the genuine
P
P
6
obtained outside the human brain support the present
result. Skinner et al. [24] obtained the axial diffusivity of
2.16±0.22 µm2/ms in the normal rat spinal cord using the
linear water mobility filter (double PFG) in the direction
orthogonal to the spinal cord. This should underestimate
D0 as illustrated in the middle panel in Fig. 1. The under-
estimation can be evaluated using Eq. (1) with the twice
reduced σ0, which gives 2.16±0.03 µm2/ms using D0 and
σ0 obtained in the present study. This perfect agreement
should not be over-interpreted in view of the difference
in the investigated tissues. Jelescu et al. [25] obtained
the axial axonal diffusivity close to 1.7 µm2/ms using Gd
injection in the rat brain. Since this figure does not ac-
count for the axonal dispersion, it should be compared
with the reduced diffusivity according to Eq. (1) with the
full value of σ0. This gives 1.87± 0.06 µm2/ms indicating
a reasonable agreement between the results. Jespersen et
al. investigated fixed pig spinal cord for variable diffusion
time and found the long-time intra-axonal diffusivity to
be larger than the axial extra-axonal one and close to 1/2
of the free water diffusivity [26].
The approximately two-fold reduction in the intra-
axonal diffusivity relative to the cytoplasm value was also
obtained using intra-neurite reporter molecules, NAA, in
the rat brain with the reduction factor 0.5 [27] and tNAA
in the human brain with the reduction factor 0.7 [28].
Equal reduction in diffusivity for different molecules sug-
gests the purely geometric (not chemical) mechanism of
this effect.
The present result for D0 does not agree with the
branch selection made for WMTI [21] and the fixed values
assumed by NODDI [29].
The obtained width of the fiber orientation distribu-
tion of θ0 = 24 ± 2◦ agrees reasonably with the values
around 19◦ obtained from histology [28, 30] and MRI-
based studies [9, 28, 30].
The present method has a few assumptions about the
white matter microstructure. Its robustness with respect
to deviations from these assumptions and minor correc-
tions to the obtained values are discussed below.
More about microstructural features
The majority of white matter models treat extra-
axonal space as hindered by axons, but otherwise struc-
tureless. In reality, white matter is 'structurally crowded'
being comprised of oligodendrocytes, astrocytes, mi-
croglia and vasculature. Both astrocytes and oligoden-
drocytes possess processes. The astrocyte processes are
smooth, thin and have relatively less branching but ex-
tend more than 100 micrometers. The oligodendrocyte
processes wrap around the axons to form the myelin
layers.
It is quite plausible that glial processes have
anisotropic diffusion properties. The assumption of
Gaussian extra-axonal space is based on the effective
P
FIG. 5. Dependence of the fitted intra-axonal diffusivity on
the fitting interval, from bmin
upwards. Systematic deviation
for small values of bP are due to the extra-axonal contribu-
tions not accounted for in the models. The large error for
large bP follow from too short fitting intervals. The final se-
lection is shown with the horizontal lines. Note that any other
reasonable selection affects the result values within their error
bars.
width of the axonal orientation distribution, σ0, due to
the additional orientation dispersion introduced by the
voxel selection. This effect can be corrected assuming the
uniform distribution of voxels within the aperture of 30◦
and the independence of microscopic parameters from the
fiber orientation. The corrected values, σ0 = 0.41 ± 0.03
with the corresponding angle, θ0 = 24 ± 2◦.
IV. DISCUSSION
The obtained intra-axonal diffusivity of D0 =
2.25 ± 0.03 µm2/ms is the factor 0.75 smaller than the
free water diffusion coefficient at the body temperature.
These rather high value resolves the bi-modality of pa-
rameter estimation arising from multiple single-direction
measurements [8]. The high precision of our result should
not be over-interpreted, since it applies to a specific signal
averaging; investigation of regional variations was beyond
the scope of this study.
The present result agrees reasonably well with the in-
terval [1.9, 2.2] µm2/ms found from Da obtained at ultra-
high b, using the 1/b scaling of the apparent orienta-
tional dispersion in single-fiber populations according to
Eq. (7) in the limit when only the intra-axonal signal was
present [9]. Similar values although with large regional
variations were found by studying the combined echo-
and diffusion time dependencies [22] and using the ro-
tational invariants of signal weighted in multiples single
directions and b-values [23].
Although not a subjected for direct comparison, data
01234bminF(ms/µm2)usedforfitting22.12.22.32.42.5D0(µm2/ms)D0=2.38D0=2.25 from isotropic measurement from parallel measurement coarse-graining of structural features for long diffusion
times [31, 32]. The course-graining of the whole extra-
axonal space within the experimental diffusion time could
be effectively hindered by the low permeability of cell
membranes [33, 34].
In such a scenario, the advantage
of the present planar filter over the linear filter (Fig. 1)
becomes crucial.
Partial volume effect
We chose 4 mm isotropic resolution to increase the
SNR, which is crucial when an essential fraction signal
is suppressed by application of strong planar filter. How-
ever, the question still retains anatomical relevance only
when we can clearly resolve not just white and gray mat-
ter but also find single bundle WM voxels where the fibers
are coherently aligned.
Owing to high diffusivity in the cerebrospinal fluid,
moderate planar water mobility filter efficiently sup-
presses any partial voluming due to this compartment.
Inclusion of more than one fiber bundle in the selected
voxels is limited by the selection of voxels with high
anisotropy and by the efficient suppression of possible
sub-dominant fiber bundles by the planar filter. Again,
the difference between the planar and linear filters is cru-
cial for the accurate measurement. Some admixture of
gray matter in the selected voxels is possible, but limited
by the selection criteria of high anisotropy.
Deviations of axonal geometry from ideal cylinders
The values of D0 from isotropic measurement is larger
than that from the linear measurement by the value
0.13 ± 0.04 µm2/ms, which remains to be explained. We
speculate that this difference can be attributed to devia-
tions of axon geometry from that of ideally straight cylin-
ders. Such deviations should be effective over the water
diffusion length. A simple estimate shows that the cur-
vature with the typical radius 100 µm would explain the
difference. Assigning such a curvature to all axons does
not sound realistic, but a large contribution from a rela-
tively small sub-population cannot be excluded. In this
context, the value D0 = 2.25 ± 0.03 µm2/ms obtained
with the linear weighting is interpreted as the diffusiv-
ity along the axons, while the isotropic weighting adds
about 0.07 µm2/ms for each transverse direction due to
deviations of axons from the ideal cylindrical form.
7
duration of the applied magnetic field gradients. In more
detail, the overestimation results from the finite width of
the gradient power spectrum in the following signal form
for weak diffusion weighting,
(cid:90) dω
2π
ln S = −
q(ω)2D(ω) ,
(10)
where q(ω) is the Fourier transform of the time-integrated
gradient, g(t) = γG(t) and 2D(ω) is the autocorrelation
function of molecular velocity, which is directly related
to the conventionally defined diffusion coefficient [35, 36].
The magnitude of this effect depends of the form of
D(ω) for small ω. We use the theoretical result sup-
ported by experimental evidences that this dependence
takes the form D(ω) ≈ D∞ + Const ω1/2 [31, 37]. The
calculated correction for the gradients used in our exper-
iments is about D0 − D∞ = 0.25 µm2/ms. Note that this
small correction is only noticeable due to the dependence
of D(ω) on the square root of frequency. A linear depen-
dence would result in a negligible difference D0 − D∞.
Comparison with other filter techniques
Our suppression technique is akin to the filter-ex-
change method [33, 38]. The filtered-dPFG [38] method
uses two pairs of gradient pulses perpendicular to each
other [39] and has been used to show local anisotropy in
macroscopically isotropic material including gray matter
[40].
Assuming that extra-axonal compartment is an ax-
ially symmetric tensor, both the filtered-PFG method
and the planar filter would have equivalent suppression
efficiency for the extra-axonal compartment. However,
these two methods differ in their signal suppression for
the dispersed axons. While suppression due to the pla-
nar filter only depends on the polar angle θ between the
axonal direction and normal to the plane, for filtered-
PFG the suppression also depends on the azimuthal angle
φ, Eq. (4). Therefore, the suppression efficiency filtered-
PFG is lower for dispersed axons. Hence, even in the
absence of extra-axonal compartment, parallel ADC ob-
tained after with filtered-PFG method produces a greater
downward bias of the true intra-axonal diffusivity than
that obtained with the planar filter. This probably ex-
plains why we obtain a constant increase in parallel ADC
with increasing filter weighting but application of filtered-
PFG with similar filter weighting showed less effect [24]
as discussed above.
Correction for finite diffusion time
Outlook
The values of diffusivities found in this study slightly
overestimate the genuine long-time values due to the final
In order to fully realize the clinical potential of diffu-
sion MRI, it is important to understand how the resulted
microstructural measures change in response to pathol-
ogy. This was not however the aim of the present study
that focused on providing an accurate estimate of the
intra-axonal water diffusion coefficient. This quantity is
central for biophysical modeling that mediates the evalu-
ation of microstructural parameters from the dMRI sig-
nal. While creation a solid ground for modeling efforts
is the main focus of this study, testing the sensitivity of
intra-axonal diffusivity to diverse neurological diseases
remains the aim of future work.
8
Acknowledgement
We thank Dmitry S. Novikov for useful discussions.
[1] ME Moseley, J Kucharczyk, J Mintorovitch, Y Cohen,
J Kurhanewicz, N Derugin, H Asgari, and D Norman,
"Diffusion-weighted mr imaging of acute stroke: cor-
relation with t2-weighted and magnetic susceptibility-
enhanced mr imaging in cats." American Journal of Neu-
roradiology 11, 423–429 (1990).
[2] DJ Werring, CA Clark, GJ Barker, AJ Thompson, and
DH Miller, "Diffusion tensor imaging of
lesions and
normal-appearing white matter in multiple sclerosis,"
Neurology 52, 1626–1626 (1999).
[3] Kinuko Kono, Yuichi Inoue, Keiko Nakayama, Miyuki
Shakudo, Michiharu Morino, Kenji Ohata, Kenichi
Wakasa, and Ryusaku Yamada, "The role of diffusion-
weighted imaging in patients with brain tumors," Amer-
ican Journal of Neuroradiology 22, 1081–1088 (2001).
[4] Peter J Basser, Sinisa Pajevic, Carlo Pierpaoli, Jeffrey
Duda, and Akram Aldroubi, "In vivo fiber tractography
using dt-mri data," Magnetic resonance in medicine 44,
625–632 (2000).
[5] David S Tuch, "Q-ball imaging," Magnetic resonance in
medicine 52, 1358–1372 (2004).
[6] J-Donald Tournier, Fernando Calamante, David G Ga-
dian,
and Alan Connelly, "Direct estimation of the
fiber orientation density function from diffusion-weighted
mri data using spherical deconvolution," NeuroImage 23,
1176–1185 (2004).
[7] Marco Reisert, Irina Mader, Constantin Anastasopoulos,
Matthias Weigel, Susanne Schnell, and Valerij Kiselev,
"Global fiber reconstruction becomes practical," Neu-
roimage 54, 955–962 (2011).
[8] Ileana O Jelescu, Jelle Veraart, Els Fieremans,
and
Dmitry S Novikov, "Degeneracy in model parameter es-
timation for multi-compartmental diffusion in neuronal
tissue," NMR in Biomedicine 29, 33–47 (2016).
[9] J. Veraart, E. Fieremans, and D. S. Novikov, "Univer-
sal power-law scaling of water diffusion in human brain
defines what we see with MRI," ArXiv e-prints (2016),
arXiv:1609.09145 [physics.bio-ph].
[10] Edward O Stejskal and John E Tanner, "Spin diffu-
sion measurements:
spin echoes in the presence of a
time-dependent field gradient," The Journal of Chemi-
cal Physics 42, 288–292 (1965).
[11] Francisco Aboitiz, Arnold B Scheibel, Robin S Fisher,
and Eran Zaidel, "Fiber composition of the human corpus
callosum," Brain research 598, 143–153 (1992).
[12] Bibek Dhital, Elias Kellner, Valerij G Kiselev,
Marco Reisert, "The absence of
pool
10.1016/j.neuroimage.2017.10.051.
in brain white matter," Neuroimage
and
restricted water
(2017),
[13] Samo Lasic, Filip Szczepankiewicz, Stefanie Eriks-
and Daniel Topgaard, "Mi-
son, Markus Nilsson,
croanisotropy imaging: quantification of microscopic dif-
fusion anisotropy and orientational order parameter by
diffusion mri with magic-angle spinning of the q-vector,"
Frontiers in Physics 2 (2014), 10.3389/fphy.2014.00011.
[14] Filip Szczepankiewicz, Samo Lasic, Danielle van Westen,
Pia C Sundgren, Elisabet Englund, Carl-Fredrik Westin,
Freddy Stahlberg, Jimmy Latt, Daniel Topgaard, and
Markus Nilsson, "Quantification of microscopic diffusion
anisotropy disentangles effects of orientation dispersion
from microstructure: Applications in healthy volunteers
and in brain tumors," NeuroImage 104, 241–252 (2015).
[15] Peter J Basser, James Mattiello, and Denis LeBihan,
"Mr diffusion tensor spectroscopy and imaging." Bio-
physical journal 66, 259 (1994).
[16] Roberto Caminiti, Filippo Carducci, Claudia Piervin-
cenzi, Alexandra Battaglia-Mayer, Giuseppina Con-
falone, Federica Visco-Comandini, Patrizia Pantano, and
Giorgio M Innocenti, "Diameter, length, speed, and con-
duction delay of callosal axons in macaque monkeys and
humans: comparing data from histology and magnetic
resonance imaging diffusion tractography," J Neurosci
33, 14501–11 (2013).
[17] Dmitry S Novikov and Valerij G Kiselev, "Effective
medium theory of a diffusion-weighted signal," NMR
Biomed 23, 682–97 (2010).
[18] Joao P de Almeida Martins and Daniel Topgaard, "Two-
dimensional correlation of isotropic and directional dif-
fusion using nmr," Physical review letters 116, 087601
(2016).
[19] Elias Kellner, Bibek Dhital, Valerij G Kiselev,
and
Marco Reisert, "Gibbs-ringing artifact removal based on
local subvoxel-shifts," Magnetic resonance in medicine
76, 1574–1581 (2016).
[20] C-F Westin, Stephan E Maier, Hatsuho Mamata, Arya
Nabavi, Ferenc A Jolesz, and Ron Kikinis, "Processing
and visualization for diffusion tensor mri," Medical image
analysis 6, 93–108 (2002).
[21] Els Fieremans, Jens H Jensen, and Joseph A Helpern,
"White matter characterization with diffusional kurtosis
imaging," Neuroimage 58, 177–188 (2011).
[22] Jelle Veraart, Dmitry S Novikov, and Els Fieremans,
"TE dependent diffusion imaging (TEdDI) distinguishes
between compartmental T2 relaxation times," Neuroim-
age (2017), 10.1016/j.neuroimage.2017.09.030.
[23] Dmitry S. Novikov, Jelle Veraart, Ileana O Jelescu, and
Els Fieremans, "Mapping orientational and microstruc-
tural metrics of neuronal integrity with in vivo diffusion
mri," https://arxiv.org/abs/1609.09144 (2016).
[24] Nathan P Skinner, Shekar N Kurpad, Brian D Schmit,
L Tugan Muftuler, and Matthew D Budde, "Rapid in
vivo detection of rat spinal cord injury with double-
9
28, 1489–1506 (2015).
[39] PT Callaghan and ME Komlosh, "Locally anisotropic
motion in a macroscopically isotropic system: displace-
ment correlations measured using double pulsed gradient
spin-echo nmr," Magnetic Resonance in Chemistry 40,
S15–S19 (2002).
[40] ME Komlosh, F Horkay, RZ Freidlin, U Nevo, Y Assaf,
and PJ Basser, "Detection of microscopic anisotropy in
gray matter and in a novel tissue phantom using dou-
ble pulsed gradient spin echo mr," Journal of magnetic
resonance 189, 38–45 (2007).
diffusion-encoded magnetic resonance spectroscopy,"
Magnetic resonance in medicine 77, 1639–1649 (2017).
[25] Ileana Ozana Jelescu, Nicolas Kunz, Analina Raquel
Da Silva, and Rolf Gruetter, "Intra- and extra-axonal ax-
ial diffusivities in the white matter: which one is faster?"
in Proc. 25th Annual Meeting of ISMRM (Honolulu,
2017) p. 0001.
[26] Sune Nørhøj Jespersen, Jonas Lynge Olesen, Brian
Hansen,
and Noam Shemesh, "Diffusion time depen-
dence of microstructural parameters in fixed spinal cord,"
Neuroimage (2017), 10.1016/j.neuroimage.2017.08.039.
and
Dmitriy A Yablonskiy, "On the nature of the naa diffu-
sion attenuated mr signal in the central nervous system,"
Magnetic resonance in medicine 52, 1052–1059 (2004).
[27] Christopher D Kroenke, Joseph JH Ackerman,
[28] Itamar Ronen, Matthew Budde, Ece Ercan, Jacopo An-
nese, Aranee Techawiboonwong,
and Andrew Webb,
"Microstructural organization of axons in the human cor-
pus callosum quantified by diffusion-weighted magnetic
resonance spectroscopy of n-acetylaspartate and post-
mortem histology," Brain Structure and Function 219,
1773–1785 (2014).
[29] Hui Zhang, Torben Schneider, Claudia A Wheeler-
Kingshott, and Daniel C Alexander, "Noddi: practical in
vivo neurite orientation dispersion and density imaging
of the human brain," Neuroimage 61, 1000–1016 (2012).
[30] Trygve B Leergaard, Nathan S White, Alex de Crespigny,
Ingeborg Bolstad, Helen D'Arceuil, Jan G Bjaalie, and
Anders M Dale, "Quantitative histological validation of
diffusion MRI fiber orientation distributions in the rat
brain," PLoS One 5, e8595 (2010).
[31] Dmitry S Novikov, Jens H Jensen, Joseph A Helpern,
and Els Fieremans, "Revealing mesoscopic structural uni-
versality with diffusion," Proc Natl Acad Sci U S A 111,
5088–93 (2014).
[32] Dmitry S Novikov, Sune N Jespersen, Valerij G Kiselev,
and Els Fieremans, "Quantifying brain microstructure
with diffusion mri: Theory and parameter estimation,"
arXiv preprint arXiv:1612.02059 (2016).
[33] Markus Nilsson, Jimmy Latt, Danielle van Westen, Sara
Brockstedt, Samo Lasic, Freddy Stahlberg, and Daniel
Topgaard, "Noninvasive mapping of water diffusional ex-
change in the human brain using filter-exchange imag-
ing," Magnetic resonance in medicine 69, 1572–1580
(2013).
[34] Donghan M Yang, James E Huettner, G Larry Bret-
thorst, Jeffrey J Neil, Joel R Garbow,
and Joseph
J H Ackerman, "Intracellular water preexchange lifetime
in neurons and astrocytes," Magn Reson Med (2017),
10.1002/mrm.26781.
[35] Dmitry S Novikov and Valerij G Kiselev, "Surface-to-
volume ratio with oscillating gradients," J Magn Reson
210, 141–5 (2011).
[36] Valerij G Kiselev, "Fundamentals of diffusion MRI
physics," NMR Biomed 30 (2017), 10.1002/nbm.3602.
[37] Els Fieremans, Lauren M Burcaw, Hong-Hsi Lee, Gre-
gory Lemberskiy, Jelle Veraart, and Dmitry S Novikov,
"In vivo observation and biophysical interpretation of
time-dependent diffusion in human white matter," Neu-
roimage 129, 414–27 (2016).
[38] Nathan P Skinner, Shekar N Kurpad, Brian D Schmit,
and Matthew D Budde, "Detection of acute nervous sys-
tem injury with advanced diffusion-weighted mri: a sim-
ulation and sensitivity analysis," NMR in Biomedicine
|
1407.2765 | 4 | 1407 | 2015-11-12T09:54:47 | Collective dynamics of actomyosin cortex endow cells with intrinsic mechanosensing properties | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.CB"
] | Living cells adapt and respond actively to the mechanical properties of their environment. In addition to biochemical mechanotransduction, evidence exists for a myosin-dependent, purely mechanical sensitivity to the stiffness of the surroundings at the scale of the whole cell. Using a minimal model of the dynamics of actomyosin cortex, we show that the interplay of myosin power strokes with the rapidly remodelling actin network results in a regulation of force and cell shape that adapts to the stiffness of the environment. Instantaneous changes of the environment stiffness are found to trigger an intrinsic mechanical response of the actomyosin cortex. Cortical retrograde flow resulting from actin polymerisation at the edges is shown to be modulated by the stress resulting from myosin contractility, which in turn regulates the cell size in a force-dependent manner. The model describes the maximum force that cells can exert and the maximum speed at which they can contract, which are measured experimentally. These limiting cases are found to be associated with energy dissipation phenomena which are of the same nature as those taking place during the contraction of a whole muscle. This explains the fact that single nonmuscle cell and whole muscle contraction both follow a Hill-like force-velocity relationship. | physics.bio-ph | physics | Collective dynamics of actomyosin cortex
endow cells with intrinsic mechanosensing properties
Jocelyn ´Etienne,∗†‡ Jonathan Fouchard, D´emosth`ene Mitrossilis, Nathalie Bufi,
Pauline Durand-Smet, and Atef Asnacios§
Accepted for publication in the
Proceedings of the National Academy of Sciences of the USA
in a revised version
5
1
0
2
v
o
N
2
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
4
v
5
6
7
2
.
7
0
4
1
:
v
i
X
r
a
Abstract
Living cells adapt and respond actively to the mechanical properties of their environment.
In ad-
dition to biochemical mechanotransduction, evidence exists for a myosin-dependent, purely mechanical
sensitivity to the stiffness of the surroundings at the scale of the whole cell. Using a minimal model of
the dynamics of actomyosin cortex, we show that the interplay of myosin power strokes with the rapidly
remodelling actin network results in a regulation of force and cell shape that adapts to the stiffness of the
environment. Instantaneous changes of the environment stiffness are found to trigger an intrinsic mechan-
ical response of the actomyosin cortex. Cortical retrograde flow resulting from actin polymerisation at the
edges is shown to be modulated by the stress resulting from myosin contractility, which in turn regulates
the cell size in a force-dependent manner. The model describes the maximum force that cells can exert
and the maximum speed at which they can contract, which are measured experimentally. These limiting
cases are found to be associated with energy dissipation phenomena which are of the same nature as those
taking place during the contraction of a whole muscle. This explains the fact that single nonmuscle cell
and whole muscle contraction both follow a Hill-like force -- velocity relationship.
When placed in different mechanical environ-
ments, living cells assume different shapes [1, 2, 3,
4, 5]. This behaviour is known to be strongly de-
pendent on the contractile activity of the actomyosin
network [6, 7, 8, 9, 10]. One of the cues driving the
cell response to its environment is rigidity [11]. Cells
are able to sense not only the local rigidity of the
material they are in contact with [12], but also the
one associated with distant cell-substrate contacts :
that is, the amount of extra force needed in order
to achieve a given displacement of microplates be-
tween which the cell is placed [13], Fig. 1c, of an
AFM cantilever [14, 15], or of elastic micropillars
[16]. This cell-scale rigidity sensing is totally depen-
dent on Myosin-II activity [13]. A working model
of the molecular mechanisms at play in the acto-
myosin cortex is available [17], where myosin contrac-
tion, actin treadmilling and actin crosslinker turnover
∗Univ. Grenoble Alpes, LIPHY, F-38000 Grenoble, France
†CNRS, LIPHY, F-38000 Grenoble, France
‡to whom correspondence should be addressed; E-mail: [email protected]
§Universit´e Paris-Diderot -- CNRS, Laboratoire Mati`ere et Syst`emes Complexes (MSC), UMR 7057, Paris, France
are the main ingredients. Phenomenological models
[18, 19] of mechanosensing have been proposed, but
could not bridge the gap between the molecular mi-
crostructure and this phenomenology. Here, we show
that the collective dynamics of actin, actin crosslink-
ers and myosin molecular motors are sufficient to ex-
plain cell-scale rigidity sensing. The model deriva-
tion is analogous to the one of rubber elasticity of
transiently crosslinked networks [20], with the addi-
tion of active crosslinkers, accounting for the myosin.
It involves four parameters only: myosin contractile
stress, speed of actin treadmilling, elastic modulus,
and viscoelastic relaxation time of the cortex, which
arises because of crosslinker unbinding.
It allows
quantitative predictions of the dynamics and statics
of cell contraction depending on the external stiffness.
The crucial dependence of this behaviour on the fact
that crosslinkers have a short life time is reminiscent
1
of the model of muscle contraction by A. F. Huxley
[21], in which the force dependence of muscle con-
traction rate is explained by the fact that for lower
muscle force and higher contraction speed, the num-
ber of myosin heads contributing to filament sliding
decreases in favour of those resisting it transiently,
before they unbind. While it is known that many
molecules associated with actomyosin exhibit stress-
dependent dynamics [22], collective effects alone can
explain observations in both Huxley's muscle contrac-
tion model and the present model. We have pre-
viously evidenced the similarity of single nonmuscle
cell and muscle contraction rate [13].
In spite of
very dissimilar organisation of actomyosin in mus-
cles, where it forms well-ordered sarcomeres, and in
nonmuscle cell cortex, where no large-scale pattern-
ing is observed, we show that corresponding mech-
anisms explain their similar motor properties. The
collective dynamics we describe are consistent with
the fact that the actin network behaves as a fluid at
long times. We show how myosin activity can con-
tract this fluid at a given rate that depends on exter-
nal forces resisting cell contraction, arising e.g. from
the stiffness of the environment. This, combined with
actin protrusivity, results in both a sustained retro-
grade flow and a regulation of cell shape. In addition,
this explains the elastic-like behaviour observed in
cell-scale rigidity-sensing and justifies ad hoc models
based on this observation [18].
Results
Intrinsic rigidity sensing of actomyosin
The actin cortex of nonmuscle cells is a disor-
dered network situated at the cell periphery. Actin
filaments are crosslinked by proteins such as α-
actinin, however these crosslinkers experience a rapid
turnover, of order 10 s e.g. for α-actinin [23], and
actin itself has a scarcely longer turnover time
[24]. The actin network is thus only transiently
crosslinked. Following [20], we describe the behaviour
of such a network by a rubber-like model. Up to the
first order, this model leads to a stress-strain rela-
tionship of a Maxwell viscoelastic liquid in which the
relaxation time is a characteristic unbinding time τα,
(cid:79)
σ +σ − 2ταE ε = 0,
τα
(1)
(cid:79)
where ε is the rate-of-strain tensor and
σ the objec-
tive time-derivative of the stress tensor σ, see Ap-
pendix S2. In the linear setting, σ = 2E(β2 (cid:104)RR(cid:105) −
I), where E is the elastic modulus of the crosslinked
actin network, β is inversely proportional to the
Kuhn length, and R is a basic element of this net-
work, namely the strand vector spanning the distance
between any two consecutive actin -- actin bonds, Fig.
1a. As long as these two bonds hold (for times much
shorter than τα), this basic element behaves elasti-
cally, and the stress tensor σ grows in proportion
with the strain. When a crosslinker unbinds, the fil-
aments can slide somewhat relative to one another,
and the elastic tension that was maintained via this
crosslinker is relaxed: this corresponds to an effec-
tive friction, and happens at a typical rate 1/τα. In
sum, the actin network behaves like an elastic solid of
modulus E over a time shorter than τα, as crosslink-
ers remain in place during such a solicitation, and it
has a viscous-like response for longer times with an
effective viscosity ταE, since the network yields as
crosslinkers unbind.
Such a viscoelastic liquid is unable to resist me-
chanical stresses [25]. However living cells are able
to generate stress themselves [26] thanks to myosin
bipolar filaments, which act as actin crosslinkers but
are in addition able to move by one step length
along one of the filaments they are bound to, us-
ing biochemical energy. Let us call αmyo the frac-
tion of crosslinkers which are myosin filaments, and
effectuate a power-stroke of step length (cid:96) at fre-
quency 1/τmyo. The power-stroke corresponds to a
change of the binding location of the myosin head
along the actin filament, and thus affects the length
of the neighbouring strands R, Fig. 1a. Supple-
mentary Eq. 4 includes the additional term that
describes this myosin-driven evolution of the net-
work configuration. When this equation is integrated
to give the local macroscopic stress tensor σ, this
term results in a contractile stress Σ proportional to
ταEαmyo((cid:96)β)2/τmyo (see Appendix S2):
(cid:79)
σ +σ − 2ταE ε = Σ.
(2)
τα
The three-parameter model obtained (τα, E, Σ) is in
line with early continuum models [27] and the active
gel theory [28], however we do not supplement this
active stress with an elastic stress, unlike previous
models of mechanosensitive active gels [19, 16] where
cells are treated as viscoelastic solids (Appendix S4).
The interpretation of this equation is that the con-
tractile stress Σ gives rise either to the build-up of
a contractile tension σ (if clamped boundary con-
ditions allow no strain) or a contractile strain rate
Σ/(2ταE) (if free boundary conditions allow strain
but not tension build-up, this is e.g. the case of super-
precipitation in vitro [29]), or a combination of these.
We then asked whether this simple rheological
law for the actomyosin cortex could explain the be-
2
a
c
b
Figure 1: Model of the actomyosin behaviour and experimental setup. (a) Sketch of a transiently crosslinked
actin network with myosin bipolar filaments. Unbinding of a crosslinker releases elastic tension locally, the
crosslinker will re-bind to the network, preserving its elastic properties, but after this loss of stored elastic
energy. Myosin power-strokes have the effect of modifying the equilibrium length of the adjacent strands R,
this results in an increase of the tension (see Supplementary Discussion S2). (b) Model of the mechanical
components of the cell and microplate system. Microplates impose that the vertical force equilibrating the
cortical tension is linked to the cell half-height h with F = 2k(h0 − h). Tension along the actomyosin
cortex (green surface) is anisotropic and has values σ and σ⊥ along directions es and eφ. Actin treadmilling
provides a boundary condition at the cell leading edge, the actin cortex undergoes a retrograde flow away
from the plates. (c) Transmission image of a cell and setup.
a
b
c
Figure 2: Predictive modelling of the stiffness-dependent cell mechanical response. (a) The force and cell
height at equilibrium are adapted to the stiffness of the environment up to a maximum force at high stiffness,
and, for vanishing stiffness, there is a well defined equilibrium length He (independent of microtubules, see
Fig. S3). Circles, experimental results [13] for force (black) and height (green), black line, force predicted by
the 3D model, green line, height predicted by the 3D model, blue curve and shaded area, force and confidence
interval in the 1D model. Two out of the four parameters of the model adjusted in this plot, using the force
at infinite stiffness and the height at zero stiffness. (b) During the transient part of the experiment, the
rate of loading of the cell is adapted to the stiffness of the environment. Boxes, experimental results [13],
green curve and shaded area, force and confidence interval in the 1D model. One parameter of the model
is fitted in this plot, the last one is adjusted on figure 4a. (c) Instantaneous adaptation to a change of the
microplate stiffness k. Red line, stiffness imposed using a feedback loop, black dots, force measured [30],
blue line, 1D model prediction of force, using the stiffness changes imposed in the experiment (red line),
and the four parameters obtained in (a) and (b), without any further adjustment.
3
time tvector Rpassive crosslinkactive crosslink (myosin)power strokeunbindingtime t+δt haviour of cells in our microplate experiments, Fig.
1c. To do so, we investigated the equilibrium shape
and force of a thin actomyosin cortex surrounding a
liquid (the cytosol), in the three-dimensional geom-
etry of the experimental setup in Fig. 1b, Appendix
S5. Surprisingly, the rigidity sensing property of cells
is adequately recovered by this simple model: while
a fixed maximum force is predicted above a certain
critical stiffness kc = ΣS/H0 of the microplates, the
actomyosin-generated force is proportional to k < kc,
Fig. 2a. Here H0 is the initial plate separation and S
is the section area of the actin network. Thus the
contractile activity of myosin motors is enough to
endow the viscoelastic liquid-like actin cortex with
the spring-like response to the rigidity of its environ-
ment [31, 13], a property which was introduced phe-
nomenologically in previous models [18]. In order to
get a clear understanding of the mechanism through
which this was possible, we simplified the geometry to
a one-dimensional problem (Fig. 3a) and found that
the spring-like behaviour of the contractile fluid was
retained, Fig. 3c and Appendix S3.
Indeed, for an
environment (external spring) of stiffness k beyond
the critical value kc, the contractile fluid is unable
to strain beyond some plate deflection smaller than
H0, and equilibrium is reached as it exerts its maxi-
mal contractility Σ. For an external stiffness k below
the critical value, the tension σ which is balanced by
microplate force is smaller than Σ for any admissible
deflection. Thus there is a nonzero rate of contrac-
tion,
ε < 0, as long as a maximal deflection is not
reached, which is H0 with the hypotheses done so
far. The next section discusses actin polymerisation
as limiting this deflection, thus setting an equilibrium
cell size.
The force finally achieved is proportional to k --
just as a prestretched spring of stiffness kc would do
(Fig. 3b), or, alternatively, just as cells do when ex-
hibiting a mechanosensitive behaviour [18, 16] (Ap-
pendix S4). This is supported by our previous re-
port [30] that, when the external spring stiffness is
instantaneously changed in experiments, cells adapt
their rate of force build-up dF/dt to the new con-
ditions within 0.1 s. This observation was repeated
using an AFM-based technique [14]. In [15], an over-
shoot of the rate adaptation, which relaxed to a
long-term rate within 10 s, was noted in addition
to the initial instantaneous change of slope. While
this instantaneity at the cell scale is not explained by
mechanochemical regulation, this behaviour is fully
accounted by the mechanical model proposed here
(see Appendix S3.4, Fig. 2b,c). Thus, the actomyosin
cortex is mechanosensitive by essence: its peculiar ac-
tive viscoelastic nature, which arises from collective
effects, provides a built-in system of adaptation to
changes of the mechanical environment.
Force-dependent regulation of cell size
In microplate experiments, cells are observed to
spread laterally simultaneously as they deflect mi-
croplates. These two processes both affect the arc
distance between the cell adhesions on each plate
(Fig. S1). Cell spreading is known to be mediated
by actin treadmilling [32, 33], which controls the ex-
tension of the lamellipodium [34]. An effect of tread-
milling is that there is a net flow of filamentous actin
from the lamellipodial region into the proximal part
of the actomyosin cytoskeleton, between adhesions
[35, 33], which persists even if the cell and its ad-
hesions are immobile [17]. Thus the size of the cell
is regulated by the combination of the myosin-driven
contraction rate − ε of the cytoskeleton (the retro-
grade flow described above) and the speed vt at which
newly polymerised actin is incorporated into the cor-
tex. This feature can be included in the model as
a boundary condition, prescribing a difference vt be-
tween the speed of the cell edge and the one of the
actin cortex close to the edge (Appendix S3.2 and
S5). We find that this reduces the maximum tension
that the actomyosin network can develop, however
the shape of the dependence versus the external stiff-
ness k is little altered (Fig. 3c).
In particular, the
force continues to be linearly dependent on k for low
stiffnesses, albeit with a reduced slope: this is a direct
consequence of a mechanical regulation of cell height,
which maintains the microplate deflection H0 − He
when k varies, thus F = k(H0 − He) varies linearly
with k in this range.
Indeed, the equilibrium height of the cell is ob-
tained in the classical situation when the polymerisa-
tion of actin at the cell edge is in balance with the ret-
rograde flow which drives the actomyosin away from
the cell edge (Fig. 3d). However, the interpretation
here is not that polymerisation generates this flow,
but that myosin contraction is at its origin, and that
the equilibrium height is reached when the force bal-
ance between myosin contraction and external forces
acting on the cell is such that retrograde flow exactly
balances polymerisation speed. In the case when they
do not balance, the cell edge will move at a speed
which is the difference between the speed of poly-
merisation and the retrograde flow at the edge, until
equilibrium is reached. In the case of our set-up, it
is found that retrograde flow is initially faster than
polymerisation, leading to a decrease of cell height,
and, because of the resistance of the microplates to
4
a
b
c
d
Figure 3: A one-dimensional simplification of the 3D-model preserves the essential mechanisms and results.
(a) Sketch of the one-dimensional model, crosslinkers and myosin unbind and rebind so that the bulk prop-
erties are constant. (b) Spring model used elsewhere in the literature and to which the present model is
compared. (c) Equilibrium state of models and cells as a function of microplate stiffness k, normalised by
critical stiffness kc and maximum force Fmax in each case. Dashes: tension of the spring model, green:
tension calculated in the 1D model for different values of the speed of treadmilling vt, blue: experimen-
tal results plotted with kc = Σ/(H0 + 4ταvt) = 41 nN/µm (no adjustment, see Supplementary Discussion
S3.5). The stiffness-dependence of force in experiments is well-matched both by the spring model and the
1D model. Insert, tension calculated in the 1D model for different values of the speed of treadmilling vt,
normalised by Fmax in the absence of treadmilling, vt = 0: treadmilling reduces the force transmitted to the
microplates. (d) A dynamic equilibrium is attained when the retrograde flow exactly balances the speed of
polymerisation at the boundaries, here sketched in the 1D model geometry.
cell contraction, the tension σ increases.
In turn,
this higher tension reduces the retrograde flow un-
til it is exactly equal and opposite to the polymeri-
sation speed. Treadmilling and myosin contraction
thus work against one another, as has been noted
for a long time [32] and is specifically described by
Rossier et al. [17].
These phenomena regulate cell size. For low ex-
ternal force, myosin-driven retrograde flow is high as
the tension that opposes it is small, and the balance
between retrograde flow and polymerisation speed
is obtained when the cell has significantly reduced
its height, He in Fig. 2a. This height He is thus
a trade-off between the speed at which actin tread-
milling produces new cortex vt (Appendix S3) and
the rate of myosin-generated contractile strain 1/τc =
Σ/(2ταE) = αmyo((cid:96)β)2/(2τmyo), which is itself the
result of the frictional resistance of crosslinkers to
myosin contraction. Indeed, in the one-dimensional
toy problem, the equilibrium height of the system is
2τcvt for k (cid:28) kc, and thus the force developed by
the external spring is F = k(H0 − 2τcvt).
In the
three-dimensional full model of the cell cortex, this
equilibrium height is slightly modified as a function
of its geometry, but is still proportional to the prod-
uct τcvt (Appendix S5.3). This equilibrium height
is reached when the treadmilling speed balances ex-
actly the speed at which myosin in the bulk con-
tracts the boundaries of the existing cortex, via a
retrograde flow that involves the whole of the cor-
tex but is maximum in distal regions (Fig. 3d). This
type of competition between the protrusive contri-
bution of actin polymerisation and the contractile
contribution of myosin is of course noted in crawling
cells [36], it is also observed in immobile cells where
centripetal movement of actin monomers within fila-
ments is noted even when adhesive structures are lim-
ited to a fixed location on a micropattern [17]. Thus,
the cell-scale model and experiments allow to deter-
mine the speed of treadmilling, which is a molecular-
scale quantity. We find vt = 6.5 ± 1.5 nm/s in
the 1D model, and 4 nm/s for the 3D model, val-
ues which are in agreement with the literature [17],
4.3 ± 1.2 nm/s. The 1D model also allows to obtain
the relaxation time of the crosslinked actomyosin net-
work, τα = 1186 ± 258 s, consistent with elastic-like
behaviour for frequencies higher than 10−3 Hz, the
contractile characteristic time τc = 521 ± 57 s, con-
sistent with a 24-minute completion of actin super-
precipitation [29] and ΣS = (2.0 ± 0.9) · 103 nN, see
Appendix S3.6. These values fit both the plateau
(v = 0) force vs. stiffness experimental results, Fig. 2a
5
0.01 0.1 1 0.01 0.1 1 10 100Fp/Fmax (logscale)k/kc (logscale)vt 0.01 0.1 1 0.01 1 100 a
b
c
Figure 4: Liquid-like motor properties of cells. (a) The rheological model leads to a Hill-type equation (3)
which matches quantitatively the experimental data both during loading (blue curve at H0 − H = 1µm:
force-velocity in 1D model, boxes: experiments) and at equilibrium (surface intercept with v = 0: 1D
model, red curve: 3D model, circles: experiments). Dotted lines correspond to the force-distance relation-
ship imposed by a given microplate stiffness k. Same parameters used as in Fig. 2. (b) Sketch of (a). (c)
Power usage in a microplate experiment as a function of the external load F : only a small part of the
load-independent myosin power is being transmitted to the cell environment as mechanical power, the rest
is dissipated internally or compensates the antagonistic role of polymerisation. Boxes, experimental results,
to be compared to the dashed line, mechanical power at H0 − H = 1 µm, solid lines correspond to H = H0.
Same parameters used as in Fig. 2.
and the dynamics of the experiments, Fig. 2b. With-
out further adjustment, they also allow to predict the
dynamical adaptation of the loading rate of a cell be-
tween microplates of variable stiffness [30], Fig. 2c,
and to plot the force -- velocity -- height phase-portrait
of the experiments, Fig. 4a.
From an energetic point of view, it may seem very
inefficient to use up energy for these two active phe-
nomena that counterbalance one another. However,
in a great number of physiological functions such as
cytokinesis and motility, either or both of actin poly-
merisation and myosin contraction are crucial. It is
therefore highly interesting that, combined together,
they provide a spring-like behaviour to the cell while
preserving its fluid nature, endowing it with the same
resilience to sudden mechanical aggression as the pas-
sive mechanisms developed by some organisms, such
as urinary-tract bacteria [37] and insects [38].
Single cells have similar energetic expenses
to muscles
These antagonistic behaviours of polymerisation and
myosin contractility entail energy losses, which de-
fine a range of force and velocity over which the
actomyosin cytoskeleton is effective. The study of
the energetic efficiency of animal muscle contraction
was pioneered by A. V. Hill [39], who determined
a law relating force F and speed of shortening v:
(F + a)(v + b) = c, where a, b and c are numeri-
cal values which depend on two values specific of a
given muscle, namely a maximum force and a maxi-
mum speed, and a universal empirical constant. Hill's
law was then explained using a model based on the
muscle molecular structure by A. F. Huxley [21]. Re-
cently, we have shown that a law of the same form
describes the shortening and force generation of cells
in the present setup [13].
In particular, the max-
imum attainable force and velocity are due to en-
ergy losses. The model can shed light on the molec-
ular origin of these losses, and leads to the quanti-
tative force-velocity diagrams in Fig. 4a,b. Indeed,
in terms of F and v, Eq. 2 yields (Appendix S3.5):
+ E
(v + 2vt + vα) = (Σ + E)vα − H F
2S
.
(3)
(cid:19)
(cid:18) F
S
Here vα = H/(2τα) interprets an internal creep veloc-
ity. The right-hand side corresponds to the source of
power (minus the internal elastic energy storage term
H F /2), the left-hand side is the power usage (up to
a constant, Evα, added to both sides). The formal
similarity of this law with Hill's law for muscles is
not a surprise when one compares the present model
with Huxley's model of striated muscle contraction.
Indeed, the main components in both models are an
elastic structure with transient attachments and an
ATP-fuelled 'pre-stretch' of the basic elements of the
systems which, upon release, generates either tension
or contraction, or a combination of the two. Tread-
6
milling vt in our model superimposes an effect similar
to the ones already present.
It is easiest to under-
stand this law in the extreme cases of zero speed or
zero load, which correspond respectively to maximum
contractile force and maximum contraction speed.
The case of zero load, F = 0, corresponds to the
highest velocity. In the case of muscle contraction,
the fastest sliding of actin relative to myosin fila-
ments is limited in Huxley's model by the rate at
which myosin heads detach after their stroke. This is
because myosin heads which remain bound to actin
will get entrained and exert an opposing force to the
motion. This transient resistance is similar to fric-
tion. In our nonmuscle actomyosin model, crosslink-
ers also need to unbind so that the actomyosin net-
work, which is elastic at short times, fluidises and
flows. The velocity it reaches is thus a decreasing
function of the relaxation time τα.
Zero speed, v = 0, corresponds to microplates of
infinite stiffness. If in addition protrusion via actin
treadmilling is blocked, vt = 0, there is no net defor-
mation of the network (or sliding of the filaments in
Huxley's model), however energy is still being dis-
sipated:
indeed, myosin motors will still perform
power strokes and generate tension in the network,
but nearby crosslinkers and myosins themselves will
also detach at the rate 1/τα. This detachment will
result in the local loss of the elastic energy that had
been stored as tension of the network without result-
ing in a global deformation, corresponding to some
internal creep. This time, the maximum force is an
increasing function of relaxation time τα. In nonmus-
cle cells, actin treadmilling is still present when the
cell edge is immobile [17], and reduces the maximum
force that can be attained, because part of the myosin
power will be used to contract this newly formed cor-
tex.
In the intermediate regimes where neither F nor
v are zero, the term F v corresponds to an actual me-
chanical work performed against the external load,
Fig. 4c. This work uses up the part of myosin energy
that is not dissipated by internal creep, by effective
friction or by working against the actin-driven edge
protrusion.
Discussion
The model described here is based on a simple de-
scription of collective dynamics of actin and myosin
that is consistent with observations at the protein
scale [17], but does not include molecular sensitiv-
ity of the dynamics or actively driven reorganisation
of the actomyosin cortex. We find that the linear
7
rheological law that arises from this description al-
lows to predict accurately the rigidity sensing exper-
iments that we carry over two decades of external
rigidity. The dynamics of cell pulling are also re-
covered, and we show that their similarity with the
dynamics of muscle contraction [39] are due to the
parallelism that exists between actomyosin dynam-
ics in nonmuscle cells and the dynamics of thin and
thick filaments in muscle [21]. Because the model is
based on the collective dynamics of myosin and actin
filaments, it allows to understand their role in rigid-
ity sensing: myosin provides a contractile stress Σ
which will in turn generate traction forces at the cell --
substrate contact area. In cases when Σ is in excess
to these traction forces (which resist cell contraction),
a retrograde flow is generated, the cell contracts and
deforms its environment. This retrograde flow is lim-
ited by the time needed by the actin network to flu-
idise (viscoelastic relaxation time τα), and is force-
dependent. Retrograde flow also works against actin
protrusion at the cell edge: we hypothesise that this
antagonism regulates cell shape, and show in the case
of our parallel microplates setup that this regulation
of cell height determines the rigidity sensing features
of cells. The existence of a deformation set-point had
already been speculated on micropillar array exper-
iments [31] and used as a hypothesis in modelling
work [18], here we shed light on its relationship with
molecular processes and retrograde flow. Because the
deformation set-point is obtained as a balance be-
tween protrusion and contraction, it is versatile and
is likely to be tuned by the many pathways known
to affect either of these. Undoubtedly, it is a loss of
energy for the cell to have such antagonistic mech-
anisms, which we quantify in Fig. 4c. However, the
total power of myosin action that we calculate for a
single cell is of the order of 10 fW, which is less than
one thousandth of the total power involved in cell
metabolism [40]. It is thus not surprising that this
energy expense is not optimised in nonmuscle cells, as
the evolutionary pressure on this cost can be deemed
very low, while on the other hand the same structure
confers to the cell its mechanical versatility and re-
activity to abrupt mechanical challenges. In this, the
cell may be likened to a wrestler ready to face a sud-
den struggle: the wrestler maintains a high muscle
tone, having his own muscles work against one an-
other. Maintaining this muscular tone is an expense
of energy, but the benefit of resisting assaults widely
exceeds this cost.
Materials and methods
Cell culture, fibronectin coating and drugs
Rat embryonic fibroblasts-52 (Ref52) line with YFP-paxillin, kindly
provided by A. Bershadsky, Weizmann Institute, and C2-7 myogenic
cell line, a subclone of the C2 line derived from the skeletal muscle
of adult CH3 mice, kindly provided by D. Paulin and Z. Xue, Uni-
versit´e Paris-Diderot, were cultured and prepared both using the
procedure described in [13]. Glass microplates and, in the case of
the experiment shown in Fig. S1b, the glass coverslip at the bottom
of the chamber were coated with 5µg/mL as described in [13]. In
experiments described in Appendix S1 and Fig. S3, 1 µM Colchicin
was used.
Side-view experimental procedure
Ref52 or C2-7 cells were used in microplate experiments as de-
scribed in [30] with the same equipment and reagents. Cells were
then suspended in a temperature-controlled manipulation chamber
filled with culture medium and fibronectin-coated microplates were
brought in contact with a single cell as described in [13]. After a
few seconds, the two microplates were simultaneously and smoothly
lifted to 60 µm from the chamber's bottom to get the desired config-
uration of one cell adherent between two parallel plates. One of the
plates was rigid, and the other could be used as a nanonewton force
sensor [41]. By using flexible microplates of different stiffness values,
we were able to characterise the effect of rigidity on force generation
up to a stiffness of about 200 nN/µm. In order to measure forces
at an even higher stiffness, we used a flexible microplate of stiffness
(cid:39) 10 nN/µm but controlled the plate -- to -- plate distance using a feed-
back loop, maintaining it constant regardless of cell force and plate
deflection [41, 30]. Concurrently, we visualised cell spreading un-
der brightlight illumination at an angle perpendicular to the plane
defined by the main axis of the two microplates. For conditions re-
ferred to as 'low stiffness', such experiments with n = 5 Ref52 cells
and n = 4 C2-7 cells were analysed. For conditions referred to as
'infinite stiffness', again n = 5 Ref52 cells and n = 4 C2-7 cells were
analysed. In both cases, the distribution of values obtained for the
different cell types was not significantly different, see Fig. S4.
Bottom-view experimental procedure
Additionally, n = 4 Ref52 cells were used in experiments using an-
other experimental procedure allowing to image the adhesion com-
plexes at one of the plates. The objective of the TIRF-equipped
microscope (Olympus IX-71, with 488 nm wavelength laser) was put
in contact with the fibronectin-coated rigid glass plate at the bottom
of the chamber. A fibronectin-coated flexible plate is put in contact
with the cell after sedimentation. The deflection of the flexible plate
is measured by imaging its tip with a custom-made microscope. The
latter is composed of a computer-controlled light source, an optical
tube (InfinitubeStandard, EdmundOptics), a long working-distance
objective (20X, Mitutoyo), a prism which orients the beam in the di-
rection of the flexible plate tip, and, eventually, a mirror that reflects
back light towards the prism. Thus the flexible plate tip is imaged,
through the objective and the optical tube, on a photo-sensitive de-
tector (S3931, Hamamatsu). A feedback procedure is applied as in
[13] in order to mimic an infinite stiffness of the flexible microplate.
Image analysis and geometric reconstruction
Images were treated with ImageJ software (National Institutes of
Health, Bethesda, MD). For side-view experiments, 6 geometrical
points were identified at each time position, corresponding to the 4
contact points of the cell surface with the microplates and to the
2 extremities of the cell 'equator', i.e.
the mid-points where cell
surface is perpendicular to microplates. Assuming a symmetry of
revolution, these points define uniquely the cell equatorial radius
Rc, the average radius at the plates Rp, the cell half-height h, and
the average curvature of the cell surface κ (average of the inverse
of the radii of the circles shown in Fig. S1a). For bottom-view ex-
periments, only the radius Rp at the bottom plate can be acquired
dynamically. The initial radius of cells when still spherical was mea-
sured using transmission image microscopy, it was used as the value
h0 and was found to be consistent with side-view measurements of
h0. In the case of infinite stiffness, we assumed that the curvature of
fully spread cells viewed from the bottom behaved as the curvature
of side-viewed cells, κh0 = 0.90 ± 0.03 (n = 9). This allowed to esti-
mate the fully-spread radius at the equator, Rc, and to use n = 13
experiments for identifying data in the fully spread configuration for
infinite stiffness.
Data analysis and model resolution
Microplate deflection time series were converted to force time series
as described in [30]. Force data was then re-gridded onto the time
positions of the microscopy images using the in-house open-source
software DataMerge, based on the LOESS implementation of the
GNU Scientific Library. Analytical calculations were assisted by the
open-source computer algebra system Maxima. Numerical simula-
tions of ordinary differential equations derived from the model were
simulated using the GNU Octave scientific computing programming
language for figure 2. Analytical implicit equations were plotted
using the open-source Gnuplot software in figures 2a, 3c and 4a.
Acknowledgments
J.E. thanks especially John Hinch, Claude Verdier, Martial Balland,
Karin John, Philippe Marmottant and Cyril Picard for fruitful dis-
cussions. J.E. wishes to acknowledge funding by R´egion Rhone-
Alpes (Complex systems institute IXXI and Cible), ANR "Trans-
mig" and Tec21 (Investissements d'Avenir - grant agreement n ANR-
11-LABX-0030). The experimental work was supported in part by
ANR funding (ANR-12-BSV5-0007-01, "ImmunoMeca").
References
[1] F. R. Turner and A. P. Mahowald. Scanning electron mi-
croscopy of drosophila melanogaster embryogenesis: II. Gas-
trulation and segmentation. Develop. Biol., 57:403 -- 416, 1977.
[2] A. Engler, Lucie Bacakova, C. Newman, A. Hategan, M. Griffin,
and D. E. Discher. Substrate compliance versus ligand density
in cell on gel responses. Biophys. J., 86:617 -- 628, 2004.
[3] T. Yeung, P. C. Georges, L. A. Flanagan, B. Marg, M. Ortiz,
M. Funaki, N. Zahir, W. Ming, V. Weaver, and P. A. Janmey.
Effects of substrate stiffness on cell morphology, cytoskeletal
structure, and adhesion. Cell Motil. Cytoskeleton, 60:24 -- 34,
2005.
[4] A. Saez, M. Ghibaudo, A. Buguin, P. Silberzan, and B. Ladoux.
Rigidity-driven growth and migration of epithelial cells on mi-
crostructured anisotropic substrates. Proc. Natl. Acad. Sci.
USA, 104:8281 -- 8286, 2007.
[5] J. Zhong, J. B. Baquiran, N. Bonakdar, J. Lees, Y. Wooi Ching,
E. Pugacheva, B. Fabry, and G. M. O'Neill. NEDD9 stabilizes
focal adhesions, increases binding to the extra-cellular matrix
and differentially effects 2D versus 3D cell migration. PLoS
one, 7:e35058, 2012.
[6] P. E. Young, A. M. Richman, A. S. Ketchum, and D. P. Kiehart.
Morphogenesis in Drosophila requires nonmuscle myosin heavy
chain function. Genes Dev., 7:29 -- 41, 1993.
[7] R. J. Pelham Jr. and Y. Wang. Cell locomotion and focal ad-
hesions are regulated by substrate flexibility. Proc Natl Acad
Sci USA, 94:13661 -- 13665, 1997.
[8] M. E. Chicurel, C. S. Chen, and D. E. Ingber. Cellular control
lies in the balance of forces. Curr. Opin. Cell Biol., 10:232 -- 239,
1998.
8
[9] A. L. Zajac and D. E. Discher. Cell differentiation through tis-
sue elasticity-coupled, myosin-driven remodeling. Curr. Opin.
Cell Biol., 20:609 -- 615, 2008.
[10] Y. Cai, O. Rossier, N. C. Gauthier, N. Biais, M.-Antoine
Fardin, X. Zhang, L. W. Miller, B. Ladoux, V. W. Cornish,
and M. P. Sheetz. Cytoskeletal coherence requires myosin-IIA
contractility. J. Cell Sci., 123:401 -- 423, 2010.
[11] D. E. Discher, P. Janmey, and Y. Wang. Tissue cells feel and
respond to the stiffness of their substrate. Science, 310:1139 --
1143, 2005.
[12] V. Vogel and M. Sheetz. Local force and geometry sensing
regulate cell functions. Nat. Rev. Mol. Cell Biol., 7:265 -- 275,
2006.
[13] D. Mitrossilis, J. Fouchard, A. Guiroy, N. Desprat, N. Ro-
driguez, B. Fabry, and A. Asnacios. Single-cell response to
stiffness exhibits muscle-like behavior. Proc. Natl. Acad. Sci.
USA, 106:18243 -- 18248, 2009.
[14] K. D. Webster, A. Crow, and D. A. Fletcher. An afm-based
stiffness clamp for dynamic control of rigidity. PLoS ONE,
6:e17807, 2011.
[15] A. Crow, K. D. Webster, E. Hohlfeld, W. P. Ng, P. Geissler,
and D. A. Fletcher. Contractile equilibration of single cells to
step changes in extracellular stiffness. Biophys. J., 102:443 -- 451,
2012.
[16] L. Trichet, J. Le Digabel, R. J. Hawkins, S. R. K. Vedula,
M. Gupta, C. Ribrault, P. Hersen, R. Voituriez, and B. Ladoux.
Evidence of a large-scale mechanosensing mechanism for cellu-
lar adaptation to substrate stiffness. Proc. Natl. Acad. Sci.
USA, 109:6933 -- 6938, 2012.
[17] O. M. Rossier, N. Gauthier, N. Biais, W. Vonnegut, M.-A.
Fardin, P. Avigan, E. R. Heller, A. Mathur, S. Ghassemi, M. S.
Koeckert, J. C. Hone, and M. P. Sheetz. Force generated by ac-
tomyosin contraction builds bridges between adhesive contacts.
EMBO J., 29:1033 -- 1044, 2010.
[18] A. Zemel, F. Rehfeldt, A. E. X. Brown, D. E. Discher, and S. A.
Safran. Optimal matrix rigidity for stress-fibre polarization in
stem cells. Nature Phys., 6:468 -- 473, 2010.
[19] P. Marcq, N. Yoshinaga, and J. Prost. Rigidity sensing ex-
plained by active matter theory. Biophys. J., 101:L33 -- L35,
2011.
[20] M. Yamamoto. The visco-elastic properties of network struc-
I. General formalism. J. Phys. Soc. Jpn, 11:413 -- 421,
ture:
1956.
[21] A. F. Huxley. Muscle structure and theories of contraction.
Prog. Biophys. Biophys. Chem., 7:255 -- 318, 1957.
[22] M. Kov´acs, K. Thirumurugan, P. J. Knight, and J. R. Sellers.
Load-dependent mechanism of nonmuscle myosin 2. Proc. Natl.
Acad. Sci. USA, 104:9994 -- 9999, 2007.
[23] S. Mukhina, Y. Wang, and M. Murata-Hori. α-actinin is re-
quired for tightly regulated remodeling of the actin cortical
network during cytokinesis. Dev. Cell, 13:554 -- 565, 2007.
[25] A. Vaziri and A. Gopinath. Cell and biomolecular mechanics
in silico. Nature Mater., 7:15 -- 23, 2008.
[26] A. K. Harris, P. Wild, and D. Stopak. Silicone rubber sub-
strata: a new wrinkle in the study of cell locomotion. Science,
208:177 -- 179, 1980.
[27] X. He and M. Dembo. On the mechanics of the first cleavage
division of the sea urchin egg. Exp. Cell Res., 233:252 -- 273,
1997.
[28] K. Kruse, J.F. Joanny, F. Julicher, J. Prost, and K. Sekimoto.
Generic theory of active polar gels: a paradigm for cytoskeletal
dynamics. Eur. Phys. J. E, 16:5 -- 16, 2005.
[29] M. Soares e Silva, M. Depken, B. Stuhrmann, M. Korsten, F. C.
MacKintosh, and G. H. Koenderink. Active multistage coars-
ening of actin networks driven by myosin motors. Proc. Natl.
Acad. Sci. USA, 108:9408 -- 9413, 2012.
[30] D. Mitrossilis, J. Fouchard, D. Pereira, F. Postic, A. Richert,
M. Saint-Jean, and A. Asnacios. Real-time single cell response
to stiffness. Proc. Natl. Acad. Sci. USA, 107:16518 -- 16523, 2010.
[31] A. Saez, A. Buguin, P. Silberzan, and B. Ladoux. Is the me-
chanical activity of epithelial cells controlled by deformations
or forces? Biophys. J., 89:L52 -- L54, 2005.
[32] T. J. Mitchison and L. P. Cramer. Actin-based cell motility
and cell locomotion. Cell, 84:371 -- 379, 1996.
[33] M. F. Fournier, R. Sauser, D. Ambrosi, J.-J. Meister, and A. B.
Verkhovsky. Force transmission in migrating cells. J. Cell Biol.,
188:287 -- 297, 2010.
[34] T. D. Pollard, L. Blanchoin, and R. D. Mullins. Molecular
mechanisms controlling actin filament dynamics in nonmuscle
cells. Annu. Rev. Biophys. Biomol. Struct., 29:545 -- 576, 2000.
[35] A. Ponti, M. Machacek, S. L. Gupton, C. M. Waterman-Storer,
and G. Danuser. Two distinct actin networks drive the protru-
sion of migrating cells. Science, 305:1782 -- 1786, 2004.
[36] J. V. Small and G. P. Resch. The comings and goings of actin:
coupling protrusion and retraction in cell motility. Curr. Opin-
ion Cell Biol., 17:517 -- 523, 2005.
[37] E. Fallman, S. Schedin, J. Jass, B.-E. Uhlin, and O. Axner.
The unfolding of the P pili quaternary structure by stretching
is reversible, not plastic. EMBO rep., 6:52 -- 56, 2005.
[38] W. Federle, E. L. Brainerd, T. A. McMahon, and B. Hlldobler.
Biomechanics of the movable pretarsal adhesive organ in ants
and bees. Proc. Natl. Acad. Sci. USA, 98:6215, 2001.
[39] A. V. Hill. The heat of shortening and the dynamic constants
of muscle. Proc. R. Soc. Lond. B, 126:136 -- 195, 1938.
[40] G. B. West, W. H. Woodruff, and J H. Brown. Allometric
scaling of metabolic rate from molecules and mitochondria to
cells and mammals. Proc. Natl. Acad. Sci. USA, 99:2473 -- 2478,
2002.
[24] M. Fritzsche, A. Lewalle, T. Duke, K. Kruse, and G. Charras.
Analysis of turnover dynamics of the submembranous actin cor-
tex. Mol. Biol. Cell, 24:757 -- 767, 2013.
[41] N. Desprats, A. Guiroy, and A. Asnacios. Microplates-based
rheometer for a single living cell. Rev. Sci. Instrum., 77:055111,
2006.
9
Appendix
S1 Role of microtubules
It has been reported that in addition to the substrate, part (of
order 13%) of the cortical tension could be balanced by the
resistance to compression of microtubules [55]. We have thus
controlled whether this was the case in our setup, in particular
for low external stiffness, which corresponds to lower height of
cells and thus are geometrically more likely to involve micro-
tubule compression. The results, shown in Fig. S3, indicate
that there is no such influence within experimental error. This
allows us to neglect the role of microtubules compared to ac-
tomyosin tension and microplate resistance to bending in the
modelling that follows.
S2 Model derivation
As stated in the text of the article, we are looking for the
simplest model consistent with the fact that the actin plus
crosslinkers network in vivo is not able to resist extensional
stress in the long term. This is consistent with a dominant
loss modulus at low frequencies in cell-scale rheological prob-
ing [56] and in vitro studies [29], and is linked with the fact that
crosslinkers in vivo are transient with a short residence time
[24]. Basic models of transiently crosslinked networks based
on rubber-like models were first explored by Green and Tobol-
sky [57] and Yamamoto [20], and their nonlinear properties are
still being investigated [58]. Their response, up to the first or-
der, turns out to be the same as the one of polymer solutions,
that is, their stress-strain relationship is governed by Maxwell
constitutive Eq. 1,
(cid:79)
σ +σ − 2ταE ε = 0,
τα
with σ = 2E(cid:0)β2 (cid:104)RR(cid:105) − I(cid:1) ([59, p.
116], the parameter
β2 = 3/(2NK b2
K ) is related to bK the length and NK the num-
ber of Kuhn steps between two crosslinks). Here the upper
(cid:79)
σ= σ − ∇vTσ − σ∇v
convected Maxwell tensor derivative
takes into account the affine stretching of the strand vectors
R, the basic units of the network, by the velocity gradient ∇v.
The difference is that, for polymer solutions, the time τα is
the ratio of solvent viscosity to polymer elasticity, because this
is the characteristic time at which the polymers can deform
relatively to an affine global deformation of their surroundings
(the solvent, [59, p. 123]), while, in the case of transiently
crosslinked networks, τα is the characteristic unbinding time of
the crosslinks. Thus, the product ταE which has the dimension
of a viscosity is only some apparent viscosity at the macroscopic
scale, and corresponds in fact to an elastic energy dissipation
at rate 1/τα. If there is a large number of crosslinkers present
along a single filament, there will not be a single relaxation
time τα but several [60].
In the present work, we choose to
investigate the properties of the single-relaxation time model
above because this allows to calculate analytically the model
solution while retaining the essence of a long-time viscous-like
and short-time elastic-like material.
A fraction αmyo of the crosslinkers considered are myosin
bipolar filaments. In addition to their crosslinking role, they
may effectuate a power-stroke at a frequency 1/τmyo, which re-
sults in "sliding" the corresponding crosslinker by the myosin
step length (cid:96). If ψ(ρ) is the orientational distribution function
[59], this appears as additional sink and source terms in the
right hand side of the probability balance equation,
∂ψ
∂t
(cid:17)
+ ∇ρ ·(cid:16) Rψ
(cid:18)
=
αmyo
τmyo
(cid:39) αmyo(cid:96)2
2τmyo
∂2ψ
∂ρ2
−ψ(ρ) +
ψ(ρ − (cid:96)ρ/ρ) +
1
2
ψ(ρ + (cid:96)ρ/ρ)
1
2
(cid:19)
(4)
which, multiplied by ρρ and integrated over all configurations,
yields an additional term of contractility
(cid:79)
σ +σ − 2ταE ε = Σ = ΣA,
τα
αmyo((cid:96)β)2 is proportional to the myosin con-
where Σ = E τα
τmyo
centration and power-stroke frequency. The tensor A is the
local orientation tensor of the actin fibres,
(cid:90)
A =
ρρ
ρ2 ψ dρ.
ρ
The ratio of the apparent viscosity ταE and contractile stress
Σ provides us with another characteristic time, τc = 2ταE/Σ =
2τmyoα−1
myo((cid:96)β)−2, which characterises the dynamics of shrink-
ing of an actomyosin network in the absence of crosslinkers, as
is the case in in vitro experiments [29].
Note that both characteristic times of crosslinker unbind-
ing τα and of myosin power-stroke τmyo have been taken as con-
stants, independent of the stress or strain they are submitted
to. It is of course well established that there is a dependence
of these parameters on stress and strain [61], which would in-
troduce a nonlinearity in the model. An important, unsolved
question is to determine whether these dependences are or not
major players in the cell-scale mechanical behaviour of acto-
myosin, e.g. through a stress-driven ripping of crosslinks. As
their effect is a nonlinear variation of the above model, the
standard modelling procedure is to study the linear response
first (with constant characteristics times), before considering
the full nonlinear model.
It is found in the sequel that the
linear model is sufficient to reproduce the experimental data,
that is, for our experiments, collective effects explain the ob-
served behaviour by themselves1. In these conditions, the effect
of additional nonlinear terms could not be distinguished from
this linear baseline, and were therefore not introduced in the
present work.
The model obtained is thus the one of a viscoelastic liquid.
This is in line with the seminal model by He and Dembo [27]
who modelled the cytoskeleton as a viscous fluid and used it in
numerical simulations of the cytokinesis. This is also similar
to the actin dynamics part of the model used by [62] to simu-
late keratocyte migration. We provide above a microstructure-
based derivation of this class of models, which allows us to
interpret the dissipation in terms of molecular behaviours (see
the discussion on Hill's law, Appendix S3.5). This type of
model, to the best of our knowledge, was never used to study
mechanosensing behaviours of single cells. In the sequel (Ap-
pendix S4), we compare it to models that have been used to
analyse cell-scale mechanosensing, but will first investigate its
basic predictions.
S3 One-dimensional problem
S3.1 In the absence of treadmilling
In this section we investigate the behaviour of a material mod-
elled by the constitutive law, Eq. 2,
in a simplified, one-
dimensional geometry described in Fig. 3a. The actomyosin
1This is also the case for Huxley's model of muscle contraction [21], where the author observes, p. 290, that assuming
specific force-dependent binding kinetics only tunes the system's efficiency, not affecting its ability to fit Hill's force-velocity
experimental observations.
10
network is assumed to occupy an infinite cuboid between two
horizontal plates, one of which is fixed and the vertical posi-
tion of the other governed by a spring of stiffness k/S (per unit
area) and equilibrium distance with the other plate H0. If the
current distance between the plates is H, the force exerted by
the top plate on the material is thus F/S = k(H0 − H)/S per
unit area.
We assume that the bulk forces acting on the actomyosin
network (such as the friction with the cytosol) are negligible
compared to the force developed by myosin contraction, this
writes:
∇ · σ = 0,
(5)
with the boundary condition σez = (F/S) ez at the upper
plate; and corresponds to neglecting friction with the cytosolic
fluid in the balance between the actomyosin stress and the force
at the plates. By symmetry, only the vertical component σzz
of the stress tensor σ in the actomyosin material is nonzero,
we denote it ζ, from the above we have ∂zζ = 0 and thus the
boundary condition imposes ζ = F/S = k(H0 − H)/S at every
z, the stress is fully transmitted through the material.
On the other hand, the rate-of-strain tensor ε is also lim-
ited to a vertical component ε = H/H. Using these equalities,
the constitutive Eq. 2 describes the complete dynamics of the
ζ = 0) is
system.
reached for
In particular, a permanent regime ( ε=0,
(cid:115)(cid:18) k + kc
kc
(cid:19)2 − 4k
kc
−
k + kc
(cid:26) kH0
ΣS
2
kc
ΣS
if k < kc,
else,
(6)
Fp = ζpS =
=
for k < kc, with kc = ΣS/H0. The actomyosin model's re-
sponse is thus very close to the behaviour of a spring of stiff-
ness kc and free length 0, when put in series with an external
spring of stiffness k: Indeed, the asymptotes are the same for
k → 0 and k → ∞ (Fig. 3c), which is not obvious since in the
permanent regime the model only includes viscous dissipation
and contractility. We have thus shown that a contractile fluid
behaves like a spring in these conditions.
Indeed, when the cell is pulling on a spring of stiffness k,
the maximum deflection it can impose to the spring is H0, e. g.
the initial cell height and, consequently, the maximum height
it can shorten. However, this maximum deflection is achieved
only if the maximum force generated by the cell ΣS is larger
than the external spring force at maximum deflexion kH0, in
other words only if k is lower than kc = ΣS/H0. In that case,
the force reached is kH0, thus proportional to k. In the other
case k > kc, the deflection is less than H0 and is set by having
an equal tension ΣS in both the external equivalent spring k
and the cell.
S3.2
In the presence of treadmilling
We now introduce the fact that actin filaments in vivo are
constantly being polymerised from one end ('plus' end) and
depolymerised, mostly from the other ('minus' end), which re-
sults in the so-called treadmilling phenomenon [32]. Assuming
that this treadmilling is at steady-state in the cell on average
at the relevant time scale of our experiment, the effect of tread-
milling in the bulk does not affect the modelling assumptions
done in Appendix S2: the elastic modulus E of the crosslinked
network at any instant will have a constant average. However,
at the boundary, some filaments will have their 'plus' end ori-
ented towards the boundary and thus polymerisation of these
will entail a net extension of the material (before stress equili-
bration, depending on the boundary conditions).
In the framework of our one dimensional toy problem, this
introduces a drift between the deformation of the material and
the displacement of the boundary, which can be expressed as:
H = H ε + 2vt
(7)
with vt = (cid:96)a/τa the treadmilling speed, that is, the extra length
added by a monomer polymerising divided by the characteris-
tic time for an ATP-fuelled monomer addition. The factor 2 is
due to this effect taking place at both plates.
When injected into the constitutive model, this modifies
the level of force (and height Hp) in the permanent regime:
k + kc
kc
(cid:115)(cid:18) k + kc
kc
(cid:19)2 − 4k
ΣS
−
(H0 − He)
Fp = ζpS =
ΣS
2
,
(8)
(cid:96)a
with a modified critical stiffness kc = ΣS/(H0 + Hα). Here
Hα = 4ταvt = 4τα(cid:96)a/τa is a characteristic elastic length of
newly polymerised material. The critical stiffness kc is low-
ered in proportion with the corresponding relative increase of
height. We have also introduced another new parameter, the
length He = 2τcvt = 4 τmyo
. This length is discussed
τα
in the main text, it is a trade-off between the rate at which
actin contracts under the effect of myosin, and the speed at
which actin network expands by polymerisation at its edges.
It defines the shortest height that the cell achieves in the 1D
model, when the stiffness of the plates is vanishingly small: in
that case, all of the work of myosins is spent in contracting the
part of actin network newly polymerised.
Fp ∼ k(H0 − He), and for a large stiffness,
For a low stiffness, this expression has for asymptote
((cid:96)β)2αmyo
Fp → ΣS
H0 − He
H0 + Hα
(9)
This does not change the qualitative spring-like response of the
material, only the equilibrium length of the equivalent spring
is not zero anymore but He. This means that as the device
stiffness goes to zero, the height of the model meshwork will
tend to He = 2τcvt, which is a dynamic balance between the
speed 2vt at which new material is added at the two boundaries
(modelling polymerisation), and the rate 1/τc at which the ex-
isting material is contracted by the myosin motors. Inside this
model material, even once the final length He is reached and
the boundaries are immobile, there remains a continuous cen-
tripetal flow (Fig. 1b) that is very reminiscent of the retrograde
actin flow observed in both crawling [32] and immobile spread
cells [17]. Exactly as in these cases, this is made possible by
depolymerisation inside the model material. This is an addi-
tional source of energy dissipation in steady state, when the
cell is apparently at equilibrium with constant length He and
force Fp.
S3.3 Dynamics
The above constitutive and force balance equations can be writ-
ten in terms of one equation only with the tension as the un-
known,
∂ζ
∂t
=
Σ (k(H0 − He)/S − ζ) + ζ (ζ − k(H0 + Hα))
τcΣ + τα(ζ + kH0/S)
.
(10)
One can analyse the rate of tension increase when the system
begins to pull and ζ is still much smaller than Σ, for a low
stiffness k (cid:28) ES/H0,
∂F
∂t
=
k(H0 − He) − F (t)
τc
∼ Fp(k) − F (t)
τc
.
(11)
And for high stiffness k (cid:29) ES/H0,
∂F
∂t
=
=
ΣS(H0 − He) − F (t)(H0 + Hα)
(Fp(k) − F (t)) (H0 + Hα)
ταH0
.
τα H0
(12)
(13)
11
Experiments for very low values of the microplate stiffness
k allow to identify the parameters τc and vt of the model.
From n = 9 experiments with k ≤ 1.6 nN/µm, we find
2τcvt = He = 6.6±1.1 µm and H0 = 13.2±0.1 µm. This is con-
sistent with the fact that ∂F/(∂k) = H0− He = 6.8± 0.4 µm in
[13]. Additionally, the maximum force attained Fp, the current
force F (t) and the rate of growth of this force ∂F /∂t allow to
calculate τc using Eq. 11, τc = (Fp−F (t))(∂F /∂t)−1 = 521±57
s. This yields vt = 6.5± 1.5 nm/s, which is consistent with the
literature [17] as stated in the main text.
S3.4 Response to a step-change of external
stiffness
In [30], we are able to vary instantaneously (within 0.1 s) the
stiffness k of the external spring, while ensuring that there is
no instantaneous change of the force F felt by the cell or of
the microplate spacing H. In the modelling, this corresponds
to an instantaneous change of both k and H0 at time t∗ such
that F and H are continuous. We can thus write the following
relations, ensured by the experimental setup:
(cid:26) k0, t < t∗,
k1, t > t∗,
k(t) =
F (t) = F ∗ + φ(t), H(t) = H∗ − φ(t)
k(t) ,
where φ(t) is the variation of force around the force at time t∗,
φ(t∗) = 0.
Changing only the stiffness in this manner ensures that the
cell does not feel any step-change: indeed, force and geometry
are preserved, the only change is the response of the microde-
vice to a variation of the force applied to it. The experimental
result is that, despite the fact that none of the physical observ-
ables have been changed for the cell, its rate of loading ∂F /∂t
of the device is instantaneously modified. In [30], it is found
that, when going from a stiffness k0 to a stiffness k1, the new
rate of loading matches the rate of loading that the same cell
type exhibits in a constant stiffness k = k1 experiment, at a
time such that F = F ∗. In [15], this experiment is repeated
with another cell type and it is also found that the rate of load-
ing is instantaneously changed. However, the value ∂F /∂t for
t (cid:38) t∗ exhibits an overshoot:
it is initially different from the
value of the corresponding constant stiffness experiment, and
exhibits a relaxation towards it.
We can rewrite Eq. 10 around ζ∗ = F ∗/S, we find:
∗
(t
∂F
∂t
) = k(t)
H∗(ΣS − F ∗) − HαF ∗ − HeΣS
τcΣS + τα(k(t)H∗ + 2F ∗)
.
(14)
The result is thus a step-function, whose value for t (cid:38) t∗ is
exactly the rate of growth predicted by the model for a con-
stant stiffness experiment with k = k1. This corresponds to
the experimental results in both [30] and [15], except for the
overshoot found in the latter.
The step-change of the rate of loading can thus be ex-
plained by a purely intrinsic property of the cell's cytoskeleton,
as developed in the main text.
In [15], a model is proposed that accounts for a step-change
and relaxation, as they observe experimentally. However this
model has a limited validity around t∗ and breaks down at long
times, predicting infinite force. Here, our model originally aim-
ing at describing the long-time behaviour of the cell depending
on stiffness predicts the main feature of the experiments (the
step-change of the rate of loading), but not the overshoot found
in one of the experiments. Our model however may produce
this type of overshoot if more than a single relaxation time is
used. This would correspond to replacing the spring k1 in the
model in [15] by our viscoelastic constitutive equation. This is
found not to be necessary to explain the data studied here.
In Fig. 2c, we give a numerical simulation result that cor-
roborates the step-change of ∂F /∂t found in Eq. 14 and can be
compared to an experimental curve without any parameter ad-
justment. The time-profile of the microplate effective stiffness
is imposed to be the same in the numerical simulation as in the
experiment. The profile of force increase obtained presents the
same instantaneous change of slope as the stiffness is varied,
and the overall profiles match quantitatively.
S3.5 Hill's law
We examine the power dissipation predicted by the one-
dimensional model between two boundaries. The boundaries
move towards the sample at velocity v (which is thus positive
when the sample contracts) and feel a force F exerted by the
sample (positive in the direction of contraction).
Hill's historical experiment [39] takes place at a constant
level of force F . Our experiments, on the other hand, prescribe
a relationship between F and v through the microplate stiff-
ness, ∂F /∂t = −kv. Both can scan the F -- v relationship by
varying F for the former or k for the latter (in the case of [13],
we needed to investigate this relationship for a fixed H = H0−δ
where δ is small), Fig. 4a and b. The calculation below does
not require to use either of these experimental ways to scan
the F -- v relationship. It relies only on the material properties
of the sample.
As above, the mechanical equilibrium of the sample im-
poses that the vertical component of the stress tensor ζ is equal
to F/S. Also, the treadmilling produces a mismatch between
the plate velocity v = H and the recoil of the existing network
at the plate −H ε, expressed by v = −H ε + 2vt (see Eq. 7).
These relations can be injected into the constitutive Eq. 2:
(cid:16) F − 2 εF
(cid:17)
τα
+ F − 2ταES ε = ΣS.
We recover a virtual work formulation by multiplying this by
the velocity H/(2τα). Adding the constant ESH/(2τα) to both
sides, this work can now be factorised in the manner of Hill's
law:(cid:18) F
S
(cid:19)(cid:18)
(cid:19)
(cid:18)
+ E
v + 2vt +
H
2τα
= (Σ + E)
H
2τα
− H
2S
F .
(15)
The meaning of some of these terms is explained in the main
text.
In the case when the velocity v is zero, the force generated
is finite as calculated above, Eq. 9, which can also write:
Fmax
S
= Σ
1 − E + Σ
Σ
2vt
vα + 2vt
(cid:19)
Even if both v and vt are zero, and thus the actomyosin does
not contract macroscopically ( ε = 0), the force generated re-
mains finite since it does work at a molecular scale. Actin
polymerisation, by adding new material at the edges, intro-
duces a 'boundary creep', which consumes additional work in
conditions of fixed length (v = 0).
As stated in the main text, apart from the polymerisation
'boundary creep', this is the same dissipative mechanism as in
the model of muscle contraction by Huxley [21]. In this model,
myosin 'elastic tails' are prestretched preferentially in one di-
rection before binding the actin thin filament. When there is no
net sliding (v = 0), they eventually unbind without having had
the opportunity to provide work, that is, they have conserved
this level of stretching. Although this is not explicitly written
in the 1957 paper, the prestretch that had been bestowed on
the myosin 'elastic tail' is thus lost.
Zero force F = 0 condition is a (theoretical) limit cor-
responding to zero-stiffness of the microplate. This is not at-
tainable experimentally, as cells do not spread on both plates if
their displacement does not generate any external force -- which
corresponds to the impossibility to apply any normal force to
the plates. This effect probably has to do with the mechanism
12
of force reinforcement of adhesions. In the model however, the
limit can be studied, and yields a maximum velocity,
vmax + 2vt =
Σvα
E
=
H
τc
.
This corresponds to the power injected by the myosin mo-
tors, divided by the elastic modulus of the crosslinked network
(minus the treadmilling contribution):
indeed, in this limit,
the myosins work against the elasticity of the actomyosin it-
self. The maximum velocity is thus limited by the rate τα
at which the actin network fluidises thanks to the detachment
of crosslinkers.
In comparison to the case of muscles, actin
treadmilling reduces the maximum speed of shortening by 2vt,
as the receding speed of the edge of the actin network must
compensate for this speed of protrusion. Quantitatively, using
the values obtained in Appendix S3.3, we predict vmax (cid:39) 12.3
nm/s, this is very close to vmax = 13 nm/s published in [13].
Again, the dissipation here is of the same nature as for mus-
cles in the model by Huxley: in his case, the only crosslinker
between thin and thick filaments are myosin heads, and zero
force is obtained at the finite velocity at which the work rate
performed by pulling myosin heads is exactly balanced by the
work rate needed to deform myosin heads that have not yet
detached from the actin filament. They will eventually detach,
thus dissipating at a fixed rate the elastic energy that has been
transferred to them.
When neither F nor v are zero, there is a nonzero produc-
tive work performed on the plates. Because of the existence of
maximum force and velocity, it is necessarily written
(cid:18)
(cid:19)
.
F v
Fmaxvmax
= r
1 − F
Fmax
− v
vmax
In both the experiments by A. V. Hill [39] and ours [13], r is
found to be mostly independent of F and v, and r (cid:39) 0.25.
In Huxley's model of muscles, r is a signature of the preferen-
tial pre-stretch of myosins that models the power-stroke, nor-
malised by the detachment rate [21, 63]. In our model of cells,
2r (cid:39) τc/τα is the ratio of the characteristic times of contraction
and of stress relaxation through crosslinker unbinding. This
does not explain the coincidence of finding the same value of r
in both cells and muscles, however we note that this parameter
has a similar signification in both models.
Specifically,
(cid:16)
E
1 − E+Σ
Σ
r =
Σ
(16)
2vt
vα+2vt
(cid:17) .
(cid:18) τc
2τα
(cid:19)2
Thus in our case, Hill's parameter r is such that
E
Σ
=
τc
2τα
≤ r ≤ τc
2τα
+
,
(17)
depending on vt.
Note that these equations can be applied to the acto-
myosin pushing against an obstacle, with 0 ≤ −F ≤ ES and
0 ≤ −v ≤ vt.
S3.6 Quantitative analysis
Since r (cid:39) 0.25 in experiments [13], τc/τα has to be in the range
0.4 to 0.5, thus since τc = 521 ± 57 s, we have τα = 1186 ± 258
In turn, using also vt = 6.5 nm/s (see S3.3), we find
s.
ΣS = Fmax(H0 +Hα)/(H0−He) = (2.0±0.9)·103 nN (n = 13).
Thus the four parameters of the 1D model, namely τα, τc, vt
and ΣS are identified using only the average maximum force
(equilibrium of infinite stiffness experiments), maximum ve-
locity (dynamics of experiments with very low stiffness), and
shortest height at equilibrium (experiments with very low stiff-
ness), plus the shape of the Hill-type law (parameter r).
The other experimental data (stiffness-dependence of the
force, Fig. 3c; dynamics as the microplate stiffness is mod-
ified, Fig. 2c; and values of F (t) at H0 − H = 1µm, Fig.
4a) are matched by the model without any adjustable pa-
rameter. E.g., the values found yield a critical stiffness kc =
ΣS/(H0 + 4ταvt) = 41 nN/µm, which is consistent with the
experimental results, Fig. 3c.
Moreover, the values for τc, τα and especially vt are inde-
pendently measurable and are consistent with the literature,
see text.
S4 Comparison with other models featur-
ing cell-scale mechanosensing
In [18, 64], the cell is modelled as a linear elastic body hav-
ing an intrinsic equilibrium shape, and which is prestretched
to some maximum strain, in our notation:
σ = 2E(ε − ε0),
with ε0 an equilibrium shape that the cell would take in the ab-
sence of external stresses. They supplement this model with a
phenomenological feedback on the elastic modulus, on the time
scale of hours or days, which corresponds to the phenomenon
of stress-fibre polarisation [65]. This long term effect is out of
the scope of our model and experiments, thus our model com-
pares to theirs before this feedback comes into play. In a one-
dimensional setting, their model thus corresponds to the pre-
stretched spring in Fig. 3b, which yields a stiffness-dependence
closely mimicking experimental data and our model predic-
tions, Fig. 3c. Our model in addition provides a microstructure
basis for this qualitative behaviour and a quantitative reading
of experiments, in particular the equilibrium shape ε0 and cell
stiffness kc in [18] are explicited, respectively, as proportional to
the product of a characteristic contraction time and the tread-
milling speed, τcvt, and as the ratio of the myosin contractile
stress and a length, kc = ΣS/(H0 + 4ταvt).
In [16], they introduce a model also in the framework of
active gel theory. However, they resort to the stress-strain
relationship of [18] so as to avoid to calculate the orientation
tensor Qij of the cytoskeleton and use a 1D linear elastic stress-
strain relationship, their equation (S7) writes in our notation
σ = 2E(H − He).
In [19], they present a 1D visco-elastic solid model, com-
bining their equations (2) and (5) writes in our notation:
σ = 2E(H − He) + 2η H.
C − FS/kC in their notation. To the difference
with He = l0
to the previous models, the viscous term added in this model
allows to study the dynamics of the cell shortening. However,
the equilibrium shape in all three models is a static elastic bal-
ance between the environment resistance to deformation and a
phenomenological internal elasticity, which corresponds to the
spring model described in Fig. 3b.
S5 Three dimensional problem
In this section, we present a full three-dimensional model of the
cell mechanics as a contractile visco-elastic thin shell, obeying
the rheological Eq. 2, and enclosing an incompressible cytosol.
Forces are transmitted to the microplates at the contact line
between the cell boundaries and the microplates.
S5.1 Geometry
It is seen from experimental observations that the cell bound-
aries connecting the plates are in most cases very well approx-
imated by an arc of circle (Fig. 1b, insert). This had already
been shown in the case of cells spread on a microneedle array
having reached a stationary shape [66], in the present setup we
find that it is also true while the cell is spreading (Fig. S1a).
13
When observing cells from the side, we assume that the cell
shape is cylindrical. This is supported by other experiments
where cells are observed from the bottom in TIRF, Fig. S1b.
Thus, along the axis z orthogonal to the microplates, whose
location is parametrised as z = ±h where h is the half-height
of the cell, we can fit experimental results using the law:
(cid:16)
1 −(cid:112)1 − (κz)2
(cid:17)
,
(18)
r(z) = Rc +
1
κ
where Rc is the radius at the cell equator (z = 0) and κ the
signed curvature in the vertical plane, Fig. 1b. The curvature
κ evolves in time from a positive curvature (t = 10 s in Fig.
S1a) to a negative one (t = 120 s and later in Fig. S1a).
To the physicist it may be a surprise that the cell shape is
not well fitted by a minimal surface such as a catenoid. Indeed,
although each boundary seen on Fig. S1a can be reasonably fit-
ted with an hyperbolic cosine function, the asymptotes of these
fits do not match: the cell shape is close to a "minimal" sur-
face with a different weight on its curvatures, κσ + κ⊥σ⊥ = 0,
where κ⊥ is the curvature in a plane orthogonal to the side
view. In light of Laplace law, these weights can be interpreted
as tensions of different magnitude in the longitudinal (es in
Fig. 1b) and orthoradial (eφ = en × es) directions. The reason
for these different tensions and details of this Laplace law are
given in the next section. We cannot fit the cell shape with an
analytical function matching this law, firstly because functions
solving the corresponding differential equation have never been
investigated and do not have the same properties as the hyper-
bolic cosine, secondly because the tensions σ and σ⊥ actually
vary in some measure along the vertical direction.
From this we can calculate the volume of the cell through
time. Here again, the physicist is surprised to find that the
volume defined by Eq. 18 and the plate positions z = ±h
is not a constant, Fig. S2. However, this volume is not the
volume of the cell itself but the volume enclosed by the lat-
eral cell boundaries:
indeed, it is observed in side views of
cells presenting such a large enclosed volume increase that the
cell detaches from the microplates in the central region of the
contact area (Fig. S2), forming a "pocket" between the cell
membrane and the microplate, which has every reason to be
filled by culture medium seeping between the adhesions seen
in Fig. S1b. Although it was not possible to track the volume
of these pockets through time and compare it to the enclosed
volume variations, we can estimate the energetic cost of the
corresponding water flow: adhesions are more than 1 µm apart
over 10 to 20 µm length between the periphery and the central
region where medium pocket is being formed. Assuming a low
estimate of N > 30 passages of height 0.1 µm through which
medium can flow from the periphery to the cell interior, we
obtain that the pressure needed to drive the flow noted in Fig.
S2 is about 10 Pa, and that the power needed for this is of the
order 10−17 W : that is, 2 orders of magnitude smaller than the
actomyosin power transmitted to the microplate and measured
in the experiments. The formation of such pockets is thus very
plausible, and indeed is observed in most experiments when
the force is large.
The data we present does not allow to check whether the
totality of the change of apparent volume is due to this water
seepage. Therefore, there may be also some volume variations
due to a regulation of cell volume [67] superimposed to the one
due to the formation of the pocket. While the accurate mea-
surement of these variations would be important to understand
the spreading dynamics of the cells in our setup, the model be-
low does not require to assume volume conservation in order
to predict the vertical deflection dynamics of the microplates.
two microplates with an arbitrary stiffness, see Fig. 1, Fig. S1
and S2. In order to check whether the rheological model devel-
oped above can explain the observed cell behaviour, we need
first to calculate the state of stress within the actin cortex from
the experimental observables.
Since it was shown in [13] that the force generation in
these experiments is due to actomyosin contraction, we model
the cells as an actomyosin surface (shell) surrounding an in-
compressible but passive cell body (cytosol, nucleus and non-
cortical cytoskeleton), whose mechanical action is solely rep-
resented by a homogeneous internal pressure difference with
the medium outside, P = Pcell − Pmedium. Actomyosin be-
ing considered here as a thin structure, we perform here the
calculations in terms of a surface tension wσ, where w is the
thickness of the actomyosin cortex. σ is assumed to be a tensor
tangential to the cortex and to have no variation across w.
Because inertia is irrelevant at this scale, the spring force
F = 2k(h0 − h) of the microplate device needs to be balanced
at any instant by the combination of the cell body pressure
force and the tension force in the cell cortex,
F = 2πRp wσz=h − πR2
p P.
(19)
Here, Rp is the radius of the cell at a microplate, and σ is
the tension of the actomyosin cortex along the vertical direc-
tion. It is dependent on z, and corresponds to the component
σ along eses of the stress tensor of the actomyosin tensor, Fig.
1b. Because of the symmetry and of the assumption of a thin
actomyosin cortex, this tensor can only have one other nonzero
component, along the orthoradial direction, σ⊥eφeφ. Using
curvilinear coordinates, we can show that the force balance in
the es and eφ directions at any height z writes,
(cid:18) P − κwσ − sin θ
w ∂σ
∂s + cos θ
r wσ⊥
r w (σ − σ⊥)
(cid:19)
(20)
0 = P en + ∇ · σ =
where the first line is Laplace law written with different ten-
sions in the vertical and orthoradial directions, and the second
describes the equilibrium along direction es.
In order to solve these equations, we need to specify the
geometry of the actomyosin walls using Eq. 18. One can then
use power series, and get σ = σ0 + σ1z2 and σ⊥ = σ0⊥ + σ1⊥z2
in a closed form depending on geometrical parameters (h, Rc,
κ) and on the pressure difference P :
(cid:18)
(cid:18) 1
(cid:19)
(cid:19)
(cid:18)
− P
w
κRc − 1
2
(cid:19)
(21a)
(21b)
σ = σ0 + z2 κ
2
σ0
+ κ
Rc
σ⊥ = Rc(P/w − κσ0) + z2κP/w
with
wσ0 =
1
2
RcP +
F
2πRc
The presence of P in these equations means that we cannot
read directly the state of stress of the actin cortex from its
geometry and the measurement of F . However, if we have
e.g. an indication on the orthoradial stress σ⊥, which is possi-
ble when the shape is stationary and thus no dissipation takes
place in the orthoradial direction, then both P and σ can be
determined. Note that in this section we have not made use
of any assumption on the rheology of actomyosin, in particular
we have not used the constitutive Eq. 2 yet: the experimental
observations fitted by a geometry are sufficient to describe the
state of stress in the cortex.
S5.2 State of stress of the actin cortex in ex-
periments
Experiments of single cell stretching [13] allow to track simul-
taneously the geometry and force generated by cells between
S5.3 Equilibrium height and force
We apply the rheological model Eq. 2 in order to predict the
rate of strain ∂vs/∂s in the actin cortex along es, and obtain
its value as a power series in z and in function of Fp, the ge-
ometry and the model parameters τc, τα and Σ. For this, we
14
(cid:18)
(cid:19)(cid:18)
(cid:19)
(cid:18)
need to write the tensorial constitutive Eq. 2 in curvilinear
coordinates, assuming that A = eses + λeφeφ, i.e. that the
contractile stress in the orthoradial direction eφ is a fraction λ
of the contractile stress orthogonal to the plates:
+ vs
− 2
∂σ
∂s
(cid:18) ∂vs
∂s
−2ταE
(cid:19)
(cid:18) ∂vs
+ κvn
∂s
(cid:19)
σ
(cid:19)
+ σ
(cid:19)
τα
(cid:18) ∂σ
(cid:18) ∂σ⊥
∂t
∂t
τα
− 2
r
(vs cos θ + vn sin θ) σ⊥
+ σ⊥
− 2ταE
r
(vs cos θ + vn sin θ) = λΣ.
(22b)
Let us consider a cell that has reached an equilibrium
shape, such as the cell in Fig. S2 at time t = 2000 s. The
force plateaus at a value Fp, the curvature κ has reached a
stable negative value, and the equator length 2πRc is steady.
Thus h = 0,
Rc = 0, vn = 0, and the force does not evolve
either ( F = 0 and hence ∂σ
∂t = 0), these constitutive equations
simplify to:
ταvs
∂σ
∂s
+
1 − 2τα
∂vs
∂s
σ − τcΣ
= Σ
∂vs
∂s
σ⊥ = λΣ
Using the force balance, Eq. 21, we can calculate vs(z) as a
function of Fp :
− φ
τc
+
1 − φ
2τα
1 +
κz2
6Rc
κRc +
2τα + τc
τc
φ
(23)
(cid:19)(cid:19)
(cid:18)
vs =z
where
2 − Rcκ
(cid:16)
2 − Rcκ + 2 τα
τc
λ + Fp
πRcΣ
(cid:17) ∈ (0, 1]
φ =
This flow is the balance between a contractile term propor-
tional to 1/τc and an extensional term in 1/(2τα). In practice,
it is always negative: it corresponds to a retrograde flow that
vanishes at the cell's equator for obvious symmetry reasons,
and increases in magnitude with z. It is modulated by geomet-
ric factors, but also by the force Fp. This retrograde flow is
present for all values of the external stiffness.
In order to reach an equilibrium we need the retrograde
flow to compensate exactly the addition of new cortex through
polymerisation at z = h, which means that
vs(hp) = −vt.
(24)
Thus we have a relation for hp when the geometry is known in
terms of Rc and κ. When the force Fp is low (in the case of
vanishing k), there is an asymptote value he for hp provided
that it is much smaller than Rc, which is found to be the case.
Experiments provide a redundant reading of he, since the force
has to be 2k(h0 − he) at low k values. Thus we also have,
lim
k→0
∂Fp
∂k
= 2(h0 − he).
These consistently give he/h0 = 0.46 ± 0.06.
In Eq. 23, he
is proportional to τcvt as in the 1D model, but is modulated
both by the curvature (which is observed) and the contractility
in the orthoradial direction, which cannot be accessed. Using
the values τc = 521 ± 57 s and τα = 1186 ± 258 s obtained
from the dynamics of the 1D model (see S3.3), it is found that
the model can predict the cell behaviour only if the orthoradial
contractility Σ⊥ = λΣ is significantly lower than Σ, λ (cid:46) 0.5 --
else the pressure build-up in the cytosol prevents the cell from
contracting. We thus take λ = 0.5 and vt = 4 nm/s which is
+ κvn
= Σ,
(22a)
References
close to the value found in 1D (6.5 ± 1.5) and in the literature
(4.3 ± 1.2 nm/s, [17]).
There remains one last free parameter in the 3D model, the
(surfacic) contractility wΣ. This can be assessed in the limit of
infinite microplate stiffness k, we find wΣ = 15 nN/µm. Using
these values, the 3D model yields a plateau force very close to
the one of the 1D model, see Fig. 2a.
[55] Stamenovi´c D, Mijailovich SM, Toli´c-Nørrelykke IM, Wang N
(2002) Cell prestress. II. Contribution of microtubules. Am J
Physiol Cell Physiol 282:C617 -- C624.
[56] Wottawah F, et al. (2005) Optical rheology of biological cells.
Phys Rev Lett 94:098103.
[57] Green MS, Tobolsky AV (1946) A new approach to the theory
of relaxing polymeric media. J Chem Phys 14:80 -- 92.
[58] Vaccaro A, Marrucci G (2000) A model for the nonlinear rhe-
ology of associating polymers. J Non-Newtonian Fluid Mech
92:261 -- 273.
[59] Larson RG (1999) The structure and rheology of complex fluids.
Topics Chem. Engng (Oxford Univ. Press).
[60] Broedersz CP, et al. (2010) Cross-link governed dynamics of
biopolymer networks. Phys Rev Lett 105:238101.
[61] Debold EP, Patlak JB, Warshaw DM (2005) Slip sliding
away: Load-dependence of velocity generated by skeletal mus-
cle myosin molecules in the laser trap. Biophys J 89:L34 -- L36.
[62] Rubinstein B, et al. (2009) Actin-myosin viscoelastic flow in the
keratocyte lamellipod. Biophys J 97:1853 -- 1863.
[63] Williams WO (2011) Huxley's model of muscle contraction with
compliance. J Elas 105:365 -- 380.
[64] Zemel A, Rehfeldt F, Brown AEX, Discher DE, Safran SA
(2010) Cell shape, spreading symmetry, and the polarization
of stress-fibers in cells. J Phys: Condens Matter 22:194110.
[65] Curtis A, Aitchison G, Tsapikouni T (2006) Orthogonal (trans-
verse) arrangements of actin in endothelia and fibroblasts. J R
Soc Interface 3:753 -- 756.
[66] Bischofs I, Klein F, Lehnert D, Bastmeyer M, Schwarz U (2008)
Filamentous network mechanics and active contractility deter-
mine cell and tissue shape. Biophys J 95:3488 -- 3496.
[67] Jiang H, Sun SX (2013) Cellular pressure and volume regulation
and implications for cell mechanics. Biophys J 105:609 -- 619.
List of SI figures
S1 (a) Light transmission image of a cell spreading on mi-
croplates with infinite stiffness k seen from the side.
The sequence of shapes assumed by the cell walls in
the course of an experiment can be described by arcs
of circles. (b) TIRF visualisation of fluorescent paxillin
in a cell spreading on microplates with infinite stiffness
k seen from the bottom. The spreading is isotropic,
which supports the axial symmetry hypothesis. Adhe-
sion zones are clearly apart from one another, along
a circular region of interest we find N (cid:39) 40 adhe-
sion zones separated by paxillin-free passages of average
width 1.5 µm (see Appendix S5.1).
15
S2 Time evolution of the force and geometry of a single cell
spreading between microplates of intermediate stiffness
k = 176 nN/µm. Top, the force grows until it reaches a
maximum value. Center, concurrently with the force
increase, the cell spreads on the microplates, Rp in-
creases. The radius at the equator Rc also increases
after a transient decrease. Both stabilise when the force
is maximal. As the cell deflects the microplates, its
half-height h decreases, however this decrease does not
compensate the spreading in terms of (apparent) vol-
ume, and the volume V enclosed by the lateral cell sur-
faces increases more than two-fold. Bottom, transmis-
sion images show that this apparent volume increase
happens concurrently with the formation of 'pockets'
(arrow heads) away from the peripheral cell adhesions
(Fig. S1b) where the cell locally detaches from the mi-
croplate. See Appendix S5.1 for details.
S3 Microtubules have a negligible influence on stiffness-
dependent force generation. Blue boxes, plateau force
exerted by cells in microplate experiments in presence
of 1 µM Colchicine. Red crosses, control. See Appendix
S1 for a discussion.
S4 Plateau force measured for two different cell types,
Ref52 fibroblasts and C2-7 myoblasts. (A) Plateau force
divided by external stiffness, F/k, in nN/(nN/µm), for
k < kc. Welch two-sample t-test cannot discriminate
them, p-value 0.083 > 0.05.
(B ) Plateau force for
k > kc, in nN (includes both experiments with side-
and bottom-view for Ref52 cells). Welch two-sample t-
test cannot discriminate them, p-value 0.17 > 0.05.
16
a
b
Supplementary Fig. S1
17
Supplementary Fig. S2
Supplementary Fig. S3
18
05101520250100020003000Radius, height (µm)Apparent volume V (µm3)Time t (s) 050010001500200001002003004005006000500100015002000Force F (nN) RchRp 0 50 100 150 200 250 0 5 10 15 20 25Force F (nN)Stiffness k (nN/µm) Fp/k
Fp
a
b
Supplementary Fig. S4
19
C2−7Ref52051015lllllllllC2−7Ref520100200300400500lllllllllllll |
1312.5447 | 1 | 1312 | 2013-12-19T09:03:58 | Estimations of local thermal impact on living organisms irradiated by non-thermal microwaves | [
"physics.bio-ph",
"cond-mat.mes-hall"
] | Pennes' differential equation for bioheat transfer and the heat transfer equation are solved for the temperature distribution in a living tissue with spherical inclusions, irradiated by microwave power. It is shown that relative temperature excess in a small inclusion in the tissue in some cases is inversely proportional to its radius and does not depend on the applied power. In pulsing RF fields the effect is amplified proportionally to the ratio of the pulse period to the pulse duration. The local temperature rise significantly outpaces the averaged one and therefore the Watt to Weight SAR limits may be insufficient to estimate the safety of RF radiation and the conventional division of the biological effects of electromagnetic fields on the thermal and non-thermal needs to be revised. | physics.bio-ph | physics | ESTIMATIONS OF LOCAL THERMAL IMPACT ON LIVING ORGANISMS
IRRADIATED BY NON-THERMAL MICROWAVES
Vladimir Shatalov*
Department of Biophysics, Donetsk National University, Donetsk, Ukraine
Pennes’ differential equation for bioheat transfer and the heat transfer equation are solved for the temperature
distribution in a living tissue with spherical inclusions, irradiated by microwave power. It is shown that relative
temperature excess in a small inclusion in the tissue in some cases is inversely proportional to its radius and does not
depend on the applied power. In pulsing RF fields the effect is amplified proportionally to the ratio of the pulse period
to the pulse duration. The local temperature rise significantly outpaces the averaged one and therefore the Watt to
Weight SAR limits may be insufficient to estimate the safety of RF radiation and the conventional division of the
biological effects of electromagnetic fields on the thermal and non-thermal needs to be revised.
Key words: non-thermal; electromagnetic field; bioheat transfer; inhomogeneity; temperature rise
INTRODUCTION
Traditionally, the effects of electromagnetic fields (EMF) on living organism are divided into
ionizing (non-ionizing) and thermal (non-thermal). They are extensively investigated to date [e.g.,
NIEHS, 1999]. However, when considering effects of weak EMF, an important problem is to find
and validate clear criteria of weakness of the impact [Challis, 2005; Shatalov, 2012]. For thermal
effects of the EMF the presently accepted criteria are based upon application of physiologically
acceptable limits to the increase of the average temperature of the samples.
Non-thermal effects defined in [ICEMS, 2010] are biological mechanisms that are not
connected to the temperature increase, which is assumed to be less than 0.01degrees C (living
organism), 0.001(cells) or 0.0005 (sub-cellular). By comparison, ANSI, WHO, IEEE & ICNIRP
consider that exposures below 0.05 degrees C (0.4W/kg) are safe for workers, and exposures below
0.01 C (0.08 W/kg) are negligible for the public. Any biological effects below these levels of
heating are considered by these organizations to have no biological significance and to be
reversible. These criteria are used for estimation of maximum permissible power of domestic radio
equipment (cell phones, Bluetooth, Wi-Fi, etc.). In particular, the effect of mobile phone radiation
on human health is the subject of recent interest and study [Vecsei et al., 2013] due to the enormous
* Correspondence to: V.M. Shatalov, Biological Faculty, DonNU,
Schorsa st., 46, Donetsk, 83050, Ukraine
E-mail: [email protected]
increase in mobile phone usage throughout the world. Any biological effect of environmental RF
EMFs is usually referred as non-thermal.
Another important application of these criteria lies with the development of non-lethal
microwave weaponry (such as «Active Denial System», developed by the U.S. Department of
Defense) to evaluate whether their effects are indeed reversible.
One well-understood effect of microwave radiation is dielectric heating, in which any polar
solution (such as living tissue) is heated by rotations of the polar molecules induced by the
electromagnetic field. In the case of a person using a cell phone, most of the heating effect occurs at
the surface of the head, causing its temperature to increase by a fraction of a degree. The maximum
power output from a mobile phone is regulated by the mobile phone standard and by the regulatory
agencies in each country. In the USA, the Federal Communications Commission (FCC) has set a
Specific Absorption Rate (SAR) limit of 1.6 W/kg, averaged over a volume of 1 gram of tissue, for
the head. In Europe, the limit is 2 W/kg, averaged over a volume of 10 grams of tissue. SAR values
are heavily dependent on the size of the averaging volume.
Non-thermal effects can also arise due to the low-frequency pulsing of the mobile phones
carrier signal. Biological significance of these modulations is subject to a recent debate [Foster and,
Repacholi, 2004]. Some researchers argue that so-called "non-thermal effects" can be reinterpreted
as a normal cellular response to an increase in temperature. Others believe the stress proteins are
unrelated to thermal effects, since they occur for both extremely low frequencies and radio
frequencies (RF), which have very different energy levels [Blank and Goodman,. 2009]. Another
study that was conducted using fluorodeoxyglucose injections and positron emission tomography
concluded that exposure to radiofrequency signal waves within parts of the brain closest to the cell
phone antenna results in increased levels of glucose metabolism, but the clinical significance of this
finding is unknown [Volkow et al., 2011].
The aim of our paper is to give a mathematical model of the local thermal effects of the EMF
by specifically taking into account inhomogeneity of the heating (resulting from non-uniformity of
living tissue). The idea of the non-uniform heating, of course, is not new. However, that is hard to
find specific estimates, therefore it was suggested by Challis [2005] as an open problem. Our main
goal is to provide such analytical estimations and to give physical background of thermal effects
related to inhomogeneity in space and time. We show that in some cases, the local temperature rise
significantly outpaces the average absorbed radiation power and therefore the Watt to Weight SAR
limits may be insufficient. The possibility of such a local heating (with the negligible average
temperature increase) challenges the common division of EMF effects into thermal and non-
thermal. While the effects, connected to the local heating, are thermal, the average heating may be
negligible, making them classified conventionally as non-thermal.
Our consideration is not limited to any particular frequency band of the electromagnetic field.
It is only important that the field is heating the conducting medium locally, be it via the eddy
currents in the medium of non-uniform conductance, the polarization-induced vibrations of the
molecules in the medium of non-uniform polarizability, or any other local mechanism. There are
several different mechanisms, which lead to RF power absorption and have different penetration
depths in different media.
It is obvious that the temperature distribution in any bounded sample heated by penetrated
radiation has to be inhomogeneous. Firstly, it is due to the spatially inhomogeneous heating by the
exponentially decaying RF power. It is well known that the temperature distribution has maximum
under the sample surface if the heat transfer inside exceeds that one outside the sample (for
example, due to pure transparency of the sample surface). The maximum value is limited by RF
power decay depth that actually varies from fractions of a millimeter to several centimeters.
Therefore, this maximum is flat and does not exceed much the surface temperature value. Unlike
the case described, we will examine the temperature peaks at the micro inclusions having high
electrical conductivity, which are located far enough away from the cooling surfaces or heat sinks.
TEMPERATURE DISTRIBUTION IN THE CASE OF RANDOMLY
DISTRIBUTED POINT SOURCES
It is well known that living tissue is substantially heterogeneous on micro and nano-scales
with some of its components having increased electrical conductivity. For example: 1) electrical
conductivity of the cytoplasm of living cells is somewhat higher than that of extracellular media due
to the Na-K asymmetry [Lyashchenko and Lileev, 2010]; 2) electric double layer, forming on the
surface of colloidal inclusions (such as air nano-bubbles in water), consists of a diffuse cloud of
ions with superior conductivity compared to that of the solvent [Shatalov et al., 2012]; 3) axon
membranes have extra high conductivity, etc. There are also numerous instances of nano-scale
conductance inhomogeneity in inanimate nature exemplified by any colloidal solution (liquid or
solid) of conductive nanoparticles in weakly conductive medium. Since the absorbed microwave
and RF power is proportional to the electrical conductivity, conducting inclusions are heated by
radiation more intensively than the surrounding media.
To describe this process of non-uniform heating in the medium with thermal regulation
(present in all living tissue on all scales) let us use the classical Pennes’ differential equation
[Pennes, 1948]:
Θ
Θ∂
t
τ
∂
k
2
−Θ∇=
(1)
+
W
VC
,
,
−
'
−=
–
(2)
r
'd
3
q
r
)(
q
r
)(
r
)(
2
−Θ∇
where Θ is temperature excess [K] above the “normal” temperature, t – time [s], k – temperature
conductivity [m2 s–1] (for example, in water k =0.15mm2/s at normal conditions),
– Laplace
operator, τ is characteristic temperature relaxation time [s], W – absorbed power [J m–3 s–1], Cv –
volumetric heat capacity [J/(m3 K)]. The second term on the right hand side describes thermal
regulation, whatever the mechanism is. In the absence of heat sources (W=0) the medium relaxes
towards zero temperature excess (Θ), or, to its “normal” temperature. In steady state, when time
derivative is zero, Eq. (2) reduces to the well-known Helmholtz equation (or screened Poison
equation):
r
)(
Θ
2
R
τ
where Rτ=(k τ)1/2 is a characteristic size of the temperature relaxation region,
kCWq
=
V
renormalized absorbed heat [K/m2]. This equation is linear and for an arbitrary
)(rq
its solution can
formally be written with the help of Green’s function as
e
rr
R
/'
−−
τ
r
r
4
π
(3)
.
Next we assume that heat is supplied in the form of randomly distributed point sources of the
density n and power W. The average temperature excess in the sphere of radius Ri around each
source can be obtained by direct integration as
W
3
τ
(
)
1
,
RC
4
π
i
V
RRx
/
.
=
i
τ
r
)(
=Θ
=Θ
i
∫∫∫
V
=Θ 0
(4)
The average equilibrium temperature excess, measured in the whole medium, is then
Wnτ
VC
(5)
,
Assuming that temperature conductivity of the media and inclusions are the same and
identifying Ri with inclusion size, the relative temperature excess in a small inclusions (x << 1) is
inversely proportional to its radius and does not depend on the input power:
3
Θ
Rnk τπ8
Θ
i
(6)
.
For small low-density inclusions in a living tissue with weak thermal regulation and low
temperature conductivity, this ratio can be extremely high. Furthermore, if the heating is not
stationary but consists of pulses of the width τw and the period τp > τw, it is easy to show that the
local heating during the pulse can also be represented by (6) with an additional factor of τp / τw. That
is, the pulsing EMF source leads to even higher local increase of the relative temperature.
=
0
i
−
xe
−
x
−
x
−
e
3
HEATING OF A SINGLE SPHERICAL INCLUSION IN A SPHERICAL
SAMPLE
Pennes’ equation (1) assumes that the total energy exchange between tissue and the flowing
blood can be modeled as a non-directional heat outlet, whose magnitude is proportional to the
volumetric blood flow and the difference between local tissue and major supply arterial
temperatures. This approximation is valid for large enough samples, containing numerous blood
vessels. To investigate the temperature distribution at a smaller scale in this section, we explore a
different model, in which a heated center indirectly interacts with the cooling surface. Such a model
might describe the temperature distribution in a close vicinity of a blood vessel that we refer as a
thermostat. To simplify the consideration, the living tissue will be modeled as a sphere. Starting
from such a spherical horse, we mean to explain the origin of temperature spikes in nano-inclusions
contained in any tissue of any form.
Stationary temperature distribution
Let us consider a spherical inclusion of radius Ri in the center of the irradiated spherical
sample with radius R0, placed in a thermostat. The inclusion and the surrounding media have
different electrical and the same temperature conductivities. The stationary equation of heat transfer
in this case looks like:
r
r
)(
)(
2
2
Θ∂
Θ∂
r
r
r
2
∂
∂
rq
0)(
=
+
+
(7)
.
Here r-dependence of the renormalized absorbed heat has the form:
q
Rr
0,
<<
i
i
yq
R
Rr
,
<<
i
i
0
rq
)(
=
⎧
⎨
⎩
2
+
−
C
r
)(
=Θ
(8)
,
where factor y points to the abovementioned difference of electrical properties inside and outside
the inclusion.
Obviously, any solution of (7) has the form:
rB
A
r
6
(9)
.
where the constants A, B, and C differ for the inclusion and outer space. Putting (9) into (7) we get
Bi = qi and B0 =yqi. Then from the condition of zero flow at r=0 we get Ai =0. Next, the flow
continuity condition at the boundary of the inclusion gives the value of A0, the value C0 we get by
putting Θ(R0)=0. Finally, the equality of temperatures at the boundary of the inclusion gives us the
value Ci. Thus we get the solution of (7), that allows us to calculate the average temperature of the
inclusion Θi and the outer space Θ0 and its ratio:
=
2
)
3
x
−
4
x
)
(
1
Θ
Θ
[
x
12
2
x
10
+
x
++
(
x
1
−
(
y
x
x
315
10
3
+
−
[
]4
(
)
x
y
x
212
5
3
2
+
+
+
,
where x= Ri /R0 and y= q0 /qi. For the case under interest x«1 and y=0 we get:
R0
12
Θ
R
5
Θ
i
]
)
x
2
x
5
3
−
+
−
=
i
0
i
2
3
x
(10)
0
=
(11)
.
So, the ratio of average temperatures of the inclusion to that one of the outer space is reversly
proportional to its radii. This qualitatively coincidences with (6) outlined in previous section and
differs just in coefficient.
In the case of a homogeneous medium, when properties of the inclusion and the surrounding
environment are the same (y=1), then for x«1 we get:
5
Θ i
2
Θ
0
.
(12)
Thus, the temperature of a micro-inclusion exceeds the average temperature 2.5 times just because
the inclusion is far enough from the thermo-stabilized outer boundary of the sample.
Finally, we may conclude that there exist two reasons giving temperature increase in a little
inclusion. First is the increased heat production, and second – the remote distance from the cooling
boundary.
Non-stationary temperature distribution
Another source of the temperature non-homogeneity is time inhomogeneous heating. In this
subsection, we explore the non-stationary case for the spherical model from previous section.
At the starting point, a short RF pulse heats the spherical inclusion and surrounding media and
trΘ
. To get the distribution we solve
we monitor the temperature distribution in space and time
),(
numerically the non-stationary equation of heat transfer (1) for a simple case of two co-centered
spherical media with different electrical and the same temperature conductivities. We are interested
in the time range t«τ therefore the term that describes thermal regulation in (1) may be omitted. So,
in the spherical coordinates centered in the cell center the equation (1) takes the form:
tr
tr
tr
),(
),(
),(
2
2
Θ∂
Θ∂
Θ∂
r
r
t
r
2
∂
∂
∂
trQ
),(
=
+
+
,
(13)
correspondently. We state the absorbed power
where r and t are given in units R0 and
R 2
0=τ
k
distribution
trQ
),(
=
2
trqR
),(
0
as
(
0
(
R
i
(
=
)
τ
w
,
)
τ
w
q
i
sin
q
0
sin
(14)
trq
),(
Rr
≤≤
i
⎞
,
⎟⎟
⎠
⎞
,
⎟⎟
⎠
)
t
)
(
t
0
≤≤∩
(
)
t
r
0
≤≤∪<
t
⎛
π
⎜⎜
τ
⎝
w
t
⎛
π
⎜⎜
τ
⎝
w
(
<
τ
w
⎧
⎪
⎪
⎪⎪
⎨
⎪
⎪
,0
⎪
⎪
⎩
where x= Ri /R0 and τw is the RF pulse duration time,
0q – heated power inside and outside
iq and
of the inclusion, its ratio is proportional to the ratio of correspondent electric conductivities. The
initial and boundary conditions for the eq. (13) have the form:
r
r
0,0
)0,(
;
∞<<
=
Θ
⎧
⎪⎪
tr
),(
Θ∂
⎨
r
∂
⎪
0 tR
),
Θ
⎪
⎩
The results of our calculations of the relative temperature dynamics in a sample irradiated by
a single RF pulse with τw = 0.2τ. for the size ratio x=0.2 and different heated power ratios y = 0, 1,
2, 4 and ∞ are shown in Figure 1.
(15)
=
.0
r
=
;0
=
,0
Figure 1. Relative average temperature of the inclusion with Ri = 0.2R0 as a
function of time (in units of τ) in a sample irradiated by RF pulse (14)
with τw = 0.2τ. Numbers over the curves are qi/q0 ratios of RF power
absorption inside and outside the inclusion.
In the case of dielectric inclusion in conducting media (y = 0) the temperature of media
exceeds the temperature of inclusion Θi /Θ0 < 1 only at initial moments, but soon after this ratio
rapidly increases to some asymptotic value close to 3. For larger y the ratio Θi /Θ0 exceeds 1 at all
times. For the conducting inclusion in dielectric media (y = ∞) the relative temperature Θi /Θ0 is
very high in initial moments because the media is heating via heat transfer from the inclusion. After
the heating pulse finished all the curves go to the asymptotic value.
The curves on Figure 1 were calculated for the case of a single RF pulse but it is obvious if
the time scale expands to a sequence of pulses then every next pulse will start at increased values of
relative temperature of the inclusion and that amplifies the effect of inhomogeneous heating of the
sample.
CONCLUSIONS
The results obtained in the paper have to draw attention to the obvious and at the same time
unexpectedly large spatial inhomogeneity of heating of the irradiated samples, regardless of the
intensity of irradiation. Until now, it seemed to be obvious that RF power is the main criteria, which
refers any RF exposure to the thermal or non-thermal one. However, we have shown that relative
temperature rise in micro inclusions is RF power independent and may be very high because it is
reversely proportional to the size of the inclusions. This is the case when the radiation is absorbed
mainly in the inclusions that situated sufficiently far from the heat outlets. Moreover, qualitatively
this conclusion is an obvious consequence from Fourier’s law and the heat transfer boundary
conditions, and this is a long-solved problem in physics that leaves no questions.
To prove the results experimentally one has to measure the local temperature inside nano-
objects that is not so easy but may be very informative. For example, by coating their samples with
molecular layers with well-defined melting temperatures, Samsonov and Popov, [2013] measured
the temperature in the immediate vicinity of individual cells during exposure to the EMF. They
detected a rapid temperature jump in the experimental chamber and rapid establishment of the
steady-state temperature. This allowed them to compare, in a quantitative manner, the cellular
effects of the EMF with those of the temperature jump elicited by conventional heating.
There exist many experimental works with inanimate samples, the results of which, we
believe, can be interpreted in terms of local heating (or overheating) of irregularities in the
irradiated samples. For example, Bunkin et al., [2010] very carefully examined the effect of
dissolved gases on some of the properties of distilled water. They showed that in presence of
dissolved air IR laser irradiation heats the microbubbles of air to plasma temperatures that increases
absorption of the radiation, while in the absence of dissolved gases, the water is essentially
transparent due to low probability of the multiphoton absorption mechanism. In the work of
Doroshkevych et al. [2012] it was shown that weak pulsed magnetic field speed up self-
organization processes in the nanopowder dispersed systems based on compacted ZrO2. The list of
examples may be easily continued. We believe that all these effects may be consequences of the
inhomogeneous heating in micro- and nano-scales.
There exists a lot of contradictory evidence for non-thermal effects of electromagnetic
radiation on living organisms. Some of them were pointed in Introduction [Challis, 2005; Shatalov,
2012; ICEMS, 2010]. According to our paper, many of the interpretations need to be revisable. So,
we are confident enough to judge the dispute in the works Foster and Repacholi, [2004]; Blank and
Goodman, [2009] in favor of the thermal origin of the observed effects. Another example,
D’Andrea et al., [2003] pointed that effects of low-level RF exposure on the blood–brain barrier are
controversial and the effects have been generally accepted for exposures that are thermal. Studies at
these levels have observed effects on norepinephrine, dopamine, and serotonin. Now we can say
that thermal mechanisms have to be favorable in explanation of any effect of MW on
neurochemistry. Concerning discussions in Internet on safety of the non-lethal microwave weapon,
it should be pointed to possible uncontrolled temperature rise in under skin inclusions.
So, the local temperature can rise significantly outpaces the averaged one and therefore the
Watt to Weight SAR limits may be insufficient to account the safety of RF radiation and the
conventional division of the biological effects of electromagnetic fields on the thermal and non-
thermal needs to be revised.
ACKNOWLEDGMENTS
The author thanks Prof. Andrew Lyaschenko for discussions that served as a stimulus to the
writing this article. He is also very grateful to Dr. Konstantin Metlov for reading and discussing the
manuscript and to Prof. Nicholas Bunkin for helpful comments.
REFERENCIES
Blank M, Goodman R. 2009. Electromagnetic fields stress living cells. Pathophysiology 16 (2–3):
71–78.
Challis LJ. 2005. Mechanisms for interaction between RF fields and biological tissue.
Bioelectromagnetics; Supplement 7:S98-106.
D’Andrea JA, Chou CK, Johnston SA, Adair ER. 2003. Microwave effects on the nervous system.
Bioelectromagnetics Supplement. 6:S107-S147.
Doroshkevych OS, Shylo AV, Saprukina OV, Danilenko IA, Konstantinova TE, Ahkozov LA.
2012. Impedance spectroscopy of concentrated zirconia nanopowder dispersed systems
experimental technique. World Journal of Condensed Matter Physics. 2(1):1-9.
Foster KR, Repacholi MH. 2004. Biological effects of radiofrequency fields: does modulation
matter? Radiation Research 162 (2): 219–244.
Glaser R. 2005. Are thermoreceptors responsible for "non-thermal" effects of RF fields? Ed.
Wissenschaft (Bonn, Germany: Forschungsgemeinschaft Funk) (21). OCLC
179908725.
ICEMS. 2010. Non thermal effects and mechanisms of interaction between electromagnetic fields
and living matter. ICEMS, eds. Guiliani L. & Soffritti M.: Ramazzini Institute,
European J of Oncology, Library Vol. 5.
Lyashchenko AK, Lileev AS. 2010. Dielectric relaxation of water in hydration shells of ions. J.
Chem. Eng data. 55:2008-2016.
Bunkin NF, Ninham BW, Babenko VA, Suyazov NV, Sychev AA. 2010. Role of dissolved gas in
optical breakdown of water: differences between effects due to helium and other gases.
J. Phys. Chem. B 114:7743–7752.
NIEHS. 1999. Report on health effects from exposure to power-line frequency electric and
magnetic fields. National Institute of Environmental Health Sciences of the U.S.
National Institutes of Health. NIH Publication No. 99-4493.
Pennes HH. 1948. Analysis of tissue and arterial blood temperature in the resting human forearm. J.
Appl. Physiol. 1:93–122.
Samsonov A, Popov SV. 2013. The effect of a 94 GHz electromagnetic field on neuronal
microtubules. Bioelectromagnetics, 34:133–144.
Shatalov VM. 2012. Mechanism of the Biological Impact of Weak Electromagnetic Fields and the
In Vitro Effects of Blood Degassing. Biophysics 57:808-813.
Shatalov VM, Filippov AE, Noga IV. 2012. Bubbles induced fluctuations of some properties of
aqueous solutions. Biophysics. 57(4):421–427.
Vecsei Z, Csathó A, Thuróczy G, Hernádi I. 2013. Effect of a single 30 min UMTS mobile phone-
like exposure on the thermal pain threshold of young healthy volunteers.
Bioelectromagnetics. 34(7):530-541.
Volkow ND, Tomasi D, Wang G-J, Vaska P, Fowler JS, Telang F, Alexoff D, Logan J, et al. 2011.
Effects of cell phone radiofrequency signal exposure on brain glucose metabolism.
JAMA 305 (8): 808–821.
|
1803.03942 | 3 | 1803 | 2018-06-20T23:03:52 | Self-organized system-size oscillation of a stochastic lattice-gas model | [
"physics.bio-ph",
"cond-mat.stat-mech",
"q-bio.SC"
] | The totally asymmetric simple exclusion process (TASEP) is a paradigmatic stochastic model for non-equilibrium physics, and has been successfully applied to describe active transport of molecular motors along cytoskeletal filaments. Building on this simple model, we consider a two-lane lattice-gas model that couples directed transport (TASEP) to diffusive motion in a semi-closed geometry, and simultaneously accounts for spontaneous growth and particle-induced shrinkage of the system's size. This particular extension of the TASEP is motivated by the question of how active transport and diffusion might influence length regulation in confined systems. Surprisingly, we find that the size of our intrinsically stochastic system exhibits robust temporal patterns over a broad range of growth rates. More specifically, when particle diffusion is slow relative to the shrinkage dynamics, we observe quasi-periodic changes in length. We provide an intuitive explanation for the occurrence of these self-organized temporal patterns, which is based on the imbalance between the diffusion and shrinkage speed in the confined geometry. Finally, we formulate an effective theory for the oscillatory regime, which explains the origin of the oscillations and correctly predicts the dependence of key quantities, as for instance the oscillation frequency, on the growth rate. | physics.bio-ph | physics | Self-organized system-size oscillation of a stochastic lattice-gas model
Mareike Bojer,1, 2, ∗ Isabella R. Graf,1, ∗ and Erwin Frey1, †
1Arnold Sommerfeld Center for Theoretical Physics and Center for NanoScience,
Department of Physics, Ludwig-Maximilians-Universitat Munchen,
Theresienstrasse 37, D–80333 Munchen, Germany
2Department of Physics, Technische Universitat Munchen, D–85748 Garching, Germany
The totally asymmetric simple exclusion process (TASEP) is a paradigmatic stochastic model for
non-equilibrium physics, and has been successfully applied to describe active transport of molecular
motors along cytoskeletal filaments. Building on this simple model, we consider a two-lane lattice-
gas model that couples directed transport (TASEP) to diffusive motion in a semi-closed geometry,
and simultaneously accounts for spontaneous growth and particle-induced shrinkage of the system's
size. This particular extension of the TASEP is motivated by the question of how active transport
and diffusion might influence length regulation in confined systems. Surprisingly, we find that the
size of our intrinsically stochastic system exhibits robust temporal patterns over a broad range of
growth rates. More specifically, when particle diffusion is slow relative to the shrinkage dynamics,
we observe quasi-periodic changes in length. We provide an intuitive explanation for the occurrence
of these self-organized temporal patterns, which is based on the imbalance between the diffusion and
shrinkage speed in the confined geometry. Finally, we formulate an effective theory for the oscillatory
regime, which explains the origin of the oscillations and correctly predicts the dependence of key
quantities, as for instance the oscillation frequency, on the growth rate.
I.
INTRODUCTION
Understanding collective transport phenomena is an
important challenge in theoretical physics, with possi-
ble implications for biology and materials science. One-
dimensional, asymmetric simple exclusion processes form
a prominent class of idealized theoretical models that are
amenable to detailed mathematical analyses; see for in-
stance Ref. [1] for a review.
Interestingly, these mod-
els appeared simultaneously in the mathematical litera-
ture as conceptual models with which to study interact-
ing Markov processes [2] and in the biological literature
as idealized models for ribosomes moving along mRNA
during translation [3]; for recent reviews see Ref. [4, 5].
The simplest version of such a model is the totally
asymmetric simple exclusion process (TASEP). In this
one-dimensional stochastic lattice-gas model, particles
move step-wise and uni-directionally from lattice site to
lattice site at a constant (hopping) rate, provided that
the next site is vacant. Models of this class have been
used to study the collective, directed transport of molec-
ular motors along microtubules.
In that context, the
TASEP has been extended to include the exchange of
particles between the lattice (microtubules) and the sur-
rounding environment (cytosol) in terms of Langmuir ki-
netics [6–9]. The traffic jams predicted by these models
have recently been observed experimentally [10, 11], sug-
gesting that these idealized lattice gases are indeed suit-
able for describing the collective dynamics of molecular
motors.
In a further interesting line of research, extensions of
the TASEP to dynamic lattices have been developed [12–
∗ These authors contributed equally to this work.
† Please send correspondence to [email protected].
28]. On the one hand, motivated by the transport of
vesicles along microtubules that facilitate growth of fun-
gal hyphae, or by growth of flagellar filaments, TASEP
models have been considered in which a particle that
reaches the end of the lattice may extend it by a sin-
gle site [12, 14, 15, 17, 22]. On the other hand, in efforts
to quantify experimental observations of motor-mediated
microtubule depolymerization in vitro, dynamic lattice-
gas models have proven useful for probing the regula-
tion of microtubule length by motors that show uni-
directional [16, 18, 29, 30] or diffusive motion [31–33].
Recently, these models for depolymerizing molecular mo-
tors have been extended towards dynamic microtubules,
in order to study the interplay between lattice growth
and shrinkage [19–21, 23, 24, 26], and to understand the
basic principles underlying cellular length control mech-
anisms [34, 35].
There are many possible extensions of these models,
which are both interesting in their own right and can help
us to understand important biological processes. Exam-
ples include large networks of biofilaments [36–38], lim-
ited protein resources [6, 39–45], the fact that proteins
in the cytosol do not form a spatially uniform reservoir
because their dynamics is limited by diffusion [6, 46–51],
and that proteins may be spatially confined, as they are
in fungal hyphae or filopodia [6, 9, 46, 51, 52].
In this paper our goal is to study the interplay be-
tween diffusive motion and directed transport as a pos-
sible mechanism for length regulation under confinement
[Fig. 1(a)]. This relationship is of great interest because,
in contrast to diffusion, directed transport is an intrin-
sically non-equilibrium process.
It leads to currents of
motors directed towards the growing/shrinking end (tip)
and so to a strong interaction between the motors and
the growing/shrinking end. The combination of trans-
port with diffusion in a semi-closed geometry has recently
8
1
0
2
n
u
J
0
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
3
v
2
4
9
3
0
.
3
0
8
1
:
v
i
X
r
a
2
FIG. 1. (a) Illustration of the dynamics in cellular protrusions. Movements of molecular motors are indicated by
black arrows, and are restricted to the cell body and the protrusion by the cell membrane. On the filament the motors move
unidirectionally towards the protrusion tip, while their motion in the surrounding cytoplasm is diffusive. (b) Illustration of
the two-lane lattice-gas model. We consider a two-lane lattice-gas model consisting of a TASEP/transport lane (TL, upper
lane, occupied by orange (light gray) particles) and a diffusive lane (DL, lower lane, occupied by blue (dark gray) particles) with
hopping rates ν ≡ 1 and , respectively. The lanes are coupled by attachment and detachment kinetics at rate ω, respecting
exclusion for transfer from the DL to TL. Entry and exit occurs via the first DL site only, at rates α and , respectively. The
system spontaneously grows by simultaneously appending a site to the TL and DL tip at rate γ, while both lanes shrink by a
site, at rate δ, if the TL tip is occupied by a particle. In the latter case, particle conservation is ensured by shifting all particles
of the previous TL and DL tip site to the new DL tip site. (c) Illustration of the particle currents and density profiles.
The density profile on the TL (DL), ρ(x) (η(x)), is displayed in orange (light gray) in the upper panel (blue (dark gray) in the
lower panel). The density ρ(x) is discontinuous at the last TL site, with ρ− referring to the left and ρ+ to the right limit. The
currents (black arrows) come from entry α, exit η0, diffusion D, attachment and detachment J D→T , directed movement J T ,
and detachment due to depolymerization ρ+δ.
been studied with a conceptual model [51]. This model
assumes a fixed length for the system and suggests an im-
portant role for diffusion in the transport of motors to the
tip. While biologically motivated exclusion in this model,
and also more generally, can change the dynamics qual-
itatively, here we here focus on the low-density regime
where exclusion only has a minor quantitative influence.
Instead we extend the previous model by including length
regulation. This is motivated by polymerization and de-
polymerization of filaments in highly dynamic cellular
protrusions. For the particular choice of the growth and
shrinkage dynamics, we draw our inspiration from exper-
imental studies of microtubules, in which motor-induced
depolymerization [29, 30, 53–55] and growth by attach-
ment of tubulin heterodimers [56, 57] were found. Other
choices such as the "opposite" scenario where polymer-
ization is motor-dependent and depolymerization is spon-
taneous, or a system with two types of motors, namely
polymerizing and depolymerizing ones, are also expected
to give rise to interesting phenomena but are out of the
scope of the present paper.
While our motivation originates from specific biological
processes, we do not want to study a particular biological
system. Rather our lattice gas model (Fig. 1) provides
us with an exemplary model to examine the combined
role of diffusion and active transport for length regula-
tion under a confined geometry. Unexpectedly, we find
that the size of our intrinsically stochastic system shows
periodic behavior when diffusion is slow compared with
the growth and shrinkage dynamics. This indicates that
diffusion-limited transport can be an important ingredi-
ent for the occurrence of (self-organized) oscillations.
This paper is organized as follows.
In Section II we
explain the processes incorporated into the stochastic
lattice-gas model and show analytical calculations for the
simplest possible scenario, the stationary state, to gain a
basic understanding. To check these results and explore
a broader parameter regime, we continue in Section III
with numerical simulations. We determine the depen-
dence of the stationary length on the growth rate and
find a parameter regime in which length oscillations oc-
cur. For this oscillatory behavior we then develop an
intuitive explanation. Finally, in Section IV we derive an
effective theory from this intuitive explanation, and com-
pare its predictions to the results from stochastic simu-
lations. We conclude with a summary and discussion in
Section V. Readers that are primarily interested in the
phenomenology may want to skip the more technical part
of Section II B. It aims at giving a mathematical intuition
about the processes constituting the presented model.
II. MODEL DEFINITION AND
MATHEMATICAL ANALYSIS
A. Stochastic Lattice-Gas Model
[Fig. 1(b)], and extend previous work [51] by combin-
ing it with a length-regulation mechanism. One lane, the
TASEP/transport lane, TL, emulates the directed trans-
port along filaments in cellular protrusions in terms of
a totally asymmetric simple exclusion process (TASEP)
As outlined in the Introduction, we consider a
in a semi-closed geometry
two-lane lattice-gas model
{{εεεαωγν=1δfilamentscell bodycytoplasmmotorsεη0αJT(x) JD→T(x) D(x)η(x)ρ(x)ρ-ρ+ρ+δ0xLL-1(a)(b)(c){shrinkgrowTLDL [3, 58, 59]. It is characterized by a rate ν at which parti-
cles hop unidirectionally along the lattice, from the base
towards the growing/shrinking end (tip). Particles ex-
clude each other, i.e., there can be at most one particle
at any lattice site and, consequently, particles can only
hop forward if the site ahead of them is empty. Later we
will see that exclusion is not essential for the qualitative
findings discussed in this paper. We measure all rates in
units of ν and thus set ν ≡ 1.
The second lane, the diffusion lane, DL, mimics diffu-
sive transport of motors in the cytosol [Fig. 1(b)], and
describes it as effectively one-dimensional: Particles per-
form a symmetric random walk with hopping rate to
the left and right. As the density of motor proteins in
the cytosol is small, we assume no particle exclusion on
the DL. Hence the hopping probability is not influenced
by the occupancy of the neighboring sites.
Moreover, molecular motors constantly cycle between
the filaments and the surrounding cytosol by attaching
to the filaments and detaching into the cytosol. This
motion is represented as follows: At a rate ω, a particle
from the DL can attach to the corresponding TL site, if
it is vacant, and a particle from the TL can always attach
to the corresponding DL site.
Particles can enter the system only from a reservoir
via the first DL site, corresponding to motors entering
the protrusion from the cell body, and similarly can only
leave the system via that same site. Entry occurs at rate
α, and particles diffuse out at a rate equal to the hopping
rate . We do not model the dynamics in the cell body ex-
plicitly, as diffusion in the cell body is three-dimensional
and we expect that, as a result, entry and exit events
should be roughly uncorrelated. We thus approximate
the cell body as an infinite reservoir.
The lanes grow by the spontaneous addition of a TL
site to the TL tip at rate γ, accompanied by the simul-
taneous extension of the DL by one site. Motor-induced
depolymerization is realized by cutting off the TL tip site
at rate δ. The cytoskeletal filament is considered to span
the protrusion, meaning that with shrinking filament the
length of the cytosolic volume shrinks as well. So, when
the TL shrinks by one site the DL is simultaneously re-
duced by one site and all leftover particles, including the
one responsible for the shortening event, are shifted to
the new DL tip site. Thus, the DL tip site can be easily
populated by several particles at once. Since the motors
can neither penetrate the membrane nor leave the sys-
tem at the tip, they remain in the cytosol at the tip even
when the system shrinks.
In summary, a typical particle journey would start by
the particle's entry into the system at the first DL site,
followed by diffusion on this lane until it attaches to the
TL and begins to hop towards the tip. Once there, it
eventually cuts off the site it is occupying and joins the
other particles from the previously 'lost' DL site on the
new DL tip site. Each of these particles then diffuses on
the DL until it reattaches to the TL or leaves the system
at the DL's first site.
3
B. Mathematical Analysis: Adiabatic Limit
To gain a better quantitative understanding of the sys-
tem, we analyzed the stochastic dynamics of the lattice-
gas model in terms of a set of Master equations, and
employed a mean-field approximation to derive a set of
rate equations for the density of motors on the TL and
DL. The analysis follows Refs. [51] and [19], and is dis-
cussed in detail in Appendix A. Here, we will discuss
the main results and their interpretation, focusing on the
low-density limit and the limit of slow length change com-
pared with particle movement, i.e. ν ≡ 1 (cid:29) γ.
To begin with, let us introduce a set of random vari-
ables to describe the state of the system: L(t) denotes
the lattice length at time t, configurations on the TL are
indicated by a tuple of random variables (ni)l
i=0, with ni
describing the occupancy of lattice site i. Each lattice
site occupancy can assume the value ni = 1 (occupied)
or ni = 0 (empty) due to mutual exclusion. We use l to
denote the actual value of L(t) at a specific time. The
random variables (mi)l
i=0 representing the DL occupancy
can take values in N0 (no exclusion).
The dynamics of the two-lane model
is a difficult
stochastic many-body problem, in which the bulk dy-
namics and the size of the system are mutually coupled.
In the limit where the bulk dynamics is much faster than
the length changes, we may however assume that on the
time scale over which the length of the lattice changes,
the distribution of particles on the lattice is stationary
(adiabatic assumption). Thus we can decouple the equa-
tions for the length change and particle movement, which
simplifies the mathematical analysis considerably. Using
a mean-field approximation (see Appendix A) one ob-
tains occupancy densities. We denote these as ρi = (cid:104)ni(cid:105)
and ηi = (cid:104)mi(cid:105), where averages are ensemble averages.
In the adiabatic limit, the stochastic dynamics of the
lattice length is a simple birth-death (polymerization
- depolymerization) process. Thus the system length
changes as
∂tL(t) = γ − δρ+(L) ,
(1)
with the TL tip density denoted by ρ+. Spontaneous
polymerization occurs at rate γ and motor-induced de-
polymerization at rate δ. L now refers to the average
length and is no longer a stochastic variable. In the fol-
lowing, we will only consider the stationary case (and de-
note the stationary length by L). Thus the length change
equation (1) yields a condition on the TL tip density:
ρ+ =
γ
δ
.
(2)
In the remaining part of this Section we will formulate
the current-balance equations for both lanes to derive a
length-dependent expression for the particle density at
the tip. Solving for the length L yields the main result
of this Section, Eq. (13). From the analysis it becomes
apparent that the relevant length scale, denoted by λ
[Eq. (7)], corresponds to the average distance a particle
diffuses on the DL, before it attaches to the TL. Fur-
thermore, apart from the adiabatic assumption, meaning
that the particle occupancy equilibrates fast in compar-
ison to the length dynamics, we make use of three more
approximations: first, a mean-field approximation ne-
glecting correlations between the occupancies at different
lattice sites, justified by the low-density regime, second,
the continuum limit requiring that the number of lattice
sites is large and, third, a mesoscopic limit implying that
the total attachment and detachment rate over the entire
lattice are comparable to the hopping rate on the TL. A
reader not interested in the mathematical details of the
dynamics may want to skip the remaining part of this
Section.
The density profiles on the TL and DL bulk, ρi and ηi
for i = 1, . . . , L, are determined by the current balance
for each lane and site [see also Fig. 1(c)],
i − J T
+ (J T
i+1) ,
(3a)
0 = + J D→T
(TL)
(DL) 0 = − J D→T
i
i
+ Di ,
(3b)
where we have defined the transport current on TL as
:= ρi−1(1−ρi), and the exchange current between TL
J T
:= ω(1−ρi)ηi−ωρi. Moreover, diffusion
i
and DL as J D→T
on the DL is described by Di := (ηi+1−ηi)−(ηi−ηi−1).
At the left boundary (base of the protrusion) which is
i
coupled to the reservoir one finds
0 = +J D→T
0 = −J D→T
0
0
− J T
1 ,
+ (η1 − η0) − η0 + α .
(4a)
(4b)
The density current onto the TL's first site is due to
particle transfer from the first site of the DL, J D→T , and
the transport current on the TL, J T . For the first site of
the DL there is the diffusive current onto the neighboring
DL site as well as the exchange with the first site of the
TL. Furthermore, at rate α particles enter the first site of
the DL from the reservoir. This gives the corresponding
influx current α. At diffusion rate particles also exit
the system from the first site of the DL, which leads to
a current of −η0 out of the system.
To solve these equations, we employ a continuum ap-
proximation, assuming that the lattice spacing is smaller
by far than the lattice length. In other words, we perform
a Taylor expansion in the ratio of lattice spacing a ≡ 1
to system size, and only keep terms up to second order.
In this way, we obtain the following continuous currents
with x ∈ [0, L],
J T (x) = [ρ(x) − ∂xρ(x)][1 − ρ(x)] ,
J D→T (x) = ω[1 − ρ(x)]η(x) − ωρ(x) ,
D(x) = ∂2
xη(x) ,
(5a)
(5b)
(5c)
and rewrite the flux balances accordingly. From the flux
balances for the first sites η(0), ∂xη(0) and ρ(0) are de-
termined to be
η(0) =
α
,
having defined the length scale
∂xη(0) = λ−2 η(0) ,
ρ(0) = ω η(0) ,
(cid:114)
.
ω
λ ≡
4
(6a)
(6b)
(6c)
(7)
∝ √
Thus the motor density at the first DL site, η(0), equals
the ratio of the particle influx rate to the particle out-
flux rate. ρ(0) is given by the DL density at the first
site from which transfer to the first TL site occurs. The
length scale λ can be interpreted as the average distance
(in units of the lattice spacing) covered by a particle on
the DL by diffusion before it attaches to the TL, and it
is closely related to the root mean square displacement
t after the typical attachment time scale t = 1/ω.
It will turn out that λ is the intrinsic length scale of the
system and most distances on the lattice will be mea-
sured with respect to this quantity. The three boundary
conditions [Eq. (6)] will now be used as initial conditions
for the bulk equations.
First, in the low-density limit, ρ (cid:28) 1, ρ (cid:28) η, we decou-
ple the two equations [Eq. (3)]. Note that ρ (cid:28) 1 implies
that (1 − ρ) ≈ 1, which is equivalent to lifting the parti-
cle exclusion. With the two initial conditions, Eqs. (6a)
and (6b), we solve the resulting second-order differential
equation, ∂2
xη(x) = η(x)/λ2, to give
η(x) = η(0)
sinh
+ cosh
.
(8)
(cid:18) 1
λ
(cid:16) x
(cid:17)
λ
Sorting the bulk current balance on the TL by orders of
1/L implies that the TL density is the integral of the DL
density that has attached to the TL, ω
η(y)dy = ρ(x),
yielding
ρ(x) = ρ(0) +
α
λ
(cid:18) 1
λ
(cid:16) x
(cid:17)
λ
0
+ sinh
(cid:16) x
(cid:17)(cid:19)
λ
cosh
.
(9)
The resulting density profile for the low-density phase
has a similar functional form as the density profile found
in Ref. [51] although a static lattice was considered in
that case. In particular, the exponential density increase
toward the tip can be reproduced.
Regarding the last site, we expect a discontinuity in
the density profile, as the hopping rules change discon-
tinuously to accommodate growth and shrinkage. The
left limit ρ− [see also Fig. 1(c)] is determined by the bulk
density, while the right limit ρ+ is fixed by the stationar-
ity condition on the length, i.e. ρ+ = γ/δ, Eq. (2). The
system is closed everywhere except at the first site, and
consequently, the flux to the last site has to equal the flux
out of the TL onto the DL, which is ρ+δ to first order,
J T (L) = ρ+δ .
(10)
(cid:16) x
(cid:17)(cid:19)
λ
x(cid:82)
This equality gives us an implicit condition on the system
length L.
The equations for the tip dynamics become more
transparent when formulated in the co-moving reference
frame, as otherwise the last site is not necessarily L. In
this frame two additional currents add to the bulk cur-
rent in the previously used reference frame, the currents
from relabeling due to a growth or a shrinkage event:
J T (x) = ρ(x)(1 − ρ(x)) − γρ(x) + δρ+ρ(x) .
(11)
Solving the flux balance (10) yields
(1 −(cid:112)1 − 4γ) ≡ ρ− ,
ρ(L) =
1
2
(12)
where ρ− can be interpreted as the left limit of the den-
sity at the last site. Approximating the hyperbolic func-
tions as exponential functions with positive argument
(λ (cid:28) 1), we obtain
(cid:20)
(cid:21)
L = λ ln
2
.
(13)
(ρ− − ρ(0))
λ
α
The higher the particle density on the TL, the faster the
system depolymerizes. Hence, a smaller value of λ results
in a smaller steady-state length. This reasoning not only
applies for the prefactor but also for the numerator of
the argument in the logarithm. Here the influx into the
diffusive lane (α) is weighted by 1/λ. ρ− corresponds
to the critical density that depolymerizes the system at
exactly the speed that is necessary in order for polymer-
ization and depolymerization to be balanced on average.
The bigger the critical density, the higher the stationary
particle density on the TL and the longer it takes to fill
the system. Thus the system has more time to grow.
III. NUMERICAL ANALYSIS
So far, we derived analytical expressions for the limit
of slow length change with respect to particle density
equilibration. Now we want to explore the full regime,
which informs us about the phenomenology of the model
beyond the adiabatic regime. We therefore perform
stochastic simulations of the lattice-gas model defined
in Section II A employing Gillespie's algorithm [60]. The
numerical results will also be used to check the approxi-
mate analytical description for the adiabatic case, which
was obtained by using a mean-field analysis.
A. Choice of Parameter Space and Numerical
Method
For the numerical analysis of the system we focus on
the dependence on the growth rate γ, while keeping the
other parameters fixed. The variation of γ causes a quali-
tative change in the dynamics: For small γ, the adiabatic
5
assumption should be valid, and we expect a well-defined
length, whereas for large γ Ref. [19] suggests that length
regulation is no longer possible [61]. We want to focus
the analysis on what happens in an intermediate regime
of γ: The initial length L0 was set to L0 = 100. We
fix as attachment and detachment rate ω = 1/L0, as in-
flux rate α = 0.1, as diffusion rate = 5.0 [62], and as
depolymerization rate δ = 1.0; in each case these param-
eters are expressed in terms of the hopping rate on the
TL ν ≡ 1. The choice of Ω = ωL0 ≡ 1 to be of the or-
der of the other rates is motivated by the processivity of
the molecular motors, which can walk over long distances
along the cytoskeletal filament before detaching [63]. It
is also the theoretically interesting case as it guarantees
that the number of attachment and detachment events
over the length of the system competes with the other
rates [7, 8]. For simplicity, we choose the same rate for
attachment and detachment. However, we do not expect
the qualitative results to change for different attachment
and detachment rates as long as both are still taken to be
small. Moreover, α and together are chosen such that
the density at the first site of the DL is rather small. As
shown in Ref. [19], length control in their system, that is
the system neither shrinks to zero size nor grows with-
out bound, is only feasible in the low-density parameter
regime. Lastly, δ is chosen to be equal to the hopping
rate.
We only took into account simulations where the sys-
tem did not shrink to zero length but a stationary state
was reached. Accordingly, we also chose the interval for
the growth rate in such a way that most simulations ful-
filled this criterion. The choice of L0 (for fixed ω = Ω/L0)
did not influence the results in any way, as we discarded
the initial behavior before the stationary state.
B. Stochastic Simulations and Model
Phenomenology
1. Mean Length
We tested the analytical
insights described in Sec-
tion II B by comparing them to the results of stochas-
tic simulations. To begin with, we determined the mean
length of the system as a function of the growth rate γ,
as shown in Fig. 2. For small growth rates, the length
increases sub-linearly with the growth rate, up to an in-
flection point from which it then increases super-linearly.
As expected, the numerical result agrees nicely with the
analytical results in the adiabatic limit as the growth rate
tends to zero: γ → 0 (i.e. when the growth and shrink-
age dynamics are slow relative to the particle dynamics).
However, the simulation results deviate strongly from the
predictions for larger growth rates γ. As the adiabatic
assumption was the only critical assumption in the the-
oretical analysis [64], the numerical simulations tell us
that this approximation cannot be valid for larger growth
rates. On the contrary, with increasing growth rates, the
6
FIG. 2. Mean system length. Mean length of the system
is plotted as a function of the growth rate γ. The analyti-
cal result (red, solid line) agrees well with the results from
stochastic simulations (gray, filled circles) for small growth
rates, γ (cid:28) 1, where the adiabatic assumption is expected to
hold. For increasing growth rates the numerical data show an
inflection point at which they begin to deviate strongly from
the results in the adiabatic limit. This indicates that the
dynamics shows qualitatively new behavior for large growth
rates.
particle configuration on the lattice no longer equilibrates
on the time scale of the length changes. As a result, there
must be a time lag between the length change and the
equilibration of the motor configuration, and this could
possibly lead to interesting dynamics. To explore this
further, we next study the length distribution.
2. Length Distribution
Figure 3 shows the length histograms for different val-
ues of the growth rate γ. In the inset, we also show the
minimal and maximal lengths in comparison to the aver-
age length and the standard deviation of the length. We
observe that for larger growth rates the length distribu-
tions become broader, while all are right-skewed. This
right-skewness implies that we cannot approximate them
as Gaussian distributions as was done in Ref. [19], and so
it is not feasible to use a van Kampen system-size expan-
sion to obtain higher moments of the length distribution
analytically.
From the analysis of the numerical results, we make
the following observations: The standard deviation of
the length increases with the growth rate. Moreover,
the maximum length attained also increases with growth
rate, namely faster than linearly. In contrast, the min-
imum length reached remains rather constant. This is
surprising as, intuitively, a larger growth rate should also
lead to a larger minimal length. Might this be connected
with the suspected time lag between the length change
and the equilibration of the motor configuration? To an-
swer this question, we looked at a simple temporal quan-
tity first, namely the autocorrelation function.
FIG. 3. Length histograms for different growth rates γ
and a simulation time of 107. The larger the growth rate,
the longer the average length and the broader the length dis-
tribution. The distributions are right-skewed in contrast to
a Gaussian. Inset: The average length (squares), the stan-
dard deviation (bars) of the average length, and the maxi-
mum length reached (right-pointing triangles) increase non-
linearly with larger growth rates, while the minimum length
attained (left-pointing triangles) remains essentially constant.
The shaded areas (green (gray), orange (light gray), and red
(dark gray)) correspond to the value of the growth rate γ in
the corresponding length histograms.
3. Autocorrelation
Figure 4 shows the ensemble autocorrelation function
for different values of the growth rate γ. It can be stated
in terms of the covariance between lengths at times τ and
t + τ , Cov(L(τ ), L(τ + t)), as follows
C(t) := (cid:104)Cov (L (τ ) , L (τ + t))(cid:105)/σ2 ,
(14)
FIG. 4. Autocorrelation function. The autocorrelation
functions, each of an ensemble of 1000 runs, for several growth
rates γ are compared. The autocorrelation function for the
smallest growth rate γ = 0.005 (purple line with squares) al-
most immediately decays to zero, while the autocorrelation of
γ = 0.14 (blue line with triangles) oscillates with a frequency
comparable to the observed length oscillations.
0.000.020.040.060.080.100.120.140.16growth rate 0100200300mean system length Lanalyticsimulation050001000015000200002500030000time0.20.00.20.40.60.81.0autocorrelation=0.14=0.12=0.06=0.02=0.005 7
FIG. 5. Time traces for filament length, total particle number, DL occupancy, and TL occupancy for a full simulation time of
107. (a) System length (gray) and total particle number (red (dark gray)) dynamics for a long time interval and small growth
rate γ = 0.01: Both the length and the total particle number change stochastically. (b) System length (gray) and total particle
number (red (dark gray)) dynamics for a large growth rate γ = 0.14: We observe length oscillations and a sawtooth-shaped
behavior of the total particle number. (c) Zoom in for large γ = 0.14: Upper panel: System length (gray) and total particle
number (red (dark gray)) dynamics. Middle panel: occupancy of the TL (orange (light gray), lower line) and DL (blue (dark
gray), upper line) tip neighborhood, which is chosen to consist of 20 sites from the tip. Lower panel: occupancy of the TL
(orange (light gray), lower line) and DL (blue (dark gray), upper line) bulk, which corresponds to the whole lane except the
tip neighborhood. We observe that the tip neighborhood is densely occupied compared with the bulk.
where (cid:104). . .(cid:105) denotes the ensemble average and σ is the
standard deviation of the length.
at individual time traces of the system length for small
and large growth rates.
In general, we would expect the autocorrelation func-
tion C(t) to decay exponentially with time, yielding an
autocorrelation time that is equal to the typical internal
relaxation time, i.e. the time scale on which a pertur-
bation in length influences the length dynamics. This
is indeed the case for a small growth rate (γ = 0.005
in Fig. 4). However, for larger growth rates, while still
being enveloped by an exponential decay, the autocorre-
lation function oscillates with an oscillation period that
increases with the growth rate. This indicates that for
large growth rates the length is oscillating and that there
might be two qualitatively different limits for the length-
changing dynamics, namely for small and large growth
rates, respectively. To study this issue further, we looked
4. Time Traces
Visual inspection of the time traces (Fig. 5) confirms
the impression gained from the autocorrelation function
that for small growth rates the length of the system fluc-
tuates stochastically. In contrast, for large growth rates,
the fluctuations in length are very small with respect to a
dominant underlying quasi-periodic length-changing pat-
tern, which shows roughly the same oscillation frequency
as the corresponding autocorrelation function. This is
striking, as one would not automatically assume that
enhancing the spontaneous growth rate could lead to a
quasi-periodic pattern.
What might account for such behavior? The first ques-
tion that comes to mind is whether the system is actu-
ally in stationary state and, if that were the case, how
could it be reconciled with an oscillatory behavior.
In
this respect, the most obvious quantity to look at is the
total number of particles that are either on the TL or
on the DL. Is this quantity noisy or does it also show
oscillatory behavior for large growth rate γ? For small
γ, the total particle number behaves highly stochasti-
cally, as expected [Fig. 5(a)]. For large polymerization
rates γ, we observe that not only the length but also the
total particle number shows oscillatory behavior. Sur-
prisingly, however, the time trace of the total particle
number looks very different from the time trace of the
system length: Instead of being rather symmetric within
one period, the time trace for the total particle number
has a sawtooth-like shape, i.e. the total particle number
increases steadily almost during the whole period before
abruptly and drastically decreasing [Fig. 5(b)]. Hence,
the influx of particles dominates the outflux for most of
the time and, in addition, the total particle number does
not change synchronously with the length. Rather, the
dynamics of the total particle number is time-delayed
with respect to the length dynamics – contrary to what
one would expect if the density on the DL were more or
less equilibrated.
This suggests that the DL occupancy is far from homo-
geneous and that there is an intricate interaction between
the motors and the length dynamics: From the equation
of motion for the length L (here considered as a stochas-
tic variable), ∂tL = γ − δn+ with n+ being the particle
number at the TL tip, one expects that the instantaneous
value of n+ should be a key quantity for the length dy-
namics. It is determined by the currents along the TL
and from the DL tip back to the TL tip or to the base. To
garner information about these currents, we determined
not only the total number of motors but also the num-
ber of motors located in the immediate vicinity of the
tip on both the TL and DL; for specificity we chose the
size of the "tip neighborhood" to be 20 sites. We refer
to the number of motors in the tip neighborhood and in
the remaining part of the lane as "tip occupancy" ("tip
occ."), and "bulk occupancy" ("bulk occ."), respectively.
These quantities are shown in Fig. 5(c) for one oscillation
period.
Based on the numerical results we can make several ob-
servations. First, since there are typically more particles
in the DL tip region of only 20 sites than on the remain-
ing part of the DL, the DL tip density is far higher than
the DL bulk density, indicating a considerable crowding
of particles at the tip. Secondly, the DL tip occupancy
in particular increases over almost the whole oscillation
period before drastically decreasing only at the very end
(similarly to the total particle number). Hence, although
the system is already shrinking, the DL tip density con-
tinues to increase. This suggests that there is no com-
munication between the DL tip density and the reservoir
throughout most of the shrinkage phase: as diffusion is
finite, there is no instantaneous equilibration between the
(higher) density at the tip and the reservoir density. Only
when the system is already very short, the cluster at the
tip is released into the reservoir.
This suggests the following mechanism (see Fig. 6):
Diffusion of the particles is slow relative to shrinkage, so
that as shrinkage proceeds the particles cluster more and
more at the tip and do not come into contact with the
reservoir at the left end [Fig. 6(e)]. Hence, they can-
not leave the system as long as its length is not yet
sufficiently short for diffusion to be competitive. Only
when the length of the system falls below a critical value
8
FIG. 6. Intuitive picture for the occurrence of length
oscillations. Starting from a short and empty system (a),
the only two processes possible are growth and influx of par-
ticles from the reservoir into the system (b). Once attached
to the TL particles start walking towards the tip, away from
the reservoir, and the system grows while new particles enter
(c). Since growth is slow compared with transport of parti-
cles on the TL, the particles on the TL "catch up" with the
tip. Furthermore, due to the finite diffusion and the closure
at the tip, the particles then begin crowding at the tip, turn-
ing the growth phase into a shrinkage phase (d). During the
shrinkage phase more and more particles accumulate at the
tip as new particles still enter from the reservoir on the left
while the system shrinks from the right (e). Only when the
system has become very short, is diffusion of particles fast
enough that particles which accumulate at the tip can leave
the system by exiting into the reservoir (f), leaving behind a
short and empty system (a), from which the next oscillation
cycle can begin anew.
[Fig. 6(f)], can the motors diffuse fast enough to reach the
first site of the DL and get out of the system. This then
happens quickly, as the reservoir particle density is very
low and many motors have accumulated at the DL tip
that all exit the system at around the same time, equi-
librating the DL tip density with the reservoir density.
Following this reasoning, this critical length should then
depend on the diffusion rate together with the effective
shrinkage speed, as these two parameters determine the
typical length that the particles can move away from the
tip before the system further shrinks. If the system then
becomes depleted of particles [Fig. 6(a)], there are no
more particles at the TL tip and, as shrinkage is assumed
to be particle-induced, the system can only grow. Since
even for "large" growth rates γ, growth is considerably
slower than the TL hopping rate, γ (cid:28) 1, particles begin
to move toward the tip as the system grows [Fig. 6(b)]
and finally reach the tip and accumulate there [Fig. 6(c)],
turning the growth phase into a shrinkage phase (parti-
cles "catch up" with the TL tip) [Fig. 6(d)].
Notably, this mechanism, which is based on the par-
ticle accumulation at the DL tip, heavily relies on the
particle conservation, since particles can leave the tip re-
gion only via the diffusive lane. If this were not the case,
particles could simply leave the tip region via an exit
rate, effectively reducing the clustering at the tip and so
shortening the extended shrinkage phase.
(a)(b)(c)(d)(e)(f) IV. EFFECTIVE THEORY FOR THE
OSCILLATORY REGIME
So far, we have built up a heuristic mechanism from
an analysis of the numerical data. To examine the valid-
ity of the suggested heuristic mechanism, and to gain a
more quantitative understanding of the oscillations in the
parameter regime considered, we now construct an effec-
tive, semi-phenomenological theory. The effective theory
incorporates the main ideas of the heuristic picture, and
we will check how closely its predictions fit the numerical
results.
The theory is based on an effective description of the
diffusion lane, and on the idea that, depending on where
particles detach from the TL, they either reattach to it
after an average time 1/ω (which is the inverse of the
attachment rate), or leave the system. We first divide
our system qualitatively into four regions (see Fig. 7).
From base to tip, these are the "in-region", the "bulk",
the "tip neighborhood" and the "tip":
• The in-region is close to the base: Here, newly en-
tered particles attach to the TL (via DL), and de-
tach from the TL at rate ω.
• The tip: The last site on the TL at which growth
and shrinkage (together with detachment of the
triggering particle) occur.
• The tip neighborhood: Here, particles that have
previously detached from the tip reattach to the
TL. We neglect detachment and further reattach-
ment, as we assume that particles which detach in
the tip neighborhood reattach in the same region,
balancing each other out.
• The bulk: This merely serves as a linker region
between the "in-region" close to the base and the
"tip neighborhood" close to the tip. Here, we as-
sume that attachment and detachment of particles
balance each other out (particles that detach there
also reattach there).
In summary, we assume that particles that enter the
system, and do not immediately leave it again, attach to
the TL in the in-region. They then either detach there
again and return to the reservoir, or they walk on the TL
towards the tip. Furthermore, particles that detach at
the tip reattach to the TL in the tip neighborhood after
an average time 1/ω. Moreover, growth and shrinkage
occur at the tip.
Note that the division of the system into those regions
is motivated by key components of the system dynamics
such as the coupling to the reservoir at the base, the par-
ticle dynamics on and between the lanes and the length-
changing dynamics at the tip. It is however a theoretical
construct and instead of fixed boundaries there will be
continuous transitions between the different regimes in
the real system.
9
FIG. 7. Schematic of the effective model. We split the
system into four regions, and use an effective description of
the DL, restricting our analysis to the TL. In the in-region
we have attachment at an effective in-rate αeff and detach-
ment at rate ω. The bulk region links the in-region to the
tip neighborhood and we assume that in the bulk attachment
and detachment balance. For the tip neighborhood we assume
that particles which have detached at the tip a time 1/ω be-
fore reattach to TL in the tip neighborhood and need another
time ∆ − 1/ω to reach the TL tip again, yielding a recursion
relation for the tip density ρ+
it . Finally, at the tip we have
detachment at rate δρ+, and the corresponding shrinkage of
the system, and spontaneous growth. The tip and the tip
neighborhood are described in the co-moving frame.
As we have seen in Section III B, the total number
of particles in the system increases almost throughout
the oscillation period, including the greater part of the
shrinkage phase. As a first step, we determine the effec-
tive rate at which particles enter the system, and then
attach to the TL. This rate will not equal the "bare" in-
rate α, as particles can also leave the system again before
attaching.
What is the probability, Prob(leaving), that a particle
that has just entered leaves the system again before at-
taching to the TL? To answer this question we assume
that the length of the system is considerably larger than
the length of a typical journey of a particle on the DL be-
fore it attaches to the TL, and discuss the influence of a
short length separately below. By carefully keeping track
of all possible exit paths we determine Prob(leaving) as
Prob(leaving) = 1 −
+ O(ω)
√
ω
(see Appendix, C 1), where we allow the exit rate from
the system, , to be different from the diffusion rate .
As a result, the effective on-rate onto the TL is given by
√
αeff ≈ α
ω
(15)
to lowest order in ω. It is proportional to the ratio of par-
ticle influx α to particle outflux from and back into the
reservoir itself, which can be interpreted as the density in
the reservoir. Furthermore, the effective on-rate onto the
αDLTLρit+δ0αeffωρ=ρit-1+δtip neighb.bulkin-regiontip////γLεη0~~ρit+δtime ∆ TL increases with the attachment rate ω, as expected,
and with the diffusion rate , since for a higher diffusion
rate (compared with the exit rate ) particles diffuse fur-
ther into the system. Using this effective entrance rate
we now proceed to our effective TASEP model.
First, we estimate the length of the in-region lI , since
- due to attachment and detachment here - its length in-
fluences the density. To do so we model a typical particle
on the DL (which does not leave the system immediately)
until it attaches to the TL, as a symmetric random walker
with reflecting boundary at x = 0. Attachment to the TL
follows a Poisson process at rate ω. Assuming that the
particle starts at x = 0, and diffuses with diffusion con-
stant (lattice spacing 1), we find that the average lattice
site until which the particle has diffused when attaching
to the TL is given by (cid:104)x(cid:105)± σ(x) =(cid:112)/ω±(cid:112)/ω = λ± λ
(see Appendix, C 2). Here λ =(cid:112)/ω is a characteristic
length scale of the system (see Section II B).
Since we assume the in-region to extend from the base
into the system, we will approximate it as the region
[0, 2λ] on symmetry grounds. The left (right) boundary
corresponds to the average distance a particle travels be-
fore attaching to the TL (λ) minus (plus) the standard
deviation (also λ). The length of the in-region is deter-
mined by the characteristic length scale λ:
lI ≈ 2λ .
(16)
With this, we now determine the density profile in the
in-region, which we assume to equilibrate quickly on the
time scale of the oscillations. Furthermore, we assume
that attachment is evenly distributed over the whole in-
region, yielding an attachment rate of αeff /lI per site.
In the low-density and continuum limit, together with
the hopping transport on the TASEP and detachment
of particles at rate ω, this yields a density profile in the
in-region ρ(x) = αeff [1 − e−ωx] / (lI ω), x ∈ [0, lI ] (see
In particular, the density at the right
Appendix C 2).
end of the in-region is given by
(cid:104)
ω(cid:105)
ρ(lI ) =
α
2
√
1 − e−2
.
(17)
It increases with the density in the reservoir, α/, and
also with both the diffusion rate and the attachment
and detachment rate ω. Note that we measure time in
units of the hopping rate ν ≡ 1 on the TASEP, and length
in units of the lattice spacing a ≡ 1.
We introduced the bulk region in order to interpo-
late between the densities in the in-region and in the
tip neighborhood. Since the bulk region is sufficiently
far from the reservoir and from the tip (at least when
the length of the system L ≥ 4λ) we assume that at-
tachment and detachment approximately balance, and
so the density is approximately constant and equal to
ρ(lI ) [Eq. (17)].
For the analysis of the dynamics in the tip neighbor-
hood and at the tip, we switch to a different reference
frame, namely starting at the tip and reaching into the
10
tip neighborhood, co-moving with the tip. The tip neigh-
borhood represents that part of the system within which
particles that have detached from the TL tip typically
diffuse on the DL before reattaching to the TL. Thus, we
assume that the tip neighborhood has the same length
as the in-region lT = lI = 2λ as for both the average
distance traversed before attaching to the TL is essen-
tial. We will now substantiate the idea that particles
that have detached from the tip reattach back to the TL:
We suppose that particles that detach at the tip reattach
to the TL on average after time 1/ω. Furthermore, they
then walk to the tip during an additional average time
lT /2 = λ, since on average they attach to the TL at a dis-
tance lT /2 away from the tip and take one directed step
during time 1/ν = 1. So, the tip density at time t, ρ+(t),
influences the tip density at time t + 1/ω + λ ≡ t + ∆,
ρ+(t + ∆). We determine ρ+(t + ∆) as the steady-state
of the dynamics in the tip neighborhood and at the tip
that results from the usual TASEP dynamics in the low-
density and continuum limit combined with growth and
shrinkage, and attachment at rate δρ+(t)/lT per site in
the tip neighborhood (see Appendix C 3).
In summary, we imagine that particles that enter the
tip region start "cycling" there: They detach at the tip,
diffuse in the tip neighborhood, reattach to the TL, walk
back to the tip, detach again and so on (Fig. 7). As
long as < 1/ω, the average distance λ to the tip after
reattaching to the TL is less than the average walking
distance 1/ω on the TL, so most particles that reattach
to the TL reach the tip.
This procedure yields a recursion relation for the tip
it at times tit = it×∆ (see Appendix C 3 for an
densities ρ+
explicit formula) that could, in principle, be used to de-
termine the time evolution of the tip density. So far, how-
ever, we have implicitly assumed that the length of the
system, l, is infinitely long, l (cid:29) λ, and we have not con-
sidered how the physics changes for comparatively short
system lengths. In particular, we ignored the fact that
the shorter the system, the less likely particles that have
previously detached from the tip are to reattach to the
TL, as they may now leave the system beforehand. So,
there will be some minimal length at which the majority
of particles that had previously been in the tip region has
left the system. From about this point the system starts
growing again.
To estimate this minimal length, we consider a 1D sys-
tem with injection of particles (=detachment) at rate r
at site l (tip), symmetric diffusion at rate within the
system, outflux (=reattachment) of particles at rate ω ev-
erywhere, and an additional outflux of particles at rate
at site 0. In the steady-state and with a continuum ap-
proximation, the reattachment probability of a particle
detaching at the tip at length l can be approximated as
preattach(l) ≈ 1 − F exp
(18)
(cid:19)
(cid:18)
− l
λ
for l (cid:29) λ (see Appendix, C 4), where
F =
2λ3
ϕ + 2ϕλ + (1 + ϕ)λ2 + λ3 .
(19)
Here, ϕ = / is the ratio between the diffusion rate
and the exit rate. As expected, the reattachment prob-
ability decreases with decreasing length l, and has the
characteristic length scale λ. Furthermore, F increases
with decreasing ϕ, and so the reattachment probability
decreases with decreasing ϕ. As a result, for a larger
exit rate compared with the diffusion rate (small ϕ), the
reattachment probability is small.
We have chosen the time interval ∆ in such a way that
during time ∆ a given particle that has detached at the
tip diffuses in the DL, reattaches and walks back on the
TL to the tip. So, in order for a particle to remain in
the system, it needs to reattach to the TL each time it
has detached and so, it needs to reattach back for all
lengths lit the system attains at times tit = it × ∆. We
11
have psurvival ({lit}it=1,...,n) = (cid:81)n
it=1 preattach(lit) after a
series of lengths {lit}it=1,...,n. We further define the min-
imal length as the length where approximately 50% of
the particles that were in the system at maximal length
have left it. Making the rough assumption that the sys-
tem shrinks at a constant velocity v ≈ γ/2, which is half
the maximal growth speed, we find
(cid:20) 2λF
(cid:21)
γ∆ ln(2)
lmin ≈ λ ln
,
(20)
with F as defined before, Eq. (19) (see Appendix C 4).
This means that, to leading order, the minimal length is
determined by the typical length scale λ. The weak loga-
rithmic dependency on the inverse growth rate 1/γ arises
from the fact that the growth (and shrinkage) speed
scales with γ.
Taking these considerations together, we find the fol-
it and the
lowing recursion relation for the tip densities ρ+
lengths lit at times tit = it × ∆:
(cid:26)(cid:2)ρ+
it−1δ2 1(lit−1−lmin) − A(cid:3) +
ρ+
it =
lit = lit−1 + γ∆ − ρ+
it−1δ∆ ,
(cid:113)(cid:2)ρ+
it−1δ2 1(lit−1−lmin) − A(cid:3)2
+ 2B(cid:2)ρ+
it−1δ 1(lit−1−lmin) + C(cid:3)(cid:27)(cid:46)
B ,
(21a)
(21b)
with initial condition ρ+
0 = 0 and l0 = 0 (which, however,
does not influence the long-term behavior, as in the case
of the stochastic simulation). Furthermore,
A = δ (1 − γ) + γ (1 − γ) − δ ρ(lI ) (2 − γ) ,
B = 2δ [δ(1 − ρ(lI )) + γ] ,
C = (1 − γ)ρ(lI ) ,
where we use ρ(lI ),
lmin and ∆ as defined before.
Eq. (21b) derives from the growth and shrinkage dy-
namics (constant growth at rate γ and motor-induced
shrinkage at rate δρ+) during the time interval ∆, and 1
denotes the Heaviside step function.
(22)
(23)
(24)
Solving this recursion relation numerically, we now
compare the predictions of our effective theory to the
outcomes of simulations. To begin with, let us look at
the result of the recursion relation, Eq. (21), itself, which
is shown in Fig. 8. In line with the stochastic simulations
[Fig. 5(c)], the length changes periodically with relatively
symmetrical growth and shrinkage phases, while oscilla-
tions of the tip density, in contrast, follow a sawtooth
pattern.
For a more quantitative comparison, we have numer-
ically determined several quantities from the recursion
relation and compared them to the results from simula-
tions. In accordance with the stochastic simulation, we
find that the minimal length is largely independent of
the growth rate γ (Fig. 9) with a tiny decrease in mini-
mal length for increasing growth rate in both stochastic
FIG. 8. Solution of the recursion relation for the tip
density (orange, solid line) and the length (gray, dashed line)
as a function of the iteration step, for γ = 0.14. Both show
periodic behavior, but while the growth and shrinkage phases
are rather symmetric for the length dynamics, the tip density
exhibits a sawtooth shape.
simulations and the analytic prediction. This is what
we would expect, as the turning point from shrinkage
to growth should mainly be determined by the point at
which diffusion (rate ) is fast enough relative to the
shrinkage dynamics to enable the tip cluster to equili-
brate with the reservoir, and thus the system to quickly
deplete.
0100200300400500iteration (it)0.000.050.100.150.200.250.30TL tip densityAnalytic +it0100200300400length litAnalytic lit 12
FIG. 9. Minimal and maximal length. The average min-
imal (maximal) length per oscillation period from stochastic
simulations is compared with the prediction from the effec-
tive theory. From the stochastic simulations we determined
the minimal and maximal length for each oscillation period,
and the average minimal (maximal) length is depicted with
orange squares (green circles), with error bars representing
the corresponding standard deviation. Note that the average
minimal length is approximately independent of the growth
rate, in contrast to the average maximal length. The pre-
diction from the effective theory is shown with red lines: As
in the stochastic simulations the maximal length (solid line)
increases with the growth rate γ, whereas the minimal length
(dashed line) is only weakly dependent on the growth rate,
decreasing slightly with increasing growth rate.
Second, not only the turning points from shrinkage
to growth, but also the inflection points from growth to
shrinkage are important. In numerical simulations, not
only the maximally reached length during the full sim-
ulation (see again Fig. 3 for more details) but also the
average maximal length of the system per oscillation pe-
riod increases faster than linearly with the growth rate
(Fig. 9). This behavior is reproduced by our effective
theory insofar as it also exhibits a faster than linear in-
crease in the maximal length per oscillation period with
the growth rate γ over the parameter range considered.
Comparing the prediction of the effective theory with the
average maximal length per period from simulations, we
find quite good quantitative agreement.
Apart from its amplitude (difference between maximal
and minimal length), the oscillation is also character-
ized by its frequency. Only with the suggested intuitive
mechanism in mind, it is not clear a priori how the fre-
quency should depend on the growth rate γ: There are
two possible, opposing mechanisms. On the one hand,
growth (and shrinkage [65]) increase with larger growth
rate γ, so the oscillation period (frequency) should de-
crease (increase) with growth rate γ. On the other hand,
for larger growth rate, the amplitude increases as well,
namely faster than linearly, and so, the oscillation period
(frequency) should increase (decrease). Furthermore, it
is not clear how fluctuations in length influence the os-
cillation frequency. In summary, it is difficult to predict
FIG. 10. Oscillation frequency. The oscillation frequency
from stochastic simulations is compared with the prediction
from the effective theory. We determined the length oscilla-
tion frequency from the stochastic simulations, first, by de-
termining the autocorrelation oscillation frequency (freq via
autocorr, purple squares), second, by evaluating the distri-
bution of duration between two adjacent minima (freq via
hist, orange circles) and, third, by performing a Lomb Scar-
gle analysis comprising a sine fit (freq via L. S., blue left-
pointing triangles). For the distribution approach, where a
bound on the maximal frequency is used as explained in the
Appendix B, the distribution average and standard deviation
(error bars) are depicted. For the Lomb Scargle analysis the
most probable frequency is shown. Clearly, for all methods
the oscillation frequency decreases with larger growth rates.
For very small rates noise masks the oscillation, such that
the methods employed cannot determine the frequency cor-
rectly. Thus, results from the stochastic simulations are only
shown for growth rates ≥ 0.06. The predicted oscillation fre-
quency from the effective theory is shown as a solid red line.
It displays the same qualitative behavior as the result from
stochastic simulations.
from the intuitive picture alone how the oscillation fre-
quency depends on the growth rate.
We can, however, use our effective theory and the re-
cursion relation to numerically determine the "analyti-
cal" oscillation frequency. We find that the analytical os-
cillation frequency decreases with increasing growth rate
γ (Fig. 10). As mentioned above, visual inspection of the
autocorrelation functions already suggests that the same
is true for the stochastic simulations, and this is con-
firmed by different methods to determine the oscillation
frequency from the stochastic simulations (Fig. 10): In
both the simulation results and the analytical prediction,
the oscillation frequency at γ = 0.14 is around half of its
value at γ = 0.08. Note that for smaller growth rate γ it
is very hard to determine an oscillation frequency from
the stochastic simulations as the oscillation is largely ob-
scured by stochastic noise.
All in all, in the parameter regime considered, our ef-
fective theory agrees nicely with the results from stochas-
tic simulations (Sec. III), supporting the intuitive picture
on which the effective theory is built.
0.060.080.100.120.140.16growth rate 0200400600system length LAnalytic LminAnalytic Lmaxmin. L per periodmax. L per period(Lmin)(Lmax)0.060.080.100.120.140.16growth rate 0.000050.000100.000150.00020frequencyAnalytic freq freq via autocorrfreq via histfreq via L. S. 13
as well [69, 70]. Based on our analysis, we believe that
it would be interesting to further explore how time de-
lays can emerge intrinsically in a spatially extended non-
equilibrium system, and under what conditions this leads
to robust oscillations.
ACKNOWLEDGMENTS
We thank Silke Bergeler, Matthias Rank, Emanuel
Reithmann, and Patrick Wilke for critical reading of this
manuscript and for helpful comments. This research was
supported by the German Excellence Initiative via the
program "NanoSystems Initiative Munich" (NIM). MB
and IRG are supported by a DFG fellowship through
the Graduate School of Quantitative Biosciences Munich
(QBM). MB and IRG contributed equally to this work.
V. SUMMARY AND DISCUSSION
In summary, we have studied a semi-closed system con-
sisting of two coupled lanes, a TASEP lane and a diffusive
lane which, at the tip, spontaneously grow, and shrink
when a particle reaches the tip of the TASEP lane. We
find two qualitatively different regimes for small and large
growth rates, respectively, which differ in the dynamics of
length change: For small growth rates, length change is
mainly stochastic, while for large growth rates oscillatory
patterns dominate.
The occurrence of those oscillatory patterns relies on
the accumulation (crowding) of particles at the dynamic
tip during the shrinkage phase [Fig. 6(d)]. This crowd-
ing leads to a positive feedback mechanism for shrinking
[Fig. 6(e)], as each particle that reaches the TASEP lane
tip further shrinks the system. The crowding is resolved
only after a time delay, namely when the system size be-
comes comparable to the finite diffusion length. Then
exchange of particles can occur between the tip region
and the reservoir at the base, and the tip density equili-
brates with the reservoir density [Fig. 6(f)], finally turn-
ing the shrinkage phase into a growth phase [Fig. 6(a)].
As transport on the TASEP lane is fast compared with
the growth of the system, particles entering the system
from the reservoir [Fig. 6(b)] "catch up" with the growing
tip, and start accumulating there [Fig. 6(c)]. As soon as
the crowding reaches a critical value, the whole process
begins over again.
We provide a deeper quantitative understanding of the
length oscillations by formulating an effective theory. It
relies on the intuitive explanation we propose for the oc-
currence of the oscillations, namely cumulative crowding
of motors at the tip due to finite diffusion, and correctly
predicts the dependence of the oscillation frequency and
amplitude on the growth rate, validating our intuitive
picture.
From this intuitive picture it is evident that the emer-
gence of the periodic behavior crucially depends on the
finite diffusion speed, which - together with particle con-
finement - enables crowding of particles. To our knowl-
edge, oscillatory patterns have not been observed in any
similar lattice-gas model. We attribute this to the fact
that in those models diffusion had not been taken into
account explicitly, or only in terms of a homogeneous
reservoir, corresponding to infinitely fast diffusion.
In our system, in the limit of infinitely fast diffusion,
the equilibration between the DL tip and the reservoir
takes place infinitely fast, and the density on the DL is
homogeneous. So, in this limit our model reduces to the
model discussed in Ref. [19].
On a broader perspective, the time delay due to a finite
diffusion speed in a confined geometry also seems to be
crucial for the occurrence of oscillatory behavior in other
systems, such as in recent models for the Par or Pom
protein systems [66, 67] and for mass-conserving reaction-
diffusion systems [68]. In general, delay times have been
associated with periodic behavior in well-mixed systems
Appendix A: Analytic approach
In the following we perform in detail the calculations
leading to the steady state density profiles as sketched
in Sec. II A. In particular, we will elaborate on the used
approximations, i.e. the adiabatic assumption, the mean
field approximation, the continuum limit and the meso-
scopic limit. We start with some comments on the used
notation.
We denote the first, i.e. the leftmost, site by "0" and the
last site by "L". Indices will be used to denote site num-
bers. Moreover, the results below are stated in terms of
ρj(t)(t), i.e. using the index at time t, not t + dt. This is
necessary to clarify as site indices change due to length
changes. Occupancy numbers n will be approximated
14
by occupancy densities ρ (η) on the TL (DL). Often we
simply denote (cid:104)L(cid:105) by "L". "(cid:104)(cid:105)" represents the ensemble
average.
We begin with the adiabatic assumption which allows us
to decouple length change and particle dynamics. We
perform the argument exemplarily for the TL. The first
step is to write down the probability for a certain lattice
site to be occupied.
Any tuple of length l with entries zero (=empty) or
one (=occupied) describes a possible state of the TL with
length l, e.g. (n0 = 1, n1 = 0, n2 = . . . , . . . , nl = 1). Let
us denote the complete set of such tuples as Ω(l), and
the number of elements it contains by Ω(l). (ni)l
i=0,j
describes the j-th element of this set. In this notation
the probability of a site i to be occupied with one particle
at time t + dt can be written as
P(cid:0)ni(t+dt)(t + dt) = 1(cid:1) =
∞(cid:88)
Ω(l)(cid:88)
l=0
j=1
P(cid:2)ni(t+dt)(t + dt) = 1 (ni)l
i=0,j(t), L(t) = l(cid:3) P(cid:2)(ni)l
i=0,j(t) L(t) = l(cid:3) P [L(t) = l] .
(A1)
is the Kronecker delta. On the other hand, in Eq. (A2)
for the length changing dynamics the actual occupancy
nT
l can be replaced by its time average. The time aver-
age is equivalent to the ensemble average, ρl at length l,
that is the average tip occupancy at length l (in contrast
to the average occupancy at site l for arbitrary length or
the one for average length (cid:104)L(cid:105)). We find
P(cid:0)ni(t+dt)(t + dt) = 1(cid:1) =
Ω((cid:104)L(cid:105))(cid:88)
(cid:104)
P
ni(t+dt)(t + dt) = 1 (ni)
(cid:104)L(cid:105)
i=0,j(t)
(cid:105)
(cid:104)
(ni)
P
(cid:105)
(A3)
(cid:104)L(cid:105)
i=0,j(t)
,
The first factor is the probability that site i is occupied
at time t + dt under the condition that the system was l
sites long at time t and its state was (ni)l
i=0,j. The second
factor gives the probability that the system was in state
(ni)l
i=0,j at time t under the condition that its length was
l and the last term corresponds to the probability that
the system was l sites long. Every possible state at fixed
length and any length could contribute, hence the sums.
The difficulty is that the length distribution P(L(t) = l)
itself again depends on the occupancy numbers {ni}, in
particular on the TL tip occupancy nT
l :
∂tP(L = l) =δnT
l+1P(L = l + 1)
+ γP(L = l − 1) − (δnT
l + γ)P (L = l),
(A2)
j=1
where the first two terms describe the probability gain
due to a shrinkage or growth event of a longer or shorter
length, respectively, while the last term represents the
corresponding probability loss.
To tackle this problem analytically, we assume that the
length changing dynamics happens at a far longer time
scale than the particle hopping. Thus both dynamics can
be decoupled. We refer to this simplification as adiabatic
assumption. It is untenable for large growth rates, as con-
firmed in the simulations, but suitable for small growth
rates. In this regime the assumption implies that we can
take the particle densities to adapt instantaneously to
the current length and correspondingly that we can re-
place the (changing) length by a constant length when
describing the particle occupancy dynamics. Thus, for
the occupancy number dynamics, Eq. (A1), we choose
the, by this assumption constant, length to equal the
average lattice length. Mathematically this can be ex-
pressed by setting P (L(t) = l) ∝ δ(l,(cid:104)L(cid:105)) where δ(i, j)
and
∂tP(L = l) =δρl+1P(L = l + 1) + γP(L = l − 1)
− (δρl + γ)P (L = l).
(A4)
So, applying the adiabatic assumption, we can decou-
ple the occupancy number and length dynamics and pro-
ceed.
From now on, we will furthermore restrict ourselves to
the stationary state of the system,
∂t(cid:104)ni(cid:105) != 0
and L ≡ (cid:104)L(cid:105). The next approximation to solve the cou-
pled set of occupancy equations is to eliminate the cor-
relations between occupancies at different sites by using
the mean-field approximation
(cid:104)ninj(cid:105) ≈ (cid:104)ni(cid:105)(cid:104)nj(cid:105) ≡ ρiρj.
The equations for the occupancy dynamics at any site
are then given by
0 = ∂tρ0 = −νρ0 (1 − ρ1) − ω (ρ0 − µ0 + ρ0µ0) ,
0 = ∂tρi = ν (ρi−1 (1 − ρi) − ρi (1 − ρi+1))
− ω (ρi − µi + ρiµi) ,
0 = ∂tρL = −γρL + δρL (ρL−1 − 1) + νρL−1 (1 − ρL) ,
0 = ∂tµ0 = α − (2µ0 − µ1) + ω (ρ0 + ρ0µ0 − µ0) ,
0 = ∂tµi = (µi+1 + µi−1 − 2µi) + ω (ρi + ρiµi − µi) ,
0 = ∂tµL = −γµL + δρL (1 + µL−1) + (µL−1 − µL) ,
with i denoting any bulk site. The corresponding flux
balances are
0
0 = +J D→T
(TL)
(DL) 0 = −J D→T
0 = +J D→T
(TL)
(DL) 0 = −J D→T
0
i
i
+ J T
1 ,
+ (η1 − η0) − η0 + α,
+ (J T
i − J T
i+1) ,
+ Di ,
(A5)
(A6)
(A7)
(A8)
for the first site and the bulk respectively.
Moreover, although the lattice is growing and shrink-
ing, its length L is typically 100 up to 1000 times larger
than the remaining parameters and densities. Thus it is
justified to consider the limit where the lattice spacing
ξ tends to zero when the total length of the system is
rescaled to 1.
The second step is thus to apply the continuum limit
by replacing the lattice by a smooth interval [0, 1]. Note
that this is different to the choice in Sec. II B, where
the interval is set to [0, L] in order to keep the notation
cleaner. Here we want to clearly see the orders of the
following Taylor expansion. We further define the occu-
pancy density (also named ρ) to be the smooth function
satisfying ρ (ξi/L) = ρi with i = 1, . . . , L − 1 denoting
the lattice site index. We set ξ = 1 in order to rescale
the system size to 1). ρ can then be Taylor-expanded in
the limit 1/L → 0:
(cid:18)
ρ
(cid:19)
x ± 1
L
= ρ(x) ± 1
L
∂xρ(x) +
1
2L2 ∂2
xρ(x) + O
For the currents this expansion gives
(cid:20)
(cid:21)
J T (x) =
[1 − ρ(x)]
J D→T (x) = ω [1 − ρ(x)] η(x) − ωρ(x)
ρ(x) − ∂x
L
ρ(x)
D(x) =
∂2
x
L2 η(x) .
(cid:18) 1
(cid:19)
L3
(A9)
(A10)
(A11)
Moreover, we focus on the mesoscopic limit [7, 8] of ω.
This implies that ω = Ω/L0, with L0 denoting the initial
length, is treated as order 1/L. Consequently, J D→T (x)
has no 0th order contribution.
As the DL is the only source of particles on the TL, we
will at first solve the equation for the diffusive lane and
15
use it to obtain the TL density profile. We begin at the
left boundary.
(DL)
0 = −η(0) + α − J D→T (0) +
∂x
L
η(0) ,
(A12)
thus, to 0th order in the lattice spacing we are left with
0 = −η(0) + α, concluding
η(0) =
α
.
For the TL we have
0 = ω [1 − ρ(0)] η(0)− ωρ(0)− ρ(0)
(cid:20)
1 − ρ(0) − ∂x
L
(A13)
(cid:21)
ρ(0)
,
(A14)
implying 0 = −ρ(0)(1 − ρ(0)) + O(1/L). This equation
has two solutions, either the first site is always occupied
or always empty. To lowest order, as we only treat the
low density limit, the site has to be empty, i.e. ρ(0) = 0.
To first order, we obtain 0 = ωη(0) − ρ(0), thus
ρ(0) = ωη(0) .
(A15)
The first order equation for the DL is
ω [(1 − ρ(0)) η(0) − ρ(0)] =
1
L
∂xη(0).
(A16)
With ρ(0) = 0 to 0th order, we obtain
η(0) = λ−2η(0) ,
∂x
L
with
λ ≡
(cid:114)
ω
(A17)
.
(A18)
Having solved the boundary equations, we apply these
results to solve the bulk equations. By adding the DL
bulk dynamics equation corresponding to Eq. A8
(DL)
0 = −J D→T +
∂2
x
L2 η(x)
(A19)
to the TL bulk dynamics equation derived from Eq. A7
.
(TL)
0 = J D→T (x) − ∂x
L
ρ(x)(2ρ(x) − 1) ,
(A20)
we obtain the first order TASEP bulk equation 0 =
∂xρ(x)(2ρ(x) − 1). As we are in the low density limit,
the solution ρ(x) = 1/2, corresponding to the maximal
current solution, can be ruled out, thus ∂xρ(x) = 0.
Using our results from the left boundary as initial val-
ues, we conclude that the constant density equals ρ(x) =
ρ(0) = 0 to first order. We conclude that the occupancy
is constant and thus equals the occupancy at the first
site. To first order, it has been determined to equal zero
(Eq. A15), thus ρ(x) = ρ(0) = 0. Inserting this result to
the second order DL equation gives
∂2
x
L2 η(x) = ωη(x) ,
(A21)
which is solved by
η(x) = A sinh
xL
λ
+ B cosh
xL
λ
.
(A22)
Using our boundary conditions, η(0) = α/ (Eq. A13)
and (∂x/L)η(0) = λ−2η(0) (Eq. A17), gives
η(x) = η(0)
.
(A23)
sinh
xL
λ
+ cosh
xL
λ
λ
(cid:18) 1
(cid:19)
We continue with the second order equations of the TL
bulk,
0 = ω(1−ρ(x))η(x)−ωρ(x)+
ρ(x)(2ρ(x)−1) . (A24)
∂x
L
Upon employing that the first order value of ρ is zero
(Eq. A15), we are left with
0 = ωη(x) − ∂x
L
ρ(x) ,
(A25)
which is solved by
ρ(x) = ρ(0) +
α
λ
(cid:18) 1
cosh
xL
λ
+ sinh
xL
λ
λ
(cid:19)
.
(A26)
In summary, we have found analytic expressions for the
steady state TL and DL occupancy densities in the adi-
abatic limit.
Appendix B: Oscillatory behavior
In section III we have learned about the existence of a
parameter regime where the length change exhibits oscil-
latory behavior. Following up on these investigations, we
want to further discuss the methods used. Moreover we
want to examine the occupancy densities at the system
tip when the system switches from growth to shrinkage.
As the time intervals between two events in the sim-
ulation are not uniformly spaced, we performed a Lomb
Scargle analysis instead of a Fourier analysis [71] to de-
termine the average oscillation frequency (see Fig. 10).
The algorithm essentially fits a sine function to the data
and checks which frequency matches the data best. We
deduced the frequency with the smallest false alarm prob-
ability as well as the second and third best choice. For
larger γ values, the frequency decreases with increasing
growth rate and reassuringly, the three best frequencies
agree quite well. For small growth rates, the results
should not be taken seriously, as there are also no vis-
ible oscillations in the time traces.
Moreover, we determined the minima and maxima of a
time-series of the length. This was done by cutting off the
data of length for lengths larger than the initial length
L0 (which is 2-3 times larger than the minimum average
length and smaller than the average length). Within each
of the remaining intervals we determined the minimal
length, while sorting out all minima that occurred very
16
quickly after each other, i.e. in less time than a threshold
∆T . This threshold excludes small fluctuations around
L = L0 and is chosen in a way to minimize artefacts of
chopping off the length at L0. We used ∆T = 800. Note
that our choice of ∆T does influence the frequency results
as it restricts the maximal frequency. Between each two
minima, we then determined the maxima. The respec-
tive averages and standard deviations for the maximally
and minimally obtained system length during an oscil-
lation period are plotted in Fig. 9. The maxima clearly
increase with larger growth rates, whereas the minima
remain rather constant. The latter further supports our
intuition of a particle cluster at the DL tip, which equili-
brates with the reservoir only when the system length is
small enough for diffusion to be comparable to shrinkage.
From the temporal distance of the minima, the oscillation
frequency was deduced (see Fig. 10). The values agree
with the result of the Lomb Scargle analysis mentioned
before. As a third method to determine the oscillation
frequency we extracted the frequency from the autocor-
relation function (Eq. 14). We searched for the first 2-4
maxima and minima of the autocorrelation function and
averaged their distance. For smaller grow rates we had to
reduce the number of maxima and minima, as the num-
ber of oscillations reduced from > 4, to 2 and even 1 in
the case of γ = 0.005. The extracted frequencies are also
shown in Fig. 10.
Fig. 11 shows the TASEP tip neighborhood (i.e. 20 tip
sites) occupancy at the oscillation maxima correspond-
ing to the turning point between a phase of growth and
a phase of shrinkage. The blue upper line represents the
mean of the TASEP tip neighborhood occupancy (red
squares). The results vary strongly, thus we checked re-
lated observables. But also the maximal tip neighbor-
hood occupancy (gray left-pointing triangles) within ten
timesteps - five before the maximum is reached and five
thereafter - and the average (yellow right-pointing trian-
gles) fluctuate. Nevertheless we see that nearly all mea-
surements of the maximal tip neighborhood occupancy
(gray left-pointing triangles) lie above the critical den-
sity (purple lower line), being γ times the tip neighbor-
hood size (here 20 sites) as the length change is given
by ∂tL = γ − δρ+ (as δ = 1). When we compare the
critical density to the mean density for a time interval
covering more than one oscillation period, and not just
at the time points where the amplitude is maximal, the
values coincide. It can further be noted that none of the
observables of the turning point tip occupancy increases
for larger amplitudes, i.e.
system lengths (for a fixed
growth rate). These observations further support our in-
tuition that the length grows until a critical occupancy
density at the tip (depending solely on the growth rate)
has been reached, triggering the switch to the shrinking
phase. As shown in the time trace plot, Fig. 5, in section
III, the total occupancy density follows a sawtooth-like
trajectory. This is due to a constant influx from the
reservoir during growth phase and most of the shrinkage
phase. Only at the end of the shrinkage phase the cluster
17
Section C 3, and the minimal length in Section C 4.
1. Effective in-rate αeff
In this subsection, we comment on how we determine
the effective rate at which particles enter the system and
then attach to the TL. This rate will not equal the "bare"
in-rate α as particles can also leave the system before at-
taching. What is the probability that a particle that
enters from the reservoir leaves the system again, be-
fore attaching to the TL? To answer this question, let
us consider a situation where the length of the system is
considerably larger than the length of a typical journey
of a particle on the DL before attaching to the TL, the
latter of which we estimate as λ ± λ (see later). Here,
λ =(cid:112)/ω is the characteristic length scale of the system.
In this case of large length, the probability that a given
particle that enters the DL leaves back into the reservoir
before attaching to the TL, Prob(leaving), is given by:
Prob(leaving) =
pqjAj ,
(C1)
j=0
where p = / ( + + ω) is the probability that a particle
exits from the first DL site back into the reservoir [72].
The quantity q = / ( + + ω) is the probability that a
particle proceeds to diffuse into the protrusion and A is
the probability that a particle that starts at the first site
of the DL returns to the first site of the DL without at-
taching to the TL in between. Since returning to the first
site of the DL without attaching to the TL in between
can only happen after an even number of steps on the
DL, this probability A comprises the probabilities that
the particle diffuses back to the first site of the DL in ex-
actly 2j steps, j ∈ N, without attaching to the TL. The
latter are given by the product of the probability that
a symmetric random walker returns back to its starting
point after exactly 2j steps and the probability that the
particle stays on the DL in each step. Taken together we
determine A as:
∞(cid:88)
FIG. 11. Occupancies of the TASEP lane tip neighborhood,
i.e. here the 20 last sites, at the time point or time interval
when the length reaches its maximum during one oscillation
period. They have been measured for several periods and are
shown for a growth rate of 0.14. The occupancy at these time
points (red squares) and their average (upper blue line) as well
as the occupancy average (yellow right-pointing triangles) and
maximum (gray left-pointing triangles) over a time period
of ten events - five before and five after the maximum - is
depicted. Moreover, the critical density for switching from
growth to shrinkage (lower purple line) is shown.
at the tip communicates with the reservoir and is quickly
emptied.
Appendix C: Detailed calculations for the effective
theory
In this section we elaborate on the mathematical de-
tails for the effective theory. First, in Section C 1 we
determine the effective in-rate from the reservoir onto
the TL. In order to estimate the density in the bulk, we
infer the length of the in-region, see Section C 2. Finally,
we deduce the recursion relation for the tip density in
A = Prob(return to site 1 w/o attaching to TL) =
Prob(return to site 1 in exactly 2j steps w/o attaching to TL) =
Prob(return to site 1 in exactly 2j steps not attaching to TL during the 2j steps)×
j=1
∞(cid:88)
∞(cid:88)
∞(cid:88)
j=1
j=1
=
=
=
× Prob(not attaching to TL during the 2j steps) =
(cid:18) 2
(cid:19)2j−1
f2j
2 + ω
(cid:32)
(cid:112)ω(4 + ω)
(cid:33)
2 + ω
=
2 + ω
2
1 −
where f2j = (cid:0)2j
j
(cid:1)/(cid:2)(2j − 1)22j(cid:3) is the probability that
a symmetric 1D random walker returns to its starting
point for the first time in exactly 2j steps. Note that
the probability that the particle does not attach during
the 2j steps only has 2j−1 terms 2/ (2 + ω) as the first
step into the protrusion is already accounted for by the
probability q in Prob(leaving), Eq. (C1). Combining the
result for A with the explicit formulas for p and q we find
Prob(leaving) = 1 − ω +(cid:112)ω(4 + ω)
ω +(cid:112)ω(4 + ω) + 2
.
We approximate this formula for small ω by Taylor ex-
panding up to first order in ω:
Prob(leaving) = 1 −
√
2 −
22 ω + O(ω3/2) .
+
ω
The effective in-rate is given by the "bare" in-rate α
weighted by the probability that a particle that enters
from the reservoir attaches to the TL. The latter prob-
ability is just 1− Prob(leaving). This implies that the
effective in-rate is given by
ω +(cid:112)ω(4 + ω)
ω +(cid:112)ω(4 + ω) + 2
αeff = α
√
ω
≈ α
(C2)
to lowest order in ω.
2. Length of the in-region and density in the bulk
√
To continue we now estimate the length of the "in-
region" lI since - due to the attachment and detach-
ment - the density at the end of the in-region depends
on the length. For this, we look at a symmetric ran-
dom walk with reflecting boundary at x = 0 as we
want to find out the typical journey of a particle on
the DL that is eventually attaching to the TL (and thus
not leaving the system again). Using the initial condi-
tion p(x, t = 0) = δ(x), the probability distribution of
such a process is given by p(x, t) = e−x2/(4t)/
πt for
x ≥ 0 where is the diffusion constant (lattice spacing
1). To determine the average distance a particle trav-
els on the DL before attaching to the TL, (cid:104)x(cid:105)attach, and
its standard deviation, σattach, we need to take two pro-
cesses into account. First, we need to find out how the
time at which the particle attaches to the TL is dis-
tributed and, second, how far a particle travels until a
certain time point. Using that the attachment process
is a Poisson process of rate ω, where the time until an
attachment event happens has the probability distribu-
tion f (t) = Prob(T = t) = ωe−ωt, we calculate the
0 dt f (t)(cid:104)x(t)(cid:105) and
σ2
(cid:104)x2(t)(cid:105)−(cid:104)x(t)(cid:105)2 are the mean and variance of the travelled
distance of the symmetric random walk until time t. For
mean and variance as (cid:104)x(cid:105)attach = (cid:82) ∞
0 dt f (t)(cid:0)(cid:104)x2(t)(cid:105)−(cid:104)x(t)(cid:105)2(cid:1). Here, (cid:104)x(t)(cid:105) and
attach =(cid:82) ∞
those quantities we find (cid:104)x(t)(cid:105) = 2(cid:112)t/π and (cid:104)x(t)2(cid:105) =
18
result, (cid:104)x(cid:105)attach =(cid:112)/ω = λ and σattach = λ = (cid:104)x(cid:105)attach.
2t from the above probability distribution p(x, t). As a
Therefore, (cid:104)x(cid:105)attach ± σattach = λ ± λ, which means that
the standard deviation is the same as the mean. We will
thus approximate the length of the in-region as twice the
average distance a particle travels before attaching to the
TL:
lI ≈ 2λ.
(C3)
With this relation, we now determine the density
profile in the in-region which we assume to equilibrate
quickly on the time scale of the oscillations. Let us de-
note by ρi the density at site i from the base. Then,
we approximate the time evolution of the density at site
i = 1, . . . , LI as
0 = ∂t ρi ≈ ρi−1 − ρi +
− ω ρi
αeff
LI
in the low-density limit with the boundary condition ρ0 ≈
0. Here, LI = Round(lI ) is the integer length of the in-
region within which we assume homogeneous attachment
(at rate αeff /LI per site). Performing a continuum limit
i → x, x ∈ [0, lI ], and considering only the first derivative
with respect to x, we have:
0 = −∂x ρ(x) +
− ω ρ(x)
αeff
lI
with the solution ρ(x) = αeff (1 − e−ωx)/(lI ω). So, in
particular, we obtain for the particle density at the end
of the in-region
(cid:16)
ω(cid:17)
,
ρ(lI ) =
α
2
√
1 − e−2
where we combined all the above results.
3. Recursion relation for the tip density
As mentioned in Section IV, for the analysis of the tip
neighborhood and the tip, we go to a different reference
frame, namely starting at the tip and reaching into the
tip neighborhood, co-moving with the tip. By ρ0 ≡ ρ+
we denote the density at the tip and by ρi the density
at the i-th site from the tip. Since we defined the tip
neighborhood to be the region where motors that have
previously detached at the tip reattach to the TL, we as-
sume that the tip neighborhood has the same length as
the in-region lT = lI = 2λ as for both the average dis-
tance before attaching to the TASEP becomes essential.
Note that we ignore the influence of the growth and
shrinkage dynamics on the average distance before at-
taching. This, however, should be legitimate in our pa-
rameter regime: Assume that we look at a symmetric
random walk on a lattice with one reflecting boundary
at the left end where also new particles are injected. At
each site, the particles can leave the system at rate ω, and
the reflecting boundary moves at rate v > 0 to the right
(or, in case of v < 0, at rate −v to the left). Then, in the
co-moving frame (moving with the reflecting boundary),
the steady-state profile is proportional to e−x/¯λ with the
v2 + 4ω(cid:1) which corresponds
length scale ¯λ = 2/(cid:0)v +
√
to the average travelled distance before leaving the sys-
tem via ω. In our case, the velocity is not constant but if
we assume that the velocity is homogeneously distributed
in [−γ, γ], we get an average length scale which is very
close to λ for our choice of parameters.
Let us now go back to the densities in the tip neigh-
borhood and right at the tip. By taking into account the
reattachment of motors that have detached at the tip
an (average) time 1/ω before, the growth and shrinkage
dynamics, and the usual hopping, we find for the time
evolution of the density ρi in the low-density limit:
∂tρi=ρi+1−ρi+
δρ+
before
lT
+γ(ρi−1 − ρi)+δρ+(ρi+1 − ρi),
(C4)
and for the tip density ρ0 = ρ+
19
homogeneously reattach to the TL in the tip neighbor-
hood. To proceed we now make the following ansatz: We
assume that for a tip density ρ+
before at time t we can de-
termine the tip density at time t(cid:48) = t + 1/ω + λ = t + ∆
by solving Eqs. (C4-C5) for ρ+ in the steady-state. The
idea behind this is that a particle that has detached at
the tip needs on average 1/ω to reattach to the TL, and
then has to walk on average λ sites to get back to the tip
(we measure time in units of ν ≡ 1, and length in units
of the lattice spacing a ≡ 1). Using the continuum ap-
proximation in Eq. (C4) and considering only zero- and
first-order terms, we find for the density in the tip neigh-
borhood
ρ(x) = ρ(lI ) +
δρ+
before
lT
1
1 + δρ+ − γ
(lT + 1 − x) ,
where we used the boundary condition ρ(lT + 1) = ρ(lI ).
As a result, the density at the site next to the tip is given
by
∂tρ+=ρ1−γρ+−δρ+(1−ρ1).
(C5)
ρ1 = ρ(1) = ρ(lI ) +
δρ+
before
1 + δρ+ − γ
.
(C6)
Note, however, that in the last equation for the tip den-
sity we take exclusion into account explicitly by assum-
ing that the occupancy at the tip is exactly 1 in case of a
shrinkage event (last term). If exclusion was lifted, there
could be more than one particle at the tip and several
particles would then be simultaneously released into the
cytosol in case of a shrinkage event.
For the time evolution of ρi, Eq. (C4), we assume that the
particles that have previously detached at the tip (at rate
δρ+
before),
before, corresponding to a previous tip density ρ+
Combining this with Eq. (C5) we solve for ρ1 and find an
equation for the tip density ρ+ in terms of the previous
tip density ρ+
before:
= (1 + δρ+)(cid:2)(1 − γ+δρ+)ρ(lI ) + δρ+
(δ + γ)ρ+(1 − γ + δρ+) =
(cid:3) .
before
Bearing in mind that the tip density should be positive,
this equation is solved by
(cid:20)
δ2ρ+
before − A +
(cid:113)(cid:0)δ2ρ+
before − A(cid:1)2
+ 2B(cid:0)δρ+
before + C(cid:1)(cid:21)
/B
ρ+ =
(cid:16)
√
1 − e−2
ω(cid:17)
where ρ(lI ) = α
/ (2) (see above) and
4. Minimal length
A = δ (1 − γ) + γ (1 − γ) − δ ρ(lI ) (2 − γ) ,
B = 2δ [δ(1 − ρ(lI )) + γ] ,
C = (1 − γ)ρ(lI ).
So, this equation relates the previous tip density ρ+
before
at time t to the tip density ρ+ at time t + ∆. Iterating
this procedure, we find a recursion relation for the tip
densities ρ+
it at times tit = it × ∆:
(cid:113)(cid:0)δ2ρ+
it−1−A(cid:1)2
+2B(cid:0)δρ+
it−1+C(cid:1)(cid:21)
(cid:20)
ρ+
it =
δ2ρ+
it−1−A+
/B.
(C7)
So far, we have considered the situation where the
length of the system is much longer than the average
distance a particle typically travels on the DL. However
- if the system is too small - the particles do not reattach
to the TL as they leave the system too quickly. As a re-
sult, most of the particles will have left the system before
the system is shrunk to zero, and the system will regrow
from a minimal length larger than zero. To estimate this
minimal length, let us consider a 1D system of length l
with injection of particles at rate r at site l, symmetric
diffusion at rate within the system, outflux of particles
at rate ω everywhere, and an additional outflux of par-
ticles at rate at site 0. In the steady-state and with a
continuum approximation we thus have
Those equations are solved by
xp(x, t) − ωp(x, t) ,
0 = ∂tp(x, t) = ∂2
0 = ∂tp(0, t) = ∂xp(0, t) − (ω + )p(0, t) ,
0 = ∂tp(l, t) = −∂xp(l, t) − ωp(l, t) + r .
20
p(x) =
(l−x)
λ
re
ω
ϕλ
1 + e 2x
λ
λ (2ϕ + λ2)
1 + e 2l
λ
(cid:16)
(cid:16)
+(cid:0)ϕ + λ2(cid:1)(cid:16)−1 + e 2x
(cid:17)
(cid:17)
+ (ϕ + (ϕ + 1) λ2)
(cid:17)
(cid:16)−1 + e 2l
λ
λ
(cid:17) ,
where ϕ = / is the ratio between the diffusion and the
exit rate. So, we determine the (steady-state) probability
that a particle that enters the system at site l (the tip)
exits it via the rate (back into the reservoir) and not via
ω (attaching to the TL) as pexit = p(0)/r, which yields
pexit =
λ (2ϕ+λ2)
(cid:16)
1+e 2l
λ
(cid:17)
2λ3e l
λ
+ (ϕ+ (ϕ+1) λ2)
(cid:16)
(cid:17) .
λ −1
e 2l
As a result, the reattachment probability for a particle
detaching at the tip at length l is approximated as
preattach(l) ≈ 1 − F e− l
λ
for l (cid:29) λ. Here, F = 2λ3/(cid:0)ϕ + 2ϕλ + (ϕ + 1) λ2 + λ3(cid:1).
This means that the probability that a particle has not
yet left the system after a series of lengths {lit}it=1,...,n
is given by
(C8)
psurvival ({lit}it=1,...,n) =
n(cid:89)
it=1
the "survival" probability until time t as
ln [psurvival (t)] ≈
≈
=
ln [preattach(l0 − v∆k)] ≈
dk ln [preattach(l0 − v∆k)] =
dt(cid:48) ln [preattach(l(t(cid:48)))] =
t
∆
k=0
∆(cid:88)
(cid:90) t
(cid:90) t
(cid:90) l0
(cid:90) ∞
0
1
∆
l(t)
0
l(t)
=
1
v∆
≈ 1
v∆
dl ln [preattach(l)] ≈
dl ln [preattach(l)] ,
(C9)
where we used the coordinate transformations t(cid:48) = k∆,
l(t(cid:48)) = l0 − vt(cid:48), and approximated the maximal length
of the system l0 by ∞ as for maximal length the reat-
tachment probability should be close to 1. Approximat-
ing the logarithm as ln [preattach(l)] ≈ −1 + preattach for
preattach ≥ 0.9 we thus find
ln [psurvival (t)] ≈ − 1
v∆
dl [1 − preattach(l)] .
(cid:90) ∞
l(t)
preattach(lit)
Finally, using Eq. (C8) for 1 − preattach(l) for l (cid:29) λ we
get
or, equivalently,
ln [psurvival ({lit}it=1,...,n)] =
n(cid:88)
it=1
ln [preattach(lit)] .
Assuming that the system shrinks at constant velocity v:
l(t) = l0 − vt, and taking into account that in our effec-
tive system each length is realized for time ∆ (during this
time a particle that has detached potentially reattaches
and walks back to the tip), we identify the length dy-
namics until time t with {l0, l0 − v∆, . . . , l0 − v∆(t/∆)}.
Approximating the sum as an integral, we then deduce
ln [psurvival (t)] ≈ − λF
v∆
e− l(t)
λ .
Defining the average minimal length as the length where
the probability that a particle that was in the system at
maximal length has left the system is just 0.5 we find
lmin ≈ λ ln [λF/ (v∆ ln(2))]. For the (constant) velocity
we make a very crude approximation, namely v ≈ γ/2,
and we have
(cid:20) 2λF
γ∆ ln(2)
(cid:21)
.
lmin ≈ λ ln
21
[1] R A Blythe and M R Evans, "Nonequilibrium steady
states of matrix-product form: a solver's guide," J. Phys.
A 40 (2007), 10.1088/1751-8113/40/46/R01.
tor molecules on the dynamics of treadmilling filaments,"
Phys. Rev. E 86, 051906 (2012).
[22] S Muhuri, "Scale-invariant density profiles of a dynami-
[2] F Spitzer, "Interaction of Markov processes," Adv. Math.
cally extending TASEP," EPL 101, 38001 (2013).
5, 246–290 (1970).
[3] C T MacDonald, J H Gibbs, and A C Pipkin, "Kinetics
of biopolymerization on nucleic acid templates," Biopoly-
mers 6, 1–25 (1968).
[4] T Chou, K Mallick, and R K P Zia, "Non-equilibrium
statistical mechanics: from a paradigmatic model to bi-
ological transport," Rep. Prog. Phys. 74, 116601 (2011).
[5] C Appert-Rolland, M Ebbinghaus, and L Santen, "Intra-
cellular transport driven by cytoskeletal motors: General
mechanisms and defects," Phys. Rep. 593, 1–59 (2015).
[6] R Lipowsky, S Klumpp, and T M Nieuwenhuizen, "Ran-
dom Walks of Cytoskeletal Motors in Open and Closed
Compartments," Phys. Rev. Lett. 87, 108101 (2001).
[7] A Parmeggiani, T Franosch, and E Frey, "Phase Coexis-
tence in Driven One-Dimensional Transport," Phys. Rev.
Lett. 90, 086601 (2003).
[8] A Parmeggiani, T Franosch, and E Frey, "Totally asym-
metric simple exclusion process with Langmuir kinetics,"
Phys. Rev. E 70, 046101 (2004).
[9] S Klumpp and R Lipowsky, "Traffic of Molecular Motors
through Tube-Like Compartments," J. Stat. Phys. 113,
233–268 (2003).
[10] C Leduc, K Padberg-Gehle, V Varga, D Helbing, S Diez,
and J Howard, "Molecular crowding creates traffic jams
of kinesin motors on microtubules," Proc. Natl. Acad.
Sci. 109, 6100–6105 (2012).
[11] R Subramanian, S-C Ti, L Tan, S A Darst, and T M
Kapoor, "Marking and Measuring Single Microtubules
by PRC1 and Kinesin-4," Cell 154, 377–390 (2013).
[12] M R Evans and K E P Sugden, "An exclusion process for
modelling fungal hyphal growth," Physica A Stat. Mech.
Appl. 384, 53–58 (2007).
[13] S A Nowak, P W Fok, and T Chou, "Dynamic bound-
aries in asymmetric exclusion processes," Phys. Rev. E
76, 1–13 (2007).
[14] K E P Sugden, M R Evans, W C K Poon, and N D Read,
"Model of hyphal tip growth involving microtubule-based
transport," Phys. Rev. E 75, 1–5 (2007).
[15] K E P Sugden and M R Evans, "A dynamically extending
exclusion process," J. Stat. Mech. Theory Exp. 2007,
P11013 (2007).
[16] L E Hough, A Schwabe, M A Glaser, J R McIntosh,
and M D Betterton, "Microtubule Depolymerization by
the Kinesin-8 Motor Kip3p: A Mathematical Model,"
Biophys. J. 96, 3050–3064 (2009).
[17] M Schmitt and H Stark, "Modelling bacterial flagellar
growth," EPL 96, 28001 (2011).
[18] L Reese, A Melbinger, and E Frey, "Crowding of molec-
ular motors determines microtubule depolymerization,"
Biophys. J. 101, 2190–2200 (2011).
[19] A Melbinger, L Reese, and E Frey, "Microtubule length
regulation by molecular motors," Phys. Rev. Lett. 108,
1–5 (2012).
[20] D Johann, C Erlenkamper, and K Kruse, "Length regu-
lation of active biopolymers by molecular motors," Phys.
Rev. Lett. 108, 1–5 (2012).
[23] H-S Kuan and M D Betterton, "Biophysics of filament
length regulation by molecular motors," Phys. Biol. 10,
036004 (2013).
[24] L Reese, A Melbinger, and E Frey, "Molecular Mech-
anisms for Microtubule Length Regulation by Kinesin-8
and XMAP215 Proteins," Interface Focus 4, 1–21 (2014).
[25] J De Gier and C Finn, "Exclusion in a priority queue,"
J. Stat. Mech. Theory Exp. 2014 (2014), 10.1088/1742-
5468/2014/07/P07014.
[26] C Arita, A Luck, and L Santen, "Length regulation of
microtubules by molecular motors: Exact solution and
density profiles," J. Stat. Mech. Theory Exp. 2015, 1–14
(2015).
[27] C Schultens, A Schadschneider, and C Arita, "Effective
ergodicity breaking in an exclusion process with varying
system length," Physica A Stat. Mech. Appl. 433, 100–
106 (2015).
[28] M Sahoo, J Dong,
and S Klumpp, "Dynamic block-
age in an exclusion process," J. Phys. A 48 (2015),
10.1088/1751-8113/48/1/015007.
[29] V Varga, J Helenius, K Tanaka, A A Hyman, T U
Tanaka, and J Howard, "Yeast kinesin-8 depolymerizes
microtubules in a length-dependent manner," Nat. Cell
Biol. 8, 957–962 (2006).
[30] V Varga, C Leduc, V Bormuth, S Diez, and J Howard,
"Kinesin-8 Motors Act Cooperatively to Mediate Length-
Dependent Microtubule Depolymerization," Cell 138,
1174–1183 (2009).
[31] G Klein, K Kruse, G Cuniberti, and F Julicher, "Fila-
ment Depolymerization by Motor Molecules," Phys. Rev.
Lett. 94, 108102 (2005).
[32] J Helenius, G J Brouhard, Y Kalaidzidis, S Diez, and
J Howard, "The depolymerizing kinesin MCAK uses lat-
tice diffusion to rapidly target microtubule ends." Nature
441, 115–119 (2006).
[33] E Reithmann, L Reese, and E Frey, "Nonequilibrium
Diffusion and Capture Mechanism Ensures Tip Local-
ization of Regulating Proteins on Dynamic Filaments,"
Phys. Rev. Lett. 117, 078102 (2016).
[34] W F Marshall, "Cellular length control systems." Ann.
Rev. Cell Dev. Biol. 20, 677–693 (2004).
[35] L Mohapatra, B L Goode, P Jelenkovic, R Phillips, and
J Kondev, "Design Principles of Length Control of Cy-
toskeletal Structures," Annu. Rev. Biophys. 45, 85–116
(2016).
[36] I Neri, N Kern, and A Parmeggiani, "Totally asymmetric
simple exclusion process on networks," Phys. Rev. Lett.
107 (2011), 10.1103/PhysRevLett.107.068702.
[37] I Neri, N Kern,
and A Parmeggiani, "Modeling Cy-
toskeletal Traffic: An Interplay between Passive Diffu-
sion and Active Transport," Phys. Rev. Lett. 110 (2013),
10.1103/PhysRevLett.110.098102.
[38] I Neri, N Kern, and A Parmeggiani, "Exclusion pro-
cesses on networks as models for cytoskeletal trans-
port," New Journal of Physics 15 (2013), 10.1088/1367-
2630/15/8/085005.
[21] C Erlenkamper, D Johann, and K Kruse, "Impact of mo-
[39] D A Adams, B Schmittmann, and R K P Zia, "Far-
22
[57] J Howard and A A Hyman, "Growth, fluctuation and
switching at microtubule plus ends." Nat. Rev. Mol. Cell
Biol. 10, 569–574 (2009).
[58] J Krug, "Boundary-Induced Phase Transitions in Driven
Diffusive Systems," Phys. Rev. Lett. 67, 1882–1885
(1991).
[59] B Derrida, "An exactly soluble non-equilibrium system:
The asymmetric simple exclusion process," Phys. Rep.
301, 65–83 (1998).
[60] D T Gillespie, "A general method for numerically simu-
lating the stochastic time evolution of coupled chemical
reactions," J. Comput. Phys. 22, 403–434 (1976).
[61] For γ > 1 the system grows indefinitely as then trans-
port on the TL is too slow to keep up with the growth
dynamics.
[62] The value of corresponds to a diffusion constant that is
of the order of molecular motor diffusion in cytosol.
[63] S M Block, L S B Goldstein, and B J Schnapp, "Bead
movement by single kinesin molecules studied with opti-
cal tweezers," Nature 348, 348–352 (1990).
[64] Since we focus on the low-density limit for the TL and
since there is no exclusion on the DL the mean-field ap-
proximation should be valid.
[65] The critical density given by Eq. (2) increases linearly
with the growth rate, so for larger γ the shrinkage speed
should increase as well.
[66] J C Walter, J Dorignac, V Lorman, J Rech, J Y Bouet,
M Nollmann, J Palmeri, A Parmeggiani, and F Geniet,
"Surfing on Protein Waves: Proteophoresis as a Mech-
anism for Bacterial Genome Partitioning," Phys. Rev.
Lett. 119, 1–6 (2017).
[67] S Bergeler and E Frey, "Regulation of Pom cluster dy-
namics in Myxococcus xanthus," arXiv (2018).
[68] J Halatek and E Frey, "Rethinking pattern formation
(2018),
in reaction–diffusion systems," Nature Physics
10.1038/s41567-017-0040-5.
[69] A Vilfan and E Frey, "Oscillations in molecular mo-
tor assemblies," J. Phys. Condens. Matter 17 (2005),
10.1088/0953-8984/17/47/018.
[70] B Nov´ak and J J Tyson, "Design principles of biochemical
oscillators." Nat. Rev. Mol. Cell Biol. 9, 981–91 (2008).
[71] W H Press, Numerical recipes in C : the art of scientific
computing (Cambridge University Press, 1992) p. 994.
[72] We allow here that the rate at which a particle leaves into
the reservoir, , can be distinct from the usual diffusion
rate .
from-equilibrium transport with constrained resources,"
J. Stat. Mech. Theory Exp. 2008 (2008), 10.1088/1742-
5468/2008/06/P06009.
[40] L J Cook and R K P Zia, "Feedback and fluctuations in
a totally asymmetric simple exclusion process with finite
resources," J. Stat. Mech. Theory Exp. , P02012 (2009).
[41] L J Cook, R K P Zia, and B Schmittmann, "Competi-
tion between multiple totally asymmetric simple exclu-
sion processes for a finite pool of resources," Phys. Rev.
E 80, 1–12 (2009).
[42] C A Brackley, L Ciandrini, and M C Romano, "Multi-
ple phase transitions in a system of exclusion processes
with limited reservoirs of particles and fuel carriers," J.
Stat. Mech. Theory Exp. 2012 (2012), 10.1088/1742-
5468/2012/03/P03002.
[43] P Greulich, L Ciandrini, R J Allen, and M C Romano,
"Mixed population of competing totally asymmetric sim-
ple exclusion processes with a shared reservoir of parti-
cles," Phys. Rev. E 85, 011142 (2012).
[44] L Ciandrini, I Neri, J C Walter, O Dauloudet, and A
Parmeggiani, "Motor protein traffic regulation by supply-
demand balance of resources," Phys. Biol. 11 (2014),
10.1088/1478-3975/11/5/056006.
[45] M Rank, A Mitra, L Reese, S Diez, and E Frey, "Limited
resources induce bistability in microtubule length regu-
lation," Phys. Rev. Lett. (2018, in press).
[46] M J I Muller, S Klumpp, and R Lipowsky, "Molecu-
lar motor traffic in a half-open tube," J. Phys. Condens.
Matter 17 (2005), 10.1088/0953-8984/17/47/014.
[47] K Tsekouras and A B Kolomeisky, "Parallel coupling of
symmetric and asymmetric exclusion processes," J. Phys.
A 41 (2008), 10.1088/1751-8113/41/46/465001.
[48] J Tailleur, M R Evans, and Y Kafri, "Nonequilibrium
phase transitions in the extraction of membrane tubes by
molecular motors," Phys. Rev. Lett. 102, 1–4 (2009).
[49] M R Evans, Y Kafri, K E P Sugden, and J Tailleur,
"Phase diagrams of two-lane driven diffusive systems,"
J. Stat. Mech. Theory Exp. 2011 (2011), 10.1088/1742-
5468/2011/06/P06009.
[50] B Saha and S Mukherji, "Coupling driven exclusion
lanes: Boundary
and diffusion processes on parallel
induced phase transitions and boundary layers," J.
Stat. Mech. Theory Exp. 2013 (2013), 10.1088/1742-
5468/2013/09/P09004.
[51] I R Graf and E Frey, "Generic Transport Mechanisms
for Molecular Traffic in Cellular Protrusions," Phys. Rev.
Lett. 118, 128101 (2017).
[52] I Pinkoviezky and N S Gov, "Traffic jams and shocks of
molecular motors inside cellular protrusions," Phys. Rev.
E 89, 52703 (2014).
[53] M L Gupta, P Carvalho, D M Roof,
and D Pell-
man, "Plus end-specific depolymerase activity of Kip3,
a kinesin-8 protein, explains its role in positioning the
yeast mitotic spindle," Nat. Cell Biol. 8, 913–923 (2006).
[54] P M Grissom, T Fiedler, E L Grishchuk, D Nicastro, R R
West, and J R McIntosh, "Kinesin-8 from Fission Yeast:
A Heterodimeric, Plus-Enddirected Motor that Can Cou-
ple Microtubule Depolymerization to Cargo Movement,"
Mol. Biol. Cell 20, 963–972 (2009).
[55] X Su, R Ohi, and D Pellman, "Move in for the kill:
Motile microtubule regulators," Trends Cell Biol. 22,
567–575 (2012).
[56] T Mitchison and M Kirschner, "Dynamic instability of
microtubule growth," Nature 312, 237–242 (1984).
|
1608.01491 | 1 | 1608 | 2016-08-04T10:47:03 | Micelles Hydrodynamics | [
"physics.bio-ph",
"cond-mat.soft"
] | A micelle consists of monolayer of lipid molecules containing hydrophilic head and hydrophobic tail. These amphiphilic molecules in aqueous environment aggregate spontaneously into monomolecular layer held together due to hydrophobic effect by weak non-covalent forces. Micelles are flexible surfaces that show variety of shapes of different topology, but remarkably in mechanical equilibrium conditions they are spherical in shape. The shape and size of a micelle are functions of many variables such as lipid concentration, temperature, ionic strength, etc. Addressing the question, why the shape of micelles is sphere in mechanical equilibrium conditions, analytically proved to be a difficult problem. In the following paper we offer the shortest and elegant analytical proof of micelles spheroidal nature when they are thermodynamically equilibrated with solvent. The formalism presented in this paper can be readily extended to any homogenous surfaces, such are vesicles and membranes. | physics.bio-ph | physics | Micelles Hydrodynamics
David V. Svintradze*
*Department of Physics, Tbilisi State University, Chavchavadze Ave. 03, Tbilisi 0128, Georgia
ABSTRACT
tail. These amphiphilic molecules
A micelle consists of monolayer of lipid molecules containing hydrophilic head and
hydrophobic
in aqueous environment aggregate
spontaneously into monomolecular layer held together due to hydrophobic effect by weak non-
covalent forces. Micelles are flexible surfaces that show variety of shapes of different topology,
but remarkably in mechanical equilibrium conditions they are spherical in shape. The shape
and size of a micelle are functions of many variables such as lipid concentration, temperature,
ionic strength, etc. Addressing the question-"why the shape of micelles is sphere in mechanical
equilibrium conditions" analytically proved to be a difficult problem. In the following paper
we offer the shortest and elegant analytical proof of micelles spheroidal nature when they are
thermodynamically equilibrated with solvent. The formalism presented in this paper can be
readily extended to any homogenous surfaces, such are vesicles and membranes.
Key words: Micelle dynamics, Membrane dynamics, Equations for moving surfaces
1
1 INTRODUCTION
A micelle is an aggregate of lipid molecules dispersed in a liquid colloid. A typical micelle in
aqueous solution forms an aggregate so that the hydrophilic head regions are in contact with
the surrounding solvent, while the hydrophobic tail regions are pointed toward the aggregate
center. The aggregation is caused by the hydrophobic and hydrophilic interactions of lipids
with the surrounding water molecules [1]. In mechanical equilibrium conditions micelles are
spherical in shape. The shape and size of a micelle are a function of the molecular geometry of
its surfactant molecules and solution conditions, such as surfactant concentration, temperature,
pH, and ionic strength. In order to answer the question-"why the shape of micelles is sphere in
mechanical equilibrium conditions", it is necessary to consider the motion of micelles in fluid
induced by hydrophobic and hydrophilic interactions. Therefore, the problem is to find an exact
equation for micelle surface considered as fluid lipid membrane and to solve it analytically for
equilibrium case.
Among the remarkable aspects of fluid lipid membranes deduced from the large body of
theoretical works (see reviews [2-4]), is that the physical behavior of a membrane on the length
scale not much bigger than its own thickness, can be described with high accuracy by a purely
geometric Hamiltonian [5-7]. Unfortunately, this insight about curvature elastic models of
membrane surfaces come along with a challenging math and have not been analytically solved.
Associated Euler-Lagrange equations [8, 9], so called "shape equations", turned to be fourth
order partial nonlinear differential equations, and finding a general analytical solution is a
difficult problem, even though it has been numerically solved for some specific [10-17] and
general cases [18, 19].
In this paper we employ different approach to the problem, namely, we use tensor calculus
of moving surfaces and first law of thermodynamics to deduce the final equation for micelle
dynamics and to solve it analytically for the equilibrium case.
In fluid dynamics material particles can be treated as a vertex of geometric figure and
virtual layers as surfaces, and equations of motion for such surfaces can be searched. The
surfaces shall be called differentially variational surfaces (DVS). We propose equations of
motion of moving surfaces in aqueous solutions and apply it to analyze micelles morphology
in fluid dynamics [20].
Hydrophobic and hydrophilic interactions incorporates dispersive interactions throughout
the molecules, mainly related to electrostatics and electrodynamics (Van der Waals forces),
induced by permanent (water molecules) or induced dipoles (dipole-dipole interactions) and
possible quadrupole-quadrupole interactions (for instance stacking or London forces) plus
ionic interactions (Coulomb forces). The hydrophobic effect can be considered as synonymous
of dispersive interactivity with water molecules and the hydrophilic one as synonymous of
polar interactivity with water molecules [21, 22]. All these interactions have one common
feature and can be unified as electro-magnetic interaction's dependence on interacting bodies'
geometries, where by geometries we mean shape of the objects' surfaces [20, 23]. Analytical
solution of simplified DVS equations displayed all possible shapes of micelles spanning
spheroids, lamellas, and cylinders. The equations can be applied to the problems related to cell
motility and growth factors and show that in the mechanical equilibrium conditions with the
solvent, for homogenous surfaces, a trace of the mixed curvature tensor is a pressure across the
surface divided by membrane tension [20, 24]. Fully non-restrained, relativistic and exact
equations for moving surfaces in electromagnetic field, when the interaction with an ambient
environment is ignored, reads
2
(1)
where
is the surface mass density,
are coordinate and tangential components of
surface velocity,
is interface velocity,
for Minkowski four-dimensional space-
time ambient space,
for pseudo-Riemannian manifold (surface),
is the surface
curvature tensor,
is electromagnetic tensor,
,
is
component of
four current,
are normal and tangential components of
,
are the
normal and tangential components of the partial time derivative of the four vector potential
,
stand for surface and space integrals respectively. We don't reproduce
derivation of this set in this paper, rather just mention that first one is the consequence of mass
conservation, second and third equations come from minimum action principle of a Lagrangian
[20] and imply motion in normal direction (second equation from the set) and in tangent
direction (third equation). In case of non-relativistic motion Minkowskian space becomes
Euclidian, so that
.
and the surface is two-dimensional Riemannian manifold
In non-relativistic frame work, after modeling the potential energy as a negative volume
integral of the internal pressure and inclusion interaction with an environment, (1) further
simplifies as
(2)
where
are internal hydrodynamic and osmotic pressures, respectively (derivation of (2)
is provided in the appendix). It is noteworthy that from the set (2) only second equation differs
from the dynamic fluid film equations [25, 26]
(3)
is surface tension. (3) is only valid when the surface can be described with Laplace model
of surface tension [25, 26] meaning that the surface is homogeneous and the surface tension is
constant, while (2) does not have that restriction. Using (3) in (2) and taking into account that
in equilibrium processes internal pressure is the same as external pressure, one gets exactly the
same equation of motion in normal direction as we have in this paper.
(4)
Ideally, to address the question analytically: why the shape of micelle is sphere in
thermodynamic equilibrium condition with aqueous solution, it is necessary to be derived full
3
012()iiiiiijiijijiijiiijjiSVCBVCVCBVVFFAJVfaVVVVCCCVBdSfad,iVVC0,1,2,30,1,2iijBFFJFJJJ,iffF=F,iaaAA,S1,2,31,2i2()()0iiiiiijiijijiijiijjVCBVCVCBVVPVPVVVVVCCCVB,P(2)iijiiijiCVCBVVB(())()iiVBPVPV governing equations of motions of surfaces in solutions by including potential energy of
hydrophobic-hydrophilic interactions. As it is stated above, we have already reported such
equations (1) [20] and the solution of the simplified DVS equations (2) produce exactly same
outcome as we have in this paper, but (1, 2) are analytically much complex to digest even
though include all the information about analytical face of potential energy of hydrophobic-
hydrophilic interactions and reveal hidden geometries in potential energy, as far as right hand
side of second and third equations in (1) is proportional to gradient of potential energy, while
left hand side caries full information about the geometric motion because it includes curvature
tensor. It should be stated that (1) is fully non-restrained. It is the exact equations of motion of
surfaces in electromagnetic field and upon addition an environmental interactions explain not
only membrane dynamics, but also dynamics of macromolecular surfaces.
In present paper, we provide much efficient and shorter alternative way of deduction of
final equation for micelles' normal motion (4) and prove it in several lines without invoking
analytical face of hydration forces (1). Instead we sacrifice geometric picture of the surface full
motion and geometric description of hydrophobic-hydrophilic interactions.
2. 1 Basics of Differential Geometry
2 THEORY
In this section we provide basics of tensor calculus for moving surfaces and summarize the
theorems we used directly or indirectly to deduce equations for micelle dynamics. Differential
geometry preliminaries we used here can be found in tensor calculus text book [26].
Suppose that
(
) are the surface coordinates of the moving manifold (or the
surface)
and the ambient Euclidean space is referred to coordinates
(Figure 1).
Coordinates
are arbitrarily chosen so that sufficient differentiability is achieved in both,
space and time. Surface equation in ambient coordinates can be written as
. Let
the
position vector be expressed in coordinates as
(5)
Latin letters in indices indicate surface related tensors. Greek letters in indices show tensors
related to Euclidean ambient space. All equations are fully tensorial and follow the Einstein
summation convention. Covariant bases for the ambient space are introduced as
,
where
. The covariant metric tensor is the dot product of covariant bases
(6)
The contravariant metric tensor is defined as the matrix inverse of the covariant metric tensor,
so that
, where
is the Kronecker delta. As far as the ambient space is set to
be Euclidian, the covariant bases are linearly independent, so that the square root of the metric
tensor determinant is unit. Furthermore, the Christoffel symbols given by
vanish and set the equality between partial and curvilinear derivatives
.
4
iS1,2iSX,iSX(,)iXXtSR()(,)iRRXRtSXR/XXXXXXXX Now let's discuss tensors on the embedded surface with arbitrary coordinates
. Latin
indexes throughout the text are used exclusively for curved surfaces and curvilinear derivative
is no longer the same as the partial derivative
. Similar to the bases of ambient
space, covariant bases of an embedded manifold are defined as
and the covariant
surface metric tensor is the dot product of the covariant surface bases:
(7)
The contravariant metric tensor is the matrix inverse of the covariant one. The matrix inverse
nature of covariant-contravariant metrics gives possibilities to raise and lower indices of
tensors defined on the manifold. The surface Christoffel symbols are given by
and along with Christoffel symbols of the ambient space provide all the necessary tools for
covariant derivatives to be defined as tensors with mixed space/surface indexes:
(8)
where
is the shift tensor which reciprocally shifts space bases to surface bases, as
well
as
space metric
to
surface metric;
for
instance,
and
.
Using (7, 8), one may directly prove metrilinic property of the surface metric tensor
, from where directly follows
, meaning that
are orthogonal
vectors and as so
must be parallel to the surface normal
(9)
is a surface normal vector with unit length and
is the tensorial coefficient of the
relationship and is generally referred as the symmetric curvature tensor. The trace of the
curvature tensor with upper and lower indices is the mean curvature and its determinant is the
Gaussian curvature. It is well known that a surface with constant Gaussian curvature is a sphere,
consequently a sphere can be expressed as:
(10)
where
is some non-zero constant. When
then the surface is either plane or cylinder.
According to (9, 10), finding the curvature tensor defines the way of finding covariant
derivatives of surface base vectors and as so, defines the way of finding surface base vectors
which indirectly leads to the identification of the surface.
2. 2 Basics of Tensor Calculus for Moving Surfaces
All Equations provided above are generally true for moving surfaces. We now turn to a brief
review of definitions of coordinate velocity
, interface velocity
(which is the same as
normal velocity), tangent velocity
(Figure 1), time
-derivatives of surface tensors and
5
iSi/iiSiiSRijijSSSiijkjkSSjjjjjmmjikikikikimkikmTTXTXTTTiiXXiiSXXijijijijSSSXXXXXXX0imnS0minSSminSSinSNijijSNBNijBiiB0iiBVCiV time differentiation of the surface integrals. The original definitions of time derivatives for
moving surfaces were given in [27] and recently extended in [26].
Let's start from the definition of coordinate velocity
and show that the coordinate
velocity is
component of surface velocity. Indeed,
(11)
On the other hand the position vector
.
Taking into account partial time differential of (5) and definition of ambient base vectors, we
find
given by (5) is tracking the coordinate particle
(12)
Therefore,
normal component of the surface velocity is dot product with the surface normal, so that
is ambient component of the surface velocity
. Taking into account (12),
(13)
It is easy to show that the normal component
of the coordinate velocity, generally referred
as interface velocity, is invariant in contrast with coordinate velocity
and its sign depends
on a choice of the normal. The projection of the surface velocity on the tangent space (Figure
1) is tangential velocity and can be expressed as
Taking (13, 14) into account one may write surface velocity as
(14)
.
Graphical illustrations of coordinate velocity
, interface velocity
and tangential velocity
are given on Figure 1. Also there is a clear geometric interpretation of the interface velocity
, correspondingly.
[26]. Let the surfaces at two nearby moments of time
and
be
Suppose that
(point on
) and the corresponding point
,
has the same
surface coordinates as
(Figure 2.), then
. Let
be the point, where the unit
normal
intersects the surface
, then for small enough
, the angle
, therefore,
can be defined as
and
(15)
and can be interpreted as the instantaneous velocity of the interface in the normal direction. It
is worth of mentioning that the sign of the interface velocity depends on the choice of the
normal. Although
is a scalar, it is called interface velocity because the normal direction is
implied.
6
VXVtRiS(,)(,)iiRtSRXtSVVXtXtVVCVNVXNXVNVNCViiVVX,iiiVCVCNVSVCiVttt,tttSStAStSttBSBAABVtPtNSttSt/2APBAPVNtC0limtAPCtC 2. 3 Invariant Time Differentiation
Among the key definitions in calculus for moving surfaces, perhaps one of the most important
one is the invariant time derivative
. As we have already stated above, initial preposition for
time derivative was made in [27] and extended in [26]. In this paragraph, we just give
geometrically intuitive definition (similar to [26]).
Suppose that invariant field
is defined on the surface at all time. The idea behind the
in the normal direction. Physical
invariant time derivative is to capture the rate of change of
explanation of why the deformations along the normal direction are so important, we give
measures the rate of deformation in the
below in integration section. This is similar to how
normal direction. Let for a given point
the intersection
, find the points
and
of
and the straight line orthogonal to
(Figure 2). Then the geometrically intuitive
definition dictates that
(16)
As far as (16) is entirely geometric, it must be an invariant (free from choice of a reference
frame). From the geometric construction one can estimate value of
in point
, so that
(17)
On the other hand,
is related to
because
and are nearby points on the
surface
, then
(18)
since
shows rate of change in
along the surface and
captures the directed distance
. Determining
values from (17, 18) and putting it in (16), gives
(19)
Extension of the definition (19) to any arbitrary tensors with mixed space and surface indices
is given by the formula
(20)
The derivative commutes with contraction, satisfies sum, product and chain rules, is metrinilic
with respect to the ambient metrics and doesn't commute with the surface derivative [26]. Also
from (16) it is clear that the invariant time derivative applied to time independent scalar
vanishes. Christoffel symbol
for moving surfaces is defined by
.
7
FFCtASttBSPttStS0()()limtFPFAFtFB()()FFBFAtt()FB()FP,ttBPSttS()()iiFBFPtVFiFFitVBP(),()FAFPiiFFVFtijikiiiikkijkjjjkjjkTTVTVTVTTTtijiiijjjVCB 2. 4 Time Differentiation of Integrals
The remarkable usefulness of the calculus of moving surfaces becomes evident from two
fundamental formulas for integrations that govern the rates of change of volume and surface
integrals due to the deformation of the domain [26]. For instance, in evaluation of the least
action principle of the Lagrangian there is a central role for time differentiation of the surface
and space integrals, from which the geometry dependence is rigorously clarified.
For any scalar field
defined on a Euclidean domain
with boundary
evolving with the interface velocity
for closed surface are given by the formulas
, the evolution of the space integral and surface integral
(21)
The first term in the integral represents the rate of change of the tensor field, while the second
term shows changes in the geometry. Of course there are rigorous mathematical proofs of these
formulas in the tensor calculus text books. We are not going to reproduce proof of those
theorems here, but instead we give less rigorous but completely intuitive explanation of why
only interface velocity has to be count. Rigorous mathematical proof follows from fundamental
theorem of calculus
(22)
In the case of volume integral or surface integral it can be shown that
is replaced by
. Intuitive explanation is pretty simple. Propose there is no interface
interface velocity
velocity then surface velocity only has tangent component. Tangent velocity for each given
time (if there is no interface velocity) translates each point to its neighboring point and,
therefore, does not add new area to the surface (or new volume to the space, or new length to
the curve). As so, tangential velocity just induces rotational movement (or uniform translational
motion) of the object and can be excluded from additive terms in the integration. Perhaps, it is
easier to understand this statement for one dimensional motion. Let's assume that material
point is moving along some trajectory (some curve), then, in each point, the velocity of the
material point is tangential to the curve. Now one can translate this motion into the motion of
the curve where the curve has only tangential velocity. In this aspect, the embedded curve only
slides in the ambient plane (uniformly translates in the plane) without changing the length
locally, therefore tangential velocity of the curve does not add new length to the curve.
3 RESULTS AND DISCUSSIONS
In this section we apply basics of thermodynamics and fundamental theorems of calculus of
moving surfaces to demonstrate shortest proof of (4), describing motion of homogenous surface
(micelle) at normal direction. We consider the system consisted of aqueous media with the
formed micelle in it (Figure 3). The system is isolated with constant temperature and there is
8
(,)FFtSSCSSSSdFFddCFdSdttdFdSFdSCFdSdt()()(,)(,)()(,())btbtaadFtxFtxdxdxb'tFtbtdtt'()btC no absorbed or dissipated heat on the surface of the micelle; in other words, a process is
adiabatic. We don't ask the question of how micelle forms, instead we ask why the shape of
the micelle is sphere when it is thermodynamically equilibrated with the system. Strictly
talking, such micelle is a subsystem of the isolated system and the surface of the micelle is
closed. According to the first law of thermodynamic, as far as there is no dissipated or absorbed
heat, the change of the internal energy of the surface of the micelle must be
(23)
is infinitesimal work done on the subsystem (micelle) and
where
is infinitesimal
change of the internal energy. Because the temperature of the system is constant, the
differential of the subsystems' internal energy can be remodeled as
(24)
where
done on the subsystem is
is the total potential energy of the micelle. By the definition the elementary work
(25)
where
are external hydrodynamic and osmotic pressures applied on the surface of the
is the volume of the micelle with boundary
surface area. Now let's propose that the surface of the micelle is homogenous and can be
micelle by the surroundings correspondingly and
of
described by Laplace model of surface tension, then
(26)
is surface tension. As far as we discuss simplest case of the system consisted of aqueous
medium and single micelle, we can suggest that the surface tension is not time variable. Using
(23-26) after few lines of algebra
(27)
Assuming the surface of the micelle is moving so that (27) stays valid for any time variations,
then time differentiation of the left side must be equal to time differentiation of right integral.
As far as on the right hand side we have space integral, time differentiation can be taken into
the integral (using (21)), so that integration theorem for space integral holds and the convective
and advective terms due to volume motion are considered
(28)
To calculate time derivative of the surface integral we have to take into account the theorem
about time differentiation of the surface integral (21), from which follows that for time
invariable surface tension
(29)
9
dEWWdEdEdUU()WPd,PSdUdS()SdSPd()()()SdXPdPdCPdSdttiiSSddSCBdSdt Where
is interface velocity,
is
component of the surface normal and
is coordinate velocity,
is general coordinate and
is the trace of the mixed
curvature tensor generally known as mean curvature. After few lines of algebra putting (27-29)
together
(30)
Generalized Gauss theorem converts the surface integral of the left hand side of (30) into space
integral, so that
(31)
Combination of (30) and (31) immediately gives equation of motion for the micelle surface
(32)
, so that (32)
For equilibrium processes internal and external pressures are identical
becomes identical to (4). Also, we should note that (32) is only valid for motion of the
homogeneous surfaces at normal direction, therefore, it doesn't display any deformation in
tangent directions. (32) further simplifies when the micelle comes in equilibrium with the
solvent where divergence of the surface velocity
(stationary shape) along with
(where
becomes
) vanishes then the solution to (32), taking into account the condition (28),
(33)
Incidentally, stationary solution
to (32) dictates
corresponding to cylindrical
and lamellar surfaces. The result (33) shows that the solution is surfaces which have constant
mean curvatures (CMC). Such surfaces are rare and can be many if one relaxes the condition
we restricted the system. Namely we consider isolated system where micelle is closed sub-
system, these two preconditions mathematically mean that the micelle surface we discuss is
compact embedded surface in
. According to A. D. Alexandrov uniqueness theorem for
surfaces, a compact embedded surface in
with constant non-zero mean curvature is a sphere
(A. D. Alexandrov Amer. Math. Soc. Trans. 21, 412 (1958)). Correspondingly the solution (33)
is a sphere (as far as we have compact two-manifold in the Euclidian space). Therefore, when
(34)
the micelle is spheroid and becomes lamella or cylinder when the pressure along the surface
over the surface tension vanishes. This surprisingly simple and elegant derivation explains all
the shapes of micelles in aqueous solution in equilibrium conditions. Furthermore if the
compactness condition is relaxed then (33) predicts that all other CMC surfaces are also
possible. If one takes into account that the surface tension in general can be a function of many
variables, such as Gaussian curvature, bending rigidity, spontaneous curvature, lipid
10
CVNN/VXtXiiB()()iiSCBCPdSPVd()(())iiiiSNVBPdSVBPVd(())()iiVBPVPVPPV/PtPPiiPB0V0iiB330P concentration and etc., than (32) may predict possible deformations of differently shaped
micelles and their wide range of static shapes. In fact, if considered that surface tension, which
is defined as potential energy per unit area, can be a function of mean curvature
,
then Taylor expansion of
naturally rises terms related to Gaussian curvature,
spontaneous curvature, bending rigidity etc. Of course all these generalizations along with
taking into account temperature fluctuations can be included in the equations, which we won't
be doing in this paper, because unfavorable complication of already complicated equations
should be avoided and it should be a source for another paper.
Based on (33) we can calculate minimal value of a micelle radius. The value of the trace
of the mixed curvature tensor for a sphere is
(35)
is radius of the micelle. Let's calculate value of the surface pressure when the micelle still
can exist. Lipids in a micelle are confined in the surface by hydrophobic interactions with
average energy in the range of hydrogen bonding. As far as values of hydrogen bonding energy
are somewhat uncertain in the literature, by the first approximation we take average energy for
the hydrogen bonding energy interval and assign it to the lipid molecule. Low boundary of the
interval (minimum energy) for
unit)
unit), the low and high values are taken
and high boundary is about 161 kJ per mol (
about
according
hydrogen bond is about 1 kJ per mol (
29]. Therefore,
energy
is
average
to
references
[28,
. To estimate hydrogen bonding energy per molecule
with the undefined shape (lipid molecule) in the first approximation is to assign average energy
to it and consider the spherical shape with the gyration radius. Of course it is low level
approximation, but even such rough calculations produces reasonable results. After all these
rough estimations the pressure to move one lipid from the surface, in order to induce critical
deformations of the surface, is about average energy per the average volume of the lipid
molecule
(36)
where
is the estimated volume of a lipid molecule considered as sphere with the
gyration radius
. On the other hand, surface tension of a fluid monolayer at optimal
packing of the lipids is about
estimated micelle radius is
[30-32], using these and (35, 36) in (33) the
(37)
These calculations put the minimum radius of the micelle in nanometer scale and is in very
good agreement with experimental as well as computational frameworks [33, 34]. To further
validate the (37) result, we ran a CHARMM based Micelle Builder simulation [35, 36] for 100
phospholipid molecules
(1,2-Dimyristoyl-sn-Glycero-3-Phosphocholine, DHPC). The
Å. These
simulation result (Figure 4) generated a spherical micelle with diameter
calculations indeed indicate that even such rough estimations produce reasonable accuracy.
11
()iiB()iiB2iiBRRXHYCHCFHF20(1161)/281/1310kJmolJ2037231310/43.110/GPrNm34/3Gr1Grnm2310/Nm2072310(19.30.1)A3.110R(38.50.1) Of course, the first approximation is low level. To get more convincing estimations it is
necessary to take into account that neither lipids are spherical nor hydrophobic interactions per
lipid are average energy of single hydrogen bond. The simulation results discussed above (so
as calculations) were done on the first approximation, where lipids were considered as spherical
and hydrophobic energy per lipid was estimated as the single hydrogen bonding energy. To
produce higher level approximation and the comparison with the theory, we demonstrate all
atom simulation and estimation of the radius in the second approximation, where lipids are no
longer undefined spheres, but have well defined surfactant geometry and hydrophobic energy
unit. In
is no longer average energy of single hydrogen bond, but is 1 kJ per mol per
simulations we used DHPC lipid molecule having 12
units (Figure 5) per hydrophobic
tail, so hydrophobic energy is about
. Accurate calculation of the lipid
molecule volume using cavity, channel and cleft volume calculator [37], gives the volume
estimation of about 894 Å3. Using this value, one gets
(38)
On the other hand, using the same surface tension of a fluid monolayer at the optimal packing
of the lipids, one gets
(39)
All atom simulation also generates spherical structure with diameter
(Figure 5),
although there is still some uncertainty in this estimation because we assigned 1 kJ/mol energy
unit and we based on references [28, 29] data, while, for instance, in [38] it is
per
mentioned that the hydrophobic interactions are about 4 kJ/mol per
unit. In our
opinion, this discrepancy can be resolved if one calculates hydrophobic energy based on the
potential energy
(40)
where
is electric field per
,
is dielectric constant in the vacuum and
stands
for the volume of the lipid molecules. (40) directly emerges from
term written in (1).
In fact, for electrostatics
(41)
and one should go to the scrutiny of calculating electric field for each
units, then take
a sum of the electric field and square it (we are not going to do it in this paper). Also, one may
ask why the hydrophilic interaction energy is not taken into account in these calculations.
Hydrophilic head of the lipid molecule is in contact with water molecules and does not need
any work to be done to drag it in aqueous solution from the lipids layer. Therefore, hydrophilic
12
2CH2CH2012/1.9910kJmolJ2072271.99102.2210/0.89410PNm2072310(270.1)A2.2210R0(540.1)A2CH2CH2220()2CHCHUEd2CHE2CH0FF20012UFFdEd2CH interaction energy can be neglected. The most of the work goes on overcoming hydrophobic
interactions between lipid tails.
4 CONCLUSIONS
To summarize, we have presented a framework for the analysis of micelle dynamics using first
law of thermodynamics and calculus of moving surfaces. Final equations of normal motion (4,
32) are based on the assumption that the micelle surface is homogeneous and is restricted by
precondition to the surfaces, which can be described by time invariable surface tension.
However, (1, 2) don't have homogeneity constrain and indicate motion along normal
deformation, as well as deformation into tangent directions, but are analytically more complex.
The solution to the normal equations of motion in equilibrium conditions turned to be
surprisingly simple and displayed all possible equilibrium shapes of micelles. Micelles are
spheroids and become lamellas or cylinders when the pressure along the surface over the
surface tension vanishes. The proposed formalism was illustrated by applying it to the
estimation of micelle optimal radius and comparison to all atom simulations. The remarkable
accuracy was found even for low level approximations between theoretically calculated radius
and the one obtained from the atomistic simulations. As a final remark, the proposed theory
can be readily extended to any homogenous surfaces, such are vesicles and membranes.
ACKNOWLEDGMENTS
We were partially supported by the personal savings accumulated during visits to the
Department of Mechanical Engineering, Department of Chemical Engineering, OCMB Philips
Institute and Institute for Structural Biology and Drug Discovery of Virginia Commonwealth
University in 2007-2012 years. Limited access to Virginia Commonwealth University's library
in 2012-2013 years is also gratefully acknowledged.
APPENDIX
Here, we derive exact equations (2) for moving surfaces, where in "exact" we mean that there
have not been done any approximations while evaluating kinetic and potential energies of the
surface. Beforehand, we should mention that (2) also follows from (1) if one applies same
formalism as it is given in (23-25). Indeed, in non-relativistic framework space is three-
dimensional Euclidian, the surface is two-dimensional Riemannian and potential energy
becomes
(42)
Where
are electric and magnetic fields and
are charge density, electric potential,
magnetic vector potential and current density vector respectively. Using (23-25) one gets
,
and taking into account that the
13
22000111()(())2UFFAJdEBqAJd,EB,,,qAJ()dUPd0()1/PFFAJ pressure comes from the "normal force" applied to the surface
, and in
tangent direction
, then (1) becomes (2). It might be more helpful providing details
about (1), but, as far as it goes out of the scope of this paper, we are reluctant to do it here.
Now we turn to the derivation of (2) without using any information from (1). To deduce
the equations of motion we deduce the simplest one from the set (2) first. It is direct
consequence of generalization of conservation of mass low. Variation of the surface mass
density must be so that
, where
is surface mass with
surface density.
Since the surface is closed, at the boundary condition
, a pass integral along any
curve
across the surface must vanish (
is a normal of the curve and lays in the tangent
space). Using Gauss theorem, conservation of mass and integration formula (21), we find
(43)
Since last integral mast be identical to zero for any integrand, one immediately finds first
equation from the set (2). To deduce second and third equations, we take a Lagrangian
(44)
and set minimum action principle requesting that
. Evaluation of space integral is
simple and straightforward, using integration theorem for space integral where the convective
and advective terms due to volume motion is properly taken into account (21), we find
(45)
Derivation for kinetic part is a bit tricky and challenging that is why we do it last.
Straightforward, brute mathematical manipulations, using first equation from (2), lead
(46)
14
()PVfa0iifa/0dmdtSmdS0iivnVin0()(())(())(())iiiiSiiiiiiSiiiiSSSiiiiSvdnVdVdSVCBCBdSdVCBdSdSdSdtVCBdS2()2SVLdSPd/0Lt()()()SPdPVdCPdSt222222222222222()()222222(()))(())22222(())222(())222iiiiSSSiiiiiiiiSSiiiiSiiiiVVVVVVdSCBdSCBdStVVVVVCBVCBdSVdSVVVVVdSVVVVVVV2[())]2SiiiiSdSVVV(VVVdS At the end point of variations the surface reaches stationary point and therefore by Gauss
theorem (as we used it already in (43)), we find
(47)
is stationary contour of the surface and
is the normal to the contour, therefore interface
velocity for contour
and the integral (47) vanishes, correspondingly
(48)
To decompose dot product in the integral by normal and tangential components and, therefore,
deduce final equations, we do following algebraic manipulations
(49)
Now using Weingarten's formula
, metrinilic property of Euclidian space base
vectors
and definition of surface normal
last equation transforms
(50)
Taking into account
,
and
we have
(51)
Continuing algebraic manipulations using Thomas formula
, the formula for
surface derivative of interface velocity
and the definition of curvature tensor
yield
15
22()022iiiiSVVVdSVndin0iivnV2()2iiSSVdSVVVVdStiiijijiiijijiijijiijijVVV=VVVCVBSCVBSVVVCVBXXCVBSjjiiXBN0iXNNX)iijijiijijiiijiiijiiijiiijiiijiiijVVVCVBXXCVBSVVVCVXCVBSVVVCVXCVBSVVVCVNCVBSiiVCNVS()()jiiijVCNVS()()jjVCNVS()()()()()()iiijiiijiijiijiijiijiijijiijijjiijijjiijijVVVCVNCVBSVVCNVVSCVNCVBSVVNCVVSCVBSCNVSVNCVVSCVBSiiNCSiiNCSijijNBS
(52)
Doting (52) on
energy, so that we finally get
and combining it with (48) last derivation finally reveals variation of kinetic
(53)
Combining (43-45) and (53) together and taking into account that the pressure acts on the
surface along the surface normal, we immediately find first and the last equation of the (2). To
clarify second equation, we have
(54)
After applying Gauss theorem to the second equation the surface integral is converted to space
integral so that one gets
(55)
and, therefore, all three equations of (2) are rigorously clarified.
16
()()()()2()()()2()jiijijjiijijjiijjjiijjiijijjijiijijjijiijjiijjjijiCNVSVNCVVSCVBSCNCCSVNCVVBNVSVNCVVSVVBNCCSCVBSCNCNVNCVVBNVSVSV()(2)()jijjijjijjijiijjijjijiijiijVSVVSCCSCVBSCVCVVBNVVVCCCVBS+V2)2(2)()iiSSiijiijijiijiSijjVdSV(VVVdStCCVCVVBdSVVVVCCCVB+2()()2()()iijiijSSiijiijSCCVCBVVdSPVdCPdSCCVCBVVPdSPVd2()()iijiijVCVCBVVPVPV REFERENCES
[1] C. Tanford, The Hydrophobic Effect Formation of Micelles and Biological
Membranes (Wiley-Interscience, New York, 1973).
[2] M. Deserno, Chemistry and Physics of Lipids 185, 11 (2015).
[3] U. Seifert, and R. Lipowsky, Chapter 8: Morphology of Vesicles. In: Lipowsky, R.,
Sackmann, E. (Eds.), Structure and Dynamics of Membranes; vol. 1 of Handbook of
Biological Physics (North-Holland, Amsterdam, 1995) pp. 403-463.
[4] U. Seifert, Adv. Phys. 46, 13 (1997).
[5] P. B. Canham, J. Theor. Biol. 26, 61 (1970).
[6] W. Helfrich, Z. Naturforsch. C 28, 693 (1973).
[7] E. A. Evans, Biophys. J. 14, 923 (1974).
[8] Z.C. Ou-Yang, and W. Helfrich, Phys. Rev. Lett. 59, 2486 (1987).
[9] Z.C. Ou-Yang, and W. Helfrich, Phys. Rev. A 39, 5280 (1989).
[10] S. Svetina, and B. Zeks, Eur. Biophys. J. 17, 101 (1989).
[11] U. Seifert, and R. Lipowsky, Phys. Rev. A 42, 4768 (1990).
[12] R. Lipowsky, Nature 349, 475 (1991).
[13] U. Seifert, K. Berndl, and R. Lipowsky, Phys. Rev. A 44, 1182 (1991).
[14] F. Jülicher, and R. Lipowsky, Phys. Rev. Lett. 70, 2964 (1993).
[15] F. Jülicher, and R. Lipowsky, Phys. Rev. E 53, 2670 (1996).
[16] F. Jülicher, and U. Seifert, Phys. Rev. E 49, 4728 (1994).
[17] L. Miao, U. Seifert, M. Wortis, and H. G. Döbereiner, Phys. Rev. E 49, 5389 (1994).
[18] V. Heinrich, S. Svetina, and B. Zeks, Phys. Rev. E 48, 3112 (1993).
[19] V. Kralj-Iglic, S. Svetina, and B. Zeks. Eur. Biophys. J. 22, 97 (1993).
[20] D.V. Svintradze, Biophys. J. 108, 512 (2015).
[21] D. Chandler, Nature 437, 640 (2005).
[22] S. Leikin, V. A. Parsegian, and D. C. Rau, Ann. Rev. Phys. Chem. 44, 369 (1993).
[23] D.V. Svintradze, Biophys. J. 98, 43 (2010)
[24] D.V. Svintradze, Biophys. J. 110, 623 (2016).
[25] P. Grinfeld, J. Geom. Symm. Phys. 16, 1 (2009).
[26] P. Grinfeld, Introduction to Tensor Analyses and the Calculus of Moving Surfaces
(Springer, New York, 2010).
17
[27] J. Hadamard, Mmoire sur le problme d'analyse relatif l'quilibre des plaques
elastiques encastres (Oeuvres, Tome 2. Hermann, 1968).
[28] J. W. Larson, and T. B. McMahon, Inor. Chem. 23, 2029 (1984).
[29] J. Emsley, Chem. Soc. Rev. 9, 91 (1980).
[30] F. Jahnig, Biophys. J. 71, 1348 (1996).
[31] J. N. Israelachvili, D. J. Mitchell, and B. W. Ninham, Biochim. Biophys. Acta. 470,
185 (1977).
[32] J. N. Israelachvili, Intermolecular and Surface Forces: Revised Third Edition
(Academic Press, London, 2011).
[33] S. E. Feller, Y. Zhang, and R. W. Pastor, J. Chem. Phys. 103, 10267 (1995).
[34] E. Egberts, S. J. Marrink, and H. J. C. Berendsen, Eur. Biophys. J. 22, 423 (1994).
[35] S. Jo, T. Kim, V.G. Iyer, and W. Im. J. Comput. Chem. 29, 1859 (2008).
[36] X. Cheng, S. Jo, H. S. Lee, J. B. Klauda and W. Im, J. Chem. Inf. Model. 53, 2171
(2013).
[37] N. R. Voss, and M. Gerstein, Nucleic Acids Res. 38, W555 (2010).
[38] A. Leitmannova Liu, Volume 4. Advances in Planar Lipids Bilayers and Liposomes
(Academic Press. Elsevier, 2011).
18
FIGURE LEGENDS
FIGURE 1. Graphical illustration of the arbitrary surface and its' local tangent plane.
are local tangent plane base vectors and local surface normal respectively.
are
arbitrary base vectors of the ambient Euclidean space and
is radius vector
of the point.
is arbitrary surface velocity and
display projection of the velocity to
the
directions respectively.
FIGURE 2. Geometric interpretation of the interface velocity
and of the curvilinear time
derivative
applied to invariant field
.
curve and
is its' corresponding point on the
is arbitrary chosen point so that it lays on
is the point where
surface.
surface normal, applied on the point
, intersects the surface
. By the geometric
construction, for small enough
other hand, by the same geometric construction the field
, while from viewpoint of the
,
,
in the point
surface the
and
. On
can be estimated as
value can be
estimated as
, where
shows rate of change in
along the surface
.
and along the directed distance
FIGURE 3. (Color online) Graphical illustration of the isolated system containing aqueous
solution. Water molecules are represented as red and white sticks. The system boundary is
shown as white faces with black edges. The subsystem-micelle is closed surface, blue blob in
the center of the system.
FIGURE 4. (Color online) Simulated three dimensional coordinates of the micelle in aqueous
solution display sphere with diameter 38,5Å. (Left) 1,2-Dimyristoyl-sn-Glycero-3-
Phosphocholine molecules (DHPC phospholipids) are modeled as orange balls. (Right)
Gaussian mapping at contour resolution 8Å of the micelle shows spherical structure.
FIGURE 5. (Color online) All atom simulation of a micelle consisted from 1,2-Dimyristoyl-
sn-Glycero-3-Phosphocholine molecules. (A) The figure shows a geometry of the DHPC
surfactant molecule used in simulation and gives parametric description of volume, surface
area, sphericity and effective radius. (B) Indicates atomistic simulation result contoured by
Gaussian map and the diameter of the micelle, measured by PyMol. The diameter of the
simulated micelle appears to be 54,0 Å with the uncertainty of the measurement 0,1 Å.
19
12,,SSN123,,XXX()(,)RRXRtSV12,,CVV12,,NSSCFA()ttFSSBttSPtSAttS0t/2APBABVtAPVNtFB()()/FBFAtFtttS()FB()iiFPtVFiFFttSiBPtV
Figure 1.
Figure 2.
Figure 3.
20
Figure 4.
Figure 5.
21
|
0912.3658 | 2 | 0912 | 2010-10-19T08:47:41 | Bidirectional transport on a dynamic lattice | [
"physics.bio-ph",
"cond-mat.stat-mech"
] | Bidirectional variants of stochastic many particle models for transport by molecular motors show a strong tendency to form macroscopic clusters on static lattices. Inspired by the fact that the microscopic tracks for molecular motors are dynamical, we study the influence of different types of lattice dynamics on stochastic bidirectional transport. We observe a transition toward efficient transport (corresponding to the dissolution of large clusters) controlled by the lattice dynamics. | physics.bio-ph | physics |
Bidirectional transport on a dynamic lattice
M. Ebbinghaus,1, 2, 3, ∗ C. Appert-Rolland,4, 5, † and L. Santen3, ‡
1Laboratoire de Physique Th´eorique, Univ. Paris-Sud, Bat. 210, F-91405 Orsay Cedex, France
3Fachrichtung Theoretische Physik, Universitat des Saarlandes, D-66123 Saarbrucken, Germany
2CNRS, UMR 8627, F-91405 Orsay, France
4Laboratoire de Physique Th´eorique, UMR 8627,
Universit´e Paris-Sud XI, 91405 Orsay cedex, France
5CNRS, Orsay F-91405, France
(Dated: October 30, 2018)
Bidirectional variants of stochastic many particle models for transport by molecular motors show
a strong tendency to form macroscopic clusters on static lattices.
Inspired by the fact that the
microscopic tracks for molecular motors are dynamical, we study the influence of different types
of lattice dynamics on stochastic bidirectional transport. We observe a transition toward efficient
transport (corresponding to the dissolution of large clusters) controlled by the lattice dynamics.
Microscopic models of stochastic transport like, e.g.,
the so-called asymmetric exclusion process (ASEP) or
the zero range process (ZRP) have been extensively dis-
cussed in the past two decades [1]. The interest is partly
due to the fact that these models play a key role in devel-
oping a general framework for statistical physics far from
equilibrium [2]. Moreover these models have been applied
to many different transport problems, like road traffic [3].
More recently variants of the ASEP have been used in or-
der to describe intra-cellular transport phenomena driven
by molecular motors [4, 5], i.e., proteins that are able to
transport cargos along the filament network of biological
cells [6]. In these models particles can also attach to and
detach from the filament, leading to a finite path length
of the molecular motors [4, 5, 7]. Remarkably, if open
boundary conditions are applied, this extended model is
able to predict the experimentally observed bulk localiza-
tion of high and low density domains [8] although many
structural aspects of the filaments and motors have been
neglected.
In contrast to the success of these models describing
the unidirectional collective motion of molecular motors
in motility assays, it is still an issue to understand the rel-
evant mechanisms involved in intracellular bidirectional
transport. Several models of bidirectional transport have
been suggested that consider positional exchange of par-
ticles with different moving directions on the same track
(see, e.g., [9] and [10]). However, this family of mod-
els cannot be used to describe the motion of oppositely
moving molecular motors which cannot permeate each
other on the same track. This scenario is consistent with
recent findings for certain members of the kinesin and
dynein superfamily, moving in opposite directions on mi-
crotubules (MTs, which are polar filaments and a part of
the cytoskeleton) and sharing the same binding site [11].
∗[email protected]
†[email protected]; author to whom correspon-
dence should be addressed
‡[email protected]
Another difficulty arises from the fact that in real cells
the volume surrounding the molecular tracks, i.e., the
cytoplasm, is finite. Confinement introduces a kind of
memory effect for the motors that have detached from the
filament, i.e., they are more likely to attach again in the
vicinity of their detachment position. As a consequence,
domains of high particle densities on the track and in the
diffusive surrounding reservoir co-localize [4, 12]. Both
effects, i.e., inability of exchanges on the track and the
explicit memory of the particle position, have been con-
sidered recently for static one- and multi-lane systems.
It has been shown [12] that for generic model parame-
ters the formation of macroscopic clusters and therefore
a breakdown of the flow on the filament is observed. Re-
markably the formation of stable macroscopic clusters
does not depend on the particle density but rather on the
number of particles in the system. Therefore the natu-
ral question arises: What are the minimal prerequisites
in order to maintain bidirectional stochastic transport of
interacting particles in small volumes when they cannot
cross each other on the same track? One suggestion has
been made by Klumpp and Lipowsky [13] who considered
that particles prefer to attach themselves in the neighbor-
hood of motors of the same type, an effect that has been
observed experimentally in vitro. For high densities and
large differences in the binding affinity this leads to a
spontaneous formation of unidirectional traffic.
In the present paper, we propose a completely different
type of mechanism which could lead to efficient bidirec-
tional transport on a single track through consideration
of the filament dynamics. This extension of the model has
been inspired by the experimentally observed dynamics
of the cytoskeleton. Indeed, the MTs on which molecular
motors move are themselves highly dynamic [14], due to
nucleation, polymerization and depolymerization, which
occur on time scales similar to those involved in motor
transport and are thus likely to interfere with the motor
dynamics. Beyond the interest for intracellular traffic,
we shall give here a first example where a dynamically
driven jammed phase is hindered by the lattice dynamics.
The model [12] consists of two species of particles mov-
ing on two parallel lanes (Fig. 1) with periodic boundary
D
D
D
D
ω
d
p
ω
a
ω
a
ωp
d
FIG. 1: Schematic representation of the particle dynamics
(lattice dynamics is not included in this figure). Arrows indi-
cate possible moves with corresponding rates which are sym-
metric for both particle species. We impose hard-core inter-
action on the lower lane (filament), while the particles on the
upper lane are non-interacting. Periodic boundary conditions
are considered.
conditions. Along the lower lane, which mimics the po-
lar filament, the particles perform directed motion (rate
p) in the direction determined by the particle's species.
In the upper lane, particles diffuse freely (rate D) and
do not interact (sites on the upper lane can be multiply
occupied). Attachment to the lower lane happens at rate
ωa. The detachment rate ωd is chosen to be smaller than
the stepping rate p in order to capture the processivity of
molecular motors. This setup is motivated by the quasi-
one-dimensional geometry of axons although it largely
(over-)simplifies the real structure. Please note that one
can account for the geometry of the diffusive reservoir
by tuning the attachment rate of the particles [15]. Here
we consider only one processive lane. Counterintuitively,
this simplification reduces, for a given particle density,
the tendency to form macroscopic clusters as shown in
[12].
As a new feature we add some lattice dynamics for
the lower lane, i.e., some sites are eliminated and recre-
ated. The diffusive upper lane remains unchanged. On
the lower lane, particles can only occupy a site if this
binding site exists. The attachment moves (rate ωa) are
consequently rejected if the binding site has been elimi-
nated. Additionally, a particle will automatically switch
to the upper lane if it makes a forward step (rate p) onto
an eliminated site or if the site which is occupied by the
particle is eliminated.
(1.)
In particular:
In the following, we consider different types of lat-
a simple realiza-
tice dynamics.
(2.) Dynamics
tion of uncorrelated lattice dynamics.
which depends on occupation by particles. The results
shown were obtained from Monte Carlo simulations over
at least 106 steps with a constant set of parameters for
the particle dynamics: p = 1, ωd = 0.02, ωa = 0.33 and
D = 0.33. The particle density has been chosen high
enough (ρ±
tot = 0.5) in order to observe large clusters in
the case of a static lattice even for small system sizes.
First we consider some lattice dynamics independent
of the configuration of particles: a site of the lower lane
is randomly eliminated at rate kd and recreated at rate
kp. This choice of the lattice dynamics can be seen as
a minimal model, with a limited number of parameters.
0.015
l
s
e
c
i
t
r
a
p
l
l
a
f
o
n
o
i
t
c
a
r
F
0.01
0.005
0
0
2
kd = 0
kd = 0.01
kd = 0.04
kd = 0.1
100
Cluster length
200
300
FIG. 2: (color online). Distribution of cluster sizes in a system
of size L = 1000 with simple lattice dynamics (scenario 1).
Recreation of lattice sites occurs at kp = 1. The black line
corresponds to a static lattice.
0.08
0.06
b
l
j
s
e
c
i
t
r
a
p
f
o
x
u
F
l
0.04
0.02
0
0
0
L = 100
L = 500
L = 1000
L = 2000
L = 4000
L = 16000
1
0.8
0.6
0.4
0
0.05
0.05
0.1
0.1
0.15
0.15
0.2
0.2
Depolymerization rate kd
FIG. 3: (color online). Flux of positive particles along the
filament with simple lattice dynamics at kp = 1 for different
system sizes L and for the same density. If depolymerization
is weak, the flux depends on the system size, but changes to a
density-dependent state as large clusters disappear (see also
Fig. 2). The inset shows the same data divided by the flux in
the smallest system (L = 100).
Network dynamics induces an increase (resp. decrease)
of the effective detachment rate ωd,eff (respectively, at-
tachment rate ωa,eff), which take the form
ωa,eff = ωa
kp
kp + kd
; ωd,eff = ωd + p
kd
kp + kd
+ kd.
(1)
As the depolymerization rate kd increases -- and accord-
ingly the fraction of "holes" in the filament kd/(kp + kd)
-- large clusters become less and less dominant until they
disappear completely (Fig. 2). This disappearance of
large clusters is accompanied by a transition from a size
dependent to a size independent state (Fig. 3). For large
systems, flux vanishes below the transition, while it keeps
a finite value above the transition. As a result, the tran-
sition becomes sharper for an increasing system size.
When the depolymerization rate kd is increased, first
the flux along the filament (symmetric for both particle
space
0.1
b
0.06
static filament
dynamic filament
e
m
i
t
l
j
s
e
c
i
t
r
a
p
e
v
i
t
i
s
o
p
f
o
x
u
l
f
m
u
m
x
a
M
i
kd,max(kp)
jb(kp,kd,max(kp))
jb,static
0.08
0.06
0.04
0.02
3
x
a
m
d
,
0.05
0.04
0.03
k
e
t
a
r
n
o
i
t
a
z
i
r
e
m
y
o
p
e
d
l
a
m
l
i
t
p
O
0.02
0.01
FIG. 4: (color online). Space-time plots of the filament oc-
cupation for a static (left) and a dynamic (right) filament in
a system of size L = 1000 and density ρ±
tot = 0.5. Red resp.
green dots are particles with moving direction to the right
resp.
left, white spaces are unoccupied filament sites. The
filament dynamics clearly induces a transition from a conden-
sated to a homogenous phase. Note that considering a finite
capacity of the diffusive lane would reinforce the jamming.
species) increases too. However, note that, if kd is too
high, binding sites become increasingly sparse, and par-
ticles have fewer segments on which they can contribute
to the total flux in the system. Hence, the flux of each
particle species along the filament disappears for kd → ∞
and kp constant. This behavior entails the existence of
an optimal value for kd (not seen on Fig. 3) at which
the flux is maximized. In this optimal regime with not
too many holes in the filament, blocking situations with
at least two particles of different species still occur fre-
quently. As a consequence, the maximum fluxes that we
obtained are about one third of the flux in a comparable
single-species system [16], which is nevertheless a great
improvement compared to the static filament case.
− ρ∓
u (1 − ρ±
The degree of homogeneity can be estimated by com-
parison with a product-state.
Indeed, above the tran-
sition, particles are well-dispersed over the whole sys-
tem (Fig. 4) and are much less correlated than in the
static filament case.
In a product-state approximation
which neglects any correlations and if translational in-
variance is assumed, the stationary state leads to [12]
ωa,effρ±
b where ρ denotes the
particle density, i.e., the number of particles divided by
the system size. The indices u and b refer to the un-
bound state (upper lane) and bound state (lower lane)
and +/− signs denote the particle species moving either
In combination with
in positive or negative direction.
the conservation of particles ρ±
b + ρ±
u , the density
of particles on the filament ρb as well as the flux of one
particle species along the filament j±
−ρ∓
b )
can be calculated.
b ) = ωd,effρ±
tot = ρ±
b = pρ±
b (1 −ρ±
b
b
Below the transition, a big cluster is formed and
the product-state approximation obviously fails. Above
the transition, the product-state prediction still overesti-
mates (not shown) the flux. This indicates that clusters
made of a few particles still play an important role. It is
0
0.001
0.01
0.1
10
Polymerization rate kp
1
100
0
1000
FIG. 5: (color online). Maximum flux jb(kp, kd,max(kp)) (red
squares) and optimal depolymerization rate kd,max(kp) (black
circles) at given polymerization rates kp in a system of size
L = 1000. For comparison, the flux in a system with a static
lattice is drawn as a solid red line. Note that the horizontal
axis is in logarithmic scale and, therefore, both quantities
seem to saturate at high polymerization rates.
only for strong lattice dynamics that the product-state
solution regains validity.
In Fig. 5, simulation results for the optimal depolymer-
ization rate kd,max(kp) maximizing the flux are drawn as
a function of the polymerization rate. The maximum
flux jb(kp, kd,max(kp)) is shown as a function of kp. One
can see that the gain in transport capacity is obtained
for a wide range of values of the polymerization rate kp --
an observation which supports the general validity of our
results. Furthermore, the fraction of eliminated filament
sites kd/(kp + kd) in the optimal regime decreases with
increasing polymerization rate kp as the optimal value
kd,max saturates. This indicates that the optimal flux
depends on the time interval a site persists rather than
on the delay after which it is recreated.
In the second realization of lattice dynamics consid-
ered, a site is eliminated with rate kd only if that site is
occupied. This can be considered as a prototypical ex-
ample of a coupling between MT dynamics and transport
by molecular motors. On average, the fraction of existing
sites on the lower lane becomes kp/ (cid:2)kp + (ρ+
b )kd(cid:3)
and Eq. (1) is modified accordingly.
b + ρ−
It turns out that the results are qualitatively the same
as obtained for the first type of dynamics. The main
difference is that the flux in the homogenous phase is
higher than in the first scenario. In this phase, when a
motor is moving freely, it encounters less holes than in the
case of scenario 1, as empty sites cannot be eliminated
anymore. Processive runs along the filament are thus less
frequently interrupted. By contrast, in the condensated
phase, there is almost no difference to the first scenario
as the transport capacity is limited by the large clusters,
where all filament sites are occupied by particles anyway.
The transition to a density-dependent efficient state is
maintained if one generalizes scenario 2 by requiring that
a higher number of particles has to accumulate in order
to put enough strain on the filament to break.
Finally we mention briefly another kind of dynamics.
Biopolymers, such as actin filaments or MTs, show a
characteristic type of dynamics under certain conditions,
termed tread-milling [6], for which both ends of the fil-
ament move with the same average velocity. We have
checked that, for a fairly simplified model consisting of
regularly spaced holes in the lower lane which propagate
synchronously but stochastically through the system, the
aforementioned transition toward efficient transport is
still observed. Note that, since the holes move only in
one direction, the two species of particles are affected
differently. The flux of the two species therefore depends
on the moving direction along the filament. For both
species we remark a considerable increase of the flux and
reach a maximum current comparable to the one in the
previous scenarios.
To summarize, we have presented a model for bidi-
rectional transport on a one-dimensional dynamic lattice
coupled to a confined diffusion reservoir. While on a
static lattice persistent clusters form which inhibit effi-
cient transport [12], we have found the counterintuitive
effect that the transient suppression of filament sites dra-
matically enhances the transport capacity. Indeed, the
inhibition of large clusters leads to a transition toward
a homogenous state characterized by efficient transport.
This is a new mechanism in the phenomenology of dy-
namic phase transitions. This transition separates a size-
dependent (jammed) state from a density dependent one.
It is robust in the sense that it is obtained for quite dif-
ferent types of lattice dynamics. The actual transport
capacity of the system rather depends on the optimal
lifetime of a binding site than on the details of the fila-
ment dynamics. The lifetime has to be short enough in
order to avoid jam formation and long enough in order
to direct the transport.
Some insight could be gained from an analysis simi-
lar to the one found in [17] for the symmetric motion
of non-interacting particles in fluctuating energy land-
scapes. However, note that here the effective "potential"
landscape emerges spontaneously from the particle jam-
ming.
Finally we would also like to discuss the relevance of
our model results for axonal transport. In view of our re-
sults for bidirectional transport on static lattices, which
4
generically leads to jamming, it is rather surprising, but
of course necessary, that transport in these real one-
dimensional axonal structures would be at all stable and
efficient. This is all the more surprising given that the
model overestimates the diffusivity of the detached par-
ticles, which should be reduced in real systems due to
the size and interactions of the vesicles in the cytoplasm.
These results give strong evidence that additional effects
must come into play in order to stabilize motor driven
transport in axons and reduce the tendency to form large
particle clusters. The mechanism suggested in this work
is based on the fact that the filament dynamics limits
the size of particle clusters. A more detailed modeling
of motor driven transport in axons is difficult since the
experimental findings concerning the motor interactions
and the MT-dynamics are so far incomplete and subject
to interpretation: while it is well established that the
plus-ends of MTs undergo polymerization events toward
the synapse, the pathlength of the growing plus-ends as
well as the dynamics of the minus ends is not known yet.
Besides the robustness of our results with respect to the
lattice dynamics details, the importance of the lattice dy-
namics is supported by experimental results on transport
in axons, where the strong correlation between MT dy-
namics and vesicle transport has been demonstrated [18].
In this context, we have observed that lane formation,
which could be invoked as an alternative mechanism for
bidirectional transport [13], is impeded by the dynam-
ics of MTs. Thus it seems that it is necessary to take
into account the lattice dynamics in order to gain un-
derstanding in axonal transport. Although our picture
of bidirectional transport is qualitatively consistent with
the experimental findings it is still an issue to develop a
more quantitive description of real axonal transport. A
fully validated model could then be used in order to inves-
tigate mechanisms that are underlying clinically relevant
transport defects in nerve cells [14, 19].
Acknowledgments
ME would like to thank the DFG Research Training
Group GRK 1276 for financial support.
[1] R. A. Blythe and M. R. Evans, J. Phys. A: Math. Theor.
40, R333 (2007).
[2] J. Krug, Phys. Rev. Lett. 67, 1882 (1991).
[3] D. Chowdhury, L. Santen, and A. Schadschneider, Phys.
and P. Walter, Molecular biology of the cell (Taylor and
Francis, 2002).
[7] J. Tailleur, M. R. Evans, and Y. Kafri, Phys. Rev. Lett.
102, 118109 (2009).
Reports 329, 199 (2000).
[8] K. Nishinari, Y. Okada, A. Schadschneider,
and
[4] R. Lipowsky, S. Klumpp, and T. M. Nieuwenhuizen,
D. Chowdhury, Phys. Rev. Lett. 95, 118101 (2005).
Phys. Rev. Lett. 87, 108101 (2001).
[9] P. F. Arndt, T. Heinzel, and V. Rittenberg, J. Phys. A:
[5] A. Parmeggiani, T. Franosch, and E. Frey, Phys. Rev.
Math. Gen. 31, L45 (1998).
Lett. 90, 086601 (2003).
[10] G. Korniss, B. Schmittmann, and R. K. P. Zia, Europhys.
[6] B. Alberts, A. Johnson, J. Lewis, M. Raff, K. Roberts,
Lett. 45, 431 (1999).
[11] N. Mizuno et al., EMBO Journal 23, 2459 (2004).
[12] M. Ebbinghaus and L. Santen, J. Stat. Mech. p. P03030
Lecture Notes in Computer Science (Springer, 2010).
[16] S. Klumpp and R. Lipowsky, J. Stat. Phys. 113, 233
(2009).
(2003).
[13] S. Klumpp and R. Lipowsky, Europhys. Lett. 66, 90
[17] A. Nagar, S. N. Majumdar, and M. Barma, Phys. Rev.
(2004).
E 74, 021124 (2006).
[14] O. A. Shemesh, H. Erez, I. Ginzburg, and M. E. Spira,
[18] O. A. Shemesh and M. E. Spira, Acta Neuropathologica
Traffic 9, 458 (2008), see also Supporting Movies.
119, 235 (2010).
[15] M. Ebbinghaus, C. Appert-Rolland, and L. Santen, in To
appear in the Proceedings of the ACRI2010 conference,
[19] G. B. Stokin et al., Science 307, 1282 (2005).
5
|
1710.04100 | 1 | 1710 | 2017-10-11T15:00:18 | Multivalent Ion-Activated Protein Adsorption Reflecting Bulk Reentrant Behavior | [
"physics.bio-ph",
"cond-mat.soft"
] | Protein adsorption at the solid-liquid interface is an important phenomenon that often can be observed as a first step in biological processes. Despite its inherent importance, still relatively little is known about the underlying microscopic mechanisms. Here, using multivalent ions, we demonstrate the control of the interactions and the corresponding adsorption of net-negatively charged proteins (bovine serum albumin) at a solid-liquid interface. This is demonstrated by ellipsometry and corroborated by neutron reflectivity and quartz-crystal microbalance experiments. We show that the reentrant condensation observed within the rich bulk phase behavior of the system featuring a nonmonotonic dependence of the second virial cofficient on salt concentration c_s is reflected in an intriguing way in the protein adsorption d(c_s) at the interface. Our findings are successfully described and understood by a model of ion-activated patchy interactions within the framework of classical density functional theory. In addition to the general challenge of connecting bulk and interface behavior, our work has implications for, inter alia, nucleation at interfaces. | physics.bio-ph | physics | Multivalent Ion-Activated Protein Adsorption Reflecting Bulk Reentrant Behavior
Madeleine R. Fries1, Daniel Stopper2, Michal K. Braun1, Alexander Hinderhofer1, Fajun Zhang1, Robert
M. J. Jacobs3, Maximilian W. A. Skoda4, Hendrik Hansen-Goos2, Roland Roth2 and Frank Schreiber1
1 Institute for Applied Physics, University of Tubingen, 72076 Tubingen, Germany
2 Institute for Theoretical Physics, University of Tubingen, 72076 Tubingen, Germany
3 Department for Chemistry, Chemistry Research Laboratory,
University of Oxford, Oxford, OX1 3TA, United Kingdom and
7
1
0
2
t
c
O
1
1
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
0
0
1
4
0
.
0
1
7
1
:
v
i
X
r
a
4 Rutherford-Appleton Laboratory, ISIS Facility, Didcot, OX11 0QX, United Kingdom
(Dated: November 14, 2018)
Protein adsorption at the solid-liquid interface is an important phenomenon that often can be
observed as a first step in biological processes. Despite its inherent importance, still relatively little
is known about the underlying microscopic mechanisms. Here, using multivalent ions, we demon-
strate the control of the interactions and the corresponding adsorption of net-negatively charged
proteins (bovine serum albumin) at a solid-liquid interface. This is demonstrated by ellipsometry
and corroborated by neutron reflectivity and quartz-crystal microbalance experiments. We show
that the reentrant condensation observed within the rich bulk phase behavior of the system featur-
ing a nonmonotonic dependence of the second virial coefficient on salt concentration cs is reflected
in an intriguing way in the protein adsorption d(cs) at the interface. Our findings are successfully
described and understood by a model of ion-activated patchy interactions within the framework
of classical density functional theory. In addition to the general challenge of connecting bulk and
interface behavior, our work has implications for, inter alia, nucleation at interfaces.
The interactions of proteins, with their inherent het-
erogeneity and differently charged patches, in addition
to hydrophilic and hydrophobic regions and dispersion
forces, are very complex [1]. While obviously required
for their biological function, this complexity of the in-
teractions is very demanding for a quantitative physical
understanding. Particularly difficult is the connection
to the associated mesoscopic and macroscopic behavior,
with enormous implications for a range of rather diverse
fields. These include the understanding of protein crys-
tallization [2 -- 4], as well as various forms of aggregation
[5, 6], whether biologically desired [7 -- 9] or related to dis-
eases such as Alzheimer's [10], Huntington's, or prion
diseases (e.g. Creutzfeldt-Jakob disease) [11]. A further
level of complexity is added by the frequently heteroge-
neous environment in soft and biological systems, often
with internal interfaces, at which adsorption might take
place, coexisting with fluid (bulk-like) regions. While
in numerous studies the phase behavior of proteins has
been investigated [5, 12, 13], it remains an important
challenge to understand protein-protein interactions in a
microscopic picture and to predict the resulting macro-
scopic thermodynamic behavior of proteins in solution
and at interfaces, and how these behaviors correspond or
differ.
For the manipulation of the bulk phase behavior dif-
ferent strategies have been demonstrated. On the one
hand, the use of co-solvents such as glycerol to stabilize
a protein solution [14] can help to avoid protein aggre-
gation and cluster formation. On the other hand, enzy-
matic crosslinking [15] or the use of trivalent ions such as
yttrium cations can be employed to trigger bridge forma-
tion between globular proteins, which can lead to clus-
ter formation, reentrant condensation and liquid-liquid
phase separation (Fig. 1) [16, 17].
Protein adsorption at solid-liquid interfaces occurs in
many natural processes, and its understanding is crucial
in many fields, ranging from biotechnology, biology, phar-
macology, and medicine to environmental science and
food processing with relevance in many applications [1].
In particular, it is the first step in numerous biological
processes, such as the blood coagulation cascade, trans-
membrane signaling and adhesion of particles (bacteria
or cells) [1], and therefore plays a key role in biomedi-
cal devices, including biosensors, biochips, soft contact
lenses and biomaterials for implants [18].
Bovine serum albumin (BSA) is considered as one of
the model proteins for adsorption studies [19]. In solu-
tion, BSA is a globular protein with well-characterized
physico-chemical properties [20]. Serum albumin is the
most abundant blood protein in mammals, and its ad-
sorption has been intensely studied with different meth-
ods, under various conditions [1, 21 -- 23]. Nevertheless,
controlling the interactions and connecting to the bulk
behavior remains a challenge. In that context, the use
of multivalent ions [24 -- 26] offers a viable path, with the
unique opportunity to tailor and even invert the charge
state of proteins as well as surfaces by overcompensation
[16, 17, 27], which has been demonstrated to be a rather
universal approach [28].
In this Letter, we demonstrate the use of multivalent
ions (Y3+) to control the interaction of BSA with SiO2
interfaces. We find reentrant interface adsorption behav-
ior, reflecting in an intriguing way the bulk phase behav-
ior [Fig. 1 (b)]. Furthermore, we show that both bulk
and interface adsorption behavior can be modeled con-
sistently by statistical mechanics of ion-activated patches
[29].
2
volume fraction of 1; see Supporting Material [23] for def-
inition of d, which includes Refs. [32 -- 48]). Complemen-
tary studies were performed using neutron reflectometry
(NR) at the INTER beam line at ISIS (Rutherford Lab-
oratory, Didcot, UK) [49, 50], as well as quartz crystal
microbalance (QCM, Q-Sense Analyzer Biolin Scientific),
confirming the trends in the adsorption behavior d(cs)
[23]. For better comparability, the thicknesses extracted
from NR and QCM-D are also normalized to an assumed
BSA volume fraction of 1 [23].
Based on real-time ellipsometric data of the adsorption
kinetics, we extract d in the long-time limit (saturation
after ∼ 60 min) and plot it in Fig. 2 as a function of
cs/cp. It is convenient to use a dimensionless salt axis,
i.e. cs/cp, especially when comparing to theory. Both
BSA and SiO2 surfaces are net negatively charged in wa-
ter (no added salt). Under these conditions the electro-
static repulsion among the proteins dominates the solu-
tion compared to the repulsion between the proteins and
the solid surface leading to a minimum of the protein ad-
sorption. Evaluation of the ellipsometric data shows that
then only a d of 1.2± 0.25 nm is adsorbed. Upon increas-
ing cs to 1.3 mM still in the clear regime I as illustrated
in Fig. 1 (b), d increases to 6.29±1.02 nm (solid triangles
in Fig. 2). In our system, we assume a Rp of ∼ 3.5 nm
[23, 51] and define one monolayer equivalent (ML) to be
d ≈ 4 nm [23], corresponding in regime I at 1.3 mM to
the formation of d > 1 ML.
In regime II, d increases towards a maximum value of
9.59± 2.5 nm (> 1 ML) at cs = 4 mM (empty diamonds,
Fig. 2). At still higher cs, d decreases down to ∼ 6
nm approaching the upper boundary of regime II at c∗∗.
Note that in regime II (empty diamonds) the bulk so-
lution is centrifuged before the adsorption experiments,
which explains the jump of d in the transition region be-
tween regime II and III. This is done because the solution
in regime II is too turbid due to extensive protein cluster
formation, which causes massive bulk light scattering and
a lack of sensitivity of the ellipsometer. We show both
data sets at cs/cp = 40 (centrifuged and non-centrifuged)
to account for the experimental difference, which, impor-
tantly, does not affect the overall adsorption trend.
In regime III close to c∗∗, d is 7.28 ± 0.87 nm at cs
= 12 mM, but with increasing cs, d decreases down to a
plateau value of 4.5 nm above 30 mM (solid squares in
Fig. 2). d then corresponds to slightly less than one full
ML. These experimental results are supported by com-
plementary measurements (NR and QCM) [23]. It is in-
teresting to note that after rinsing with pure water the
surface retains an irreversibly bound layer of protein with
d = 4 nm.
To understand the adsorption behavior, it is important
to realize that the behavior of d is closely related to that
of B2/BH
S of the bulk solution (inset, Fig. 2). In regime
2
II, the value of B2/BH
S is clearly negative indicating a
2
strong overall attraction between proteins compared to
Figure 1: (a) Charge inversion of BSA as a consequence of adding
trivalent yttrium ions. (b) Schematic of the bulk phase diagram of
BSA and YCl3 (modified from Ref.
[31]) showing a liquid-liquid
phase separation (LLPS) and reentrant condensation. The dashed
red arrow indicates the path taken in the experiments.
BSA (molecular weight MW = 66 kDa) and YCl3 were
obtained from Sigma Aldrich and used as received. BSA
is net negatively charged above its isoelectric point of
pH = 4.6 [16]. Protein solutions were prepared by mix-
ing the stock solutions at temperature T = 20 ◦C. The
working protein concentration cp was set to 20 mg/mL
and the trivalent salt concentration cs ranged from 0.5 -
40 mM [depicted by the red arrow in Fig. 1 (b)]. With
increasing cs, protein solutions undergo a reentrant con-
densation (RC) phase behavior in regime III. An aggrega-
tion regime II occurs in between two salt concentrations,
c∗ and c∗∗ as illustrated in Fig. 1 (a-b). The physical
mechanisms behind the observed RC behavior are the
effective inversion of protein charge [Fig. 1 (a)] and a
cation-mediated anisotropic attraction [16, 29]. The ef-
fective interactions Veff(r) between proteins are reflected
in the behavior of the reduced second virial coefficient
B2/BH
S. B2 defines the second viral coefficient of the
2
bulk solution
(cid:90) ∞
dr r2(cid:104)
1 − e−β Veff(r)(cid:105)
B2 = 2π
.
(1)
0
S = 16πR3
The second viral coefficient of hard spheres is defined by
BH
p/3, where Rp is the radius of the protein.
2
Experimental B2/BH
S (orange inset, Fig. 2) were deter-
2
mined using small-angle X-ray scattering (SAXS) (ID02
at the ESRF in Grenoble, France) [30].
The adsorption studies were performed on standard Si
wafers with native oxide layer. Before each measurement,
all components of the liquid cell were cleaned at 50 ◦C
via ultrasonication in acetone, isopropanol, and degassed
water for 10 min in each solvent. Ellipsometry (Woollam
VASE M-2000 and Beaglehole Picometer) was employed
in situ at the Brewster angle of 68◦ (for SiO2) to extract
an effective protein layer thickness d, assuming a Cauchy
layer with density corresponding to that of pure BSA (i.e.
3
Figure 2: Individual symbols: Adsorbed protein layer thickness d extracted from ellipsometry as a function of cs/cp. c∗ and c∗∗ denote
the phase transitions of the bulk solution [16] [Fig. 1 (b)]. Note that around c∗∗, there is an experimental difference between the data
in regime III vs. regime II. The centrifuged samples in regime II reflect the adsorption trend for overall lower adsorption values due to
the removal of big clusters in bulk solution, but still follow the same adsorption trend. In addition, the top cs-axis is included showing
the absolute cs in the system (at cp = 20 mg/mL). The blue shaded area shows the approximate range of the bulk turbidity. Solid and
dashed lines: Protein adsorption based on DFT calculations as born out by the ion-activated attractive patch model, while neglecting
long-range forces, as a function of cs/cp for two different values of βε. Inset: B2/BH
S is the reduced second virial coefficient obtained
2
via SAXS measurements.
regime I and III. Note that this is not the definition of
the regimes nor its boundaries, but rather is an impor-
tant observation. The net attraction between proteins is
reflected by a sharp adsorption maximum. This observa-
tion indicates that the protein adsorption in our system
is closely related to the bulk behavior, which can suc-
cessfully be accounted for by the model for ion-activated
attractive patches as a mechanism for interactions in
protein-salt mixtures [29]. This model
is formulated
within the Wertheim theory for associating fluids [52 -- 60],
and treats proteins as hard spheres with radius Rp and
M distinct and independent binding sites (patches) [60].
These sites can be occupied by salt ions, thereby activat-
ing a given patch (ion binding). The occupation probabil-
ity of a site is given by Θ = (1+exp(βεb−βµs))−1, where
µs denotes the salt chemical potential, β = (kBT )−1, and
εb the binding energy [29]. A bond between two patches
of distinct proteins is possible only if an activated patch
meets a de-activated one (ion bridge). As a result, cs
controls the protein-protein interactions. Note, however,
that only the proteins are represented explicitly in this
model. This implies that cs as a function of µs cannot
be predicted self-consistently within this approach. We
use the location of the minimum of the experimentally
determined B2/BH
2
S in order to calibrate cs(µs).
The resulting phase diagram of the model accounts for
key features of the rather rich experimental phase dia-
gram, such as reentrant condensation and a closed-loop
LLPS binodal schematically shown in Fig. 1 (b) [29]. The
model also allows predictions of regions in the phase di-
agram which are populated by protein clusters. A quan-
titative measure for this is Φ, the fraction of proteins in
clusters. In the present study we assume that in region
II at least 20% of the proteins are part of clusters, i.e.,
Φ = 0.2 to define c∗ and c∗∗.
While the experimental results presented here suggest
that the bulk behavior dominates the adsorption trend,
the key point in the present study is the protein adsorp-
tion at a charged planar wall, which implies breaking
the translational symmetry of the system. To this end,
we employ classical density functional theory (DFT) [61]
which provides a powerful and well-established frame-
work to investigate inhomogeneous density distributions.
Within DFT one can show rigorously [61] that a func-
tional
(cid:90)
Ω[ρ] = F[ρ] +
dr ρ(r) [Vext(r) − µ]
(2)
of the inhomogeneous density profile ρ(r) exists and takes
its minimum, the grand potential, at the equilibrium den-
sity distribution.
Using a DFT formulation of the Wertheim theory [62]
based on fundamental measure theory (FMT) for hard
spheres [63, 64], we calculate d at the SiO2-water inter-
face. This interface is charged and strongly attracts yt-
trium ions, which in turn attract proteins towards the
wall [Fig. 3 (a)]. Effectively, this can be described by a
short-ranged external potential Vext(z) acting on the pro-
teins, where z is the distance normal to the SiO2 wall.
We set βVext(z) = ∞ for z < 0 in order to represent a
steric repulsion between proteins and the substrate and
βVext(z) = −βεM Θξ(z) for z ≥ 0. ξ(z) accounts for the
rather short-ranged attraction induced by the yttrium
ions condensed on the wall -- which is in-line with recent
experimental observations [26]. Here, we employ a Gaus-
sian form ξ(z) = exp(−0.5(z/Rp)2) with the range of at-
traction being roughly one protein diameter, which effec-
tively accounts for the range of the screened electrostatic
interactions between ions and the wall, and between ions
and proteins.
The strength of the external potential depends on µs
via the occupation probability Θ of the protein binding
sites. This form can be motivated by the following argu-
ments. A sketch is presented in Fig. 3. At low cs, when
Θ → 0, only a few proteins are subjected to the attrac-
tion of the wall induced by the ions. As cs increases,
more ions mediate the attractions between the wall and
the proteins. At the same time the protein-protein at-
traction increases accordingly, which leads in turn to an
increase in d. At very high cs (Θ → 1), the mechanism
for the wall attraction remains, while the protein-protein
interaction becomes weak since a majority of the binding
sites are occupied so that salt ions can no longer cause a
patchy attraction between the proteins. Therefore, one
expects from our model ∼ 1 ML of proteins to be ad-
sorbed on the wall for Θ → 1.
In Fig. 2 (solid and dashed lines), we show the value of
d in nm as a function of cs/cp for a volume packing frac-
tion η = (4π/3)ρpR3
p = 0.0078, corresponding to cp = 20
mg/mL along the path indicated by the red dashed ar-
row in Fig. 1 (b). We choose M = 4 and εb = −5 [29].
The protein adsorption is computed from the inhomo-
geneous density profile ρp(z), obtained via the DFT for
our activated patch model. In order to compare to exper-
iments, we define d as the distance from the wall where
ρp(z) is at least 50% higher than the bulk density ρp. For
suitable values of βε [1.8 (solid curve), and 1.7 (dashed
curve)], we find very good, semi-quantitative agreement
between theory and experiment. For high values of cs,
we find a finite d related to ∼ 1ML similar to the exper-
iments. Note that the fraction Φ of proteins in clusters
in the bulk system is directly related to the behavior of
the layer thickness d of proteins at the wall.
Our theoretical results confirm that ion binding at the
protein surface drives the experimentally observed non-
monotonic adsorption behavior, thereby reflecting the
underlying bulk interactions. In particular, the remark-
able agreement between experiment and theory (consid-
ering in particular the few parameters involved) empha-
sizes that our model of ion-activated attractive patchy
particles, subjected to an effective external wall poten-
tial, captures the essential effects of the protein adsorp-
tion at a charged surface in the presence of multivalent
4
Figure 3: (a) Illustration of the different interaction mechanisms of
the proteins, salt, and interface. (b) Sketch of protein adsorption
on the attractive surface by increasing cs/cp.
salt ions. Our model is kept intentionally simple with a
minimum number of parameters, which helps us to iden-
tify the key mechanism responsible for the behavior of
the system, namely the ion-activated patchy interactions
of the proteins.
Importantly, using our model we can
explore the adsorption behavior of our system in differ-
ent parts of the bulk phase diagram. As we increase the
protein concentration approaching the LLPS region the
adsorbed film thickness d increases. We find a complete
wetting regime in which d becomes even macroscopically
thick [23]. Qualitatively we find similar behavior as shown
in Fig. 2 with a maximum for c∗<cs<c∗∗.
In conclusion, we have demonstrated that multivalent
ions can be employed not only to control the bulk in-
teractions and bulk phase behavior of proteins such as
BSA, but also its adsorption behavior at a charged in-
terface such as water-SiO2. We observe reentrant effects
at the interface, which reflects the bulk behavior, mea-
sured by B2/BH
S, in an intriguing way. Furthermore, the
2
experimental data can be explained and understood by
theoretical calculations within the framework of classical
DFT based on a model of ion-activated patchy interac-
tions and their associated statistics. In addition to the
fundamental implications of the first-time demonstration
of this ion-activated patch model in the context of the
symmetry break brought about by an interface, our ap-
proach may pave the way to controlled nucleation at in-
terfaces in regime II and possibly protein crystallization
under new conditions.
Funding by the DFG and Carl-Zeiss-Stiftung is grate-
fully acknowledged.
[1] M. Rabe, D. Verdes, and S. Seeger, Adv. Colloid Inter-
face Sci 162, 87 (2011).
[2] P. Rein ten Wolde and D. Frenkel, Science 277, 1975
(1997).
[3] S. Whitelam, Phys. Rev. Lett. 105, 088102 (2010).
[4] D. Fusco and P. Charbonneau, Phys. Rev. E 88, 012721
(2013).
[5] A. Stradner, H. Sedgwick, F. Cardinaux, W. C. Poon,
S. U. Egelhaaf, and P. Schurtenberger, Nature 432, 492
(2004).
[6] P. D. Godfrin, S. D. Hudson, K. Hong, L. Porcar,
P. Falus, N. J. Wagner, and Y. Liu, Phys. Rev. Lett.
115, 228302 (2015).
[7] U. Schmidt, G. Guigas, and M. Weiss, Phys. Rev. Lett.
101, 128104 (2008).
[8] J. Yan, M. Enge, T. Whitington, K. Dave, J. Liu, I. Sur,
B. Schmierer, A. Jolma, T. Kivioja, and M. Taipale, Cell
154, 801 (2013).
[9] C. R. Monks, B. A. Freiberg, H. Kupfer, N. Sciaky, and
A. Kupfer, Nature 395, 82 (1998).
[10] A. Kakio, S.-i. Nishimoto, K. Yanagisawa, Y. Kozutsumi,
and K. Matsuzaki, J. Biol. Chem. 276, 24985 (2001).
[11] A. Aguzzi and T. O'Connor, Nat. Rev. Drug Discov 9,
237 (2010).
[12] R. P. Sear, J. Chem. Phys. 111, 4800 (1999).
[13] C. Goegelein, J. Chem. Phys. 129, 085102 (2008).
[14] V. Vagenende, M. G. S. Yap, and B. L. Trout, Biochem-
istry 48, 11084 (2009).
[15] Y. Saricay, S. K. Dhayal, P. A. Wierenga,
and
R. de Vries, Farad. Discuss 158, 51 (2012).
[16] F. Zhang, M. W. A. Skoda, R. M. J. Jacobs, S. Zorn,
R. A. Martin, C. M. Martin, G. F. Clark, S. Weggler,
A. Hildebrandt, O. Kohlbacher, and F. Schreiber, Phys.
Rev. Lett. 101, 148101 (2008).
[17] F. Zhang, R. Roth, M. Wolf, F. Roosen-Runge, M. W. A.
Skoda, R. M. J. Jacobs, M. Stzucki, and F. Schreiber,
Soft Matter 8, 1313 (2012).
[18] D. G. Castner, Biointerphases 12, 02C301 (2017).
[19] S. Schottler, G. Becker, S. Winzen, T. Steinbach,
K. Mohr, K. Landfester, V. Mailander, and F. R. Wurm,
Nat. Nanotech (2016).
[20] T. J. Su, J. R. Lu, R. K. Thomas, and Z. F. Cui, J.
Phys. Chem. B 103, 3727 (1999).
[21] J. R. Lu, X. Zhao, and M. Yaseen, Curr. Opin. Colloid
Interface Sci. 12, 9 (2007).
[22] J. J. Gray, Curr. Opin. Struct. Biol 14, 110 (2004).
[23] Supporting Material.
[24] K. Kandori, S. Toshima, M. Wakamura, M. Fukusumi,
and Y. Morisada, J. Phys. Chem. B 114, 2399 (2010).
[25] E. Anbazhagan, A. Rajendran, D. Natarajan, M. Kiran,
and D. K. Pattanayak, Colloids Surf., B 143, 213 (2016).
[26] S. S. Lee, M. Schmidt, N. Laanait, N. C. Sturchio, and
P. Fenter, J. Phys. Chem. C 117, 23738 (2013).
[27] A. Sauter, F. Roosen-Runge, F. Zhang, G. Lotze, R. M. J.
Jacobs, and F. Schreiber, J. Am. Chem. Soc 137, 1485
(2015).
[28] F. Zhang, S. Weggler, M. J. Ziller, L. Ianeselli, B. S. Heck,
A. Hildebrandt, O. Kohlbacher, M. W. Skoda, R. M. Ja-
5
cobs, and F. Schreiber, Proteins 78, 3450 (2010).
[29] F. Roosen-Runge, F. Zhang, F. Schreiber, and R. Roth,
Sci. Rep. 4, 7016 (2014).
[30] M. K. Braun, M. Wolf, O. Matsarskaia, S. Da Vela,
F. Roosen-Runge, M. Sztucki, R. Roth, F. Zhang, and
F. Schreiber, J. Phys. Chem. B 121, 1731 (2017).
[31] F. Zhang, F. Roosen-Runge, A. Sauter, M. Wolf, R. M. J.
Jacobs, and F. Schreiber, Pure Appl. Chem. 86, 191
(2014).
[32] V. Hlady and J. D. Andrade, Colloids Surf. 32, 359
(1988).
[33] R. D. Tilton, C. R. Robertson, and A. P. Gast, J. Colloid
Interface Sci. 137, 192 (1990).
[34] M. Rabe, D. Verdes, and S. Seeger, Soft Matter 5, 1039
(2009).
[35] T. J. Su, Lu, R. K. Thomas, Z. F. Cui, and J. Penfold,
J. Phys. Chem. B 102, 8100 (1998).
[36] M. Rabe, D. Verdes, J. Zimmermann, and S. Seeger, J.
Phys. Chem. B 112, 13971 (2008).
[37] H. Larsericsdotter, S. Oscarsson, and J. Buijs, J. Colloid
Interface Sci. 289, 26 (2005).
[38] C. E. Giacomelli and W. Norde, J. Colloid Interface Sci
233, 234 (2001).
[39] Y. I. Tarasevich and L. I. Monakhova, Colloid J. 64, 482
(2002).
[40] P. Roach, D. Farrar, and C. C. Perry, J. Am. Chem. Soc
127, 8168 (2005).
[41] M. Mondon, S. Berger, and C. Ziegler, Anal. Bioanal.
Chem 375, 849 (2003).
[42] H. Elwing, Biomaterials 19, 397 (1998).
[43] A. Tsargorodskaya, A. V. Nabok, and A. K. Ray, Nan-
otechnology 15, 703 (2004).
[44] A. Sauter, Crystallization and phase behavior of aqueous
beta-Lactoglobulin solutions in the presence of multivalent
cations, Diploma thesis (2011).
[45] C. Herzinger, B. Johs, W. McGahan, J. A. Woollam, and
W. Paulson, J. Appl. Phys 83, 3323 (1998).
[46] P. Schiebener, J. Straub, J. Levelt Sengers, and J. Gal-
lagher, J. Phys. Chem. Ref. Data 19, 677 (1990).
[47] M. Voinova, M. Jonson, and B. Kasemo, Biosens. Bio-
electron 17, 835 (2002).
[48] A. A. Feiler, A. Sahlholm, T. Sandberg, and K. D. Cald-
well, J. Colloid Interface Sci 315, 475 (2007).
[49] J. Webster, S. Holt, and R. Dalgliesh, Phys. B: Condens.
Matter , 1164 (2006).
[50] M. Skoda, F. Schreiber, R. Jacobs, J. Webster, M. Wolff,
R. Dahint, D. Schwendel, and M. Grunze, Langmuir 25,
4056 (2009).
[51] G. Yohannes, S. K. Wiedmer, M. Elomaa, M. Jussila,
V. Aseyev, and M. L. Riekkola, Ana. Chim. Acta 675,
191 (2010).
[52] G. Jackson, W. Chapman, and K. Gubbins, Mol. Phys.
65, 1 (1988).
[53] W. Chapman, G. Jackson, and K. Gubbins, Mol. Phys.
65, 1057 (1988).
[54] M. Wertheim, J. Stat. Phys. 35, 19 (1984).
[55] M. Wertheim, J. Stat. Phys. 35, 35 (1984).
[56] M. Wertheim, J. Stat. Phys. 42, 459 (1986).
[57] M. Wertheim, J. Stat. Phys. 42, 477 (1986).
[58] F. Romano, E. Sanz, and F. Sciortino, J. Chem. Phys.
132, 184501 (2010).
[59] J. Russo, J. Tavares, P. I. C. Teixeira, M. T. Da Gama,
and F. Sciortino, J. Chem. Phys. 135, 034501 (2011).
[60] A. B. Pawar and I. Kretzschmar, Macromol. Rapid Com-
mun. 31, 150 (2010).
[61] R. Evans, Adv. Phys. 28, 143 (1979).
[62] Y.-X. Yu and J. Wu, J. Chem. Phys. 116, 7094 (2002).
[63] Y. Rosenfeld, Phys. Rev. Lett. 63, 980 (1989).
[64] R. Roth, J. Phys.: Condens. Matter 22, 063102 (2010).
6
|
1104.0146 | 1 | 1104 | 2011-04-01T11:40:27 | A low-Reynolds-number treadmilling swimmer near a semi-infinite wall | [
"physics.bio-ph",
"nlin.CD",
"physics.flu-dyn"
] | We investigate the behavior of a treadmilling microswimmer in a two-dimensional unbounded domain with a semi-infinite no-slip wall. The wall can also be regarded as a probe or pipette inserted into the flow. We solve the governing evolution equations in an analytical form and numerically calculate trajectories of the swimmer for several different initial positions and orientations. We then compute the probability that the treadmilling organism can escape the vicinity of the wall. We find that many trajectories in a 'wedge' around the wall are likely to escape. This suggests that inserting a probe or pipette in a suspension of organism may push away treadmilling swimmers. | physics.bio-ph | physics |
A LOW-REYNOLDS-NUMBER TREADMILLING
SWIMMER NEAR A SEMI-INFINITE WALL
KIORI OBUSE∗ AND JEAN-LUC THIFFEAULT†
Abstract. We investigate the behavior of a treadmilling microswimmer in a two-
dimensional unbounded domain with a semi-infinite no-slip wall. The wall can also be
regarded as a probe or pipette inserted into the flow. We solve the governing evolution
equations in an analytical form and numerically calculate trajectories of the swimmer
for several different initial positions and orientations. We then compute the probability
that the treadmilling organism can escape the vicinity of the wall. We find that many
trajectories in a 'wedge' around the wall are likely to escape. This suggests that inserting
a probe or pipette in a suspension of organism may push away treadmilling swimmers.
1. Introduction. The locomotion of microorganisms is an active re-
search area of fluid dynamics and biology (see for instance the reviews
[10, 15]). As their motion occurs on very small length scales and speeds,
their dynamics is governed by low-Reynolds-number hydrodynamics, where
inertial forces are negligible in comparison to the viscous effects of the fluid
(Stokes flow).
Many studies deal with such dynamics in unbounded or very large do-
In reality, however, most organisms are in the vicinity
mains [1, 12, 18].
of other bodies or boundaries, where hydrodynamic interactions can have
a significant effect on their motion. The importance of boundaries has
also been suggested by experimental observations. For example, some re-
search suggests that microorganisms are attracted to solid walls [3, 17, 19].
Berke et al. [2] measured the steady-state distribution of E. Coli bacte-
ria swimming between two glass plates and found a strong increase of the
cell concentration at the boundaries. They also theoretically demonstrated
that hydrodynamic interactions of swimming cells with solid surfaces lead
to their reorientation in the direction parallel to the surfaces. Lauga et
al. [9] showed that circular trajectories are natural consequences of force-
free and torque-free swimming and hydrodynamic interactions with the
boundary. This leads to a hydrodynamic trapping of the cells close to the
surface. Drescher et al. [6] found that when two nearby Volvox colonies
swim close to a solid surface, they attract one another and can form stable
bound states in which they 'waltz' or 'minuet' around each other. These ob-
servations suggest that, in order to obtain a comprehensive understanding
of low-Reynolds-number locomotion, it is necessary to study hydrodynamic
interactions between microorganisms and boundaries.
Some of the phenomena stated above have already been confirmed
by numerical simulations [8, 16]. However, not many physical explana-
∗Research Institute for Mathematical Sciences, Kyoto University, Kyoto, 606-8502,
†Dept. of Mathematics, University of Wisconsin, Madison, WI 53706, USA. The work
Japan.
of the second author was supported in part by NSF grant DMS-0806821.
1
2
KIORI OBUSE AND JEAN-LUC THIFFEAULT
tions have been given to the locomotion of microorganism near boundaries.
Berke et al. [2] have captured the swimming microorganisms' attraction to
boundaries by modeling the swimmer as a force dipole singularity. However,
contrary to the experimental findings, the microorganism in this model
crashes into the boundary in finite time. Or and Murray [14] studied the
dynamics of low-Reynolds-number swimming organism near a plane wall.
They analyzed the motion of a swimmer consisting of two rotating spheres
connected by a thin rod as a simple theoretical model of a 'treadmilling'
swimming organism. They found that when the spheres rotate with dif-
ferent velocities their model has a solution with steady translation parallel
to the wall, and that under small perturbation the swimmer exhibits a
'bouncing' motions parallel to the wall. These results have recently been
verified experimentally on a macroscale robotic prototype swimming in a
highly viscous fluid [20]. Furthermore, Crowdy and Or [4] have proposed
a singularity model for swimming microorganisms placed near an infinite
no-slip boundary. Their model was based on a circular treadmilling organ-
ism which has no means of self-propulsion, that is, the organism doesn't
move unless it interacts with a boundary.
(For example, the organism
may be creating a feeding current.) They proposed an appropriate Stokes
singularities that represent the flow field created by this treadmilling or-
ganism. By studying the interaction between these singularities and the
no-slip wall, they formulated explicit evolution equations for the motion
of the organism, and fully characterized its motion near the wall. They
found trajectories with a periodic bouncing motion along the wall which
had remarkable similarity to the trajectories shown in Or and Murray [14].
Crowdy and Samson [5], using the point-singularity model, investigated the
dynamics of treadmilling organism near an infinite no-slip boundary with a
gap of a fixed size. They employed a conformal mapping technique to avoid
the difficulty in treating the image of the treadmilling organism on the wall.
They found that the treadmilling organism can exhibit several qualitatively
different types of trajectories: jumping over the gap, rebounding from the
gap, being trapped near the gap, and escaping the gap region even when the
organism has initial position in the gap. They also performed a bifurcation
analysis in terms of the model parameters, and demonstrated the presence
of stable equilibrium points in the gap region as well as Hopf bifurcations
to periodic bound states. This reduced model also exhibited a global glu-
ing bifurcation in which two symmetric periodic orbits merge at a saddle
point into symmetric bound states having more complex spatio-temporal
structure.
In the present paper we examine the dynamics of a treadmilling or-
ganism near a semi-infinite no-slip wall, modeled as a flat plate of zero
thickness. Though this is a special case of the model of Crowdy and Sam-
son [5], it deserves separate investigation because of the simpler equations
involved, and because the semi-infinite wall can be regarded as a probe or
pipette inserted in the system, a common situation in microbiology. We
LOW-REYNOLDS-NUMBER SWIMMER NEAR A SEMI-INFINITE WALL
3
also analyze the trajectories in a very different manner to [5], as we attempt
to quantify the probability of escape from the wall's vicinity, assuming the
treadmillers are randomly oriented.
2. Model of a treadmilling microorganism. Following previous
authors [4, 5], we consider a two-dimensional model for a microorganism in
the (x, y)-plane, which we treat as the complex plane with z ≡ x + iy. Our
derivation is a special case of [5], who considered a microswimmer near a slit
or gap in an infinite wall. Nevertheless, as mentioned in the introduction,
the semi-infinite wall is important enough to be treated separately, as it
arises in the neighborhood of a probe or a pipette.
The Stokes equations which describe the motion of an incompressible
viscous fluid are
∇p = η∆u,
∇ · u = 0,
(2.1)
where ∆ is the Laplace operator, u = (ux, uy) is the fluid velocity, and
p and η are the pressure and dynamic viscosity, respectively. As we are
considering a two-dimensional flow, we can introduce a stream function ψ,
such that the velocity is given by ux = ∂ψ/∂y, uy = −∂ψ/∂x. Then the
Stokes equations (2.1) reduce to the biharmonic equation ∆2ψ = 0. The
complex velocity is W = ux + iuy = −2i∂ψ/∂ ¯z, with
W = ux + iuy = −2i
ψ = Im[¯zf (z) + g(z)].
∂ψ
∂ ¯z
= f (z) + zf(cid:48)(z) + g(cid:48)(z),
(2.2)
(2.3)
where f (z) and g(z) are called Goursat functions; they are analytic every-
where in the flow domain, except where isolated singularities are introduced
to model swimmers. For a treadmilling swimmer, we take f (z) to have a
simple pole at z = zd, so that
µ
(cid:48)(z) =
g
µ¯zd
f (z) =
+ . . . ,
z − zd
(z − zd)2 + . . . .
(2.4)
where µ ∈ C and the form of g(cid:48) is forced by the requirement that the
complex velocity (2.2) have no higher than a z − zd−1 singularity. This
solution corresponds to a stresslet of strength µ at zd. The ellipses in (2.4)
indicate analytic terms. The expansion (2.4) is the basic solution for a
treadmilling swimmer, which does not have any self-propulsion in itself,
but can move due to its interaction with boundaries [4, 5, 11, 14, 20].
In a simple model, we assume that the treadmilling organism has a
circular body of radius , with a center at zd(t) = xd(t) + iyd(t). We also
assume that, with respect to the angle θ(t) of the head of the treadmilling
organism from the real axis, surface actuators of the treadmilling organism
induce a tangential velocity profile given by [4]
U (φ, θ) = 2V sin(2(φ − θ)),
(2.5)
4
KIORI OBUSE AND JEAN-LUC THIFFEAULT
where V is a constant and φ is the angle measured from the positive x
direction.
Next consider the treadmiller near an infinite wall along the x axis. To
satisfy the no-slip boundary condition at the wall, we make the expansion
f (z, t) =
(cid:48)(z, t) =
g
µ
z − zd(t)
(z − zd(t))3 +
+ f0 + (z − zd(t))f1 + ··· ,
(z − zd(t))2 + g0 + ··· ,
a
b
(2.6)
f (z, t) having no Stokeslet term and g(z, t) having no rotlet term implying
that the treadmilling organism is force-free and torque-free. We use the
boundary condition to find
µ = −ic,
a = µ¯zd,
(2.7)
where c(t) ≡ −iV exp(−2iθ(t)). We set the time scale of the motion by
letting V = −1 so that µ(t) = exp(2iθ(t)). The coefficients f0, f1, and g0
are given in [4].
b = µ2 − ic3 = 2µ2,
The time derivative of the position and orientation of the treadmilling
organism is obtained by equating the time rate of change of position to the
local fluid velocity, and the time rate of change of orientation to half the
local vorticity:
= −f0 + zd
¯f1 + ¯g0,
dzd
dt
= −2 Im f1.
dθ
dt
(2.8)
Equation (2.8) can then be solved as a set of three ODEs determining the
motion of the treadmiller.
Now we turn to a semi-infinite wall, extending along the negative x
axis. The conformal mapping ζ = iz1/2 maps the z plane to upper-half ζ
plane, with the negative x-axis of the z plane mapped to the real axis in
the ζ plane. In the ζ plane we can use a similar singular expansion as (2.6),
but we must take care to map the boundary conditions to the ζ plane. We
omit the lengthy details, which are similar to Crowdy and Samson [5].
See [13] for a more complete derivation. All that is required for simulating
the swimmer trajectories are the coefficients that appear in (2.8), which
are
(cid:0)2zd2 − 2¯z2
8Z 2 ¯z2
d
8Z ¯z5/2
+
d − 32(cid:1) ¯µ
(cid:0)2zd2 − 2¯z2
(cid:0)2zd2 + 2¯z2
(cid:0)2zd2 + 2¯z2
d − 32(cid:1) ¯µ
d − 32(cid:1) ¯µ
2Z 3 ¯z2
d
d
4Z 2 ¯z5/2
d
− 3z3/2
d
d − 32(cid:1) ¯µ
,
(2.9a)
(cid:19)
,
(2.9b)
f0 =
µ
4zd
− 2 ¯µ
4Z 3 ¯z3/2
d
+
(cid:18)
f1 =
1
12z2
d
− 3µ
4
+
d
9z3/2
2 ¯µ
2Z 4 ¯z3/2
d
− 3z3/2
d
LOW-REYNOLDS-NUMBER SWIMMER NEAR A SEMI-INFINITE WALL
5
Fig. 1. Examples of trajectories from the initial points (xd0, yd0) marked with
circles, with an initial orientation θ0. A treadmilling organism can escape from the
wall, such as in (a) and (d), end up above the wall as in (b), or underneath the wall as
in (c).
and
g0 = − 3µ¯zd
16z2
d
+
102µ
32z3
d
+
where Z ≡ z1/2
d + ¯z1/2
d
+
.
16Z 2 ¯z5/2
d
32 ¯µ
8Z 4 ¯zd
− (zd − ¯zd) ¯µ
(cid:0)2zd2 − 6¯z2
4Z 3 ¯z1/2
d − 32(cid:1) ¯µ
d
(cid:16)
d + Z(cid:17)
¯z1/2
,
(2.9c)
3. Results of numerical simulations. We now present the results
of numerical simulations of the governing evolution equations (2.8), to-
gether with the coefficients (2.9). The radius of the circular treadmillers
is set to = 1, giving the reference length scale. The time scale is set
by V = −1 in (2.5).
Figure 1 shows examples of swimmer trajectories for different initial
conditions for position zd0 and angle θ0. Some trajectories, such as (a) and
(d), end up far away from the wall; we refer to those as escaping trajectories.
Others, such as (b) and (c), remain close to the wall for all time, exhibiting
the 'bouncing' behavior noted in [4]. A fourth type of trajectory (not
6
KIORI OBUSE AND JEAN-LUC THIFFEAULT
(a)
(b)
Fig. 2. (a) The x -- y plane divided into five regions. The thick black line corresponds
to the wall. (b) An example of a pie chart. The different colors as a function of initial
orientation θ0 indicate which region of (a) the treadmilling organism ends up in after
a sufficiently large time.
shown) crashes into the wall, but this is an unphysical consequence of the
boundary condition at the swimmer's surface only being approximately
satisfied. It has been verified numerically that the qualitative features of
the trajectories are not affected by this approximation [7].
Both experimental observations and previous theoretical studies sug-
gest that, when there is a no-slip wall near a treadmilling organism, the
organism tends to be attracted to its own image and move towards the
wall [2, 4,5, 8, 9, 16,17, 19]. This behavior is clearly seen at an early stage in
all the trajectories in Fig. 1. Nevertheless, in the cases shown in Figs. 1(a)
and (d), the treadmilling organism moves away from the wall after it has
come close to the edge of the wall.
We now focus on the large-time behavior of treadmillers. To study
this, we first introduce five regions shown in Fig. 2(a). Regions 1 -- 4 are the
usual quadrants of the plane, but with a neighborhood of size 0.2 around
the wall removed; this removed neighborhood is region 5. The regions
1 -- 5 are colored red, blue, orange, green, and white, respectively. From
these regions, we can make a 'pie chart' around each point in the plane,
an example of which is shown in Fig. 2(b). The sectors of the pie chart
correspond to a range initial angle θ0; the color of the sector describes which
region the treadmiller ends up in for large times (here t = 1500). Region
5 corresponds to 'crashing' trajectories. Thus, for a given position, the
treadmiller may end up in different regions, and the relationship between
angle and final region is not simple.
Figure 3 shows pie charts for many initial points (xd0, yd0). The most
notable feature is the complexity of the pie charts for initial points near
the wall and with large negative x coordinates. This reflects the fact that
the treadmilling organism changes its heading direction and the quality of
123450δ=0.2εxy LOW-REYNOLDS-NUMBER SWIMMER NEAR A SEMI-INFINITE WALL
7
Fig. 3. Pie charts at t = 1500 for several initial conditions. Each pie chart is
centered on the initial condition it corresponds to. The thick black line represents the
wall. Note that region 3 only appears as very thin slivers in the pie charts along the x
axis.
its trajectory significantly when it has come to the vicinity of the edge of
the wall, x = y = 0 depending upon its (xd, yd, θ) at the time, and so even
a very small difference in initial condition can result in a huge difference
to its trajectory. (This is an example of chaotic scattering.) However, pie
charts tend to become rather simple for larger yd0 for any fixed xd0, since
the treadmiller is then less influenced by the wall.
Now let us consider the escape probability, PE(xd0, yd0), the probability
that the treadmilling organism can escape from the wall region. We define
this as the probability that the treadmilling organism be in region 1 or 4
at sufficiently large time. We assume that the initial angle θ0 is uniformly
distributed. The escape probability PE(xd0, yd0) then corresponds to the
fraction of angles in a pie chart that are red or green, for a given initial
position (xd0, yd0). The escape probability for t = 1500 for different initial
conditions is shown in Fig. 4, where each (x, y) coordinate corresponds to an
initial position of the treadmiller. Note that for almost all initial conditions
there is some finite probability of escaping or being trapped, though the
escaping probability approaches unity along the positive x axis, and goes
to zero for large negative x and large y. Observe the strong 'wedge' of
trajectories that are likely to escape (PE > 0.6), though there is also a
backwards-facing wedge of trajectories that have a reasonable change of
escaping (PE > 0.3). This suggests that a probe or pipette inserted in a
medium is likely to 'push away' treadmillers to some degree. Note also the
'tongues' of abnormally-high (PE (cid:39) 0.4) escape probability near the wall,
8
KIORI OBUSE AND JEAN-LUC THIFFEAULT
Fig. 4. Escape probability PE at t = 1500 for different initial points. Each point
corresponds to a given initial condition (xd0, yd0), integrated over all possible starting
angles θ0. A trajectory escapes if it ends up in regions 1 or 4, as defined in Fig. 2. The
excluded white area near the wall corresponds to initial conditions where the swimmer
would be partially inside the wall.
which reflects the complicated structure of the pie charts in that region.
4. Discussion and conclusions. Following [4,5], we derived the gov-
erning evolution equations for a treadmilling microorganism near a semi-
infinite no-slip wall. This can also be regarded as a two-dimensional model
of a probe or pipette inserted in the fluid. We then numerically calculated
the trajectories of the organism for different initial conditions. The tread-
milling organism was usually attracted to the wall for early times, but for
later times often escaped the wall region. Typically this happened when
the organism came close to the tip of the wall. To investigate this behavior
further, we looked at 'pie chart' diagrams, where the sectors indicate which
region a swimmer eventually ends up in as a function of its initial angular
orientation.
Initial points with larger negative x coordinates have more
complex pie charts compared to those with smaller negative x coordinates,
because they are sensitive to 'chaotic scattering' off the tip of the wall.
We then examined the escape probability PE, the probability that the
treadmilling organism can escape to the right of the wall region, assuming
that it is initially randomly oriented. There is an evident 'wedge' of initial
conditions that is likely to escape the wall. This suggests that a probe or
pipette inserted could 'push' the organisms out of the way (with the wedge
replaced by a cone in three dimensions), though since the organisms slow
down as they get further away from the wall the effect might not be very
pronounced.
Acknowledgments. The authors are grateful for the hospitality of
the Geophysical Fluid Dynamics Program at the Woods Hole Oceano-
graphic Institution (supported by NSF), and thank Matthew D. Finn for
his helpful advice and suggestions. Some of the numerical calculations for
xy −20−15−10−505101520051015200.20.40.60.8 LOW-REYNOLDS-NUMBER SWIMMER NEAR A SEMI-INFINITE WALL
9
this project were performed at the Institute for Information Management
and Communication of Kyoto University.
REFERENCES
[1] J. E. Avron, O. Kenneth, and D. H. Oakmin, Pushmepullyou: an efficient
micro-swimmer, New J. Phys., 7 (2005), p. 234.
[2] A. P. Berke, L. Turner, H. C. Berg, and E. Lauga, Hydrodynamic attraction
of swimming microorganisms by surfaces, Phys. Rev. E, 101 (2008), p. 038102.
[3] J. Cosson, P. Huitorel, and C. Gagnon, How spermatozoa come to be confined
to surfaces, Cell Motil. Cytoskel., 54 (2003), pp. 56 -- 63.
[4] D. G. Crowdy and Y. Or, Two-dimensional point singularity model of a low-
Reynolds-number swimmer near a wall, Phys. Rev. E, 81 (2010), p. 036313.
[5] D. G. Crowdy and O. Samson, Hydrodynamic bound states of a low-reynolds-
number swimmer near a gap in a wall, (2010). Preprint.
[6] K. Drescher, K. Leptos, I. Tuval, T. Ishikawa, T. J. Pedley, and R. E.
Goldstein, Dancing volvox: hydrodynamic bound states of swimming algae,
Phys. Rev. Lett., 102 (2009), p. 168101.
[7] M. D. Finn. Private communication.
[8] J. P. Hernandez-Ortiz, C. G. Dtolz, and M. D. Graham, Transport and col-
lective dynamics in suspensions of confined swimming particles, Phys. Rev.
Lett., 95 (2005), p. 204501.
[9] E. Lauga, W. R. DiLuzio, G. M. Whitesides, and H. A. Stone, Swimming
in circles: motion of bacteria near solid boundaries, Biophys. J., 90 (2006),
pp. 400 -- 412.
[10] E. Lauga and T. R. Powers, The hydrodynamics of swimming micro-organisms,
Rep. Prog. Phys., 72 (2009), p. 096601.
[11] A. M. Leshansky, O. Kenneth, O. Gat, and J. E. Avron, A frictionless mi-
croswimmer, New J. Phys., 9 (2007), p. 145.
[12] A. Najafi and R. Golestanian, Simple swimmer at low Reynolds number: Three
linked spheres, Phys. Rev. E, 69 (2004), p. 062901.
[13] K. Obuse, Trajectories of a low Reynolds number treadmilling organism near a
half-infinite no-slip wall, in Proceedings of the 2010 Summer Program in Geo-
physical Fluid Dynamics, Woods Hole, MA, 2010, Woods Hole Oceanographic
Institute.
[14] Y. Or and R. M. Murray, Dynamics and stability of a class of low reynolds
number swimmers near a wall, Phys. Rev. E, 79 (2009), p. 045302.
[15] T. J. Pedley and J. O. Kessler, Hydrodynamic phenomena in suspensions of
swimming microorganisms, Annu. Rev. Fluid Mech., 24 (1992), pp. 313 -- 358.
[16] M. Ramia, D. L. Tullock, and N. Phan-Thien, The role of hydrodynamics in-
teraction in the locomotion of microorganism, Biophys. J., 65 (1993), pp. 755 --
778.
[17] A. J. Rothschild, Non-random distribution of bull spermatozoa in a drop of sperm
suspension, Nature, 198 (1963), pp. 1221 -- 1222.
[18] A. Shapere and F. Wilczek, Geometry of self-propulsion at low Reynolds num-
bers, J. Fluid Mech., 198 (1989), pp. 557 -- 585.
[19] H. Winet, G. S. Bernstein, and J. Head, Observations on the response of human
spermatozoa to gravity, boundaries and fluid shear, J. Reprod. Fert., 70 (1984),
pp. 511 -- 523.
[20] S. Zhang, Y. Or, and R. M. Murray, Experimental demonstration of the dy-
namics and stability of a low reynolds number swimmer near a plane wall, in
Proc. Amer. Cont. Conf., 2010, pp. 4205 -- 4210.
|
1104.1337 | 1 | 1104 | 2011-04-07T14:10:10 | An elastic-network based local molecular field analysis of zinc-finger proteins | [
"physics.bio-ph"
] | We study two designed and one natural zinc-finger peptide each with the Cys2His2 (CCHH) type of metal binding motif. In the approach we have developed, we describe the role of the protein and solvent outside the Zn(II)-CCHH metal-residue cluster by a molecular field represented by generalized harmonic restraints. The strength of the field is adjusted to reproduce the binding energy distribution of the metal with the cluster obtained in a reference all-atom simulation with empirical potentials. The quadratic field allows us to investigate analytically the protein restraints on the binding site in terms of its eigenmodes. Examining these eigenmodes suggests, consistent with experimental observations, the importance of the first histidine (H) in the CCHH cluster in metal binding. Further, the eigenvalues corresponding to these modes also indicate that the designed proteins form a tighter complex with the metal. We find that the bulk protein and solvent response tends to destabilize metal-binding, emphasizing that the favorable energetics of metal-residue interaction is necessary to drive folding in this system. The representation of the bulk protein and solvent response by a local field allows us to perform Monte Carlo simulations of the metal-residue cluster using quantum-chemical approaches, here using a semi-empirical Hamiltonian. For configurations sampled from this simulation, we study the free energy of replacing Zn(II) with Fe(II), Co(II), and Ni(II) using density functional theory. The calculated selectivities are in fair agreement with experimental results. | physics.bio-ph | physics |
An elastic-network based local molecular field analysis of zinc-finger proteins
Chemical and Biomolecular Engineering, Johns Hopkins University, Baltimore, MD 21218
Purushottam D. Dixit and D. Asthagiri∗
We study two designed and one natural zinc-finger peptide each with the Cys2His2 (CCHH) type
of metal binding motif.
In the approach we have developed, we describe the role of the protein
and solvent outside the Zn(II)-CCHH metal-residue cluster by a molecular field represented by
generalized harmonic restraints. The strength of the field is adjusted to reproduce the binding
energy distribution of the metal with the cluster obtained in a reference all-atom simulation with
empirical potentials. The quadratic field allows us to investigate analytically the protein restraints
on the binding site in terms of its eigenmodes. Examining these eigenmodes suggests, consistent
with experimental observations, the importance of the first histidine (H) in the CCHH cluster
in metal binding. Further, the eigenvalues corresponding to these modes also indicate that the
designed proteins form a tighter complex with the metal. We find that the bulk protein and solvent
response tends to destabilize metal-binding, emphasizing that the favorable energetics of metal-
residue interaction is necessary to drive folding in this system. The representation of the bulk protein
and solvent response by a local field allows us to perform Monte Carlo simulations of the metal-
residue cluster using quantum-chemical approaches, here using a semi-empirical Hamiltonian. For
configurations sampled from this simulation, we study the free energy of replacing Zn2+ with Fe2+,
Co2+, and Ni2+ using density functional theory. The calculated selectivities are in fair agreement
with experimental results.
I.
INTRODUCTION
An important post-translational modification of pro-
teins involves incorporating metal ions into the protein
structure [1]. In many of these instances, metal-binding
stabilizes the folded structure or helps fold a previously
unstructured or partially structured peptide. The ver-
satility of metals as stabilizers or modifiers of protein
structure is principally due to the nature of their inter-
actions with amino acid residues: substantially strong on
a thermal energy scale and chemically intricate [2].
The prevalence of metalloproteins and the growing ap-
preciation of metal-induced (mis)folding in diseases, for
example, see Refs. [3–5], makes obtaining a molecular
level understanding of the role of metals in protein struc-
ture and function of unquestionable importance. But
the intricacies of metal-protein interactions makes this
a formidable challenge to current theory and simulation
approaches. In a step towards this larger challenge, here
we address the role of material outside the first-shell of
the metal in metal-binding and selectivity in a zinc-finger
protein.
A satisfactory description of metal-protein interactions
requires quantum chemical calculations, and these calcu-
lations, especially at a high-level of theory, are compu-
tationally demanding. Hence quantum chemical calcula-
tions are limited to a small group of residues surround-
ing the metal ion. The effect of the remaining protein
and solvent medium on the structure and dynamics of
the metal-residue cluster is typically described in one of
three ways (for example, see Refs. [6–9]): the medium
is entirely ignored, effectively simulating the cluster in
∗ Corresponding author email: [email protected]
vacuo; the medium is described as a continuum with
a dielectric constant; and in the most sophisticated of
methods, the medium is described using empirical force-
fields. Among these alternatives, only the last method
explicitly accounts for the role of the bulk in modulating
the architecture and dynamics of the cluster.
The approach we present is a rigorous reduction in the
degrees of freedom of the system and has the aim of un-
derstanding the role of the medium outside the defined
metal-residue cluster: the cluster is described in atomic
detail and its structure and dynamics are influenced by
the medium whose effect is described by a local molec-
ular field. The present approach is inspired by a recent
development in the theory of liquids [10]. The essential
idea in that development is to describe the role of the
medium external to a defined inner-shell [10] around a
solute by a molecular field whose strength is adjusted to
satisfy suitable consistency conditions, such as the mean
density of the inner-shell.
Here we obtain the local molecular field by describing
the bulk protein outside the cluster as an elastic medium
[11–19]. This development allows us to separate the sys-
tem Hamiltonian into that for the metal-residue cluster
and the remainder. We then integrate out the bulk de-
grees of freedom to obtain a molecular field acting on
the cluster. The strength of the field is adjusted self-
consistently such that the binding energy distribution of
the metal with the local residues reproduces the binding
energy distribution obtained in a simulation treating the
bulk atomically as well. (We focus on the binding en-
ergy distribution, as this is the most relevant quantity
for understanding the thermodynamics of metal binding
to the protein [20–23].)
In this initial study, the ref-
erence all-atom simulation and the simulations to de-
termine the strength of the molecular field are all per-
formed using empirical forcefields. With the local molec-
ular field determined at the coarse level, we study the
metal-residue cluster using higher level methods, here us-
ing semi-empirical and density functional methods with
large basis sets.
We use the local molecular field approach to study
Zn2+ binding to a zinc finger domain. Zinc finger do-
mains are widely distributed in cellular systems, most
notably in the transcription factor assembly [24–26]. The
isolated zinc-finger domain is partially unstructured in
the absence of the metal and achieves the correct folded
conformation only upon binding the metal [27–29].
In
the folded state, Zn2+ is coordinated tetrahedrally to n-
cysteine and 4-n histidine residues where n can be 2, 3
or 4. In the system we study, n is 2.
The rest of the article is organized as follows. In Sec. II
we derive the local molecular field. Section III collects de-
tails of the molecular simulations. Discussions and con-
clusions follow results presented in Sec. IV.
II. THEORY
Consider the metal-bound protein in a solvent medium.
We denote the conformational degrees of the protein by
X = (X1, X2), where X1 is the metal plus the neighbor-
ing amino acid residues and X2 is the remainder of the
protein. The solvent degrees are given by Xs.
In the canonical ensemble, for a given protein coordi-
nate X , we can formally integrate over the solvent de-
grees of freedom and write the effective potential on the
protein - the potential of mean force - as
2
attempts to describe the viscous damping of protein oscil-
lations as harmonic fluctuations [16], but explicit solvent-
binding site interactions are neglected. We anticipate
that these long-range contributions can be described us-
ing mean-field models (such as dielectric models) and
that they will not contribute to discriminating between
Zn2+ and a competing metal bound at the site.
Our plan is to obtain the quadratic Hamiltonian, H,
using contact topology based potentials. Since these
models are valid only up to a proportionality con-
stant [11–16, 19], to construct a physical potential energy
function, we introduce a coupling constant, α, that fixes
the strength of the harmonic interaction.
The matrix H, suppressing the dependence on β for
simplicity, is expanded as
H = (cid:20) H1 G
GT H2 (cid:21) .
(4)
Here H1 is a diagonal matrix: no two site-atoms couple
through H1 as those interactions are explicitly described
in U1. (In this regard, the present development differs
from those presented earlier [11, 13, 31].) The matrix G
couples the site atoms to the bulk, and the matrix H2
couples the bulk atoms with each other.
Under the approximations noted above, for the sys-
tem modeled by U (X1, X2; β), the excess Helmholtz free
energy, Aex, is given by
e−βAex
= Z e−βU (X1,X2;β)dX1dX2.
(5)
U(X1, X2; β) = U1(X1) + U12(X1, X2) + U2(X2) + η(X1, X2; β)(1)
Since the bulk protein coordinates X2 appear quadrati-
cally, following [31], we integrate over X2 and get
U1(X1), U12(X1, X2), and U2(X2) are site-site, site-bulk,
and bulk-bulk interactions and η(X1, X2; β) is the solvent
response. (Note that, in principle, such a decomposition
can always be made.) β = 1/kBT , where T is the tem-
perature and kB is the Boltzmann constant. We indicate
the temperature dependence of η, a thermally averaged
quantity, to emphasize the distinction from the potentials
U1, U2, and U12.
In QM/MM approaches [7–9, 30], U1 is described quan-
tum mechanically and U2 and U12 are described using
molecular mechanics. Here we approximate the latter
two quantities together with the response of the solvent
by generalized harmonic restraints around the equilib-
rium structure of the protein. Our ansatz is
U(X1, X2; β) ≈ U1(X1) + [δX1, δX2] αH(β)(cid:20) δX1
δX2 (cid:21) (2)
where
δXi = Xi − Xi,p for i = 1, 2.
(3)
Here, Xi,p for i = 1, 2 are the reference coordinates for
the binding site and the protein medium obtained from
the three dimensional structure of the protein. The ap-
proximation made in Eq. 2 is physically motivated and
e−βAex
1 αHsiteδX1)
= Z e−β(U1(X1)+δX T
≡ Z e−βU (X1;β)dX1
where
Hsite = H1 − GH−1
2 GT .
(6)
(7)
Thus, under the approximations noted above, the effec-
tive potential for the site is given by
U(X1; β) = U1(X1) + δX T
1 αHsiteδX1
= U1(X1) + φm(X1; α),
(8)
where in addition to the site-site interaction potential,
U1(X1), the effective potential contains a local molecu-
lar field, φm(X1; α) = δX T
1 αHsiteδX1, that describes the
effect of the bulk protein and solvent damping on the
site atoms. (For notational simplicity, the temperature
dependence on φm is not explicitly shown.) The field
φm acts as a restraint that limits the deflections of the
binding site away from some reference state. (A suitable
reference state can be the PDB structure.) For α = 0,
there is no coupling between the site and the bulk, and
the model reduces to a metal-residue cluster in vacuum.
A. Selectivity of the binding site
The selectivity of the zinc finger peptide for Zn2+ over
another transition metal X2+ (X2+ = Co2+, Fe2+, and
Ni2+) is determined by
X2+ − µex
Zn2+ ](S) − [µex
∆µex = [µex
X2+ − µex
Zn2+ ](aq)
= ∆µex(S) − ∆µex(aq) ,
(9)
where µex(aq) is the change in hydration free energies
and µex(S) is the corresponding quantity in the protein.
It is understood that a common reference state is used in
defining µex
X2+ in water and in the protein [20, 22].
Calculating ∆µex(S) presents significant challenges, in-
cluding the need to describe interactions quantum me-
chanically together with sampling binding site conforma-
tions. It is here that the reduction in degrees of freedom
made possible by the effective potential (Eq. 8) proves
helpful.
Following Eq. 6, ∆µex(S) is given by [20, 21, 32]
e−β∆µex(S) = he−β∆U1(X1) · e−β∆φmiZn2+
≈ he−β∆U1(X1)iZn2+ ,
(10)
where UX2+ (X1; β) is the effective potential (Eq. 8) with
X2+ in the site, ∆U1 = U1,X2+ − U1,Zn2+ , ∆φm =
φm,X2+ − φmZn2+ , and h. . .iZn2+ indicates canonical av-
eraging with Zn2+ bound to the protein. In Eq. 10, by
ignoring ∆φm we are assuming that the response of the
material outside the defined metal-residue cluster is the
same for both X2+ and Zn2+ and thus the contribution to
the selectivity free energy arises solely from interactions
within the binding site. (We comment on this approxi-
mation below.)
B. Coupling constant α
The excess chemical potential, µex
Zn2+ , is given by the
potential distribution theorem [21, 32, 33]
µex
Zn2+ = kBT log(cid:20)Z ∞
−∞
eβεP (ε)dε(cid:21) ,
(11)
where P (ε) is the probability distribution of the binding
energy of the metal ion with the surrounding material.
Since µex
Zn2+ is the essential quantity characterizing the
thermodynamics of Zn2+ binding to the site, we choose
α such that the energy distribution from the approxi-
mate model (Eq. 8) reproduces the binding energy dis-
tribution of Zn2+ with the site from all-atom molecular
simulations.
It is most economical computationally, if
these simulations are based either on an empirical po-
tential energy function or a QM/MM approach with a
less expensive quantum mechanical model. (Here we use
an empirical potential energy function.) Then with the
molecular field based on a coarse energy function, one can
refine the description of the site with high-level quantum
mechanical calculations in the presence of the molecular
field.
3
To obtain α, we minimize the Kullback-Liebler di-
vergence [34] between the distribution, Ps(ε), obtained
with simulation using a coarse model, and the distribu-
tion, P (ε; α), obtained from simulations of the site with
the effective potential (Eq. 8). Due to the exponential
weighting eβε, the high-ε tail of the energy distribution
P (ε) more sensitively determines the excess free energy
in Eq. 11. Hence we require simulations with the model
Hamiltonian (Eq. 8) to reproduce the high-ε tail. Thus,
we restrict our attention to energy values ε ≥ ¯ε, where
the mean binding energy is ¯ε. α is then obtained by
minimizing
∆(α) = logZ ∞
¯ε
Ps(ε) log
Ps(ε)
P (ε; α)
dε.
(12)
III. METHODS
A. Elastic network model
We construct the quadratic Hamiltonan (H) based on
the Gaussian network model described in Ref. [16]. The
interaction energy is expanded around the equilibrium
structure of the protein in a Taylor series and only terms
to second order are retained.
Denoting the equilibrium distance between particle i
and j as rij , the deviated distance by dij , and the devi-
ation by xij (= dij − rij ), we have
U (dij ) ≈ U (rij ) +
U ′′(rij )
2 Xµ,ν
ij
rµ
ij rν
r2
ij
xµ
ij xν
ij .
(13)
µ,ν are the labels for Cartesian components x, y, and
z. U ′′(rij ) is the second derivative of the potential en-
ergy function. For simplicity, we assume that the sec-
ond derivative is equal for all the interactions and that
only particles within 6 A of each other interact. This
cutoff value follows earlier studies using elastic network
models and other potential functions dependent on topo-
logical contact maps [11–16, 19]. We do not include U1
type interactions in constructing H as those interactions
are dealt in atomic detail. Further, the inversion of H2
(Eq. 7) is defined only within the space of eigenvectors
with non-zero eigenvalues; the remaining six zero-modes
correspond to rotations and translations and do not con-
tribute to the potential energy.
B. Molecular dynamics simulation
We consider three different zinc finger proteins to eval-
uate our model: the consensus peptide (CP1) ([29]; PDB
ID: 1MEY), the peptide (YTA) based on the human
zinc finger protein 32 (PDB ID: 2YTA), and TF3, and
the transcription factor IIIA ([35]; PDB ID: 1TF3). Of
these CP1 and YTA are designed peptides. Residues
between 139 and 162 forming the zinc-finger domain in
4
energy function that we have implemented in a collection
of Fortran codes. To calculate the field contribution in
Eq. 8, we find the deflection of a configuration from the
protein reference structure, δX1 = X1 − X1,p , and obtain
φm(X1; α) form the known matrix Hsite using
φm(X1; β) = δX1αHsiteδX1 .
For each α, the simulations consist of 15×106 sweeps of
equilibration followed by 15 × 106 sweeps of production.
Every sweep comprises three sets of moves: 1) displace-
ment of each particle; 2) rigid-body displacement of each
amino acid residue; and 3) rigid-body rotation of each
amino acid residue. Each of these moves is performed
with a probability of 0.5. Displacements are made along
randomly chosen directions and rotations are made along
each axis using randomly chosen angles. The standard
Metropolis criterion is then used to accept the sweep. In
the equilibration phase, the maximum displacements and
rotation are adjusted such that the acceptance ratio of
sweeps stabilizes at ≈ 0.25. (These maximum values are
retained without further adjustments in the production
phase.) Configurations are stored every 500 sweeps for
further analysis.
D. Monte Carlo simulations with semi-empirical
potentials
Once we have the local molecular field, we can, in prin-
ciple, examine the site with Monte Carlo or molecular dy-
namics at a high level of theory. However, even for 59 par-
ticles, our initial attempts at using B3LYP/TZV(2d+p)
in the Monte Carlo scheme proved intractable, as ex-
cessively long times were deemed necessary for adequate
convergence. For this reason, in this initial study, we
resorted to Monte Carlo simulation of the binding site
using the semi-empirical PM3 [45, 46] model in the Gaus-
sian09 package [47]. The simulations are performed with
the molecular field and the optimal value of α determined
with the coarse potential. The equilibration and produc-
tion comprises 50 × 103 and 25 × 103 sweeps respectively.
Hydrogen atoms are added on the fly to satisfy valencies
of C and N atoms. In the equilibration phase, the accep-
tance ratio is 0.25. Configurations are saved after every
5 sweeps.
To compute the free energy change (Eq. 10) in replac-
ing Zn2+ with X2+(= Co2+, Fe2+, Ni2+), we assume that
the configurations generated with the PM3 hamiltonian
well-represent the configurations that would be produced
with the B3LYP/TZV(2d+p) method, and then take
150 equally spaced configurations from the production
phase of the PM3-based simulation. For each configu-
ration, the energy change in exchanging Zn2+ (Eq. 10),
∆U1 = U1,X2+ − U1,Zn2+ , is calculated at the unrestricted
B3LYP/TZV(2d+p) level using GAMESS [48]. Consis-
tent with experiments, we model all metal complexes in
their high-spin electronic configuration [49, 50].
FIG. 1. The zinc finger domain (pdb ID:2YTA) comprises
the α helix and the β sheet. The binding site residues (Cys
and His) are shown as stick figures. These residues and the
metal (sphere) totaling 59 particles are described by the site
coordinates X1 (Eqs. 1 and 8) and are throughout represented
in atomic detail.
YTA are used in the simulations; residues outside the
zinc-finger domain are not considered. The zinc-atom
is coordinated by two cysteine thiolates and two his-
tidines. We use the CHARMM27 forcefield [36] for all
the amino acid residues, with the thiolate partial charges
from Refs.
Simulations are performed with
NAMD2 [39].
[37, 38].
We use the dummy cation Zn2+ model developed in
Ref. [40]. The key feature of this model is the presence
of four dummy sites disposed tetrahedrally at a distance
of 0.9 A from a central atom. The dummy sites have a
partial charge of 0.5e but no size. The presence of dummy
sites proves helpful in maintaining the four-coordinate
state of the metal in extended simulations [40].
Each of the proteins CP1, YTA, and TF3 are sol-
vated in TIP3P [41, 42] water molecules using the sol-
vate module in VMD [43]. After an initial minimization,
the systems are equilibrated for 1 ns at 298.15 K and 1
bar followed by a 3 ns production phase. Configurations
are saved every 0.1 ps for analysis. The temperature is
maintained by a Langevin thermostat while pressure is
maintained using a Langevin barostat [44]. The reference
binding energy distribution Ps(ε) (Eq. 12) was calculated
using code developed in-house.
C. Monte Carlo simulations
The X1 coordinates describe the Zn2+ ion and the 4 bind-
ing site residues, in all 59 atoms (Figure 1). We calcu-
late U1(X1) (Eq. 8) using the CHARMM [36] potential
2
0
−2
∆
−4
−6
−2.5
−5.0
−7.5
−10.0
)
ε
(
P
n
l
5
Reference
φm(X1; α = 1.4)
φm(X1; α = 0)
0.0
0.3
0.6
0.9
1.2
1.5
α (kBT )
−12.5
−760
−740
−720
ε (kcal/mol)
−700
FIG. 2. ∆ (Eq. 12) as a function of the coupling constant α
with (filled) and without (open) site-bulk coupling (see text).
The minimum of ∆ is at α = 1.4 kBT .
To the test the effect of suppressing protein restraints,
we also compute the free energy of replacing Zn2+ with
Co2+ or Fe2+.
50 × 103 sweeps of equilibration and
25 × 103 sweeps of production were performed with the
PM3 hamiltonian and α = 0.0. We then take 80 well
separated configurations from the production phase to
obtain a qualitative estimate of the free energy change
with energies obtained at the B3LYP/TZV(2d+p) level.
For obtaining ∆µex(aq), the change in the excess free
energy in bulk water (Eq. 9), we borrow from our earlier
results based on the primitive quasi-chemical approach
[51]. Geometry, thermal corrections, and long-range con-
tributions to the free energy are obtained from the ear-
lier study, but for consistency with the presence work,
single point energy calculations are performed at the un-
restricted B3LYP/TZV(2d+p) level.
IV. RESULTS AND DISCUSSION
Unless otherwise mentioned, we report the key results
of our model extensively only for YTA. Similar agreement
is observed for CP1 and TF3. The selective preference
for Zn2+ over competing Co2+, Ni2+, and Fe2+ has been
experimentally measured for CP1 [29] and serves as a
helpful metric to assess the present approach in quanti-
fying thermodynamics of ion binding to protein sites.
A. Validation of the quadratic model
Figure 2 shows ∆(α) (Eq. 12) between Ps(ε), the ref-
erence binding energy distribution, and P (ε; α) obtained
using the effective potential. ∆(α) is a minimum at
α = 1.4; this is the value of α chosen for all further
investigations. To examine the role of site-bulk inter-
FIG. 3. Distribution of binding energies of Zn2+ with the
site from the reference MD simulations (solid curve) and
with simulations using the effective potential (Eq. 8) with
φm(X1; α = 0) (◦) and with φm(X1; α = 1.4) (△).
actions in site-site interactions, we set G = 0 (Eq. 4);
that is, we suppress site-bulk interactions. Physically,
this corresponds to independent springs on site particles,
as opposed to a site of interconnected particles. As Fig-
ure 2 illustrates, with no site-bulk interactions ∆(α) is
close to zero for all non-zero α: the probability distri-
bution for binding energies P (ε; α) has only a minimal
overlap with the target distribution Ps(ε). Thus, in ac-
cordance with experiments indicating the importance of
second-shell interactions in tuning metal binding [28, 29]
in the zinc-finger protein, the interaction of the site with
the bulk material (protein outside the site and the sol-
vent) is important in tuning the binding energy of Zn2+
with the site.
Figure 3 shows that the site in vacuum (α = 0) pro-
duces a binding energy distribution that is substantially
different from either the MD simulations or with the op-
timal (α = 1.4) site-bulk coupling constant. In particu-
lar, in the MD simulation and for the site with the local
molecular field, ion-site interactions are shifted to more
positive values than for the site in vacuum. Thus, on
average, Zn2+ is better bound to the cluster in vacuum
than it is when bulk protein and solvent effects are con-
sidered. Physically this implies that the role of the bulk
protein and the solvent is to destabilize the binding of
the ion with the site. Or in other words, the folded ββα
conformation of the protein is stabilized by the introduc-
tion of the metal. Note that experiments suggest that
metal coordination is essential for the protein fold and
the metal free apo-peptide is unstructured [27–29].
Figure 4 shows that distribution of the potential en-
ergy of the site (excluding the Zn2+ atom). The po-
tential energy Usite includes contributions from bonding,
angles, torsions, improper angles, and non-bonded terms
−2.5
−5.0
−7.5
−10.0
)
e
t
i
s
U
(
P
n
l
Reference
φm(X1; α = 1.4)
φm(X1; α = 0)
−12.5
280
320
360
Usite (kcal/mol)
400
FIG. 4. Distribution P (Usite) of the potential energy Usite
of the site particles excluding the Zn2+ atom. Symbols as in
Figure 3. Observes that the site in vacuum is more stable
than the site including the bulk protein and solvent response.
within the site. The site in vacuum is stable by about
6 kcal/mol (∆hUsitei = −6 kcal/mol between the binding
site in vacuum and MD simulations) than the one coupled
to the protein and solvent, again emphasizing the role of
the metal in stabilizing the protein fold. Note that the
coupling constant, α, was chosen to match the binding
energy profile, P (ε) of the Zn2+ ion. The agreement in
the distribution, P (Usite), of the binding site energetics is
an independent verification of our model. Further, this
observation suggests that the ENM can be successfully
parametrized with respect to either ion binding energet-
ics or site energetics.
As we have argued before [22, 23], the conformations
that correspond to low Usite, that is conformations for
which the site is less strained energetically, are also the
ones for which the ion-site interactions are unfavorable
for the ion: the less strained conformations of the site
correspond to the high energy tail of the ion-site distri-
bution. As Eq. 11 emphasizes, these conformations are
also the ones that sensitively influence the excess chem-
ical potential of the ion in the site. In this context it is
important to note that the local molecular field is able to
capture both the low energy wing of P (Usite) (Figure 4)
and the high-energy wing of P (ε) (Figure 3).
B. Structural characterization of the binding site
In Table I we summarize the various structural param-
eters that are relevant to the geometry of the residues
around the Zn2+ atom. We find that Zn2+ is able to
maintain a tetra-coordinate state throughout the course
of the simulation: the sulphur and nitrogen atoms from
the cystine and histidine residues, respectively, are al-
6
ways coordinated with the metal.
The geometry of the binding site in vacuum (α = 0)
and with the optimal site-bulk coupling in the effective
potential (Eq. 8) agree reasonably well with MD simula-
tions. The ∠N-Zn2+-N angle is somewhat more expanded
in the MD simulations and the ∠S-Zn2+-S is somewhat
more compressed relative to the site with either α = 0
or α = 1.4. But the fluctuations are large and it is un-
clear if these differences are significant. This observation
suggests that in tuning the molecular molecular field, en-
ergies may prove more sensitive than structural parame-
ters.
C.
Investigation of the protein restraints
It is known that the folding of the zinc finger peptide
is coupled with metal binding [27–29]. Thus, residues
which are crucial in maintaining the binding site are also
expected to be important in the folding of the peptide.
It has been suggested that the β-sheet hosting the two
cysteine residues is formed in the absence of Zn2+ and
the addition of the metal induces the folding of the α-
helix [52–54]. The coordination of Zn2+ with the histi-
dine residues is thought to be essential towards folding of
the α-helix and formation of the hydrophobic core. We
next consider how analyzing the protein field can provide
insights into some of these features.
MD
rS−Zn2+
rN−Zn2+
MC α = 0 MC α = 1.4
2.17 ± 0.03 2.16 ± 0.03 2.15 ± 0.03
2.08 ± 0.04 2.1 ± 0.05 2.07 ± 0.03
∠S-Zn2+-S 112.0 ± 4.5 120.2 ± 5.9 121.9 ± 4.2
∠N-Zn2+-N 104.6 ± 4.6 102.2 ± 4.6 100.5 ± 3.2
TABLE I. Bond lengths and angles characterizing the geom-
etry of the site. All angles are in degrees and distances in
Angstroms. MD, molecular dynamics; α = 0 (vacuum) and
α = 1.4 kBT , correspond to Monte Carlo simulations with the
effective potential (Eq. 8). Standard deviations of the quan-
tities are noted.
The restraints imposed by the protein on the bind-
ing site, φm(X1; α) in Eq. 8, are collective in nature and
cannot be expressed as a sum of independent restraints
over individual atoms of the binding site. The quadratic
approximation decomposes the restraints as a sum over
mutually orthogonal collective motions of the binding site
atoms. These collective motions or vibrational modes
are along the eigenvectors of the positive definite matrix,
Hsite in Eq. 8, describing the protein field, φm(X1, ; α).
We classify the vibrational modes as either concen-
trated on a single residue or distributed on more than
one residue. Since the eigenvector has contributions from
each of the atom comprising the binding site, we consider
the contribution to the norm of the eigenvector from each
residue of the binding site to evaluate the behavior of the
modes. We call the mode concentrated if the majority of
λ
25
20
15
10
5
0
YTA
Cys Cys His His
λ
20
16
12
8
4
0
CP1
Cys Cys His His
λ
10
8
6
4
2
0
7
TF3
Cys Cys His His
FIG. 5. Vibrational modes for YTA, CP1, and TF3 arranged according to the residue of the binding site which contributes
most to its magnitude. The eigenvalue corresponding to each mode in kBT /A
is plotted for each of the four metal-binding
residues. Red lines denote concentrated modes while blue lines denote distributed ones. Observe that the strongest vibrational
modes are concentrated on the first cysteine and the first histidine of the CCHH cluster comprising the binding site.
2
the contribution to the norm is found on one residue (We
choose a threshold of 0.8 for delineating concentrated
modes from all other modes. The qualitative insights
below are independent of this threshold.) Following this
procedure for all the modes, Fig. ?? shows that the most
important protein restraints for three different zinc fin-
ger domains are in fact concentrated on the first cysteine
and the first histidine residue of the CCHH cluster com-
prising the binding site. This implies that the first C
and the first H play an important role in maintaining the
architecture of the site.
Nomura and Sugiura [55] conducted experiments on
the Sp1 zinc finger domain where they mutated one
residue at a time in the CCHH binding site to a glycine
residue, eliminating the propensity of that particular
residue to coordinate with the Zn2+ center. The authors
showed, through circular dichroism measurements, that
the CCHG mutant of the protein is able to form an α-
helix and a β-sheet structure similar to that of the wild
type protein, while no helical structure is formed with
the CCGH mutant, implying the importance of the first
histidine in maintaining the architecture of the binding
site. Our analysis of concentrated modes is consistent
with this experimental observation of the importance of
the first histidine in the CCHH cluster.
Since the β-sheet hosting the first cysteine is formed
prior to the introduction of the metal, analysis of the
modes in the presence of the metal is not sufficient to infer
the consequences of mutations in the cysteine residues.
Our analysis does suggest that the Zn2+ will be loosely
held in the GCHH finger and experience larger fluctua-
tions compared to the CGHH finger. Spectroscopic ex-
periments to probe the dynamics of the binding site can
be helpful in evaluating this suggestion.
Fig. ?? shows that the eigenvalues λ defining the
strength of the restraints on the binding site are larger
for the designed peptides CP1 and YTA than for TF3:
deflections of the binding site from the equilibrium struc-
ture require more energy for the designed peptides than
for the natural TF3 peptide. Thus we expect CP1 and
YTA to experience smaller deviations from the equilib-
rium structure as compared to TF3. Since metal coor-
dination has a stabilizing effect on the protein fold and
since deformations from the equilibrium structure will
also tend to destabilize Zn2+ binding to the protein, we
can induce that CP1 and YTA will experience a higher
stabilization due to introduction of the metal as com-
pared to TF3. It is interesting to note that CP1 was de-
signed by aligning 131 natural zinc finger sequences and
is found to have a greater affinity for zinc than the corre-
sponding natural sequences [56]. The predicted greater
stabilization of Zn2+-bound CP1 is in accordance with
this experimental observation.
Sequence specific effects on stabilization can also be
inferred. In CP1 and YTA only two residues interleave
the cysteine residues in contrast to four residues in the
natural peptide. This shorter turn between the cysteine
residues likely underlies the observed differences in bind-
ing energetics, but a thorough exploration of how such
sequence dependences influence binding is left for future
investigations. The analysis of concentrated modes de-
veloped here may prove helpful in this regard.
D. Selectivity of the site for Zn2+ over competing
transition metals
Table II summarizes the predicted values for the free
energy difference between Zn2+ and competing divalent
ions for the consensus peptide CP1. The order of selec-
tivity Co2+ < Ni2+ < Fe2+ relative to Zn2+ is consistent
with experiments [29]. The magnitude of selectivity for
Fe2+ is also in good agreement with experiments.
We first consider limitations in calculating ∆µex(aq)
in affecting the predicted selectivity, ∆µex (Eq. 9).
The aqueous component of the selectivity free energy,
∆µex(aq), is only about 1% of the absolute hydration
free energy of the metal ion [51]; thus small errors in the
absolute hydration free energy can be amplified in taking
differences. In the present study, we have used a prim-
itive quasichemical estimate [51] for ∆µex(aq): the free
energy of forming a metal ion-water cluster in vacuum is
Ion α ∆µex(S) ∆µex(aq) ∆µex (Calc.) ∆µex (Expt.)
Co2+ 1.4
0.0
Fe2+ 1.4
0.0
Ni2+ 1.4
5.2
5.2
20.4
20.4
−12.6
6.3
6.9
29.2
16.9
−7.7
1.1
1.7
8.4
−3.5
4.3
5.4
5.4
7.8
7.8
7.4
TABLE II. Free energy to replace Zn2+ Co2+, Fe2+, and
Ni2+ (Eq. 9) in the CP1 peptide. All values are in kcal/mol.
∆µex(S) is calculated using Eq. 10. Calculations for a cluster
in vacuum (α = 0) and for the cluster in a molecular field
(α = 1.4 kBT ) are reported for Co2+ and Fe2+. ∆µex(aq) was
obtained using primitive quasi-chemical theory [51]. (Please
refer to Sec. III D for details.) Experimental estimates from
Ref. [29] are provided for comparison. The calculated selec-
tivities for Co2+, Fe2+, and Ni2+, respectively, in the YTA
peptide are 1.3, 8.1, and 6.7, similar to the estimate for CP1.
We are not aware of experiments characterizing the selectivity
in YTA.
combined with a continuum dielectric correction for the
presence of the bulk material. Ignoring the role of the
bulk material in forming the metal-water cluster [57] and
potential limitations in a continuum dielectric description
of water beyond the first shell for highly charged metals
[58, 59] are important limitations in our calculation of
∆µex(aq).
Limitations in calculating ∆µex(S) is the other factor
in affecting the predicted selectivity. Examining the dis-
tribution of ∆U1 (Eq. 10), the difference in the bind-
ing energy of metal X2+ with the binding site relative
to the corresponding value for Zn2+, shows that ∆U1 is
narrowly distributed about the mean value h∆U1i (Ta-
ble III), although the distribution of the binding energy
of the metal with the cluster itself is broad (Fig. 3). Thus
the calculated free energy differences in the presence of
the field are expected to be reasonably well-converged.
Table II shows that the protein field plays a decisive
role in selectivity. The protein restraints limit the phase
space distribution of the binding site to configurations
which bind to Zn2+ strongly. Removing these restraints
allows the binding site to sample configurations which
bind favorably to Fe2+ as well. We find that the free
energy for replacing Zn2+ with Fe2+ without the protein
field (α = 0) predicts greater stabilization of the zinc-
finger in the presence of Fe2+, in complete disagreement
with experiments. The role of the protein in replacing
Zn2+ with Co2+ is predicted to be negligible. Exam-
ining the distribution of ∆U1 shows that the protein
tends to limit energy fluctuations (C2 term, Table III)
and the effect is more pronounced for metal ions that in-
teract strongly (relative to Zn2+) with the zinc-binding
residues, as the comparison of Fe2+ and Co2+ illustrates
(cf. C1 and C2 Table III). Taken together, these obser-
vations show that while the local metal-residue interac-
tion is important in selectivity, as was already inferred
in the early studies on zinc-fingers[49], the role of the
protein cannot be ignored, a result that is in accordance
8
with experimental investigations comparing two different
zinc-binding proteins [60].
An important limitation in the present study is the
small number of configurations sampled at the semi-
empirical level. The small magnitude of C2 relative to
C1 suggests that this limitation may not be severe in the
case when the protein field is present. In any case, with
a more efficient implementation of the present approach,
this limitation can always be overcome. Regardless of
these computational limitations, the important physical
point we wish to emphasize is that accounting for the
protein field is necessary for a broader characterization of
the thermodynamics of metal-ion binding in this system.
Thus, for example, the protein field will be necessary for
a proper description of excess entropies and excess ener-
gies, quantities that demand a proper description of the
role of the medium on the local ion-residue cluster [23].
6.3
6.9
Ion α ∆µex(S) C1 C2 C1 − C2
Co2+ 1.4
6.3
6.7
0.0
Fe2+ 1.4
28.7
−7.2
0.0
Ni2+ 1.4
−8.4
6.8 0.5
7.6 0.9
29.2 29.2 0.5
16.9 29.8 37.0
−7.7 −7.1 1.3
TABLE III. Comparing the exponential average (Eq. 9) with
the Gaussian model ∆µex(S) = C1 − C2, where C1 = h∆U1i
and C2 = β/2h(∆U1 − C1)2i. C1 and 2C2/β are the first
(mean) and second (variance) moments of the distribution of
∆U1 values [62]. Calculations for a cluster in vacuum (α = 0)
and for the cluster in a molecular field (α = 1.4 kBT ) are
reported for Co2+ and Fe2+. All values in kcal/mol.
Changes in the equilibrium geometry of the protein
can be expected with ion exchange owing to change in
the ion size, which, within the quadratic approximation,
will change the harmonic restraints on the binding site
and thus the estimate of the selectivity free energy. Here
∆µex(S) was calculated with the assumption that the
molecular field imposed by the protein medium and the
solvent is independent of the bound ion, ∆φ ≈ 0. For the
given coordination environment, the radii for metals con-
sidered here are expected to be similar and so assuming
∆φ ≈ 0 appears reasonable. This assumption will cer-
tainly not hold when a metal associates with the binding
site in a geometry substantially different from the one for
Zn2+ as may happen for Hg2+ [61].
V. CONCLUDING DISCUSSIONS
The architecture and conformation of the metal bind-
ing site in metalloproteins is conditioned by energetic in-
teractions of the metal cation and the binding site on
the one hand and the interaction of the binding site with
the protein and solvent media outside the binding site on
the other. In the case of the zinc finger peptide, reflecting
9
SUPPLEMENTARY INFORMATION
Complete reference to Frisch M. J. et al.
[47],
CHARMM [36], and additional data for CP1 and TF3
peptides
the fact that the peptide is unfolded in the absence of the
metal, calculations show that including the interaction of
the site with the protein material outside the binding site
results in a more destabilized metal-binding site complex
in comparison to an analogous metal-binding site com-
plex in vacuum.
The protein outside the metal-residue cluster imposes
a field on the cluster. By treating the bulk protein as
an elastic medium, we describe the molecular field by
quadratic model. This simplifies the problem of tackling
metal protein interactions by greatly reducing the num-
ber of degrees of freedom and allows us to study the effect
of the protein medium on the binding site in a trans-
parent fashion. Decomposing the protein restraints on
the binding site over mutually orthogonal collective mo-
tions of the binding site particles shows that the protein
restraints are stronger for designed zinc-finger peptides
CP1 and YTA relative to the natural zinc finger TF3.
Further, for all three zinc finger peptides, we find that
the first cysteine and first histidine in the CCHH bind-
ing site are more tightly held by the protein as compared
to the other two residues of the binding site. Our re-
sults are in accordance with the importance of the first
histidine observed on the basis of metal-induced folding
experiments with a CCGH mutant binding site. We also
predict that were a peptide with the GCHH binding motif
to fold in the presence of Zn2+, that metal-residue clus-
ter is expected to experience large fluctuations relative
to the CCHH binding motif.
Approximating the solvent and the protein response
by a quadratic term implies that we have neglected spe-
cific interactions of solvent molecules on the dynamics
of the binding site. Care would be required when the
present model is applied to ion binding sites where water
molecules play a crucial role. The molecular field ap-
proach developed here requires the equilibrium structure
of the protein in order to estimate the response of the
bulk protein due to the binding site. If this response is
not sensitively dependent on the bound ion, the molecu-
lar field obtained on the basis of a protein structure for
one ion can be used to predict the free energy change in
replacing that ion with a competing one. Such calcula-
tions on the zinc-finger peptide give reasonable estimates
of the free energy of replacing Zn2+ with Co2+, Fe2+, and
Ni2+.
ACKNOWLEDGMENTS
PD thanks Christian Micheletti for introducing him to
elastic network models. We thank Christian Micheletti
and Safir Merchant
for helpful comments on the
manuscript This research used resources of the National
Energy Research Scientific Computing Center, which is
supported by the Office of Science of the U.S. Department
of Energy under Contract No. DE- AC02-05CH11231.
10
[1] Bren, K. L.; Pecoraro, V. L.; Gray, H. B. Inorg. Chem.
[30] Reuter, N.; Dejegere, A.; Maigret, B.; Karplus, M. J.
2004, 43, 7894–7896.
[2] Lippard, S. J.; Berg, J. M. Principles of bioinorganic
chemistry; University Science Books: Mill Valley, CA,
1994.
[3] Millhauser, G. L. Acc. Chem. Res. 2004, 37, 79–85.
[4] Wilson, C. J.; Apiyo, D.; Wittung-Stafshede, P. Q. Rev.
Biophys. 2004, 37, 285–314.
Phys. Chem. A 2000, 104, 1720–1735.
[31] Ming, D.; Wall, M. E. Phys. Rev. Lett. 2005, 95, 198103.
[32] Beck, T. L.; Paulaitis, M. E.; Pratt, L. R. The potential
distribution theorem and models of molecular solutions;
Cambridge University Press, 2006.
[33] Widom, B. J. Phys. Chem. 1982, 86, 869–872.
[34] Kullback, S.; Leibler, R. A. Ann. Math. Stat. 1951, 22,
[5] Adlard, P. A.; Bush, A. I. J. Alzheimers Dis. 2006, 10,
79–86.
145–163.
[6] Siegbahn, P. E. M.; Blomberg, M. R. A. Ann. Rev. Phys.
Chem. 1999, 50, 221–249.
[7] Ryde, U. Curr. Opin. Chem. Biol. 2003, 7, 136–142.
[8] Lin, H.; Truhlar, D. G. Theor. Chem. Acc. 2007, 117,
185–199.
[9] Senn, H. M.; Thiel, W. Angew. Chem. Intl. Ed. 2009,
48, 1198–1229.
[35] Foster, M. P.; Wuttke, D. S.; Radhakrishnan,
I.;
Case, D. A.; Gottesfeld, J. M.; Wright, P. E. Nat. Struct.
Biol. 1997, 4, 605–608.
[36] MacKerell, Jr., A. D. et al. J. Phys. Chem. B 1998, 102,
3586–3616.
[37] Foloppe, N.;
J.;
Nordstrand, K.;
Berndt, K. D.; Nilsson, L. J. Mol. Biol. 2001, 310,
449–470.
Sagemark,
[10] Pratt, L. R.; Ashbaugh, H. S. Phys. Rev. E 2003, 68,
[38] Calimet, N.; T.Simonson, J. Mol. Graph. Model. 2006,
021505.
24, 404–411.
[11] Zheng, W.; Brooks, B. R. Biophys. J. 2006, 88, 3109–
3117.
[12] Zheng, W.; Liao, J. C.; Brooks, B. R.; Doniach, S. Pro-
teins: Struc. Func. Bioinfo. 2007, 67, 886–896.
[13] Woodcock, H. L.; Zheng, W.; Ghysels, A.; Shao, Y.;
Kong, J.; Brooks, B. R. J. Chem. Phys. 2008, 129,
214109.
[14] Atilgan, A. R.; Durell, S. R.;
Jernigan, R. L.;
Demirel, M. C.; Keskin, O.; Bahar, I. Biophys. J. 2001,
80, 505–515.
[39] Phillips, J.; Braun, R.; Wang, W.; Gumbart, J.; Tajkhor-
shid, E.; Villa, E.; Chipot, C.; Skeel, R.; Kale, L.; Schul-
ten, K. J. Comp. Chem. 2005, 26, 1781–1802.
[40] Pang, Y. P. Proteins: Struc. Func. Bioinfo. 2001, 45,
183–189.
[41] Jorgensen, W.; Chandrasekhar, J.; Madura, J. D.; Im-
pey, R. W.; Klein, M. L. J. Chem. Phys. 1983, 79, 926–
935.
[42] Neria, E.; Fischer, S.; Karplus, M. J. Chem. Phys. 1996,
105, 1902–1921.
[15] Bahar, I.; Reader, A. J. Curr. Opin. Struc. Biol. 2005,
[43] Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graphics
15, 586–592.
1996, 14, 33–38.
[16] Micheletti, C.; Carloni, P.; Maritan, A. Proteins: Struc.
[44] Feller, S.; Zhang, Y.; Pastor, R.; Brooks, B. J. Chem.
Func. Bioinfo. 2004, 55, 635–645.
Phys. 1995, 103, 4613–4621.
[17] Pontiggia, F.; Colombo, G.; Micheletti, C.; Orland, H.
Phys. Rev. Lett. 2007, 98, 048102.
[18] Capozzi, F.; Luchinat, C.; Micheletti, C.; Pontiggia, F.
[45] Stewart, J. J. P. J. Comp. Chem. 1989, 10, 209–220.
[46] Stewart, J. J. P. J. Comp. Chem. 1989, 10, 221–264.
[47] Frisch, M. J. et al. Gaussian 09 Revision A.1. Gaussian
J. Proteome Res. 2007, 6, 4245–4255.
Inc. Wallingford CT 2009.
[19] Zen, A.; Carnevale, V.; Lesk, A. M.; Micheletti, C. Pro-
tein Science 2009, 17, 918–929.
[20] Asthagiri, D.; Pratt, L. R.; Paulaitis, M. E. J. Chem.
Phys. 2006, 125, 24701.
[21] Pratt, L. R.; Asthagiri, D. In Free energy calcula-
tions: Theory and applications in chemistry and biol-
ogy; Chipot, C., Pohorille, A., Eds.; Springer series in
Chemical Physics; Springer, 2007; Vol. 86; Chapter 9, pp
323–351.
[48] Schmidt, M. W.; Baldridge, K. K.; Boatz, J. A.; El-
bert, S. T.; Gordon, M. S.; Jensen, J. J.; Koseki, S.;
Matsunaga, N.; Nguyen, K. A.; Su, S.; Windus, T. L.;
Dupuis, M.; Montgomery, J. A. J. Comp. Chem. 1993,
14, 1347–1363.
[49] Berg, J. M.; Merkle, D. L. J. Am. Chem. Soc. 1989, 111,
3759–3761.
[50] Krizek, B. A.; Berg, J. M. Inorg. Chem. 1992, 31, 2984–
2986.
[22] Dixit, P. D.; Merchant, S.; Asthagiri, D. Biophys. J.
[51] Asthagiri, D.; Pratt, L. R.; Paulaitis, M. E.; Rempe, S. B.
2009, 96, 2138–2145.
J. Am. Chem. Soc. 2004, 126, 1285–1289.
[23] Dixit, P. D.; Asthagiri, D. Biophys. J. 2011, 100, 1542–
[52] Shi, Y.; Beger, R. D.; Berg, J. M. Biophys. J. 1992, 64,
1549.
749–753.
[24] Klug, A.; Schwabe, J. W. FASEB J. 1995, 9, 597–604.
[25] Berg, J.; Shi, Y. Science 1996, 271, 1081–1085.
[26] Wolfe, S. A.; Nekludova, L.; Pabo, C. O. Ann. Rev. Bio-
phys. Biomol. Struc. 2000, 29, 183–212.
[27] Frankel, A. D.; Berg, J. M.; Pabo, C. O. Proc. Natl. Acad.
Sc. USA 1987, 84, 4841–4845.
[28] Michael, S. F.; Kilfoil, V. J.;
Schmidt, M. H.;
Amann, B. T.; Berg, J. M. Proc. Natl. Acad. Sc. USA
1992, 89, 4796–800.
[53] Miura, T.; Satoh, T.; Takeuchi, H. Biochim. Biophys.
Acta 1998, 1384, 171–179.
[54] Hanas, J. S.; Larabee, J. L.; Hocker, J. R. In Zinc Fin-
ger Proteins: From atomic contact to cellular Function;
Iuchi, S., Kuldell, N., Eds.; Springer series on Life and
Biomedical Sciences; Kluwer Academic/Plenum Publish-
ers: Boston, 2005; Chapter 8.
[55] Nomura, A.; Sugiura, Y. Inorg. Chem. 2002, 41, 3693–
3698.
[29] Berg, J.; Godwin, H. A. Ann. Rev. Biophys. Biomol.
[56] Krizek, B. A.; Amann, B. T.; Kilfoil, V. J.; Merkle, D. L.;
Struc. 1997, 26, 357–371.
Berg, J. M. J. Am. Chem. Soc. 1991, 113, 4518–4523.
[57] Merchant, S.; Asthagiri, D. J. Chem. Phys 2009, 130,
195102.
[58] Asthagiri, D.; Pratt, L. R.; Ashbaugh, H. S. J. Chem.
Phys. 2003, 119, 2702–2708.
[60] Lachenmann, M. J.; Ladbury, J. E.; Dong, J.; Huang, K.;
Carey, P.; Weiss, M. A. Biochemistry 2004, 43, 13910–
13925.
[61] Sakharov, D. V.; Lim, C. J. Am. Chem. Soc. 2009, 30-2,
[59] Fatmi, M. Q.; Hofer, T. S.; Randolf, B. R.; Rode, B. M.
191–202.
J. Chem. Phys. 2005, 123, 054514.
[62] Hummer, G.; Pratt, L. R.; Garcia, A. E. J. Phys. Chem.
A. 1998, 102, 7885–7895.
11
|
1608.00542 | 2 | 1608 | 2016-08-03T08:18:59 | SMILE Microscopy : fast and single-plane based super-resolution volume imaging | [
"physics.bio-ph",
"physics.optics"
] | Fast 3D super-resolution imaging is essential for decoding rapidly occurring biological processes. Encoding single molecules to their respective planes enable simultaneous multi-plane super-resolution volume imaging. This saves the data-acquisition time and as a consequence reduce radiation-dose that lead to photobleaching and other undesirable photochemical reactions. Detection and subsequent identification of the locus of individual molecule (both on the focal plane and off-focal planes) holds the key. Experimentally, this is achieved by accurate calibration of system PSF size and its natural spread in off-focal planes using sub-diffraction fluorescent beads. Subsequently the identification and sorting of single molecules that belong to different axial planes is carried out (by setting multiple cut-offs to respective PSFs). Simultaneous Multiplane Imaging based Localization Encoded (SMILE) microscopy technique eliminates the need for multiple z-plane scanning and thereby provides a truly simultaneous multiplane super-resolution imaging. | physics.bio-ph | physics | SMILE Microscopy : fast and single-plane based super-resolution volume imaging
Partha Pratim Mondal∗
Nanobioimaging Laboratory,
Instrumentation and Applied Physics,
Indian Institute of Science, Bangalore, INDIA
(Dated: July 17, 2021)
Fast 3D super-resolution imaging is essential for decoding rapidly occurring biological pro-
cesses. Encoding single molecules to their respective planes enable simultaneous multi-plane super-
resolution volume imaging. This saves the data-acquisition time and as a consequence reduce
radiation-dose that lead to photobleaching and other undesirable photochemical reactions. De-
tection and subsequent identification of the locus of individual molecule (both on the focal plane
and off-focal planes) holds the key. Experimentally, this is achieved by accurate calibration of
system PSF size and its natural spread in off-focal planes using sub-diffraction fluorescent beads.
Subsequently the identification and sorting of single molecules that belong to different axial planes
is carried out (by setting multiple cut-offs to respective PSFs). Simultaneous Multiplane Imaging
based Localization Encoded (SMILE) microscopy technique eliminates the need for multiple z-plane
scanning and thereby provides a truly simultaneous multiplane super-resolution imaging.
Fast 3D super-resolution imaging is essential for visu-
alizing rapidly occurring events in mammalian cells and
microorganisms with sub-diffraction resolution [1][3][7][2].
With the advent of single molecule based super-resolution
imaging techniques, it is now possible to visualize and follow
biophysical processes (protein dynamics and single molecule
binding) within single mammalian cell, E. Coli and bacteria
[3][2][8]. Prominent techniques that are capable of super-
resolution include, photoactivated localization microscopy
(PALM), fluorescence PALM (fPALM), stimulated emis-
sion depletion (STED), stochastic optical reconstruction
microscopy (STORM), super-resolution optical fluctuation
imaging (SOFI) and saturated structured illumination
microscopy (SSIM) [9]. Simultaneous Multiplane Imaging
based Localization Encoded (SMILE) microscopy use PSF
dimension to encode the location of single molecules in the
respective z-planes. SMILE microscopy achieves this by
determining the natural spread of PSF at greater depths,
followed by single molecule sorting to enable simultaneous
multiplane imaging. To construct volume image, existing
techniques employ, light-sheet [11][12], Bessel beam [13][14]
and cylindrical lens [15]. These techniques require sophisti-
cated optical components (scanning mirror and Z-axis piezo
stage) to render volume image [1]. Moreover, some of these
techniques are known to cause photobleaching and optical
aberrations. A single-plane technique has the capability to
eliminate these undesirable effects by simply reducing the
radiation-dose and eliminating moving components. Here,
we propose a single-plane super-resolution technique that
utilizes the information (PSF size and number of emitted
photons) from off-focal molecules. When integrated with
the focal plane information it results in the reconstruction
of 3D volume stack. We harness the information encoded
in both diffraction-limited PSF and off-focal larger PSFs
(Gaussian approximated). Due to diffraction nature of light,
a point source (sub-diffraction nano-beads or fluorescent
molecules) in focal plane appear as a diffraction-limited
spot (∆xy ≈ 1.22λ/2N A = 263nm) [16]. However the
spot size (PSF) broadens for point sources that are located
in the off-focal planes (far from coverslip). This is due
∗Corresponding author: [email protected]
to diffraction and the spherical aberration (around the
spherical beads) that results in bead size way larger than
∆xy [17].
Larger distances (typically, > 1.5µm) gives
rise to non-linear airy-disc like pattern that cannot be
approximated by Gaussian function. Hence, we choose to
work close to coverslip (< 1.5µm from the coverslip). For
generating simultaneous multiple planes from a single data
set, we carried out a thorough calibration to experimentally
obtain multiple cut-offs for system PSF size with desired
inter-plane separation and plane thickness. These cut-offs
decide the distance between adjacent Z-planes.
This
distinct size Wz (as seen by the detector in the image
plane) forms the basis for simultaneous multiplane imaging.
Based on the calibration experiment, the following selection
criterion (for selecting molecules Mi
in specific z-planes
Zl) is considered, Mi ∈ Zl;
l∆xy < Wz < (l + 1)∆xy for
l-planes and molecule Mi with a size Wz. This criterion is
applied on every detected single molecule (after localizing
them) and a corresponding z-plane tag is assigned to the
molecule.
The schematic diagram of SMILE microscopy is shown
in Fig.1A. For calibration, sub-diffraction size nano-beads
(< ∆xy) embedded in Agarose transparent gel-matrix
are imaged (Fig.1B). Nanobeads (point emitters) situated
near the coverslip (focal plane of the detection objective)
has diffraction-limited spot while those situated in other
planes (slightly above the focal plane) have size > ∆xy.
We study actin (labeled with a photoactivatable CAGE
552 dye) distribution in NIH 3T3 mouse fibroblast cells
[18]. Experimentally recorded single molecule signatures
can be visualized in the supplementary video. The study
covers a distance of 1400 nm with a interplane separation
of 200 nm starting from 600 nm above the coverslip. The
region < 600 nm remain inaccessible due to negligible
size (area < 3 × 3 camera pixels) of the system PSF
that cannot be reliably approximated by a 2D Gaussian
function. Probing regions very close to coverslip (< 600nm)
require high precision piezo z-stage that makes the system
costly and complicated. We decided to keep the system
simple yet reliable for 3D volume imaging. The separation
is calibrated using sub-diffraction fluorescent nanobeads
(Fig.1B) that determines the cut-offs for consecutive layers.
Calibration shows a cutoff of p × p pixels (size of Gaussian
6
1
0
2
g
u
A
3
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
2
v
2
4
5
0
0
.
8
0
6
1
:
v
i
X
r
a
2
FIG. 1: SMILE Microscopy: A. Optical diagram for SMILE microscopy; B. Mechanism behind SMILE microscopy along with the
PSF size calibration curve obtained from near photobleachingless fluorescent nanobeads. The inset show the characteristic fall of
intensity far from focus (at large z); C. Simultaneous imaging of multiple planes (of two different regions, R1 and R2) separated by
∆z = 200 nm. The colored composite image (for region R2) that represent 3D positions (using plane numbers, 1-4) is also shown.
The transmission image of the NIH-3T3 cell (Axon region) is also shown. E. The characteristic increase in localization precision at
large penetration depths indicating the role of scattering and absorption of emitted photons.
approximated PSF) in the image plane (see, Fig1B). Sorting
point sources (nano-beads) that belong to specific z-planes
are then carried out based on their size using EINZEL
Matlab scripts (kindly shared by, Prof. Samuel T. Hess,
University of Maine, USA) [9][18]. Once z-dependent size
of PSF is determined, the same becomes the reference
for locating single molecules in the specimen planes. The
transmission image of a part of single fibroblast cell along
with the super-resolved images (regions, R1 and R2) and
colored composite image for R2 are shown in Fig.1C. For
both R1 and R2, 4 adjacent planes of super-resolved images
are obtained simultaneously without the need for scanning
individual planes. One can readily see change in the
distribution of single molecules over 4 planes (spaced 200
nm apart) in the axon of fibroblast cells. Although some
structures are restricted in the axial plane (i.e. only present
in the first z slice adjacent to coverslip), others persist
through planes at large depths. This is further evident from
the colored composite image that represent the 3D position
of single molecules in planes 1-4. Multiplane based SMILE
imaging do suffer from scattering and photon absorption.
This can be seen from the localization precision metric
(∆loc = ∆psf /
N , where N is the number of photons de-
tected for a single molecule) at large depths. Fig.1D shows
the average localization precision for 4 different planes up
to a depth of 1400 nm and the corresponding distribution
of localization precision is also shown (see insets in Fig1D).
The increase in average value of localization precision with
√
increasing z-distance inside the specimen is quite evident.
This reveals the fact that most of the photons emitted by
the single molecule either gets absorbed by the adjoining
nearby layers or gets scattered away thereby not reaching
the detector. As a consequence of reduced detected photons,
the average localization precision value increase at greater
depths in the specimen. The increase is almost double for a
depth (z-distance) of ≈ 600 nm. This further explains the
difficulty in realizing super-resolution at large penetration
depths. The proposed SMILE microscopy is simple yet
powerful technique that requires minimal optics and has
the ability to reduce photobleaching thereby enabling
single-plane 3D volume imaging (a step closer to temporal
super-resolution). As a consequence of these advantages,
one can now reliably image live cell in 3D and visualize
rapidly occurring dynamical events in real-time.
Acknowledgment
The author thank Prof. Samuel T. Hess (University of
Maine, Orono, USA) for discussion and for sharing EINZEL
analysis MATLAB codes that helped us develop new codes
for SMILE microscopy.
[1] Partha P. Mondal and Alberto Diaspro, Book: Fundamen-
[2] J. Errington, Nature Reviews Microbiology 13, 241248
tals of Fluorescence Microscopy, 1st Ed. (Springer, 2014).
(2015).
3
[3] C. Coltharp and J. Xiao, Superresolution microscopy for mi-
[10] A. G. Godin, B. Lounis and L. Cognet, Biophys. J. 8, 1777
crobiology, Cell Microbiol. 14, 1808 (2012).
(2014).
[4] J. G. Danzl, S. C. Sidenstein, C. Gregor, N. T. Urban, P.
Ilgen, S. Jakobs and S. W. Hell, Nature Photonics 10, 122
(2016).
[5] S. W. Hell, R. Schmidt and A. Egner, Nature Photonics 3,
381 - 387 (2009).
[6] J. Rosen and G. Brooker, Nature Photonics 2, 190 - 195
(2008).
[11] F. C. Zanacchi et al., Nat. Meth. 8, 1047 (2011).
[12] W. R. Legant, L. Shao, J. B. Grimm, T. A Brown, D. E.
Milkie, B. B. Avants, L. D. Lavis and E. Betzig, Nat. Meth.
13, 359 (2016).
[13] T. A. Planchon, L. Gao, D. E. Milkie, M. W. Davidson, J.
A. Galbraith, C. G. Galbraith and E. Betzig, Nat. Meth. 8,
417 (2011).
[7] Partha P. Mondal and A. Diaspro, Scientific Reports 1, 149
[14] S. B. Purnapatra, S. Bera and Partha P. Mondal, Sci. Rep.
(2011).
2, 692 (2012).
[8] S. B. Purnapatra, S. Bera and Partha P. Mondal, Scientific
[15] B. Huang, W. Wang, M. Bates and X. Zhuang, Science 319,
Reports 2, 692 (2012).
810 (2008).
[9] S. W. Hell and J. Wichmann, Opt. Lett. 19, 780 (1994); E.
Betzig at al., Science 313, 1642 (2006); S. T. Hess et al., Bio-
phys J. 91, 4258 (2006); M. J. Rust et al., Nat. Meth. 3, 793
(2006); T. Dertinger et al., PNAS (USA) 106, 22287 (2009);
M. G. L. Guastafsson, PNAS (USA) 102, 13081 (2005).
[16] E. Abbe, Arch. f. Mikr. Anat. 9, 413 (1873).
[17] Born, M.; Wolf, E., Principles of Optics, 8th ed.; Pergamon:
New York, 1959.
[18] M. V. Gudheti et al., Biophys J. 104, 2182 (2013).
|
1601.00478 | 1 | 1601 | 2016-01-04T12:25:58 | Thermodynamics of amyloid formation and the role of intersheet interactions | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.BM"
] | The self-assembly of proteins into $\beta$-sheet-rich amyloid fibrils has been observed to occur with sigmoidal kinetics, indicating that the system initially is trapped in a metastable state. Here, we use a minimal lattice-based model to explore the thermodynamic forces driving amyloid formation in a finite canonical ($NVT$) system. By means of generalized-ensemble Monte Carlo techniques and a semi-analytical method, the thermodynamic properties of this model are investigated for different sets of intersheet interaction parameters. When the interactions support lateral growth into multi-layered fibrillar structures, an evaporation/condensation transition is observed, between a supersaturated solution state and a thermodynamically distinct state where small and large fibril-like species exist in equilibrium. Intermediate-size aggregates are statistically suppressed. These properties do not hold if aggregate growth is one-dimensional. | physics.bio-ph | physics | Thermodynamics of amyloid formation and the role of intersheet interactions
Anders Irbäck1, a) and Jonas Wessén1, b)
Department of Astronomy and Theoretical Physics, Lund University,
Sölvegatan 14A, SE-223 62 Lund, Sweden
(Dated: 10 October 2018)
The self-assembly of proteins into β-sheet-rich amyloid fibrils has been observed to
occur with sigmoidal kinetics, indicating that the system initially is trapped in a
metastable state. Here, we use a minimal lattice-based model to explore the ther-
modynamic forces driving amyloid formation in a finite canonical (N V T ) system.
By means of generalized-ensemble Monte Carlo techniques and a semi-analytical
method, the thermodynamic properties of this model are investigated for different sets
of intersheet interaction parameters. When the interactions support lateral growth
into multi-layered fibrillar structures, an evaporation/condensation transition is ob-
served, between a supersaturated solution state and a thermodynamically distinct
state where small and large fibril-like species exist in equilibrium. Intermediate-size
aggregates are statistically suppressed. These properties do not hold if aggregate
growth is one-dimensional.
PACS numbers: 87.15.ad, 87.15.ak, 87.15.nr, 87.15.Zg
Keywords: protein aggregation, evaporation/condensation,
Carlo simulation
lattice model, Monte
6
1
0
2
n
a
J
4
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
8
7
4
0
0
.
1
0
6
1
:
v
i
X
r
a
a)Electronic mail: [email protected]
b)Electronic mail: [email protected]
1
I.
INTRODUCTION
The formation of amyloid fibrils is currently an intensely studied phenomenon.1 -- 3 Protein
aggregates of this type are found in pathological deposits in several human diseases, but
also with functional roles. In addition, they possess interesting mechanical properties, stem-
ming from their characteristic ordered cross-β organization. Insight into the mechanisms
of amyloid formation has been gained from kinetic profiles, as measured primarily by using
thioflavin T (ThT) fluorescence.4 In particular, it has been shown that kinetic data for a
broad range of systems can be well described in terms of a few basic mechanisms for the nu-
cleation and growth of fibrils, through a rate-equation approach.5 This approach can reveal
some general properties of intermediate species participating in the aggregation process, and
has proven useful for related self-assembly phenomena as well.6,7
Structure-based modeling of amyloid formation is a challenge to implement, due to the
wide range of spatial and temporal scales involved. Hence, all-atom computer simulations
with explicit solvent have focused on characterizing monomeric forms and early aggregation
events.8 By using coarse-grained models, at various levels of resolution, it has been possible
to study the formation and stability of larger assemblies9 -- 35 and also get insight into the
thermodynamic forces at play in amyloid formation.36 -- 39 However, to map out the thermo-
dynamics of amyloid formation as a function of control parameters such as temperature and
concentration is computationally demanding even in simple models.
In this article, we use cluster40 and generalized-ensemble41 -- 43 Monte Carlo (MC) tech-
niques, supplemented with a semi-analytical approximation, to investigate the thermody-
namics of a minimal model for amyloid formation.44 We study this model for three different
choices of intersheet interaction parameters. The first choice leads to aggregates with at
most two layers, and therefore an essentially 1D growth. The second choice permits ag-
gregates with more than two layers to form, but odd-layered aggregates are energetically
suppressed. This choice, inspired by evidence that the core of amyloid fibrils often has a
pairwise β-sheet organization,45,46 leads to a stepwise, quasi-2D growth. In the third and
final case, odd-layered aggregates are not suppressed, which opens up for 2D growth, al-
though slower laterally than longitudinally. Using N V T ensembles (N is the number of
peptides, V is volume, T is temperature), we investigate the equilibrium properties of these
three systems as a function of T and the concentration c = N/V . In addition, we study the
2
relaxation of the systems in MC simulations under fibril-favoring conditions, starting from
random initial states.
II. METHODS
A. Model
We use a minimal model for peptide aggregation where each peptide i is represented
by a unit-length stick centered at a site, ri, on a periodic cubic lattice with volume V =
L3.44 It is thus assumed that the internal dynamics of the peptides are fast, compared to
the timescales of fibril formation, and can be averaged out. The systems studied consist
of N identical peptides or sticks. Two peptides cannot simultaneously occupy the same
site. The orientation of a peptide is specified by two perpendicular lattice unit vectors,
bi and pi, yielding a total of 24 orientational states. The bi vector represents the N-to-C
backbone direction, whereas ±pi are the directions in which the peptide can form intrasheet
interactions. The vectors ±si = ±bi × pi represent sidechain directions, in which the
peptide can form intersheet interactions. Throughout the article, we assume units in which
the lattice spacing, the peptide mass and Boltzmann's constant have the value one.
The interaction energy is taken to have a pairwise additive form, E = (cid:80)
i<j ij, where
ij ≤ 0. The interaction geometry is illustrated in Fig. 1. Consider an arbitrary pair i and j
of peptides and let rij = rj − ri. The peptides interact (ij (cid:54)= 0) only if (i) they are nearest
neighbors on the lattice (rij = 1), (ii) their backbone vectors are aligned either parallel or
antiparallel to each other (bi · bj = 1), and (iii) bi · rij = bj · rij = 0. The interaction that
takes place when these conditions are met can be of one of three types, depending on the
relative orientation of the peptides:
1. The interaction is of intrasheet type if both pi and pj equal ±rij, and ij is then given
by
ij =
−(1 + ap)
−(1 + aap)
−1
if bi · bj = pi · pj = 1
if bi · bj = pi · pj = −1
otherwise
(rij = bi · bj = 1, pi, pj = ±rij)
(1)
3
FIG. 1. Example of a rectangular aggregate with length l = 6 and width w = 4. Along the fibril
axis, the peptides are bound together by parallel β interactions (solid blue, β). The four layers are
connected either via hh interactions (blue dashes, hh), or via pp interactions (blue dots, pp). For
potential B, pp interactions are weaker than hh interactions, which favors even-layered aggregates.
The symmetric potential A assigns equal energy to pp, hh and hp bonds. With potential C, hp and
pp interactions are missing, which makes lateral growth beyond two layers impossible.
where the first two cases represent parallel and antiparallel β-sheet structure, respec-
tively.
2. The interaction is of intersheet type if neither pi nor pj equals ±rij, which implies
that both si and sj equal ±rij. The +s and −s sides of a peptide, denoted by h and p,
respectively, are assumed to have different interaction properties. The +s, or h, side
is taken as more sticky or hydrophobic. The pair potential is given by
ij =
−(1 + bhh)
−(1 + bhp)
−(1 + bpp)
if si = −sj = rij
if si = sj
if − si = sj = rij
(rij = bi · bj = 1, si, sj = ±rij)
(2)
and is assumed lowest when the two h sides face each other (bhh ≥ bhp, bpp).
3. If the interaction is of neither of these two types (rij = bi· bj = 1, pi· pj = si·sj = 0),
the pair potential is set to ij = −1.
The intersheet interactions must be weak compared to the intrasheet interactions for
elongated fibril-like aggregates to form, but are nevertheless important. To assess the role
4
played by the intersheet interactions, we study the model using three potentials A, B and
C, which differ in the choice of the parameters bhh, bhp and bpp (Table I). The intrasheet
parameters ap and aap are the same in all three cases, namely ap = 5 and aap = 3.
Our previous study of this model was carried out using potential B.44 With this potential,
it was shown that aggregates grow in a stepwise fashion, where the major steps correspond to
changes in width. Aggregate growth may here be regarded as a quasi-2D process. Potential A
leads to less severe barriers to increases in width, and thereby to (asymmetric) 2D growth.
With potential C, there is no interaction at all (ij = 0) at hp and pp interfaces, which
prevents the formation of aggregates with more than two stacked sheets.
In this case,
aggregate growth becomes an effectively 1D process.
be conveniently defined via the inertia tensor. Specifically, we define l =(cid:112)12λ2
w = (cid:112)12λ2
In our model, aggregates can be assigned a length and a width, whereas growth in a third
dimension does not occur due to the interaction geometry. The length l and width w can
1/m + 1 and
2/m + 1, where λ1 ≥ λ2 are eigenvalues of this tensor and m is the number of
peptides in the aggregate. This definition is such that a rectangular aggregate consisting of
w stacked sheets with l peptides each (l ≥ w) is assigned exactly length l and width w.
B. MC simulations
In order to determine the thermodynamics of these systems, one needs simulations in
which large aggregates form and dissolve many times, which is challenging to achieve even
in a simple model. For our thermodynamic simulations, we therefore use a Swendsen-Wang --
type cluster move40 and a flat-histogram procedure.41 -- 43,47
The cluster move is based on a stochastic cluster construction scheme.40 The procedure
TABLE I. Our three choices of the intersheet interactions parameters bhh, bhp and bpp (Eq. 2), and
the corresponding growth behavior of aggregates.
Potential
A
B
C
bhh
0.5
1
1
bhp
0.5
0
−1
5
bpp
0.5
0
−1
Growth
2D
quasi-2D
1D
is recursive and begins by picking a random first cluster member, i. Then, all peptides j
interacting with peptide i (ij < 0) are identified and added to the cluster with probability
Pij = 1− eβij, where β is inverse temperature. This step is iterated until there are no more
peptides to be tested for inclusion in the growing cluster. The resulting cluster is subject to
a trial rigid-body translation or rotation, which is accepted whenever it does not cause any
steric clashes. Unlike simpler cluster moves, this update can split and merge aggregates.
The update fulfills detailed balance with respect to the canonical microstate distribution,
Pν ∝ e−βEν.
To further enhance the sampling, we use the multicanonical method,41 -- 43 which can be
very useful for systems with multimodal energy landscapes. Our simulation procedure con-
sists of three steps.47 First, we estimate the density of states, g(E), by the Wang-Landau
method43 (an early variant of which was proposed in Ref. 48). Second, keeping this esti-
mate, g(E), fixed, we simulate the ensemble Pν ∝ 1/g(Eν), whose energy distribution is
approximately flat.41 Finally, we calculate canonical averages via reweighting to the desired
temperature.49 Throughout these simulations, we restrict the sampling to energies above
a cutoff Emin. This cutoff is taken sufficiently high to avoid sampling of states containing
unphysical cyclic aggregates, but sufficiently low to permit unbiased studies over the temper-
ature range of interest. A more advanced, diffusion-optimized generalized-ensemble method
was recently tested on this model.50 For our present purposes, the speed-up brought by the
flat-histogram method suffices.
In our flat-histogram simulations, we use both single-peptide moves and the cluster move.
The cluster move, as defined above, does not satisfy detailed balance with respect to Pν ∝
1/g(Eν). This can be easily rectified by adding a Metropolis accept/reject step with the
acceptance probability given by Pacc(ν → ν(cid:48)) = min[1, g(Eν)e−βEν /g(Eν(cid:48))e−βEν(cid:48) ]. Note that,
in this context, β is a tunable algorithm parameter, entering in the cluster construction,
rather than a physical parameter. In our simulations, this parameter is chosen in the vicinity
of the inverse fibrillation temperature.
In addition to the thermodynamic simulations, we perform relaxation simulations under
fibril-favoring constant-temperature conditions. Here, motivated by experimental indica-
tions that amyloid growth occurs dominantly via monomer addition,51 we use single-peptide
moves only; the elementary moves are translations by one lattice spacing and rotations of
individual peptides. Since there is no need to observe transitions back and forth between
6
states with and without large aggregates, the systems can be much larger than in the ther-
modynamic simulations (∼105 rather than ∼102 peptides).
All statistical uncertainties quoted below are 1σ errors.
C. Semi-analytical approximation
In this section, we present a semi-analytical method which can be used to estimate ther-
modynamic properties of the model for system sizes not amenable to direct simulation. Let
ξ denote a certain configuration of mξ peptides forming an aggregate, and let Nξ denote
the number of such aggregates in the system. Treating the Nξ's as independent variables,
the chemical potential for aggregates with configuration ξ may be defined as µξ ≡ ∂F/∂Nξ,
where F (T, V,{Nξ}) is a Helmholtz free energy. Imposing that the total number of peptides
(cid:80)
ξ mξNξ adds up to N, the function
(cid:32)(cid:88)
(cid:33)
F = F + λ
mξNξ − N
,
(3)
is minimized at equilibrium. After eliminating the Lagrange multiplier λ, this leads to an
equilibrium condition on the chemical potentials, namely
ξ
µξ = mξµ1,
(4)
where µ1 is the monomer chemical potential.
With a simplified grand-canonical description, the partition function for the set of aggre-
gates with configuration ξ is given by
∞(cid:88)
Nξ=0
Zξ =
eβµξNξ
Nξ!
Z (1)
ξ
Nξ = exp(cid:2)gV eβ(µξ−Eξ)(cid:3) ,
(5)
ξ = gV e−βEξ is the single-aggregate partition function, Eξ is the internal energy,
where Z (1)
and g = 24 is the number of possible spatial orientations of the aggregate. This description
neglects interactions between aggregates, but note that two adjacent ξ aggregates correspond
to one aggregate of some other type ξ(cid:48). Eq. 5 implies that the Nξ variable is Poisson
distributed with mean
= gV eβ(µ1mξ−Eξ),
(6)
(cid:104)Nξ(cid:105) =
1
β
∂ log Zξ
∂µξ
7
where Eq. 4 has been used. Given β, V and N, the monomer chemical potential can be
determined by approximating Nξ ≈ (cid:104)Nξ(cid:105) and solving
N =
mξNξ,
(7)
for µ1. Knowing µ1, one can obtain the Nξ's from Eq. 6 and compute the total energy as
(cid:88)
(cid:88)
ξ
E =
EξNξ .
(8)
ξ
The fibrillation temperature Tf may be defined as the maximum of the heat capacity, and can
therefore be estimated by numerically differentiating Eq. 8. This procedure for computing
Tf is fast and can be repeated for many different concentrations c = N/V , once the relevant
aggregates ξ and their energies Eξ are specified. Hence, it can be used to estimate the state
of the system as a function of both T and c.
The above scheme has similarities to the approach of Oosawa and Kasai,52 but has here
been derived for more general choices of included aggregates. When applying it to the
present model, we only consider rectangular aggregates with an energetically optimal internal
organization. The generic index ξ can therefore be replaced by the aggregate length l and
width w. This set of configurations turns out to be sufficient to obtain quite accurate
estimates of Tf. To respect the finite size of the systems, we limit the sums over l and w to
l ≤ lmax = min(N, L) and w ≤ wmax(l) = min [L, floor (N/l)]. For the potential C, which
leads to aggregates with at most two proper sheets, we set wmax(l) = min [2, floor(N/l)].
The energy of an aggregate with length l and width w, Elw, is a sum of intra- and
In the minimum energy configuration, the intrasheet energy is
intersheet contributions.
−(1 + ap)(l − 1)w, corresponding to a parallel organization of the l peptides, whereas the
intersheet energy depends on the parameters bhh, bhp and bpp. For the potentials A, B and
C (Table I), the respective minimum total energies are given by
E(A)
lw = −(1 + ap)(l − 1)w − 3l(w − 1)
lw = −(1 + ap)(l − 1)w − 3l(w − 1)
E(B)
lw = −(1 + ap)(l − 1)w − 2l(w − 1)
E(C)
2
2
− (1 + (−1)w)l
4
(w = 1, 2)
(9)
(10)
(11)
With potential B, the minimal energy is achieved when both outer surfaces are entirely
polar, which maximizes the number of favorable hh contacts. Likewise, with potential C,
the energy of a two-sheet aggregate is minimal if the entire interface is of hh type.
8
III. RESULTS AND DISCUSSION
We study both equilibrium and relaxation properties of the above model for the three
choices of intersheet interaction parameters listed in Table I, which correspond to 2D, quasi-
2D and 1D growth and are referred to as A, B and C, respectively. The intrasheet interac-
tions, which are stronger, stay the same in all three cases.
A. Equilibrium properties
The equilibrium properties of the model are investigated by using MC simulations and
the semi-analytical approximation described in Sec. II. We first present the MC results
which, unless otherwise stated, are obtained using N = 256 and L = 64, corresponding to a
concentration of c = 0.977 × 10−3 per unit volume.
All three systems contain large fibril-like aggregates at low T , while being disordered
at high T . The onset of fibril formation is accompanied by a peak in the heat capacity
CV = ((cid:104)E2(cid:105)−(cid:104)E(cid:105)2)/T 2 (Fig. 2a). The fibrillation temperature, Tf, may therefore be defined
as the maximum of CV , and is found to be given by T (A)
f =
0.67093 ± 0.00007 and T (C)
f = 0.6548 ± 0.0002 for systems A, B and C, respectively.
f = 0.66500 ± 0.00007, T (B)
While Tf is thus roughly similar for all three systems, there are large differences in the
height of the CV peak (Fig. 2a). This fact reflects a fundamental difference between systems
A and B, on one hand, and system C, on the other hand, as can be seen from the probability
distribution of the total system energy E (Fig. 2b). For systems A and B, with a pronounced
peak in CV , the energy distribution is clearly bimodal, showing that these systems can exist
in two distinct types of states at T = Tf. The difference between the two potentials shows
up in the location of the low-energy peak. For system C, the CV peak is broader and lower.
In this system, the onset of fibril formation is smooth. As the temperature is reduced, the
energy distribution slides toward lower values while retaining a unimodal shape.
The behavior of the systems at the fibrillation temperature can be further characterized
in terms of the aggregate-size distribution, p(m), which gives the probability for a random
peptide to be part of an aggregate with size m (Fig. 3a). For systems A and B, p(m)
is bimodal at T = Tf. Hence, whereas both small and large aggregates occur in these
systems, there is a range of suppressed intermediate sizes. Above, it was seen that the energy
9
FIG. 2. (a) Temperature-dependence of the heat capacity, CV , from MC simulations with N =
256 and L = 64 for the potentials A, B and C (Table I). The shaded bands indicate statistical
uncertainty. Our flat-histogram simulations sample energies E > Emin, where Emin is a cutoff.
Data are shown only at temperatures that are sufficiently high for the effects of this cutoff to
be negligible. (b) Distribution of energy, P (E), at T = Tf, for the same three systems. Due to
short-scale irregularities in the density of states, a moving average is used (window size ∆E = 10).
distribution is bimodal as well (Fig. 2b). States belonging to the low-energy peak contain
both small and large aggregates and contribute, therefore, to both peaks in p(m), whereas
high-energy states are dominated by small aggregates. Fig. 3b shows the contributions to
p(m) from low- and high-energy states in system B, which indeed are bi- and unimodal,
respectively. Also worth noting in this figure is that the amount of aggregates with size
between m ≈ 7 and m ≈ 35 tends to be much smaller in low-energy states than in high-
energy states. Hence, the appearance of large aggregates in low-energy states occurs, at least
in part, at the expense of these mid-size ones. For system C, the aggregate-size distribution
p(m) is fundamentally different (Fig. 3a). In this system, there is no intermediate range of
suppressed sizes m, and therefore no clear division into either small or large species.
The intersheet interactions directly influence the width of the aggregates, w (see Sec. II).
Fig. 4 shows the mass-weighted distribution of w, p(w), at T = Tf for our three systems.
As expected, p(w) decays rapidly beyond w = 2 for potential C, whereas potentials A
and B permit the formation of wider aggregates (Fig. 4). The data also confirm that the
asymmetric intersheet interactions of potential B indeed favor even-layered aggregates over
10
50 100 150 0.64 0.65 0.66 0.67 0.68 0.69CV (/103)T(a)potential Apotential Bpotential C 0 0.001 0.002 0.003 0.004 0.005 0.006-1200-1000-800-600-400-200P(E)E(b)potential Apotential Bpotential C FIG. 3. (a) Mass fraction of aggregates with size m, p(m), against m at T = Tf, as obtained from
MC simulations with N = 256 and L = 64 for the potentials A, B and C (Table I). The shaded
bands indicate statistical uncertainty. The saw-tooth-like behavior that occurs for systems B and C
is due to even-odd effects for two-layered aggregates. (b) Decomposition of p(m) for system B into
contributions from low- and high-energy states, respectively (compare Fig. 2b). Each configuration
in the simulated ensemble is classified as either low energy (E < −500) or high energy (E ≥ −500).
FIG. 4. Mass fraction of aggregates with width w, p(w), against w at T = Tf, as obtained from
MC simulations with N = 256 and L = 64 for the potentials A, B and C (Table I).
odd-layered ones.
The above discussion focused on results obtained using N = 256 and L = 64. To better
understand the sharp onset of fibril formation in systems A and B, additional simulations
were performed for a few different N, keeping the concentration approximately constant.
11
-12-10-8-6-4-2 0 0 20 40 60 80 100 120 140 160ln p(m)m(a)potential Apotential Bpotential C-12-10-8-6-4-2 0 0 20 40 60 80 100 120 140 160ln p(m)m(b)high energylow energy-10-8-6-4-2 0 2 0 1 2 3 4 5 6 7 8 9ln p(w)wpotential Apotential Bpotential C FIG. 5. Finite-size scaling for system B (Table I), at fixed concentration. The data are from
MC simulations with N = 128, 256, 512 and 1028 (L = 51, 64, 81 and 102). (a) Temperature
dependence of the specific heat, CV /N. The shaded bands indicate statistical uncertainty. (b)
Probability distribution of the energy density E/N at the specific-heat maximum Tf.
Fig. 5a shows the specific heat, CV /N, of system B for N = 128, 256, 512 and 1024. As N
increases, the peak in CV /N gets sharper. However, the height of the peak increases more
slowly than the linear growth expected at a first-order phase transition with a non-zero
specific latent heat.
Indeed, the latent heat, or energy gap, does not scale linearly with
N (Fig. 5b). Still, the gap grows sufficiently fast (faster than N 1/2) for the bimodality of
the energy distribution to become more and more pronounced with increasing N (Fig. 5b).
Therefore, fibril formation sets in at a first-order-like transition, where distinct states coexist.
Similar analyses were performed for potentials A and C, using N = 128, 256 and 512. The
results obtained with potential A are qualitatively similar to those just described for potential
B. For potential C, CV,max/N does not grow with N, thus confirming the conclusion that, in
this case, the onset of fibril formation represents a crossover rather than a sharp transition.
Our systems resemble a lattice gas at fixed particle number, albeit with asymmetric in-
teractions. For finite-volume liquid-vapor systems at phase coexistence, the formation of
droplets due to a fixed particle excess above the ambient gas concentration has been ex-
tensively investigated,53 -- 63 often by mapping to the Ising model at fixed magnetization. A
sharp transition has been shown to occur, below which the particle excess can be accommo-
dated by gas-phase fluctuations. At the transition point, a large droplet appears, whereas
intermediate-size droplets remain strongly suppressed. The volume of the droplet and the
12
0 200 400 600 800 1000 1200 1400 0.65 0.66 0.67 0.68 0.69CV / NT(a)N = 128N = 256N = 512N = 1024 0.001 0.01 0.1 1 10-5-4-3-2-1 0P(E/N)E/N(b)N = 128N = 256N = 512N = 1024 FIG. 6.
Concentration of free peptides, cs, against total concentration, c, as obtained by MC
simulations of system B (Table I) for T = 0.67093, L = 64 and different N. At this temperature,
fibrillation sets in at cf ≈ 256/643. Our definition of free peptides includes monomers and aggregates
with size m ≤ 6, so cs = c ×(cid:80)6
m=1 p(m).
latent heat scale as V 3/4. To accurately determine the corresponding scaling behavior for
our systems A and B, data over a wider range of system sizes would be required. However,
the scaling of the latent heat does seem to be faster than V 1/2 and slower than V (Fig. 5b).
At the droplet condensation transition, the gas concentration drops by an amount that
scales as V −1/4.55,59 In our systems A and B, at the threshold concentration for fibril forma-
tion, cf(T ), a similar drop occurs in the concentration of free peptides, cs, as is illustrated in
Fig. 6 by data obtained with potential B for L = 64 and different N. To test the scaling with
system size, this drop, ∆cs, was computed at the heat-capacity maxima of Fig. 5a, for system
B and four different V . These ∆cs values vary roughly as V −1/4 with V (∆cs×V 1/4 = 0.0089,
0.0079, 0.0085 and 0.0086 for N = 128, 256, 512 and 1024, respectively).
The results presented so far were obtained by MC simulations, which are bias-free but
time-consuming. To be able to study larger systems, the approximate but much faster
semi-analytical approach (Sec. II) is used. For L = 64 and N = 256, this method provides
estimates of the fibrillation temperature (T (A)
f = 0.6509) that
agree to within ∼1% with the MC results, and the ordering T (C)
is correct.
Having seen this agreement, the method is applied to estimate the threshold concentration,
cf(T ), as a function of temperature for the box size used in the relaxation simulations below,
that is L = 512. To this end, the fibrillation temperature is calculated for a large set of
f = 0.6783, T (B)
f = 0.6866, T (C)
f
f < T (A)
f < T (B)
13
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2cs ( ×103 )c ( ×103 )cs = c c = cf FIG. 7. Temperature-dependence of the threshold concentration, cf(T ), for systems A, B and C
(Table I), as obtained by the semi-analytical approximation (Sec. II) for L = 512. For system C,
cf(T ) represents a crossover rather than a sharp transition. The grey vertical bar indicates the T, c
interval studied in our relaxation simulations.
concentrations in the range 1.0× 10−6 < c < 2.3× 10−3. The resulting estimates of cf(T ) are
shown in Fig. 7. The curves for systems A and B, with closely related energies (Eqs. 9,10),
agree to within ∼1% at T = 0.65 and become even more similar at higher T . For system C,
our method estimates a higher cf(T ). As discussed above, in this system, cf(T ) represents a
crossover rather than a sharp transition. The same scheme also provides an estimate of the
heat capacity. It predicts CV to vary smoothly with T in system C but that a jump occurs
at T = Tf in systems A and B, all of which match well with our earlier conclusions based on
MC data for smaller systems.
B. Relaxation simulations
Having located the threshold concentration cf(T ) for fibril formation, we next study the
relaxation of the systems in constant-temperature MC simulations with c > cf(T ), starting
from random initial states. Assuming fibril growth to occur through monomer addition,51
the simulations are performed using single-peptide moves only. The parameters T = 0.6535
and L = 512 are the same in all these calculations, whereas N varies between 217 = 131, 072
and 300,000. The corresponding c interval is indicated in Fig. 7. To assess statistical
uncertainties, a set of eight independent runs is generated for each choice of concentration
and potential.
14
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 0.45 0.5 0.55 0.6 0.65 0.7 0.75cf (T) (×103)Tfibril + small speciessmall species onlyCrossover (C)Coexistence line (A, B) FIG. 8. MC time evolution of the mass fractions of monomers (m = 1), small aggregates (1 <
m ≤ 6), mid-size aggregates (6 < m ≤ 60) and large aggregates (m > 60) in simulations with the
potentials A, B and C (Table I), using T = 0.6535, L = 512 and N = 170, 000. At large times,
small and large species dominate in systems A and B, whereas mid-size aggregates remain present
in system C. Each MC sweep consists of N single-peptide moves. The data represent averages over
eight independent runs. The statistical uncertainties, indicated by shaded bands, are barely visible.
Fig. 8 illustrates how aggregation proceeds with potentials A, B and C, by showing
the evolution of the respective mass fractions of (i) monomers, (ii) small aggregates with
1 < m ≤ 6 peptides, (iii) mid-size aggregates with 6 < m ≤ 60, and (iv) large aggregates
with m > 60. The number of peptides is the same in all three cases (N = 170, 000).
The monomer fraction is close to unity in the random initial states, but roughly a factor
2 smaller already at the time of the first measurement, due to rapid equilibration between
monomers and small aggregates. After this point, the amounts of monomers and small
aggregates decrease monotonically toward apparent steady-state levels. The fate of the mid-
size aggregates depends on the potential. In systems A and B, these aggregates are transient
species. Examples of final configurations from the simulations of these systems can be found
in Fig. 9, both of which contain many large fibril-like aggregates but only very few mid-size
ones. In system C, there is, by contrast, a non-negligible amount of mid-size aggregates still
present in the apparent steady-state regime.
The apparent steady-state regimes in these simulations need not correspond to thermody-
namic equilibrium states. In fact, it is likely that the true equilibrium states of the systems
shown in Fig. 9 contain only one very large aggregate accompanied by surrounding small
species, as observed in our equilibrium simulations of smaller systems. However, due to the
15
0 0.2 0.4 0.6 0.8 1 0 5 10 15mass fractionMC sweeps (/106)Am = 11 < m ≤ 66 < m ≤ 60m > 60 0 0.2 0.4 0.6 0.8 1 0 5 10 15 20mass fractionMC sweeps (/106)Bm = 11 < m ≤ 66 < m ≤ 60m > 60 0 0.2 0.4 0.6 0.8 1 0 5 10 15mass fractionMC sweeps (/106)Cm = 11 < m ≤ 66 < m ≤ 60m > 60 FIG. 9. Final configurations from relaxation simulations with potentials A and B (Table I), using
T = 0.6535, L = 512 and N = 170, 000. Large (m > 60) and mid-size (6 < m ≤ 60) aggregates
are shown in blue and red, respectively. For clarity, small species (m ≤ 6) are not shown. At this
stage, the mid-size aggregates have almost disappeared (compare Fig. 8). Red circles indicate one
of ≤5 such aggregates in each configuration. Large aggregates tend to be shorter and wider with
potential A than they are with potential B.
very slow dynamics of large aggregates, the states shown in Fig. 9 are effectively frozen on
the timescales of our simulations.
Finally, we also study how the overall rate of fibril formation scales with concentration in
our relaxation simulations, focusing on systems A and B. Here, an aggregate is taken to be
a fibril if its width w exceeds 3.5, because thinner aggregates are unstable. This definition
is somewhat arbitrary, but ambiguous assemblies close to the cutoff in width are transient
species that essentially disappear as aggregation proceeds.
Fig. 10 shows the MC evolution of the total fibril mass in systems A and B for different
concentrations. The curves are sigmoidal in shape, especially at low c. As expected, as
c is increased, aggregation gets faster and the saturation level gets higher. The statistical
errors are small because our systems are large. A simple measure of the overall rate of
fibril formation is the time, t1/2, at which half the saturation level is reached. In amyloid
formation, the scaling of t1/2 with c has often, but not always,64 been found to be well
described by a power law, t1/2 ∼ cγ, where the exponent γ ≤ −0.5 depends on both the
protein and the conditions under which the fibrils grow.5 Data for t1/2 from our relaxation
simulations do not show a perfect power-law behavior, as can be seen from a log-log plot
16
potential Apotential B FIG. 10. MC evolution of the total fibril mass in relaxation simulations with the potentials A and
B (Table I) for T = 0.6563, L = 512 and different N between 217 = 131, 072 and 300,000. The data
represent averages over eight independent runs, started from random initial states. The statistical
uncertainties, indicated by shaded bands, are barely visible.
FIG. 11. Concentration-dependence of the half-time for fibril formation, t1/2, as extracted from
the simulations shown in Fig. 10. The lines are power-law fits to the data, t1/2 ∝ cγ. The fitted
exponents are γ(A) = −2.3 ± 0.2 and γ(B) = −0.8 ± 0.1 for potentials A and B, respectively.
(Fig. 11). Nevertheless, to get a measure of the overall strength of the c-dependence, power-
law fits were performed. For system B, with quasi-2D growth, the fitted exponent, γ(B) =
−0.8± 0.1, indicates a c-dependence comparable in strength to that of typical experimental
data.5 For system A, with 2D growth, the c-dependence is slightly stronger, with a fitted
exponent of γ(A) = −2.3 ± 0.2.
17
0 50 100 150 200 0 5 10 15 20fibril mass (/103)MC sweeps (/106)potential A131k150k170k190k210k250k 0 50 100 150 200 250 0 5 10 15 20 25fibril mass (/103)MC sweeps (/106)potential B150k170k190k210k250k300k 1 10 1 1.2 1.5 2t1/2 ( MC sweeps / 106 )c ( ×103 )potential Apotential B IV. SUMMARY
Amyloid formation involves a wide range of spatial and temporal scales. In this article,
we have used a minimal lattice-based model to investigate the overall thermodynamics of
amyloid formation in finite systems under N V T conditions. With 2D or quasi-2D aggregate
growth, the model exhibits a sharp transition, from a supersaturated solution state to a dis-
tinct state where small and large species exist in equilibrium. At the threshold concentration,
cf(T ), these states coexist, thus giving rise to a bimodal energy distribution. At concentra-
tions not too much higher than cf(T ), there exists, therefore, a local free-energy minimum
corresponding to a metastable solution state, in which the system can get trapped, thereby
causing fibril formation to occur after a lag period. At and above cf(T ), while both small
and large aggregates are present, intermediate-size ones are suppressed. With 1D growth,
this suppression is not observed, and the energy distribution is unimodal. Intuitively, the
first-order-like transition seen with 2D or quasi-2D growth stems from a competition be-
tween bulk and surface energies. With 1D growth, this mechanism is missing, because the
surface energy is associated with the fibril endpoints, whose size does not grow with fibril
mass. Previous work has studied the dependence of the solubility of fibrils on their width,
using different models.17,36 One study compared one-, two- and three-layered aggregates,
and showed that the stability region in the T ,c plane grows with increasing fibril width.17
This behavior suggests that fibril formation in a finite system may set in at a concentration
roughly corresponding to the solubility of the widest aggregates that occur for this system
size. Upon increasing N (at fixed c and T ), one would then expect a growth in both latent
heat and threshold concentration, as is indeed observed in our simulations.
The first-order-like onset of fibril formation that we observe with 2D or quasi-2D growth
shows similarities with the droplet evaporation/condensation transition at liquid-vapor co-
existence, which has been extensively investigated.53 -- 63 Indeed, at this transition, mid-size
droplets are suppressed and the energy distribution is bimodal. Furthermore, the specific
latent heat, which we find to decrease with system size (Fig. 5b), is known to vanish at the
droplet transition in the limit of infinite system size.
Our equilibrium findings may be used to rationalize, in part, properties observed in our
relaxation simulations. For the systems studied here, with >105 peptides, the MC evolution
of the total fibril mass turns out to be highly reproducible from run to run. The trajectories
18
are, at not too high concentrations, sigmoidal, with an initial lag phase (Fig. 10). Due to slow
dynamics of large aggregates, the apparent steady-state levels at the end of the runs need not
correspond to equilibrium states. However, as at equilibrium, intermediate-size aggregates
are suppressed in the final states (Fig. 8), which is in line with experimental findings.65.
With the droplet interpretation, this statistical suppression occurs because intermediate-
size aggregates correspond to a free-energy maximum, at which the bulk free energy and
surface energy terms balance each other. The precise shape of the aggregate-size distribution
is influenced by factors that are unlikely to be captured by our simple model, such as the
existence of specific oligomeric states with enhanced stability. One type of aggregate that
does not occur in our simulations, due to the model geometry, is closed β-barrels, which
have a potentially high stability for their size.11
ACKNOWLEDGMENTS
This work benefitted greatly from the expertise and helpfulness of the late Thomas
Neuhaus. We thank Sigurður AE. Jónsson and Stefan Wallin for useful discussions. This
work was in part supported by the Swedish Research Council (Grant no. 621-2014-4522).
The simulations were performed on resources provided by the Swedish National Infrastruc-
ture for Computing (SNIC) at LUNARC, Lund University.
REFERENCES
1F. Chiti and C. M. Dobson, Annu. Rev. Biochem. 75, 333 (2006).
2T. P. J. Knowles and M. J. Buehler, Nat. Nanotechnol. 6, 469 (2011).
3T. Härd, J. Phys. Chem. Lett. 5, 607 (2014).
4E. Hellstrand, B. Boland, D. M. Walsh, and S. Linse, ACS Chem. Neurosci. 1, 13 (2010).
5T. P. J. Knowles, C. A. Waudby, G. L. Devlin, S. I. A. Cohen, A. Aguzzi, M. Vendruscolo,
E. M. Terentjev, M. E. Welland, and C. M. Dobson, Science 326, 1533 (2009).
6F. A. Ferrone, J. Hofrichter, and W. Eaton, J. Mol. Biol. 183, 611 (1985).
7H. Flyvbjerg, E. Jobs, and S. Leibler, Proc. Natl. Acad. Sci. USA 93, 5975 (1996).
8J. E. Straub and D. Thirumalai, Curr. Opin. Struct. Biol. 20, 187 (2010).
19
9S. Auer, C. M. Dobson, M. Vendruscolo, and A. Maritan, Phys. Rev. Lett. 101, 258101
(2008).
10C. Junghans, M. Bachmann, and W. Janke, J. Chem. Phys. 128, 085103 (2008).
11A. Irbäck and S. Mitternacht, Proteins 71, 207 (2008).
12D. Li, S. Mohanty, A. Irbäck, and S. Huo, PLoS Comput. Biol. 4, e1000238 (2008).
13M. S. Li, D. K. Klimov, J. E. Straub, and D. Thirumalai, J. Chem. Phys. 129, 175101
(2008).
14G. Bellesia and J.-E. Shea, J. Chem. Phys. 130, 145103 (2009).
15Y. Lu, P. Derreumaux, Z. Guo, N. Mousseau, and G. Wei, Proteins 75, 954 (2009).
16Y. Wang and G. A. Voth, J. Phys. Chem. B 114, 8735 (2010).
17S. Auer and D. Kashchiev, Phys. Rev. Lett. 104, 168105 (2010).
18D. Kashchiev and S. Auer, J. Chem. Phys. 132, 215101 (2010).
19R. Friedman, R. Pellarin, and A. Caflisch, J. Phys. Chem. Lett. 1, 471 (2010).
20A. Rojas, A. Liwo, D. Browne, and H. A. Scheraga, J. Mol. Biol. 404, 537 (2010).
21B. Urbanc, M. Betnel, L. Cruz, G. Bitan, and D. B. Teplow, J. Am. Chem. Soc. 132, 4266
(2010).
22S. Kim, T. Takeda, and D. K. Klimov, Biophys. J. 99, 1949 (2010).
23M. Cheon, I. Chang, and C. K. Hall, Biophys. J. 101, 2493 (2011).
24B. Linse and S. Linse, Mol. BioSyst. 7, 2296 (2011).
25M. Baiesi, F. Seno, and A. Trovato, Proteins 79, 3067 (2011).
26S. P. Carmichael and M. S. Shell, J. Phys. Chem. B 116, 8383 (2012).
27N. S. Bieler, T. P. J. Knowles, D. Frenkel, and R. Vácha, PLoS Comput. Biol. 8, e1002692
(2012).
28M. R. Smaoui, F. Poitevin, M. Delarue, P. Koehl, H. Orland, and J. Waldispühl, Biophys.
J. 104, 683 (2013).
29R. Ni, S. Abeln, M. Schor, M. A. Cohen Stuart, and P. G. Bolhuis, Phys. Rev. Lett. 111,
058101 (2013).
30W. Zheng, N. P. Schafer, and P. G. Wolynes, Proc. Natl. Acad. Sci. USA 110, 20515
(2013).
31L. Di Michele, E. Eiser, and V. Foderà, J. Phys. Chem. Lett. 4, 3158 (2013).
32S. Abeln, M. Vendruscolo, C. M. Dobson, and D. Frenkel, PLoS One 9, e85185 (2014).
33A. Morriss-Andrews and J.-E. Shea, J. Phys. Chem. Lett. 5, 1899 (2014).
20
34A. Šarić, Y. C. Chebaro, T. P. J. Knowles, and D. Frenkel, Proc. Natl. Acad. Sci. USA
111, 17869 (2014).
35S. Assenza, J. Adamcik, R. Mezzenga, and P. De Los Rios, Phys. Rev. Lett. 113, 268103
(2014).
36J. Zhang and M. Muthukumar, J. Chem. Phys. 130, 035102 (2009).
37S. Auer, J. Chem. Phys. 135, 175103 (2011).
38J. D. Schmit, K. Ghosh, and K. Dill, Biophys. J. 100, 450 (2011).
39S. Auer, J. Phys. Chem. B 118, 5289 (2014).
40R. H. Swendsen and J.-S. Wang, Phys. Rev. Lett. 58, 86 (1987).
41B. A. Berg and T. Neuhaus, Phys. Lett. B 267, 249 (1991).
42U. H. E. Hansmann and Y. Okamoto, J. Comput. Chem. 14, 1333 (1993).
43F. Wang and D. P. Landau, Phys. Rev. Lett. 86, 2050 (2001).
44A. Irbäck, S. AE. Jónsson, N. Linnemann, B. Linse, and S. Wallin, Phys. Rev. Lett. 110,
058101 (2013).
45M. R. Sawaya, S. Sambashivan, R. Nelson, M. I. Ivanova, S. A. Sievers, M. I. Apostol, M. J.
Thompson, M. Balbirnie, J. J. W. Wiltzius, H. T. McFarlane, A. Ø. Madsen, C. Riekel,
and D. Eisenberg, Nature 447, 453 (2007).
46A. W. P. Fitzpatrick, G. T. Debelouchina, M. J. Bayro, D. K. Clare, M. A. Caporini,
V. S. Bajaj, C. P. Jaroniec, L. Wang, V. Ladizhansky, S. A. Müller, C. E. MacPhee, C. A.
Waudby, H. R. Mott, A. De Simone, T. P. J. Knowles, H. R. Saibil, M. Vendruscolo, E. V.
Orlova, R. G. Griffin, and C. M. Dobson, Proc. Natl. Acad. Sci. USA 110, 5468 (2013).
47S. AE. Jónsson, S. Mohanty, and A. Irbäck, J. Chem. Phys. 135, 125102 (2011).
48O. Engkvist and G. Karlström, Chem. Phys. 213, 63 (1996).
49A. M. Ferrenberg and R. H. Swendsen, Phys. Rev. Lett. 63, 1195 (1989).
50P. Tian, S. AE. Jónsson, J. Ferkinghoff-Borg, S. V. Krivov, K. Lindorff-Larsen, A. Irbäck,
and W. Boomsma, J. Chem. Theory Comput. 10, 543 (2014).
51S. R. Collins, A. Douglass, R. D. Vale, and J. S. Weissman, PLoS Biol. 2, e321 (2004).
52F. Oosawa and M. Kasai, J. Mol. Biol. 4, 10 (1962).
53K. Binder and M. H. Kalos, J. Stat. Phys. 22, 363 (1980).
54H. Furukawa and K. Binder, Phys. Rev. A 26, 556 (1982).
55M. Biskup, L. Chayes, and R. Kotecký, EPL 60, 21 (2002).
56M. Biskup, L. Chayes, and R. Kotecký, Commun. Math. Phys. 242, 137 (2003).
21
57T. Neuhaus and J. S. Hager, J. Stat. Phys. 113, 47 (2003).
58M. Biskup, L. Chayes, and R. Kotecký, J. Stat. Phys. 116, 175 (2004).
59L. G. MacDowell, P. Virnau, M. Müller, and K. Binder, J. Chem. Phys. 120, 5293 (2004).
60A. Nussbaumer, E. Bittner, T. Neuhaus, and W. Janke, EPL 75, 716 (2006).
61A. Nussbaumer, E. Bittner, and W. Janke, Prog. Theor. Phys. Suppl. 184, 400 (2010).
62B. Bauer, E. Gull, S. Trebst, M. Troyer, and D. A. Huse, J. Stat. Mech., P01020(2010).
63T. Nogawa, N. Ito, and H. Watanabe, Phys. Rev. E 84, 061107 (2011).
64G. Meisl, X. Yang, E. Hellstrand, B. Frohm, J. B. Kirkegaard, S. Cohen, C. M. Dobson,
S. Linse, and T. P. J. Knowles, Proc. Natl. Acad. Sci. USA 111, 9384 (2014).
65D. M. Walsh, A. Lomakin, G. B. Benedek, M. M. Condron, and D. B. Teplow, J. Biol.
Chem. 272, 22364 (1997).
22
|
1606.07342 | 1 | 1606 | 2016-06-23T15:17:21 | Modeling a spheroidal microswimmer and cooperative swimming in thin films | [
"physics.bio-ph",
"cond-mat.soft",
"physics.flu-dyn"
] | We propose a hydrodynamic model for a spheroidal microswimmer with two tangential surface velocity modes. This model is analytically solvable and reduces to Lighthill's and Blake's spherical squirmer model in the limit of equal major and minor semi-axes. Furthermore, we present an implementation of such a spheroidal squirmer by means of multiparticle collision dynamics simulations. We investigate its properties as well as the scattering of two spheroidal squirmers in a slit geometry. Thereby we find a stable fixed point, where two pullers swim cooperatively forming a wedge-like conformation with a small constant angle. | physics.bio-ph | physics |
Modeling a spheroidal microswimmer and cooperative swimming in thin films
Mario Theers,1, ∗ Elmar Westphal,2, † Gerhard Gompper,1, ‡ and Roland G. Winkler1, §
1Theoretical Soft Matter and Biophysics, Institute for Advanced Simulation and Institute of Complex Systems,
Forschungszentrum Julich, D-52425 Julich, Germany
2Peter Grunberg Institute and Julich Centre for Neutron Science,
Forschungszentrum Julich, D-52425 Julich, Germany
(Dated: June 24, 2016)
We propose a hydrodynamic model for a spheroidal microswimmer with two tangential surface
velocity modes. This model is analytically solvable and reduces to Lighthill's and Blake's spherical
squirmer model in the limit of equal major and minor semi-axes. Furthermore, we present an imple-
mentation of such a spheroidal squirmer by means of multiparticle collision dynamics simulations.
We investigate its properties as well as the scattering of two spheroidal squirmers in a slit geometry.
Thereby we find a stable fixed point, where two pullers swim cooperatively forming a wedge-like
conformation with a small constant angle.
I.
INTRODUCTION
Living matter exhibits a broad spectrum of unique
phenomena which emerge as a consequence of its active
constituents. Examples of such systems range from the
macroscopic scale of flocks of birds and mammalian herds
to the microscopic scale of bacterial suspensions [1, 2].
Specifically, active systems exhibit remarkable nonequi-
librium phenomena and emergent behavior like swarm-
ing [3–7], turbulence [6], and activity-induced cluster-
ing and phase transitions [8–21]. The understanding of
these collective phenomena requires the characterization
of the underlying physical interaction mechanisms. Ex-
periments and simulations indicate that shape-induced
interactions, such as inelastic collisions between elon-
gated objects or of active particles with surfaces lead to
clustering, collective motion, and surface-induced aggre-
gation [6, 22–24]. For micrometer-size biological unicel-
lular swimmers, e.g., bacteria (E. coli), algae (Chlamy-
domonas), spermatozoa, or protozoa (Paramecium), hy-
drodynamic interactions are considered to be important
for collective effects and determine their behavior adja-
cent to surfaces [1, 25–30].
Generic models, which capture the essential swimming
aspects, are crucial in theoretical studies of microswim-
mers. On the one hand, they help to unravel the relevant
interaction mechanisms and, on the other hand, allow
for the study of sufficiently large systems. A prominent
example is the squirmer model introduced by Lighthill
[31] and revised by Blake [32]. Originally, it was in-
tended as a model for ciliated microswimmers, such as
Paramecia. Nowadays, it is considered as a generic model
for a broad class of microswimmers, ranging from diffu-
siophoretic particles [33–35] to biological cells (E. coli,
Chlamydomonas, etc.) and has been applied to study
∗ [email protected]
† [email protected]
‡ [email protected]
§ [email protected]
collective effects in bulk [36–42], at surfaces [36, 43, 44],
and in thin films [20].
In its simplest form, a squirmer is represented as a
spherical rigid colloid with a prescribed surface velocity
[31, 32, 38]. Restricting the surface velocity to be tan-
gential, the spherical squirmer is typically characterized
by two modes accounting for its swimming velocity and
its force-dipole. The latter distinguishes between push-
ers, pullers, and neutral squirmers. The assumption of
a spherical shape is adequate for swimmers like Volvox,
however, the shape of bacteria such as E. coli or the time-
averaged shape of cells such as Chlamydomonas is non-
spherical. Hence, an extension of the squirmer concept
to spheroidal objects is desirable.
In 1977, Keller and
Wu proposed a generalization of the squirmer model to a
prolate-spheroidal shape, which resembles real biological
microswimmers such as Tetrahymenapyriformis, Spiros-
tomum ambiguum, and Paramecium multimicronuclea-
tum [45]. However, that squirmer model accounts for the
swimming mode only and does not include a force-dipole
mode. This is unfortunate, since the force-dipole mode
determines swimmer-swimmer and swimmer-wall inter-
actions [25, 37, 39, 46]. A route to incorporate the force-
dipole mode into the spheroidal squirmer model was pro-
posed in Ref. 44. However, to the best of our knowledge,
the resulting hydrodynamic model is not solvable analyt-
ically. In this article, we propose an alternative model for
a spheroidal squirmer, taking into account both, a swim-
ming and a force-dipole mode. The major advantage of
our approach is that the flow field can be determined
analytically (cf. Fig. 1).
Various mesoscale simulation techniques have been ap-
plied to study the dynamics of squirmers embedded in
a fluid, comprising Stokesian dynamics [39, 40, 43], the
boundary-element method [38, 44, 46–48], the multipar-
ticle collision dynamics (MPC) approach [20, 37, 49], lat-
tice Boltzmann simulations [41, 50], the smoothed pro-
file method [42], and the force-coupling approach [51]. In
the following, we will apply the MPC method. MPC is
a particle-based simulation technique which incorporates
thermal fluctuations [52–54], provides hydrodynamic cor-
relations [55, 56], and is easily coupled with other simula-
(a)
(b)
0
−1
0
−1
−2
FIG. 1. Flow field of a spheroidal puller with β = 3, (a) in
the laboratory frame, and (b) in the body-fixed frame. The
logarithm of the magnitude of the velocity field is color coded.
tion techniques such as molecular dynamics simulations
for embedded particles [53, 54]. The method has suc-
cessfully been applied in various studies of active systems
underlining the importance of hydrodynamic interactions
for microswimmers [1, 20, 24, 28, 37, 53, 57–65].
Here, we implement our spheroidal squirmer model in
MPC. More specifically, we study the resulting flow field
and compare it with the theoretical prediction. More-
over, we present results for the cooperative swimming
behavior of two spheroidal squirmers in thin films. Two
pullers exhibit a long-time stable configuration, where
they swim together in a wedge-like conformation with
a constant small angle due to the hydrodynamic interac-
tion between the anisotropic squirmers as well as squirm-
ers and walls. The cooperative and collective swimming
motion of spheroidal squirmers in Stokes flow has been
addressed in Ref. 47 by an adopted boundary-element
method. This approach neglects thermal fluctuations
and tumbling of the squirmers completely; only hydro-
dynamic and excluded-volume interactions determine the
squirmer motion. In contrast, our simulation approach
includes thermal fluctuations, which affects the stability
of the cooperative swimming motion due to the rotational
diffusion of a spheroid.
II. HYDRODYNAMIC MODEL OF A
SPHEROIDAL SQUIRMER
A. Spheroid geometry
z
θ
bz
bx
n = eτ
s = −eζ
x
z
θ
R
R
2
n = er
s = eθ
x
FIG. 2. Sketch of normal and tangent vectors of a spheroidal
(left) and spherical (right) squirmer. In the squirmer model,
self-propulsion (in z-direction) is achieved by a prescribed tan-
gential surface velocity in direction of the tangent vector s.
with bz and bx the semi-major and semi-minor axis, re-
spectively, and bz ≥ bx (cf. Fig. 2). We denote half of the
focal length by c =pb2
x, which yields the eccentric-
ity e = c/bz. Furthermore, we define a swimmer diameter
as σ = 2bz. In terms of prolate (bz > bx) spheroidal co-
ordinates (ζ, τ, ϕ), the Cartesian coordinates are given
by
z − b2
x = cpτ 2 − 1p1 − ζ 2 cos ϕ,
y = cpτ 2 − 1p1 − ζ 2 sin ϕ,
z = cτ ζ,
(2)
where −1 ≤ ζ ≤ 1, 1 ≤ τ < ∞, and 0 ≤ ϕ ≤ 2π. All
points with τ = τ0 ≡ e−1 lie on the spheroid's surface.
The intersection of the spheroid and a meridian plane,
where ϕ is constant, is an ellipse. The normal n and
tangent s to this ellipse are given by the unit vectors eτ
and −eζ, respectively, which follow by partial derivative
of Eqs. (2) with respect to the coordinates ζ and τ . For
bx = bz, the spheroid becomes a sphere. The spherical
coordinates
(x, y, z)T = r(sin θ cos ϕ, sin θ sin ϕ, cos θ)T
(3)
(2) for τ → ∞, cτ = r, and
are obtained from Eq.
In this limit, the unit vectors turn into
ζ = cos θ.
2). The Lam´e
eτ → er and eζ → −eθ (cf. Fig.
metric coefficients for prolate spheroidal coordinates are
2 (τ 2 − 1)− 1
hζ = c(τ 2 − ζ 2)
and hϕ = c(τ 2 − 1)
2 (1 − ζ 2)− 1
2 (1 − ζ 2)
2 , hτ = c(τ 2 − ζ 2)
1
1
2 .
1
1
2
We describe a nonspherical squirmer as a prolate
spheroidal rigid body with a prescribed surface veloc-
ity usq.
In Cartesian coordinates (x, y, z), the surface
equation of a spheroid, or ellipsoid of revolution, is
The squirmer is immersed in an incompressible low-
Reynolds-number fluid, which is described by the incom-
pressible Stokes equations
B. Flow field
(x2 + y2)/b2
x + z2/b2
z = 1,
(1)
η∆v − ∇p = 0, ∇ · v = 0.
(4)
Here, v(r) is the fluid velocity field, p(r) the pressure
field at the position r, and η the viscosity. In an axisym-
metric flow, the velocity field can be expressed by the
stream function Ψ as [66]
v(ζ, τ, ϕ) = curl(cid:18) 1
hϕ
Ψ(τ, ζ)eϕ(cid:19) .
The stream function itself satisfies the equation [66]
E4Ψ = 0,
with the operator [67]
E2 =
1
c2(τ 2 − ζ 2)(cid:18)(τ 2 − 1)
∂2
∂τ 2 + (1 − ζ 2)
∂2
∂ζ 2(cid:19) .
(5)
(6)
(7)
Each function Ψ in the kernel of E2 can be represented
as [67]
Ψ(τ, ζ) =
∞
4
Xn=0
Xi=1
ci
nΘi
n(τ, ζ),
(8)
with constants ci
n and the functions
Θ1
Θ3
n(τ, ζ) = Gn(τ )Gn(ζ), Θ2
n(τ, ζ) = Hn(τ )Gn(ζ), Θ4
n(τ, ζ) = Gn(τ )Hn(ζ),
n(τ, ζ) = Hn(τ )Hn(ζ).
Here, Gn(x) and Hn(x) are Gegenbauer functions of the
first and second kind, respectively (see Appendix B). The
velocity components follow from the stream function via
[66]
vτ =
1
hζhϕ
1
∂Ψ
∂ζ
= c−2(τ 2 − 1)− 1
∂Ψ
= −c−2(1 − ζ 2)− 1
∂τ
2 (τ 2 − ζ 2)− 1
∂Ψ
∂ζ
2 (τ 2 − ζ 2)− 1
2
2
vζ = −
hτ hϕ
,
(9)
∂Ψ
∂τ
.
(10)
An important feature of a squirmer is the hydrody-
namic boundary condition at its surface, which demands
v(r) = usq. For the squirming velocity usq we propose
usq = −B1(s · ez)s − B2ζ(s · ez)s
= −B1 (1 + βζ) (s · ez)s
= −B1τ0(1 − ζ 2)
2 (τ 2
1
0 − ζ 2)− 1
2 (1 + βζ) eζ.
(11)
(12)
(13)
Here, s is the tangent vector, ez = (0, 0, 1)T is the unit
vector in z-direction, B1 and B2 are the two surface
velocity modes, and β = B2/B1 (cf. Fig.
2). B1
determines the swimming velocity, while the B2 term
introduces a force-dipole, or pusher (B2 < 0) and puller
(B2 > 0) mode. Note that the spherical squirmer
introduced by Lighthill and Blake with modes B1 and
B2 [31, 32]
limit of a
spheroid, where ζ → cos(θ) = n · ez.
is recovered for the spherical
3
For B2 = 0, this model of a spheroidal squirmer was
already introduced and analysed in Refs. 45 and 68.
An additional force-dipole mode has been introduced in
Refs. 44 and 47 as usq(ζ) = −B1s · ez (1 + βn · ez) s.
However, we prefer the squirming velocity introduced in
Eq. (12), since it yields an analytically solvable bound-
ary value problem for the Stokes equation. The two ap-
proaches provide a somewhat different flow field in the
vicinity of the squirmer, but both yield the model of
Lighthill and Blake in the limit of zero eccentricity.
In the swimmer's rest frame, and with Eq. (12), the
boundary value problem becomes
(14)
(15)
(16)
U0c2(τ 2 − 1)(1 − ζ 2) for τ → ∞,
= (B1 + B2ζ)c2τ0(1 − ζ 2) for all ζ.
1
2
Ψ(τ, ζ) →
Ψ(τ0, ζ) = 0 for all ζ,
∂Ψ
∂τ (cid:12)(cid:12)(cid:12)(cid:12)τ =τ0
Equation (14) implies a constant background flow v =
−U0ez infinitely far from the squirmer, Eq. (15) guaran-
tees vτ = 0 at the spheroid surface, and Eq. (16) demands
vζ = usq(ζ) · eζ. Due to linearity of the Stokes stream
function equation (6), we can solve this boundary value
problem for B2 = 0 first, which yields the stream func-
tion Ψ1. Subsequently we solve the problem
Ψ(τ, ζ) converges for τ → ∞,
Ψ(τ0, ζ) = 0 for all ζ,
∂Ψ
1
∂τ (cid:12)(cid:12)(cid:12)(cid:12)τ =τ0
0 − ζ 2)
2 (1 − ζ 2)
= −c2(τ 2
= B2c2τ0(1 − ζ 2)ζ for all ζ.
1
2 u2(ζ),
(17)
(18)
(19)
Equation (17) imposes a vanishing velocity field infinitely
far from the squirmer, Eq. (18) again guarantees vτ = 0
at the spheroid surface, and Eq.
(19) demands vζ =
usq(ζ, B1 = 0)·eζ . We denote the solution of the problem
Eqs. (17)-(19) by Ψ2. Finally, Ψ = Ψ1 + Ψ2 solves the
initial problem (14)-(16) for arbitrary B1 and B2.
The boundary value problem Eqs. (14)-(16) for B2 = 0
can be solved by the ansatz
Ψ1(τ, ζ) = α1G2(τ )G2(ζ) + α2H2(τ )G2(ζ) + α3τ (1 − ζ 2).
(20)
Here, the third term is found by the separation ansatz
Ψ(τ, ζ) = g(τ )(1−ζ 2) for Eq. (6). Equation (14) directly
yields α1 = −2U0c2. The remaining coefficients α2 and
α3 are determined by Eqs. (15) and (16), keeping in mind
that B2 = 0. This yields
α2 = 2c2 U0(τ 2
0 + 1) − 2B1τ 2
0 + 1) coth−1 τ0 − τ0
(τ 2
0
,
(21)
α3 = c2 B1τ0(τ0 − (τ 2
0 − 1) coth−1 τ0) − U0
0 + 1) coth−1 τ0 − τ0
(τ 2
.
(22)
The boundary value problem Eqs. (17)-(19) can be solved
by the ansatz
Ψ2(τ, ζ) = α4G3(τ )G3(ζ) + α5H3(τ )G3(ζ) + α6ζ(1 − ζ 2).
(23)
As before, the third term follows by a separation ansatz
Ψ(τ, ζ) = g(τ )ζ(1 − ζ 2) for Eq. (6). Equation (17) yields
α4 = 0. The coefficients α5 and α6 are determined by
Eqs. (18)-(19) such that
z
15
10
5
0
−5
−10
−15
(a)
(b)
−0.5
−1.0
−1.5
−2.0
−2.5
4
−0.5
−1.0
−1.5
−2.0
−2.5
−3.0
α5 = c2
4B2τ0
3τ0 + (1 − 3τ 2
2/3 − τ 2
α6 = c2B2τ0
0 ) coth−1 τ0
,
(24)
0 + τ0(τ 2
3τ0 + (1 − 3τ 2
0 − 1) coth−1 τ0
0 ) coth−1 τ0
.
(25)
0
5
10
15
0
5
10
15
¯r
¯r
The total stream function Ψ = Ψ1 + Ψ2 can be trans-
formed to the laboratory frame (cf. Fig. 1) by adding
the background flow v = U0ez, which yields
1
2
Ψlab = Ψ −
U0c2(τ 2 − 1)(1 − ζ 2)
= α2H2(τ )G2(ζ) + α3τ (1 − ζ 2)
+ α5H3(τ )G3(ζ) + α6ζ(1 − ζ 2).
(26)
(27)
The force by the fluid on the spheroid is given by [66]
Fz = lim
r→∞
rΨlab
¯r2 = 8πηα3/c,
(28)
where r = px2 + y2 + z2 and ¯r = px2 + y2. As ex-
pected, Ψ2 does not contribute to the force, since it as-
sumes a constant value at infinity. Since a swimmer must
be force free, Fz = 0, which implies α3 = 0. Then,
Eq. (22) yields the swimming velocity of the squirmer
(τ0 = 1/e)
U0 = B1τ0(τ0 − (τ 2
0 − 1) coth−1 τ0),
(29)
which was already found by Keller and Wu for the case
B2 = 0 [45]. As a consequence, α2 in Eq. (21) simply
becomes α2 = 2B1c2τ0(τ 2
0 − 1).
The flow field of a point-like force-dipole is given by
[1, 48]
vF D =
P
8πη
r
r3 (cid:18) 3z2
r2 − 1(cid:19) ,
(30)
with the dipole strength P , whereas the flow field of a
source doublet is [48]
vSD = κ
1
r3 (cid:18)−ez +
3zr
r2 (cid:19) ,
(31)
with the source-doublet strength κ. Comparing the cor-
responding stream functions with Eq. (26) far from the
origin, we find
P = −8πηα6,
κ = −
cα2
6
= −
B1
3
c3τ0(τ 2
0 − 1)
(32)
(33)
FIG. 3. Fluid velocity fields of a spheroidal squirmer in the
laboratory frame for (a) B1 = 1, B2 = 0, and (b) B1 =
0, B2 = 1. The corresponding stream function is given by
Eq.
(26). The logarithm of the magnitude of the velocity
field is color coded. Note that the pusher velocity field with
B1 = 0, B2 = −1 is not shown, since it follows from that of
the puller with B1 = 0, B2 = 1 by inverting the arrows.
for our model. As expected, in the spherical limit (bz →
bx ≡ R, where R is the radius) we obtain P = −4πηB2R2
and κ = −B1R3/3.
Examples of fluid velocity fields of a spheroidal
squirmer are presented in Figs. 1 and 3.
III. MULTIPARTICLE COLLISION DYNAMICS
Multiparticle collision dynamics (MPC) is a stochas-
tic, particle based mesoscale hydrodynamic simulation
method [54]. Thereby, a fluid is modeled by N point par-
ticles with equal mass m, undergoing subsequent stream-
ing and collision steps. In the streaming step, the particle
positions ri, i = 1, . . . , N , are updated according to
ri(t + h) = ri(t) + hvi(t),
(34)
where vi are the particle velocities and h is denoted as
collision time step. In the subsequent collision step, the
particle velocities are changed by a stochastic process,
which mimics internal fluid interactions. In order to de-
fine the local collision environment, particles are sorted
into cells of a cubic lattice with lattice constant a. Dif-
ferent realizations for this stochastic process have been
proposed.[52, 69, 70] We employ the stochastic rotation
dynamics (SRD) approach of MPC with angular momen-
tum conservation (SRD+a) [71, 72], which updates the
particle velocities in a cell according to
vnew
i = vcm + R(α)vi,c − ri,c×
×hmI−1 Xj∈cell
{rj,c × (vj,c − R(α)vj,c)}i.
(35)
Here, ri,c = ri − rcm, where rcm is the center-of-mass
position of the particles in the cell, and similarly, vi,c =
vi − vcm, with the center-of-mass velocity vcm. R(α) is
the rotation matrix, which describes a rotation around
a randomly oriented axis by the angle α. The angle α
is a constant, and the axis of rotation is chosen inde-
pendently for each cell and time step. Finally, I is the
moment-of-inertia tensor of the particles in the center-of-
mass reference frame of the cell. Partition of the system
into collision cells leads to a violation of Galilean invari-
ance. To reestablish Galilean invariance, a random shift
of the collision-cell lattice is introduced at every collision
step [73, 74].
Since energy is not conserved in the collision step, we
apply a cell level canonical thermostat at temperature
T [75, 76]. The latter ensures Maxwell-Boltzmann dis-
tributed velocities. The MPC algorithm is embarrass-
ingly parallel. Hence, we implement it on a Graphics
Processing Unit (GPU) for a high performance gain [77].
The following simulations are performed with the
mean number of particles per collision cell hNci = 10,
the rotation angle α = 130◦, and the time step h =
0.02pma2/(kBT ), which yields a fluid viscosity of η =
17.8pmkBT /a4.
IV.
IMPLEMENTATION OF A SPHEROIDAL
SQUIRMER IN MPC
A spheroidal squirmer is a homogeneous rigid body
characterized by its mass M , center-of-mass position C,
orientation q, translational velocity U , and angular mo-
mentum l. Thereby, q = (q0, q1, q2, q3) is a rotation
quaternion and can be related to the rotation matrix D,
which transforms vectors from the laboratory frame to
the body-fixed frame [78]. We distinguish vectors in the
laboratory frame and body-fixed frame by a superscript,
i.e., vs is a vector in the laboratory (or space-fixed) frame
while
vb = Dvs
(36)
is the corresponding vector in the body-fixed frame. For
vectors in the laboratory frame, we will frequently omit
the superscript. The orientation vector of a spheroid is
e = DT eb = DT (0, 0, 1)T . The moment of inertia tensor
in the body-fixed frame Ib is a constant diagonal matrix
with diagonal elements Ix = (M/5)(b2
z) = Iy and
Iz = (2M/5)b2
x. When needed, the angular velocity is
x + b2
calculated as Ωs = DT (cid:0)Ib(cid:1)−1
For all simulations we choose a neutrally bouyant
spheroid, i.e., M = ρ(4π/3)bzb2
x, where ρ is the fluid
mass density.
Dls.
A. Streaming step
During the streaming step, a spheroid will collide with
several MPC particles. Since the total change in (angu-
5
lar) momentum of a spheroid during one streaming step
is small, we perform the collisions with MPC particles in
a coarse-grained way [79]:
For the streaming step at time t, we determine the
spheroid's position, velocity, orientation, and angular ve-
locity at times t + h/2 and t + h, under the assumption
that there is no interaction with MPC particles. How-
ever, steric interactions between spheroids, as well as
spheroids and walls are taken into account as described
in Sec. IV C.
Subsequently, all MPC particles are streamed,
i.e.,
their positions are updated according to ri(t + h) =
ri(t) + hvi(t). Thereby, a certain fraction of MPC par-
ticles penetrates a spheroid. To detect those particles
in an efficient way, possible collision cells intersected by
the spheroid are identified first. For this purpose, we se-
lect all those cells, which are within a sphere of radius
bz enclosing the spheroid instead of the spheroid itself,
which is more efficient, since it avoids rotating candi-
date cells into the body-fixed frame during selection. A
loop over all particles in respective collision cells iden-
tifies those particles, which are inside the spheroid and
they are labeled with the spheroid index. Then, each
particle i inside a spheroid at time t + h is moved back in
time by half a time step and subsequently translated onto
the spheroid's surface. The translation can be realized in
different ways. One possibility is to constructing a vir-
tual spheroid with semi-axes bz, bx, bz/bx = bz/bx and
ri(t + h/2) on its surface. The particle is then translated
along the normal vector of the virtual spheroid until it is
on the real spheroid's surface. Alternatively, the differ-
ence vector ri(t + h/2) − C(t + h/2) can be scaled such
that the particle position lies on the spheroid'surface. We
tried both approaches and found no significant difference.
Once the MPC particle at time t + h/2 is located on the
spheroid's surface, the momentum transfer
Ji = 2m(cid:8)vi − U − Ω × (ri − C) − DT ub
sq[D(ri − C)](cid:9)(37)
at time t + h/2 is determined, taking into account the
squirmer surface fluid velocity usq of Eq. (11) [80].
Thereby, a useful identity to determine s is given in Eq.
(8) of Ref. 45, and ζ is given by
ζ =
1
2c(cid:16)px2 + y2 + (z + c)2 −px2 + y2 + (z − c)2(cid:17) .
(38)
The velocity of the MPC particle is updated according
to v′
i = vi − Ji/m. Subsequently, the position ri(t +
h) is obtained by streaming the MPC particle for the
remaining time h/2 with velocity v′
i, i.e., ri(t + h) =
ri(t + h/2) + hv′
i/2.
As a consequence of the elastic collisions, the center-
of-mass velocity and rotation frequency of a spheroid are
finally given by
U (t + h)′ = U (t + h) + J/M,
Ω(t + h)′ = Ω(t + h) + DT (cid:0)Ib(cid:1)−1
DL,
(39)
(40)
where J = Pi Ji is total momentum transfer by the
MPC fluid and L = Pi (ri(t + h/2) − C(t + h/2)) × Ji
is the respective angular momentum transfer.
100
B. Collision step
In a first step, ghost particles are distributed inside
each spheroid [79, 81]. The number density and mass are
equal for ghost and fluid particles. The ghost particle
positions rg
i are uniformly distributed in the spheroid
and their velocities are given by
vg
i = U + Ω × (ri − C) + usq,i + vR,i.
(41)
i
)
0
(
e
·
)
t
(
e
h
10−1
0
5
15
20
10
30
t/(103pma2/(kBT ))
25
6
35
40
ance pkBT /m. The squirming velocity usq,i
i − vg
The Cartesian components of vR,i are Gaussian-
distributed random numbers with zero mean and vari-
is de-
termined by Eq. (11), with the ghost particle position
projecting onto the spheroid's surface (cf. Sec. IV A).
As a result of MPC collisions, a spheroid's linear and
angular momenta change by J g
i ) and
Lg
i are the ghost parti-
cle's velocity after and before the MPC collision. Hence,
the spheroid velocity and angular velocity become
i − C)×J g
i = m(¯vg
i , where ¯vg
i and vg
i = (rg
U ′ = U + J g/M,
Ω′ = Ω + RT (cid:0)Ib(cid:1)−1
RLg.
(42)
(43)
C. Rigid body dynamics for spheroids
During the streaming step, the spheroids move accord-
ing to rigid-body dynamics, governed by [82]
M C = F ,
1
q =
q =
dΩb
α
dt
= I −1
Ωb(cid:19)(cid:21) ,
1
2
2(cid:20)Q( q)(cid:18) 0
Ωb(cid:19) + Q(q)(cid:18) 0
Q(q)(cid:18) 0
Ωb(cid:19) ,
α + (Iβ − Iγ)Ωb
α (cid:2)T b
γ(cid:3) .
βΩb
(44)
(45)
(46)
(47)
Here, Q(q) is defined in Eq.
(A2), and F and T are
the force and torque acting on the spheroid. Forces
and torques are derived from steric interaction poten-
tials as presented in Appendix C. Equations (47) are
Euler's equations for rigid body dynamics and hold for
(α, β, γ) = (x, y, z), (y, z, x), and (z, x, y). Whenever nec-
essary, body-fixed and laboratory-frame quantities can
be related by the rotation matrix D which is given in
terms of the quaternion q in Eq. (A1).
For the numerical integration of the equations of mo-
tion, the widely applied leap-frog method [78] is not use-
ful, since velocity, angular momentum, position, and ori-
entation are required at the same point in time for the
coupling to the MPC method. Hence, we employ the Ver-
let algorithm for rigid-body rotational motion proposed
FIG. 4. Orientation correlation functions he(t) · e(0)i for pas-
sive spheroids with bz = 6a, bx = 3a (bottom blue line) and
bz = 9a, bx = 3a (top black line). The plot shows the simu-
lation data (blue and black solid lines), an exponential fit to
that data (red dashed), and the theoretical prediction accord-
ing to Eq. (55) (green dotted).
in Ref. 82. Integration for a time step τ is performed as
follows:
(i) Update C and q according to (cf. Eqs. (46) and
(47))
C(t + τ ) = C(t) + U (t)τ +
τ 2
2M
q(t + τ ) = (1 − λ)q(t) + qτ +
τ 2
2
q,
F s(t),
(48)
(49)
λ = 1 − q2τ 2/2
−p1 − q2τ 2 − q · qτ 3 − ( q2 − q4)τ 4/4.
(50)
The parameter λ is introduced to guarantee q2 = 1.
(ii) Calculate forces and torques F s(t+τ ) and T s(t+τ ).
(iii) Update U and ls according to
U (t + τ ) = U (t) +
ls(t + τ ) = ls(t) +
τ
[F s(t) + F s(t + τ )],
2M
τ
[T s(t) + T s(t + τ )].
2
(51)
(52)
V. SIMULATIONS – THERMAL PROPERTIES
AND FLOW FIELD
A. Passive colloid
For the passive spheroidal colloid (B1 = B2 = 0), we
αi as
α)2i for α ∈ {x, y, z}. Due to the equipartition
perform equilibrium simulations and determine hU 2
well as h(Ωb
of energy, we expect
hU 2
αi =
α)2i =
h(Ωb
kBT
M
kBT
Iα
,
.
(53)
(54)
We fix the aspect ratio bz/bx = 2 and vary bx in the
range bx ∈ [2a, 4a]. The simulation results agree very
well with the theoretical values (53) and (54). As ex-
pected, the deviations from theory decrease with in-
creasing spheroid size, due to a better resolution in
In general, the relative error
terms of collision cells.
theoi is larger for Ωb
σr = (hx2
α than
for Uα. We find the largest relative error for h(Ωb
z)2i,
namely σr = 9.5%, 5.3%, and 3.1% for bx = 2a, 3a, and
4a. Hence, we choose the minor axis bx ≥ 3a in the
following.
In addition, we determine the orientation correlation
function he(t) · e(0)i. The theory of rotational Brownian
motion [83] predicts
theoi − hx2
simi)/hx2
(55)
he(t) · e(0)i = exp(cid:0)−2D⊥
Rt(cid:1) ,
R + Dk
where DR = (2D⊥
R =
kBT /ξ⊥, and ξk and ξ⊥ are the parallel and perpendic-
ular rotational friction coefficients of a prolate spheroid
with respect to the major semi-axis; explicitly [84]
R = kBT /ξk, D⊥
R)/3, Dk
Simulation results for the orientational auto-correlation
function are shown in Fig. 4 for two spheroids of dif-
ferent eccentricity. The correlation functions decay ex-
ponentially. However, for the spheroid with the smaller
eccentricity, we find a somewhat faster decay than pre-
dicted by theory, whereas good agreement is found for the
larger spheroid. We attribute the difference to finite-size
effects related to the discreteness of the collision lattice.
For larger objects, discretization effects become smaller.
B. Squirmer
We determine the steady state swimming velocity of
a squirmer via he · Ui, which should be equal to U0
(cf. Eq.
(29)). Results for various eccentricities are
displayed in Fig. 5. The velocity U0 increases with in-
creasing eccentricity e in close agreement with the theo-
retical prediction of Eq. (29). We confirm that the force-
dipole parameter β does not affect the velocity of the
squirmer, as long as the Reynolds number Re is low,
i.e., Re = ρU0bz/η . 0.1. We also determine the ori-
entational correlation function and find that a squirmer
(a)
−2
(b)
4
3
4
3
ξ⊥ = 8πηb3
z
ξk = 8πηb3
z
e3(1 − e2)(2e − (1 − e2)L)−1,
e3(2 − e2)(−2e + (1 + e2)L)−1,
L = log(cid:18) 1 + e
1 − e(cid:19)
(56)
(57)
(58)
−3
−4
7
1.0
0.9
0.8
0.7
1
B
/
i
U
·
e
h
0.6
0.0
0.2
0.4
0.6
0.8
1.0
e
FIG. 5. Mean swimming velocity as function of the eccen-
tricity e for a spheroidal squirmer with B1 = 0.05pkBT /m
and B2 = 0. The solid line shows the theoretical prediction of
Eq. (29). Black dots are simulation results. The eccentricity
was varied by changing bz and keeping bx = 3a constant. For
the red triangle, we simulated a larger spheroid with bx = 6a,
which shows a better agreement with theory.
(c)
(f)
−2
−3
−4
−5
−2
−3
−4
0
−1
0
−1
−2
−3
(d)
−2
(e)
−3
−4
FIG. 6. Fluid flow fields of a spheroidal squirmer in the lab-
oratory frame with bx = 3a, bz = 6a, B1 = 0.01pkBT /m,
and β = 3 ((a),(b),(c)), and with B1 = 0.05pkBT /m, β = 0
((d),(e),(f)). The logarithm of the magnitude of the velocity
field (in units of pkBT /m) is color coded. The plots (a),
(d) show theoretical results, (b), (e) simulation results, and
(c), (f) relative errors. The relative error of the flow field is
defined as ∆vα = vtheo
α )/2]. Note,
due to the discrete representation of the velocity field, some
streamlines end abruptly.
α /[(vtheo
α − vsim
+ vsim
α
exhibits the same orientational decorrelation as the cor-
responding passive particle (cf. Fig. 4).
Moreover, we calculate the flow field from the simula-
tion data and compare it with the theoretical prediction.
As shown in Fig. 6, the two fields are in close agree-
ment. The two-dimensional flow field of the MPC fluid,
averaged over the rotation angle ϕ, is determined at the
vertices of a fine resolution mesh. The velocities at these
vertices include averages over time of an individual real-
ization as well as ensemble averages over various realiza-
tions. By the latter, we determine an estimate for the
error of the mean velocity. The median (over vertices)
of this error is approximately 5% for the parameters of
Fig. 6 (b) and 10% for that of Fig. 6 (e). Note that we
choose a smaller swimming mode B1 for the puller (Fig.
6 (b)) than for the neutral squirmer (Fig. 6 (e)). The
reason is that the agreement with theory was not satis-
factory for the puller with B1 = 0.05pkBT /m, which we
attribute to nonlinear convective effects. In Figs. 6 (c)
and (f), we observe lines of high relative errors (yellow
in the color code). They appear because theory predicts
v¯r = 0 or vz = 0 for these lines, which is difficult to
achieve in simulations. Hence, the overall agreement be-
tween simulations and theory is very satisfactory, and
the implementation is very valuable for the simulation of
squirmer-squirmer and squirmer-wall interactions, where
the details of the flow field matter.
VI. COOPERATIVE SWIMMING IN THIN
FILMS
We simulate the cooperative swimming behavior of two
squirmers in a slit geometry. The slit is formed by two
parallel no-slip walls located at y = 0 and y = Ly. The
no-slip boundary condition is implemented by applying
the bounce-back rule and ghost particles of zero mean ve-
locity in the walls [81]. Steric interactions between two
squirmers and between a squirmer and a wall are taken
into account by the procedure described in Appendix C.
The initial positions and orientations of the two squirm-
ers (i = 1, 2) are
2 (cid:19)T
C1/2 =(cid:18) Lx
2 ∓
e1/2 = (± cos(α0), 0, sin(α0))T .
dcm
2
Lz
,
,
Ly
2
,
(59)
(60)
Here, dcm is the initial center-of-mass distance and α0 =
(π−θ0)/2, where θ0 is the inital angle between e1 and e2.
The swimming mode is chosen as B1 = 0.05pkBT /m
and the force dipole mode β ∈ {−4, 0, 4}. We choose
dcm such that the squirmers are well separated and vary
θ0. The squirmers major and minor axes are bx = 3a
and bz = 6a, respectively, and the simulation box size
is Lx = Lz = 15bz, and Ly = 7a. Note that Ly & bx
which keeps the swimming orientation essentially in the
x-z plane.
8
a
/
i
s
d
h
4
3
2
1
0
1.00
0.75
0.50
0.25
i
)
θ
(
s
o
c
h
0
1
2
3
4
5
tU0/σ
FIG. 7. Average surface-to-surface distance ds and orienta-
tion of squirmers, where cos(θ) = e1 · e2, as function of time.
The solid blue, dashed black, and dotted red lines correspond
to pullers β = 4, neutrals β = 0 and pushers β = −4. The
standard deviation of the blue line (β = 4) is indicated by the
cyan shaded region.
Results for the mean surface-to-surface distance be-
tween squirmers hdsi and the mean alignment he1 · e2i =
hcos θi are shown in Fig. 7 for pushers, pullers, and neu-
tral swimmers with an inital angle θ0 = 3π/8. Due to the
setup, the squirmers initially approach each other and
collide at tU0/σ ≈ 0.5. The (persistence) P´eclet number
P e = v0/(2D⊥
Rσ) ≈ 60 is sufficiently high, such that the
squirmer orientation has hardly changed before collision.
When the neutral swimmers collide, they initially align
parallel (cos θ ≈ 1 at tU0/σ ≈ 1 in Fig. 7), but their tra-
jectories start to diverge immediately thereafter. Pushers
remain parallel for an extended time window, which is ex-
pected as pushers are known to attract each other [37],
but at tU0/σ ≈ 3 (cf. Fig. 7) their trajectories diverge as
well. This is probably due to noise, since we observe sev-
eral realizations where pushers remain parallel. Interest-
ingly, pullers, which are known to repel each other when
swimming in parallel[37], swim cooperatively and reach
a stable orientation with hcos(θ)i ≈ 0.77 shortly after
they collided (at tU0/σ ≈ 1). Thereby, their cooperative
swimming velocity is about 0.8U0. The flow field of this
stable state, determined by MPC simulations, is shown
in Fig. 8. Note that the velocity field in the swimming
(a)
(b)
−4
−3
−2
−1
FIG. 8. Flow field of two cooperatively swimming pullers
in the laboratory frame. The logarithm of the magnitude of
the velocity field (in units of pkBT /m) is color coded. We
denote the direction normal to the walls by y, the cooperative
swimming direction by z, the remaining Cartesian axis by x,
and choose the swimmers' center of mass as origin. (a) shows
the flow field at x = 0 in the zy-plane, while (b) shows the
flow field at y = 0 in the xz-plane. The black elliptical shapes
indicate the projection of the swimmers onto the considered
plane. Note that periodic boundary conditions are employed
in the MPC simulation and affect the flow field, which leads
to closed flow lines on the length scale of the periodic system.
plane is left-right symmetric, and that there is a stagna-
tion point in the center behind the swimmers. Figure 8
reveals that this point actually corresponds to a line nor-
mal to the walls. Figure 9 shows that the fixed point of
cooperatively swimming pullers is reached for nearly all
simulated initial conditions θ0 ∈ (0, π/2). Only pullers
that are nearly parallel initially (θ0 = π/8, cos θ0 ≈ 0.92
in Fig. 9), repel each other such that they will not reach
the fixed point. For P´eclet numbers P e < 60, the fixed
point remains at hcos(θ)i ≈ 0.77. However, it becomes
more likely for the swimmers to escape (or never reach)
the fixed point.
A detailed study reveals that the fixed point vanishes,
when the walls are replaced by periodic boundary condi-
tions. This is even true when we apply three-dimensional
periodic boundaries, but keep the wall potential imple-
mented, i.e., the squirmers are still confined in a narrow
9
i
)
θ
(
s
o
c
h
1.0
0.8
0.6
0.4
0.2
0.0
0
1
2
4
3
tU0/σ
5
6
FIG. 9. Time dependence of the average alignment he1 · e2i =
hcos θi of two pullers with β = 4, bz/bx = 2, and various initial
angles θ0 ∈ (0, π/2).
slit. In addition, we studied the swim behavior of spheri-
cal squirmers. Here, we observe diverging trajectories for
all squirmer types, i.e., pushers, neutral squirmers, and
pullers. Such diverging trajectories have already been re-
ported in Ref. 85 for spherical squirmers in bulk. Hence,
the stable close-by cooperative swimming of pullers is
governed by the squirmer anisotropy, by the hydrody-
namic interactions between them and, importantly, be-
tween pullers and confining surfaces.
This conclusion is in contrast to results presented in
Ref. 47, where a monolayer of spheroidal squirmers is con-
sidered, with their centers and orientation vectors fixed
in the same plane, however, without confining walls. The
study reports a stable cooperative motion for pullers with
angles θ ∈ (0, π/2) by nearest-neighbor two-body inter-
actions, where all angles between 0 and π/2 are stable.
The difference to our study is that in Ref. 47 coopera-
tive features were extracted from a simulation of many
swimmers, whereas we explicitly studied two swimmers.
Furthermore our study explicitly models no-slip walls and
includes thermal noise.
To shed light on the stability of the cooperative puller
motion, we varied the puller strength β, the aspect ratio
bz/bx, and the width of the slit Ly. Thereby, we started
from our basic parameter set bx = 3a, bz = 6a, Ly = 7a,
and β = 4.
With decreasing β, the stable alignment disappears,
i.e., the pullers' distance increases after collision. For
increasing β the fixed point remains, but the value of
cos θ decreases, i.e., the squirmers form a larger angle.
With increasing wall separation, the fixed-point value
of cos θ decreases, i.e, the angle between the swimmers
increases. For Ly/bz = 2 and higher, the fixed point
disappears.
An increase of the aspect ratio bz/bx from 2 to 3 and
4 increases the fixed-point value of hcos θi from 0.77 to
0.84 and 0.88. The more elongated shape leads to a more
parallel alignment of the squirmers. The minimal value of
β required to achieve cooperative motion depends weakly
on the aspect ratio. For bz/bx = 2, 3, and 4 the critical
values for β are ≈ 3.7, 3.6, and 3.4. Hence, a large aspect
ratio is beneficial for cooperative swimming.
VII. SUMMARY AND CONCLUSIONS
We have introduced a spheroidal squirmer model,
which comprises the swimming and force-dipole modes.
It is a variation of previously proposed squirmer mod-
els. On the one hand, it includes the force-dipole mode
as an extension to the model of Ref. 45. On the other
hand, it is an alternative approach compared to Refs. 44
and 47, with the major advantage that our model al-
lows for the analytical calculation of the flow field.
In
the present calculations we employed the Stokes stream
function equation. Very recently a full set of solutions to
Stokes' equations in spheroidal coordinates were given in
Ref. 86, which opens an alternative approach to derive
the flow field for our choice of boundary conditions.
Furthermore, we have presented an implementation of
our spheroidal squirmer in a MPC fluid. In contrast to
other frequently employed simulation approaches, MPC
includes thermal fluctuations. The comparison between
the fluid flow profile of a squirmer extracted from the sim-
ulation data with the theoretical prediction yields very
good agreement. As a consequence of the MPC approach
with its discrete collision cells, the minor axis of the
spheroid has to be larger than a few collision cells to avoid
discretization effects. The analysis of the squirmer orien-
tation correlation function shows that very good agree-
ment between theory and simulations is already obtained
for bz = 9a (major axis) and bx = 3a (minor axis).
To shed light on the cooperative swimming motion
and on near-field hydrodynamic interactions, we inves-
tigated the collision of two spheroidal squirmers in a slit
geometry. We found stable stationary states of close-
by swimming for spheroidal pullers, which is determined
by hydrodynamic interactions between the anisotropic
squirmers, and, even more important, by squirmers and
surfaces. This stationary state disappears for low puller
strengths and low eccentricities. We expect the stable
close-by swimming of pullers to strongly enhance clus-
tering in puller suspensions in thin films.
Our studies confirm that spheroidal squirmers can ac-
curately be simulated by the MPC method. The pro-
posed implementation opens an avenue to study collec-
tive and non-equilibrium effects in systems of anisotropic
microswimmers. Even large-scale systems can be ad-
dressed by the implementation of MPC and the squirmer
dynamics on GPUs.
10
ACKNOWLEDGMENTS
We thank A. Wysocki, A. Varghese, and B.U. Felder-
hof for helpful discussions. Support of this work by the
DFG priority program SPP 1726 on "Microswimmers –
from Single Particle Motion to Collective Behaviour" is
gratefully acknowledged.
Appendix A: Quaternion matrices
The rotation matrix D introduced in Eq. (36) is given
in terms of the rotation quaternion q as
1 − q2
q2
0 + q2
2 − q2
2(q2q1 − q0q3)
2(q2q1 + q0q2)
3
2(q1q2 + q0q3)
0 − q2
2 − q2
q2
2(q3q2 − q0q1)
1 + q2
3
2(q1q3 − q0q2)
2(q2q3 + q0q1)
2 + q2
q2
0 − q2
1 − q2
(A1)
3
The matrix Q(q) in Eq. (46) is given by
Q(q) =
q0 −q1 −q2 −q3
q1
q2
q0 −q3
q2
q0 −q1
q3
q0
q1
q3 −q2
.
(A2)
Appendix B: Gegenbauer functions
For n ≥ 2 and x ∈ R the Gegenbauer functions of the
first and second kind Gn and Hn, are defined in terms
of the Legendre functions of the first and second kind Pn
and Qn as [67, 87]
Gn(x) =
Pn−2(x) − Pn(x)
2n − 1
, Hn(x) =
For n = 0, 1, they are defined as
Qn−2(x) − Qn(x)
(B1)
2n − 1
.
G0(x) = −H1(x) = 1, G1(x) = H0(x) = −x.
(B2)
For the reader's convenience, we give the formula for the
Gegenbauer functions of the first kind for n = 2, 3, and
x ∈ R
G2(x) =
G3(x) =
1
2
1
2
(1 − x2),
(1 − x2)x.
(B3)
(B4)
Furthermore, the Gegenbauer functions of the second
kind for n = 2, 3, and x > 1 are given by
H2(x) =
1
2
1
2
x
(1 − x2) coth−1(x) +
2
(1 − x2)x coth−1(x) +
H3(x) =
(3x2 − 2).
Here, we used coth−1(x) = ln([x + 1]/[x − 1])/2.
,
1
6
(B5)
(B6)
Appendix C: Steric interactions
calculated analytically and are given by [88] [? ]
Here, we illustrate our implementation of the excluded-
volume interactions between spheroids and walls follow-
ing the approach provided in Ref. 88.
The spheroid's surface in the laboratory frame is given
by the quadratic form
F1 =
24ǫ0
σ0 "2(cid:18) σ0
×(cid:18) R
dR + σ0(cid:19)13
(F −1/2 − 1) −
R
dR + σ0(cid:19)7#
−(cid:18) σ0
F −3/2Xc(cid:19) ,
R
2
11
(C9)
1 = A(x) ≡ (x − C)T A(x − C),
(C1)
and
where the orientation matrix A can be expressed as
A = (1 − eeT )/b2
x + eeT /b2
z.
(C2)
For the steric interactions, we introduce a virtual safety
distance dv, which is small compared to bx and bz. When
computing steric interactions, we replace bx and bz by
bx + dv and bz + dv, respectively. In this paper we used
dv = 0.05a for all simulations.
1.
Interaction between spheroids
We introduce a repulsive interaction potential between
spheroids to prevent their overlap. The potential is given
by
U = 4ǫ0"(cid:18) σ0
dR + σ0(cid:19)12
−(cid:18) σ0
dR + σ0(cid:19)6# .
(C3)
Here, σ0 and ǫ0 correspond to a length and energy scale,
respectively. We choose ǫ0 = kBT and σ0 = 2dv. The
directional contact distance dR between two spheroids,
with orientation matrices A1, A2 and center positions
C1, C2, is an approximation to their true distance of
closest approach and is defined by
dR = R(1 − F (A1, A2)−1/2)
(C4)
Here, R = C2 − C1, R = R, and F (A1, A2) is the
elliptic contact function, defined as [88]
F (A1, A2) = max
λ
= max
λ
min
x S(x, λ)
min
(λA1(x) + (1 − λ)A2(x)) . (C6)
x
(C5)
Minimization with respect to x demands ∇S(x, λ) = 0,
and hence,
x(λ) = {λA1 + (1 − λ)A2}−1 {λA1C1 + (1 − λ)A2C2} .
(C7)
The critical value λ = λc that maximizes S(x(λ), λ) can
be found by the root finding problem
A1(x(λ)) − A2(x(λ)) = 0.
(C8)
We implement Brent's root finding approach [89]. The
forces and torques arising from the potential (C3) can be
T1 = −
12Rǫ0
σ0 "2(cid:18) σ0
dR + σ0(cid:19)13
× F −3/2(xc − C) × Xc
−(cid:18) σ0
dR + σ0(cid:19)7# (C10)
(C11)
for the first spheroid, where Xc = 2λcA1(xc − C1). The
force and torque on the second spheroid follow by New-
ton's action-reaction law, namely
F2 = −F1,
T2 = −T1 + R × F1.
(C12)
(C13)
We restrict ourselves to short-rang repulsive interactions
by setting the potential U to a constant value for dR >
( 6√2−1)σ0, which implies that F1 and T1 are zero for this
range of dR values. Note that an upper bound to dR is
R−2bz, which means that two spheroids will not interact
if R > 2bz +( 6√2−1)σ0. This inequality is checked before
a numerical calculation of dR is employed.
2.
Interaction between a spheroid and a wall
We assume that two parallel walls are positioned at
y = 0, Ly, which-taking into account the safety distance
dv-results in the effective wall positions y = dv and Ly−
dv. We propose an interaction between a spheroid and
a wall in the style of the spheroid-spheroid interaction
presented in Ref.
88. First, we find the point x on
the spheroid's surface that is closest to a wall. For the
wall at y = dv, this is achieved by minimizing the height
h(x) = ey · x − dv under the constraint A(x) = 1. Using
the method of Lagrange multipliers, we have to minimize
Λ(x, λ) = h(x) + λ(A(x) − 1). The necessary condition
for a minimum ∂Λ/∂x = 0 yields
ey + λ∇A(x) = ey + 2λA(x − C) = 0,
(C14)
and hence,
x = C − A−1ey/(2λ).
Substitution of Eq. (C15) into A(x) = 1 yields
λ = ±q(A−1)yy/2
Finally, we obtain the point closest to the wall as
x = C ∓ (A−1ey)/q(A−1)yy.
(C15)
(C16)
(C17)
Here, the minus sign has to be chosen, which can be
visualized by the example of a sphere of radius R, for
which A = R−21. This finally yields the height
and for the torque
Tα = −
∂Uw
∂h
∂h
∂ψα
,
12
(C22)
h = Cy − dv −q(A−1)yy.
We employ the Lennard-Jones potential
(C18)
with
∂h
∂ψα
=
1
p(A−1)yy (cid:0)δαx(A−1)yz − δαz(A−1)yx(cid:1) .
(C23)
Uw = 4ǫ0"(cid:18) σ0
h + σ0(cid:19)12
−(cid:18) σ0
h + σ0(cid:19)6#
(C19)
for a repulsive wall, and Uw assumes a constant value
for all h ≥ ( 6√2 − 1)σ0. We can derive the force Fα =
−∂Uw/∂Cα and torque Tα = −∂Uw/∂ψα acting on the
spheroid analytically. For the force, we find
Here, we use the relation
d
dt
B−1 = −B−1(cid:18) d
dt
B(cid:19) B−1,
(C24)
which holds for an invertible matrix B = B(t) depending
on a scalar parameter t, and Eq. (C9) from Ref. 88.
For the wall at y = Ly − dv, we have to minimize
h(x) = Ly − dv − ey · x, with x on the spheroid's surface.
This yields
F = −
∂Uw
∂h
∂h
∂Cy
ey
= −24
ǫ0
σ0 "2(cid:18) σ0
h + σ0(cid:19)13
(C20)
(C21)
−(cid:18) σ0
h + σ0(cid:19)7# ey
x = C + A−1eyh(cid:0)A−1(cid:1)yyi−1/2
.
(C25)
The formulas for torque and force do not change, except
that we have to insert h = Ly − dv − Cy −p(A−1)yy and
need to change the sign of the force.
[1] J. Elgeti, R. G. Winkler and G. Gompper, Reports on
10, 5609.
Progress in Physics, 2015, 78, 056601.
[16] G. S. Redner, M. F. Hagan and A. Baskaran, Phys. Rev.
[2] T. Vicsek and A. Zafeiris, Phys. Rep., 2012, 517, 71.
[3] M. F. Copeland and D. B. Weibel, Soft Matter, 2009, 5,
Lett., 2013, 110, 055701.
[17] Y. Fily and M. C. Marchetti, Phys. Rev. Lett., 2012, 108,
1174.
235702.
[4] N. C. Darnton, L. Turner, S. Rojevsky and H. C. Berg,
[18] R. Grossmann, L. Schimansky-Geier and P. Romanczuk,
Biophys. J., 2010, 98, 2082.
New J. Phys., 2012, 14, 073033.
[5] D. B. Kearns, Nat. Rev. Microbiol., 2010, 8, 634–644.
[6] K. Drescher, J. Dunkel, L. H. Cisneros, S. Ganguly
and R. E. Goldstein, Proc. Natl. Acad. Sci. USA, 2011,
10940, 108.
[19] V. Lobaskin and M. Romenskyy, Phys. Rev. E, 2013, 87,
052135.
[20] A. Zottl and H. Stark, Phys. Rev. Lett., 2014, 112,
118101.
[7] J. D. Partridge and R. M. Harshey, J. Bacteriol., 2013,
[21] A. Wysocki, R. G. Winkler and G. Gompper, EPL (Eu-
195, 909.
rophysics Letters), 2014, 105, 48004.
[8] J. Bialk´e, T. Speck and H. Lowen, Phys. Rev. Lett., 2012,
[22] F. Peruani, A. Deutsch and M. Bar, Phys. Rev. E, 2006,
108, 168301.
74, 030904.
[9] I. Buttinoni, J. Bialk´e, F. Kummel, H. Lowen,
C. Bechinger and T. Speck, Phys. Rev. Lett., 2013, 110,
238301.
[10] B. M. Mognetti, A. Sari´c, S. Angioletti-Uberti, A. Cacci-
uto, C. Valeriani and D. Frenkel, Phys. Rev. Lett., 2013,
111, 245702.
[23] Y. Yang, V. Marceau and G. Gompper, Phys. Rev. E,
2010, 82, 031904.
[24] J. Elgeti and G. Gompper, EPL (Europhysics Letters),
2009, 85, 38002.
[25] A. P. Berke, L. Turner, H. C. Berg and E. Lauga, Phys.
Rev. Lett., 2008, 101, 038102.
[11] I. Theurkauff, C. Cottin-Bizonne, J. Palacci, C. Ybert
[26] E. Lauga and T. R. Powers, Reports on Progress in
and L. Bocquet, Phys. Rev. Lett., 2012, 108, 268303.
Physics, 2009, 72, 096601.
[12] Y. Fily, S. Henkes and M. C. Marchetti, Soft Matter,
[27] E. Lauga, W. R. DiLuzio, G. M. Whitesides and H. A.
2014, 10, 2132.
Stone, Biophys. J., 2006, 90, 400.
[13] X. Yang, M. L. Manning and M. C. Marchetti, Soft Mat-
[28] J. Hu, A. Wysocki, R. G. Winkler and G. Gompper, Sci-
ter, 2014, 10, 6477.
entific Reports, 2015, 5, 9586.
[14] J. Stenhammar, D. Marenduzzo, R. J. Allen and M. E.
[29] R. Di Leonardo, D. Dell'Arciprete, L. Angelani and
Cates, Soft Matter, 2014, 10, 1489.
V. Iebba, Phys. Rev. Lett., 2011, 106, 038101.
[15] Y. Fily, A. Baskaran and M. F. Hagan, Soft Matter, 2014,
[30] L. Lemelle, J.-F. Palierne, E. Chatre, C. Vaillant and
13
C. Place, Soft Matter, 2013, 9, 9759.
150603.
[31] M. J. Lighthill, Commun. Pure Appl. Math., 1952, 5,
109–118.
[32] J. R. Blake, J. Fluid Mech., 1971, 46, 199–208.
[33] J. R. Howse, R. A. L. Jones, A. J. Ryan, T. Gough,
R. Vafabakhsh and R. Golestanian, Phys. Rev. Lett.,
2007, 99, 048102.
[60] M. Yang and M. Ripoll, Phys. Rev. E, 2011, 84, 061401.
[61] D. J. Earl, C. M. Pooley, J. F. Ryder, I. Bredberg and
J. M. Yeomans, J. Chem. Phys., 2007, 126, 064703.
[62] J. Elgeti, U. B. Kaupp and G. Gompper, Biophys. J.,
2010, 99, 1018–1026.
[63] J. Elgeti and G. Gompper, Proc. Natl. Acad. Sci. USA,
[34] A. Erbe, M. Zientara, L. Baraban, C. Kreidler and P. Lei-
2013, 110, 4470.
derer, J. Phys.: Condens. Matter, 2008, 20, 404215.
[64] M. Theers and R. G. Winkler, Phys. Rev. E, 2013, 88,
[35] G. Volpe, I. Buttinoni, D. Vogt, H. J. Kummerer and
023012.
C. Bechinger, Soft Matter, 2011, 7, 8810.
[65] J. Hu, M. Yang, G. Gompper and R. G. Winkler, Soft
[36] I. Llopis and I. Pagonabarraga, J. Non-Newtonian Fluid
Matter, 2015, 11, 7843.
Mech., 2010, 165, 946.
[37] I. O. Gotze and G. Gompper, EPL (Europhysics Letters),
2010, 92, 64003.
[66] J. Happel and H. Brenner, Low Reynolds number hydro-
dynamics: with special applications to particulate media,
Springer Science & Business Media, 1983.
[38] T. Ishikawa and M. Hota, J. Exp. Biol., 2006, 209, 4452–
[67] G. Dassios and P. Vafeas, Physics Research International,
4463.
2008, 2008, e135289.
[39] T. Ishikawa and T. J. Pedley, J. Fluid Mech., 2007, 588,
[68] A. M. Leshansky, O. Kenneth, O. Gat and J. E. Avron,
399–435.
New Journal of Physics, 2007, 9, 145–145.
[40] A. A. Evans, T. Ishikawa, T. Yamaguchi and E. Lauga,
[69] E. Allahyarov and G. Gompper, Phys. Rev. E, 2002, 66,
Physics of Fluids (1994-present), 2011, 23, 111702.
036702.
[41] F. Alarc´on and I. Pagonabarraga, J. Mol. Liq., 2013,
[70] H. Noguchi, N. Kikuchi and G. Gompper, EPL (Euro-
185, 56–61.
physics Letters), 2007, 78, 10005.
[42] J. J. Molina, Y. Nakayama and R. Yamamoto, Soft Mat-
[71] H. Noguchi and G. Gompper, Phys. Rev. E, 2008, 78,
ter, 2013, 9, 4923.
016706.
[43] T. Ishikawa and T. J. Pedley, Phys. Rev. Lett., 2008, 100,
[72] M. Theers and R. G. Winkler, Phys. Rev. E, 2015, 91,
088103.
033309.
[44] K. Ishimoto and E. A. Gaffney, Phys. Rev. E, 2013, 88,
062702.
[45] S. R. Keller and T. Y. Wu, J. Fluid Mech., 1977, 80,
259–278.
[73] T. Ihle and D. M. Kroll, Phys. Rev. E, 2001, 63, 020201.
[74] T. Ihle and D. M. Kroll, Phys. Rev. E, 2003, 67, 066705.
[75] C. C. Huang, A. Chatterji, G. Sutmann, G. Gompper and
R. G. Winkler, J. Comput. Phys., 2010, 229, 168–177.
[46] G.-J. Li and A. M. Ardekani, Phys. Rev. E, 2014, 90,
[76] C.-C. Huang, A. Varghese, G. Gompper and R. G. Win-
013010.
[47] K. Kyoya, D. Matsunaga, Y. Imai, T. Omori and
T. Ishikawa, Phys. Rev. E, 2015, 92, 063027.
[48] S. E. Spagnolie and E. Lauga, J. Fluid Mech., 2012, 700,
105–147.
kler, Phys. Rev. E, 2015, 91, 013310.
[77] E. Westphal, S. P. Singh, C. C. Huang, G. Gompper
and R. G. Winkler, Comput. Phys. Commun., 2014, 185,
495–503.
[78] M. P. Allen and D. J. Tildesley, Computer Simulation of
[49] A. Zottl and H. Stark, Phys. Rev. Lett., 2012, 108,
Liquids, Clarendon Press, Oxford, 1987.
218104.
[79] J. T. Padding and A. A. Louis, Phys. Rev. E, 2006, 74,
[50] I. Pagonabarraga and I. Llopis, Soft Matter, 2013, 9,
031402.
7174–7184.
[80] M. T. Downton and H. Stark, J. Phys.: Condens. Matter,
[51] B. Delmotte, E. E. Keaveny, F. Plourabou´e and E. Cli-
2009, 21, 204101.
ment, J. Comput. Phys., 2015, 302, 524–547.
[81] A. Lamura, G. Gompper, T. Ihle and D. M. Kroll, EPL
[52] A. Malevanets and R. Kapral, J. Chem. Phys., 1999, 110,
(Europhysics Letters), 2001, 56, 319.
8605–8613.
[53] R. Kapral, Adv. Chem. Phys., 2008, 140, 89–146.
[54] G. Gompper, T. Ihle, D. M. Kroll and R. G. Winkler,
in Advanced Computer Simulation Approaches for Soft
Matter Sciences III, ed. P. C. Holm and P. K. Kremer,
Springer Berlin Heidelberg, 2009, pp. 1–87.
[82] I. P. Omelyan, Phys. Rev. E, 1998, 58, 1169–1172.
[83] L. D. Favro, Phys. Rev., 1960, 119, 53–62.
[84] S. Kim and S. J. Karrila, Microhydrodynamics: Princi-
ples and Selected Applications, Butterworth-Heinemann,
2013.
[85] I. O. Gotze and G. Gompper, Phys. Rev. E, 2010, 82,
[55] E. Tuzel, T. Ihle and D. M. Kroll, Phys. Rev. E, 2006,
041921.
74, 056702.
[56] C.-C. Huang, G. Gompper and R. G. Winkler, Phys. Rev.
[86] B. U. Felderhof, arXiv:1603.08574 [physics], 2016.
[87] Z. X. Wang and D. R. Guo, Special Functions, World
E, 2012, 86, 056711.
Scientific, 1989.
[57] S. Y. Reigh, R. G. Winkler and G. Gompper, Soft Matter,
[88] L. Paramonov and S. N. Yaliraki, J. Chem. Phys., 2005,
2012, 8, 4363–4372.
123, 194111.
[58] S. Y. Reigh, R. G. Winkler and G. Gompper, PLoS ONE,
2013, 8, e70868.
[59] G. Ruckner and R. Kapral, Phys. Rev. Lett., 2007, 98,
[89] W. H. Press, Numerical Recipes 3rd Edition: The Art of
Scientific Computing, Cambridge University Press, 2007.
|
1505.06489 | 1 | 1505 | 2015-05-24T21:40:57 | Activity driven fluctuations in living cells | [
"physics.bio-ph",
"cond-mat.soft",
"q-bio.CB"
] | We propose a model for the dynamics of a probe embedded in a living cell, where both thermal fluctuations and nonequilibrium activity coexist. The model is based on a confining harmonic potential describing the elastic cytoskeletal matrix, which undergoes random active hops as a result of the nonequilibrium rearrangements within the cell. We describe the probe's statistics and we bring forth quantities affected by the nonequilibrium activity. We find an excellent agreement between the predictions of our model and experimental results for tracers inside living cells. Finally, we exploit our model to arrive at quantitative predictions for the parameters characterizing nonequilibrium activity, such as the typical time scale of the activity and the amplitude of the active fluctuations. | physics.bio-ph | physics | epl draft
Activity driven fluctuations in living cells
´E. Fodor1,∗, M. Guo2,∗, N. S. Gov3, P. Visco(a)1, D. A. Weitz2 and F. van Wijland1
1 Laboratoire Mati`ere et Syst`emes Complexes, UMR 7057 CNRS/P7, Universit´e Paris Diderot, 10 rue Alice Domon
et L´eonie Duquet, 75205 Paris cedex 13, France
2 School of Engineering and Applied Sciences, Harvard University, Cambridge MA 02138, USA
3 Department of Chemical Physics, Weizmann Institute of Science, 76100 Rehovot, Israel
∗ These authors contributed equally to this work
5
1
0
2
PACS 87.16.dj – Dynamics and fluctuations
PACS 87.16.ad – Analytical theories
PACS 87.16.Uv – Active transport processes
Abstract – We propose a model for the dynamics of a probe embedded in a living cell, where
both thermal fluctuations and nonequilibrium activity coexist. The model is based on a confining
harmonic potential describing the elastic cytoskeletal matrix, which undergoes random active hops
as a result of the nonequilibrium rearrangements within the cell. We describe the probe's statis-
tics and we bring forth quantities affected by the nonequilibrium activity. We find an excellent
agreement between the predictions of our model and experimental results for tracers inside living
cells. Finally, we exploit our model to arrive at quantitative predictions for the parameters char-
acterizing nonequilibrium activity, such as the typical time scale of the activity and the amplitude
of the active fluctuations.
y
a
M
4
2
]
h
p
-
o
i
b
.
s
c
i
s
y
h
p
[
1
v
9
8
4
6
0
.
5
0
5
1
:
v
i
X
r
a
Actin filaments are involved in a number of functions
including cell motility, adhesion, gene expression, and sig-
nalling. When fueled by ATP supply, Myosin motors
advance along these filaments by performing a directed
stochastic motion. By tracking the trajectory of a micron-
size probe embedded within the cytoskeletal network, and
by subjecting it to microrheology experiments, one can
hope to access and understand some of the properties of
the nonequilibrium activity inside the cytoskeletal net-
work. Experiments were first carried out in actin gels
without molecular motors, known as passive gels [1–5].
Some progress in the experimental field has provided new
results for tracers attached to the cortex of living cells [6],
and also for in vitro actin gels [7, 8]. In such gels, called
active gels, the tracer dynamics exhibits large excursions
corresponding to directed motion events, in addition to
the thermal fluctuations already observed in passive gels.
Due to the active processes, the actin network fluctuations
comprise a strongly nonequilibrium component. Experi-
mentally, the out-of-equilibrium nature of such activity
has been evidenced by the violation of the fluctuation dis-
sipation theorem (FDT) [9–11]. To account for nonequilib-
rium activity, a generalization of the FDT has been devel-
(a)Correspondig Author: [email protected]
oped introducing a frequency dependent effective temper-
ature [12–14]. This generalization is based on a description
of tracers dynamics at a mesoscopic scale, which can be de-
scribed using a generalized Langevin equation [15–17]. At
a macroscopic scale, the dynamics of acto-myosin networks
have been described via hydrodynamic treatments [18] or
polymer theory [19, 20].
In what follows, we present results of microrheology ex-
periments in the cytoplasm of living cells, which are char-
acterized by a highly nonequilibrium activity. Along with
experiments, we propose a model which mixes simple but
nontrivial rheology with random fluctuations due to active
processes inside the cell. We carry out a comparison with
experimental data, which allows us to directly determine
some microscopic mechanisms that drive active fluctua-
tions inside the cell. We demonstrate that our quantitive
estimation of the nonequilibrium active features is consis-
tent with different kinds of experimental measurements,
thus supporting the overall consistency of our model.
We inject sub-micron colloidal tracers in the cytoplasm
of living A7 cells, and track a two-dimensional projec-
tion of their fluctuating (3-D) motion with confocal mi-
croscopy [21]. We observe some directed motion events in
the tracers' trajectories in addition to the thermal fluctu-
p-1
´E. Fodor et al.
Fig. 1:
(Top) Typical trajectories of 200 nm PEG coated beads
in A7 cells under three conditions: control, 10 µM blebbistatin
treatment, ATP depletion. Trajectory length is about 2 min.
(Bottom) Elastic storage modulus G(cid:48) (+) and loss modulus
G(cid:48)(cid:48) (◦) from active microrheology experiments in untreated
(red), blebbistatin treated (orange), and ATP depleted (blue)
A7 cells.
ations of small amplitudes (Fig. 1), as already reported
in synthetic active gels [7]. To investigate the role of bio-
logical activity in the intracellular mechanics, we subject
cells to two treatments. We inhibit Myosin II motors by
adding 10 µM of blebbistatin to the culture medium, and
we deplete cells of ATP through addition of 2 mM sodium
azide and 10 mM of 2-deoxyglucose. We extract the one-
dimensional mean square displacement (MSD) from the
spontaneous motion of tracers for different radius sizes
a = {50, 100, 250} nm. We present the MSD multiplied
by a for the control, blebbistatin and ATP depleted con-
ditions in Fig. 2(a), showing that the MSD scales like 1/a.
The small time MSD is constant in the three conditions,
while the large time behavior is diffusive, apart for ATP
depleted cells, where it remains almost constant. Since the
time evolution of the MSD is qualitatively similar for trac-
ers of different sizes, we deduce that the tracers are bigger
than the mesh size of the cytoskeletal network, thus allow-
ing us to consider that they evolve in a continuous medium
in first approximation.
We measure the mechanical properties of the cytoplasm
via active microrheology method by using optical tweez-
ers [22]. We impose a sinusoidal oscillation on a particle
with diameter 0.5 µm within the cytoplasm. From the
resultant displacement of the bead, we extract the com-
plex modulus G∗ = 1/(6πaχ), where χ is the Fourier re-
sponse function. It reveals that it weakly depends on fre-
quency, and that the elastic contribution is significantly
larger than the dissipative one (Fig. 1), in agreement with
previous results [22, 23]. Moreover, we do not observe
a significant change in the cytoplasmic mechanical prop-
erty according to active processes. The cytoplasm is still
Fig. 2: (a) Time evolution of the one-dimensional mean square
displacement scaled with the tracer radius a = 50 (+), 100
(◦) and 250 nm (·) for control (red), blebbistatin treated (or-
ange), and ATP depleted (blue) cells, and the corresponding
best fitting curves (Eq. (2)): solid, dashed, and dot dashed line,
respectively. (Inset) FDT-ratio as a function of frequency. It
equals 1 in ATP depleted cells as for an equilibrium system,
and it deviates from it in the two other conditions at small
frequency showing that nonequilibrium processes drive the dy-
namics in this regime. (b) Time evolution of the MSD scaled
with G(cid:48) measured with tracers of radius a = 100 nm in control
cells. The G(cid:48) value increases with the percentage of PEG in-
troduced in the cell: 0% (yellow •), 3% (light green ◦), and 6%
(dark green +). The best fit curves are in solid, dashed, and
dot dashed lines, respectively. The short time scale plateau
scales like 1/G(cid:48).
(c) Time evolution of the MSD times G(cid:48)2.
The large time diffusive part scales as 1/G(cid:48)2.
mainly elastic in blebbistatin treated and ATP depleted
cells, with a storage modulus being twice as small as in
untreated cells where it equals approximately 2 Pa.
To quantify departure from equilibrium, we extract the
FDT-ratio which compares the active microrheology mea-
p-2
100101102Frequency(Hz)100Complexmodulus(Pa)G0G0010−210−1100101Time(s)10−410−310−210−1MSD·a(µm3)(a)1Equilibriumplateau10−1100101102Frequency(Hz)100101102FDT–ratioControlBlebbistatintreatedATPdepleted10−210−110010110−310−210−1MSD·G0(µm2·Pa)1(b)Control0%PEG3%PEG6%PEG10−210−1100101Time(s)10−310−2MSD·G02(µm2·Pa2)1(c)Control0%PEG3%PEG6%PEG surement with the random intracellular motion visual-
ized by tracer particles [9–11].
It is defined in terms of
the position power spectrum C and the imaginary part
of the Fourier response function χ(cid:48)(cid:48) as FDT-ratio(ω) =
−ω C(ω)/ [2χ(cid:48)(cid:48)(ω)kBT ], where T is the bath temperature.
It equals 1 for an equilibrium system, and deviates from it
otherwise. The control and blebbistatin treated cells are
out-of-equilibrium, whereas the effect of the nonequilib-
rium processes are negligible in ATP depleted cells (Inset
in Fig. 2(a)). This supports that the nonequilibrium pro-
cesses hibernate in the latter as long as no ATP supply
is provided, suggesting that there is an equilibrium refer-
ence state where the tracer particle is trapped in an elastic
cytoskeletal network. Given we can not rely on equilib-
rium physics to describe the tracer's dynamics in the two
other conditions, we offer a new model to characterize its
nonequilibrium properties.
We vary experimentally the elastic modulus G(cid:48) by
adding various amount of 300-Dalton polyethylene glycol
(PEG) into the cell culture medium 1. This results in an
osmotic compression on the cell, so that G(cid:48) increases with
the amount of PEG applied [24]. We report in Figs. 2 (b)-
(c) the MSD data multiplied by G(cid:48) and G(cid:48)2 for different
values of G(cid:48). It appears the value of the small time plateau
scales as 1/G(cid:48) while the large time diffusion constant scales
as 1/G(cid:48)2.
The cytoskeleton acts as a thermostat for the tracer par-
ticle. Provided that inertial effects are negligible in the
intracellular environment, we model the dynamics of the
tracer's position r by means of an overdamped Langevin
equation. We use a harmonic approximation to account
for the interaction of the tracer with the surrounding net-
work. The main new ingredient of our model lies in ex-
pressing the effect of nonequilibrium activity. We postu-
late that the underlying action of the active processes in-
duces local rearrangements of the network, resulting in an
active force applied on the tracers. As an example of such
nonequilibrium processes, the activity of Myosin II motors
can slide cytoskeletal filaments past each others leading to
a local deformation of the network [7]. To account for the
directed motion events observed in our experimental tra-
jectories, we consider that the active force proceeds by
a sequence of rapid ballistic jumps followed by quiescent
periods. It remains constant during intervals of average
quiescence time τ0, when the tracer is only subjected to
thermal fluctuations, and it varies during a persistence
time of order τ by a quantity fA = f n, where n is a ran-
dom direction in the three-dimensional space. We assume
that the persistence and quiescence times are exponen-
tially distributed variables as observed in synthetic active
gels [7,8,25], and that they do not depend on the network
and tracer properties. Putting these ingredients together,
we arrive at the equation for x, the one-dimensional pro-
1After the stress, cells are allowed to equilibrate for 10 min at
37◦C and 5% CO2, before we perform the imaging or optical-tweezer
measurement. The cell size and mechanics equilibrate in 2 min after
adding PEG based on our imaging and previous studies [24].
Activity driven fluctuations in living cells
Fig. 3:
Typical realization of (a) the active force fA, and
(b) the corresponding active bursts vA. fA is constant over
a quiescence time of typical value τ0, and varies linearly with
a slope uniformly distributed in [−f, f ] during a persistence
time of order τ . vA is proportional to the time derivative of
fA. (c) Schematic representation of the energetic landscape re-
arrangement due to nonequilibrium activity and its modeling
using the active burst applied on the local minimum. We depict
the network potential in black solid line, the harmonic approx-
imation in dashed red line, and the tracer particle in filled blue
circle. Nonequilibrium activity leads to a displacement vτ of
the potential, resulting in an energy gain E (cid:39) k(vτ )2 for the
tracer.
jection of r
= −kx + ξ + fA ,
γ
dx
dt
(1)
where ξ is a zero mean Gaussian white noise with correla-
tions (cid:104)ξ(t)ξ(t(cid:48))(cid:105) = 2γkBT δ(t−t(cid:48)), and fA is a random force
with typical realization described in Fig. 3(a). The spring
constant of the surrounding network is k, and γ is the
friction coefficient of the environment. Our model is asso-
ciated with a Fourier response function χ = 1/(k + iωγ),
from which we deduce that complex modulus is of the form
G∗ = 1/(6πaχ) = k/(6πa) + iωη, where η is the viscosity
of the fluid surrounding the tracer [17,26]. We neglect the
weak frequency dependence of the real part G(cid:48) as deter-
mined from active microrheology measurements, so that
the spring constant is directly given by k = 6πaG(cid:48), as al-
ready reported in other complex fluids with similar elastic
behavior [26]. Stokes' law ensures that γ is independent
of G(cid:48), and γ ∝ a.
To illustrate our model with an immediate physical pic-
ture, we introduce the variable r0 = fA/k which we regard
as the position of the local minimum of the potential in
which the tracer is trapped. The local rearrangements
of the network due to nonequilibrium activity result in a
shift of the local minimum the tracer sits in. Thus, this
position has a dynamics of its own given by a random
p-3
Timev0−vvA(b)f0−ffAττ0(a)(c)NetworkpotentialHarmonicapproximationEvτEnergyLocalrearrangementPosition ´E. Fodor et al.
active burst vA in which a burst v n is felt during the per-
sistence time, while it equals zero during the quiescence
time (Fig. 3(b)). The active force projection is simply re-
lated to the active burst projection as dfA/dt = kvA. We
assume that the typical variation f of the active force is
independent of the network properties, whereas the active
burst amplitude v = f /(kτ ) depends on the properties of
the cytoskeletal network via k.
From the Fourier transform of Eq. (1), we compute the
position autocorrelation function C(t) = (cid:104)x(t)x(0)(cid:105), and
2(C(0) − C(t)). We denote the thermal contribution to
then deduce the one-dimensional MSD as (cid:10)∆x2(cid:11) (t) =
(cid:11), and the MSD when the particle
the MSD by (cid:10)∆x2
is only subjected to motor activity by (cid:10)∆x2
(cid:11), so that:
(cid:11). The thermal MSD is the same
(cid:11) +(cid:10)∆x2
(cid:10)∆x2(cid:11) =(cid:10)∆x2
(cid:16)
(cid:11) (t) =
(cid:10)∆x2
(cid:10)∆x2
(cid:11) (t) =
as for the Ornstein-Uhlenbeck process, and we compute
the active contribution in terms of the parameters charac-
terizing the active force:
(cid:20)(cid:18) τ
1 − e−t/τr
T
A
(2a)
(cid:19)
A
T
T
A
,
(cid:17)
(cid:19)3(cid:18)
(cid:21)
2kBT
k
2kBTA/k
1 − (τ /τr)2
+e−t/τr +
t
τr
τr
− 1
1 − e−t/τ − t
τ
,
(2b)
where τr = γ/k is a microscopic relaxation time scale. In
the passive case, i.e. when TA = 0, it saturates to the value
2kBT /k within a time τr as predicted by the equipartition
theorem, meaning that the tracer is confined in the cy-
toskeleton. The active force represents the random fluctu-
ations of the cytoskeletal network induced by the nonequi-
librium activity. With such a force, the MSD exhibits a
plateau at the equilibrium value corresponding to a tran-
sient elastic confinement at times τr (cid:28) t (cid:28) τ , and then
has a diffusion-like growth on longer times with coefficient
2kBTA/γ. Provided that k ∝ G(cid:48), it follows that the equilib-
rium plateau scales like 1/G(cid:48), as we observe experimentally
(Fig. 2(b)). The energy scale kBTA = γ(vτ )2/[3(τ + τ0)]
defines an active temperature, which is related to the am-
plitude of the active fluctuations as defined by the ac-
tive burst correlations (cid:104)vA(t)vA(0)(cid:105) = kBTAe−t/τ /(τ γ).
The independence of f and τ with respect to G(cid:48) yields
v ∝ 1/G(cid:48), from which we deduce that the large time dif-
fusion coefficient scales as 1/G(cid:48)2, in agreement with our
measurements (Fig. 2(c)).
On the basis of our phenomenological picture where the
nonequilibrium dynamics is driven by an active remod-
elling of the cytoskeletal network, we propose a physical
argument for the scaling of the MSD with the tracers' size
a presented in Fig. 2(a). As presented above, we first as-
sume that k and γ scales like a. Within our model, the
active burst represents the activity-driven network defor-
mation and reorganization, which result in a change of the
tracer's local energetic landscape. During a burst event,
the local minimum is shifted by a random amount. Re-
garding this event as instantaneous, the tracer finds itself
at a distance of order vτ from the new local minimum
position after each burst. It follows that the typical en-
ergy provided by nonequilibrium activity to the particle is
E (cid:39) k(vτ )2, as depicted in Fig. 3(c). We assume that it
does not depend on the particle properties, just as τ and
τ0, thus being independent of the tracer's typical size a.
Since k ∝ a, we deduce v ∝ 1/
a, implying that TA is
independent of a. Finally, the relaxation time τr is also
independent of a, leading to a scaling of the MSD like 1/a
which agrees with our observation.
√
We use our analytic expression to fit the MSD data mul-
tiplied by a for the three conditions described above. We
assume the viscosity of the fluid surrounding the tracer is
the cytoplasm viscosity η ∼ 10−3 Pa· s [27], and we de-
duce the damping coefficient from Stokes' law: γ = 6πaη.
We estimate the k value from the small time plateau.
The only remaining parameters are the ones character-
izing nonequilibrium activity: TA/T = {2.8, 0.9} × 10−3,
τ = {0.16± 0.03, 0.39± 0.09} s, in control and blebbistatin
treated cells, respectively. The estimation error made on
TA/T is of the order of 1% in control, and 0.1% in bleb-
bistatin treated cells.
The amplitude of the active fluctuations is smaller in
blebbistatin treated cells, meaning that the inhibition of
Myosin II motors reduces the proportion of nonequilib-
rium fluctuations with respect to the thermal ones as ex-
pected. Other nonequilibrium processes drive the out-of-
equilibrium dynamics in this condition. The typical time
scale τ of the persistent motion events is enhanced in bleb-
bistatin treated cells. Assuming that each active burst
persists until the stress that accumulates in the network
causes the network to locally fail, weaker motors due to
the addition of blebistatin will contract for a longer dura-
tion until such a critical stress builds up. Provided that
1/τ is the typical frequency below which the nonequilib-
rium processes affect the dynamics, this supports that the
active fluctuations take over the thermal ones at larger
frequencies in the control cells compared with the blebbis-
tatin treated ones. Notice that TA represents the ability
of the tracer to diffuse on long times, and T quantifies
here only the motion of the bead at short times when it is
trapped within the elastic cytoskeletal network. The fact
that we find TA small compared to T does not mean that
the active processes are negligible, as they control entirely
the long-time and long-distance diffusion of the tracer. In
the absence of activity, the tracer does not diffuse at all
and remains trapped in the elastic network.
To characterize the properties of the active force, we
focus on the power spectrum of the stress fluctuations,
i.e. the Fourier transform of the time correlation function
(cid:104)fA(0)fA(t)(cid:105) [9, 11, 15]. We extract the power spectrum
of the overall force fA + ξ as the power spectrum of the
position times (6πaG∗)2 [15]. Provided that the ATP
depleted condition is in an equilibrium state, the active
force fA is negligible in these cells and the overall force
reduces to the ξ, thus providing a direct measurement of
the thermal force spectrum. We remove this equilibrium
p-4
contribution to the overall spectrum to deduce the active
force spectrum in the two other conditions. We observe
a 1/ω2 behavior at low frequency as already accounted
for on general grounds [9, 14, 15, 28], and the large fre-
quency curvature hints a crossover to another power law
(Figs. 4(a)-(b)). Our analytic prediction for the active
force spectrum reads
SA(ω) =
1
(ωτr)2
2γkBTA
1 + (ωτ )2 .
(3)
It combines properties of the network and parameters
characterizing the active force, since the effect of nonequi-
librium activity on the tracer is mediated by the network
within our model. We recover the divergence as 1/ω2 at
low frequency, and we predict a power law behavior 1/ω4
at high frequency, the crossover between the two regimes
appearing at 1/τ . We compare our prediction with the
experimental data by using the best fit parameters esti-
mated from the MSD data. Without any free parameter,
we reproduce the measured spectra (Figs. 4(a)-(b)). This
result is a strong support for our model, in which TA not
only quantifies the long time diffusion coefficient of the
tracers, it is also related to the typical amplitude of the
fluctuations generated by the nonequilibrium active force.
The study of the high frequency spectrum calls for new
experiments as it would confirm the validity of our phe-
nomenological picture.
To study in more details the properties of the active
force, we analyze the probability distribution function of
the tracer displacement (DPDF). It exhibits a Gaussian
behavior at short and long times.
In the intermediate
regime, we observe a central Gaussian part which matches
our equilibrium prediction in the absence of activity, and
exponential tails accounting for directed motion events
consistent with previous observations in synthetic active
gels [7]. Within our model, the non-Gaussian behavior
of the DPDF is a direct and unique consequence of the
non-Gaussianity of the active force. We ran numerical
simulations of the dynamics in Eq. (1) to reproduce the
time evolution of the DPDF. We set the different parame-
ter values to the one estimated previously, letting us with
only one free parameter: the average quiescence time τ0.
It quantifies the average time between two successive di-
rected motion events, thus controlling the relative impor-
tance of the exponential tails with respect to the Gaussian
central part. We adjust this parameter by matching the
exponential tails observed at different times.
With a fixed τ0 value, we manage to reproduce the evo-
lution in time of the whole experimental DPDF. This
shows that the specific form we choose for the active
process is sufficient to reproduce not only the MSD and
force spectrum data, but also to account quantitatively for
the dynamic non-Gaussian properties of the distribution
(Figs. 4(c)-(d)). We estimate τ0 = {2.5, 2.8} s in control
and blebbistatin treated cells, respectively. The extracted
values are very similar for the two conditions, showing
Activity driven fluctuations in living cells
Fig. 4:
(Top) Active force spectrum SA as a function of fre-
quency measured with tracers of radius a = 250 nm in (a) con-
trol , and (b) blebbistatin treated cells. The experimental data
are in black · , and the dashed lines correspond to Eq. (3) with
the parameter values deduced from the best fit of the MSD
data. (Bottom) Probability distribution function of the tracer
displacement (DPDF) at two different times: 1 (◦), and 5 s (·).
The DPDF is measured with tracers of radius a = 250 nm in
(c) control, and (b) blebbistatin treated cells. We present the
corresponding results from numerical simulations of Eq. (1) in
solid and dashed lines, respectively. The blue dot dashed line
is the corresponding equilibrium Gaussian. The parameter val-
ues are the same for the two lag times: (b) {TA/T, τ, τ0, k, γ} =
{2.8 × 10−3, 0.16 s, 2.5 s, 8.5 pN/µm, 4.7 × 10−3 pN · s/µm},
(c) {TA/T, τ, τ0, k, γ} = {9×10−4, 0.39 s, 2.8 s, 8.2 pN/µm, 4.7×
10−3 pN · s/µm}.
that the addition of blebbistatin does not affect the typical
time over which the tracers are only subjected to thermal
fluctuations. It suggests that this time scale is related to
the recovery of the network following a large reorganiza-
tion, thus being barely independent of the activity of the
nonequilibrium processes. Notice that the corresponding
duty ratio pon = τ /(τ + τ0) is smaller in control than
in the blebbistatin treated cells: pon = {6, 15}%, respec-
tively. It is a quantitative evidence that the exponential
tails are more pronounced in the control condition, namely
the proportion of directed motion events is increased. We
deduce the value of the typical active burst amplitude:
v = {0.86, 0.22} µm/s in control and blebbistatin treated
cells for a = 250 nm, which are compatible with velocity
scales observed in [29].
Microrheology methods have become a standard tech-
nique to explore cellular activity in living organisms [30].
p-5
10−1100Frequency(Hz)10−410−2100−2(b)Bleb.treatedExp.dataPrediction10−1100101Frequency(Hz)10−410−2100SA(pN2·s)−2(a)ControlExp.dataPrediction−0.6−0.4−0.20.00.20.40.6Displacement(µm)10−1100101DPDF(c)Control1s5s−0.6−0.4−0.20.00.20.40.6Displacement(µm)10−1100101DPDF(d)Blebb.treated1s5s ´E. Fodor et al.
In this work, we introduce a new model for characterizing
the motion of a tracer in a living cell. This model explic-
itly accounts for the elastic behavior of the cytoskeletal
network and successfully combines it with a description
of the cellular active force-a well defined non-Gaussian
colored process. By analyzing the MSD data, we quantify
two essential features of this force:
its strength, and the
typical time scale over which it is felt. Our model goes be-
yond previous modeling which treated the nonequilibrium
activity as a random noise with unprescribed character-
istics [15].
In a previous work, activity was modeled as
a trichotomous noise acting directly on the particle [12],
whereas such activity is mediated by the surrounding net-
work within our new proposal. The present model com-
bines the short time confined behavior with a long time
free diffusion which is driven by the active force, and re-
covers all the main experimental results. The model ap-
plies as long as we are in the regime of simple viscoelastic
behavior. Dressing our model with a more realistic rhe-
ology, e.g. with a power law behavior for the complex
modulus, usually observed in cell rheology [31], is concep-
tually straightforward as a future elaboration of the model.
Further generalization of our model could be used to de-
scribe active fluctuations in other non-equilibrium (living
or mechanically driven) systems that exhibits similar be-
havior [7, 32, 33].
We acknowledge several useful discussions with Julien
Tailleur and Fran¸cois Gallet. N.S.G. gratefully acknowl-
edges funding from the Israel Science Foundation (grant
no. 580/12).
REFERENCES
[1] Hou L., Luby-Phelps K. and Lanni F., The Journal of
Cell Biology, 110 (1990) 1645.
[2] Jones J. D. and Luby-Phelps K., Biophysical Journal,
71 (1996) 2742.
[3] Sanabria H. and Waxham M. N., The Journal of Phys-
ical Chemistry B, 114 (2010) 959.
[4] Apgar J., Tseng Y., Fedorov E., Herwig M. B.,
Almo S. C. and Wirtz D., Biophysical Journal, 79
(2000) 1095 .
[5] Tseng Y., Kole T. P., Lee S.-H. J. and Wirtz D.,
Current Opinion in Colloid & Interface Science, 7 (2002)
210 .
[6] Bursac P., Lenormand G., Oliver B. F. M., Weitz
D. A., Viasnoff V., Butler J. P. and Fredberg J. J.,
Nature Materials, 4 (2005) 557.
[7] Toyota T., Head D. A., Schmidt C. F. and Mizuno
D., Soft Matter, 7 (2011) 3234.
[12] Ben-Isaac E., Park Y. K., Popescu G., Brown F.
L. H., Gov N. S. and Shokef Y., Phys. Rev. Lett., 106
(2011) 238103.
[13] Prost J., Joanny J.-F. and Parrondo J. M. R., Phys.
Rev. Lett., 103 (2009) 090601.
[14] Loi D., Mossa S. and Cugliandolo L. F., Soft Matter,
7 (2011) 3726.
[15] Lau A. W. C., Hoffmann B. D., Davies A., Crocker
J. C. and Lubensky T. C., Phys. Rev. Lett., 91 (2003)
198101.
[16] Bohec P., Gallet F., Maes C., Safaverdi S., Visco
P. and van Wijland F., Europhys. Lett., 102 (2013)
50005.
[17] Mason T. G. and Weitz D. A., Phys. Rev. Lett., 74
(1995) 1250.
[18] Joanny J.-F. and Prost J., HFSP J, 3 (2009) 94104.
[19] MacKintosh F. C. and Levine A. J., Phys. Rev. Lett.,
100 (2008) 018104.
[20] Levine A. J. and MacKintosh F. C., The Journal of
Physical Chemistry B, 113 (2009) 3820.
[21] Cunningham C., Gorlin J., Kwiatkowski D.,
Hartwig J., Janmey P., Byers H. and Stossel T.,
Science, 255 (1992) 325.
[22] Guo M., Ehrlicher A. J., Jensen M. H., Renz M.,
Moore J. R., Goldman R. D., Lippincott-Schwartz
J., Mackintosh F. C. and Weitz D. A., Biophysical
Journal, 105 (2013) 1562.
[23] Fabry B., Maksym G. N., Butler J. P., Glogauer
M., Navajas D. and Fredberg J. J., Phys. Rev. Lett.,
87 (2001) 148102.
[24] Zhou E. H., Trepat X., Park C. Y., Lenormand G.,
Oliver M. N., Mijailovich S. M., Hardin C., Weitz
D. A., Butler J. P. and Fredberg J. J., Proceedings
of the National Academy of Sciences, 106 (2009) 10632.
[25] Silva M. S., Stuhrmann B., Betz T. and Koenderink
G. H., New Journal of Physics, 16 (2014) 075010.
[26] Mason T. G., Rheologica Acta, 39 (2000) 371.
[27] Mastro A. M., Babich M. A., Taylor W. D. and
Keith A. D., Proceedings of the National Academy of
Sciences, 81 (1984) 3414.
[28] Fakhri N., Wessel A., Willms C., Pasquali M.,
Klopfenstein D., MacKintosh F. and Schmidt C.,
Science, 344 (2014) 1031.
[29] Roding M., Guo M., Weitz D. A., Rudemo M. and
Sarkka A., Mathematical Biosciences, 248 (2014) 140 .
[30] Guo M., Ehrlicher A., Jensen M., Renz M., Moore
J., Goldman R., Lippincott-Schwartz J., MacKin-
tosh F. and Weitz D., Cell, 158 (2014) 822.
[31] Koenderink G. H., Dogic Z., Nakamura F., Bendix
P. M., MacKintosh F. C., Hartwig J. H., Stossel
T. P. and Weitz D. A., PNAS, 106 (2009) 15192.
[32] Weeks E. R. and Weitz D. A., Chemical Physics, 284
(2001) 361.
[33] Fodor E., Kanazawa K., Hayakawa H., Visco P. and
van Wijland F., Phys. Rev. E, 90 (2014) 042724.
[8] Stuhrmann B., Soares e Silva M., Depken M.,
MacKintosh F. C. and Koenderink G. H., Phys. Rev.
E, 86 (2012) 020901.
[9] Mizuno D., Tardin C., Schmidt C. F. and MacKin-
tosh F. C., Science, 315 (2007) 370.
[10] Betz T., Lenz M., Joanny J.-F. and Sykes C., PNAS,
106 (2009) 15320.
[11] Gallet F., Arcizet D., Bohec P. and Richert A.,
Soft Matter, 5 (2009) 2947.
p-6
|
1202.2951 | 1 | 1202 | 2012-02-14T07:05:35 | Species Diversity in Rock-Paper-Scissors Game Coupling with Levy Flight | [
"physics.bio-ph",
"cond-mat.stat-mech",
"q-bio.PE"
] | Rock-paper-scissors (RPS) game is a nice model to study the biodiversity in ecosystem. However, the previous studies only consider the nearest- neighbor- interaction among the species. In this paper, taking the long range migration into account, the effects of the interplay between nearest-neighbor-interaction and long-range-interaction of Levy flight obey the power law distance distribution with the exponent h (-0.3<h<-0.1) in spatial RPS game is investigated. Taking the probability of long range Levy flight and the power exponent as parameters, the coexistence conditions of three species are found. The critical curves for stable coexistence of three species in the parameters space are presented. It is also found that long-range-interaction with Levy flight has interesting effects on the final spatiotemporal pattern of the system. The results reveal that the long-range-interaction of Levy flight exhibit pronounced effects on biodiversity of ecosystem. | physics.bio-ph | physics | Species Diversity in Rock-Paper-Scissors Game Coupling
with Levy Flight
Dong Wang1,2, Qian Zhuang1, Jing Zhang1, Zengru Di1∗,
1. Department of Basic Science, Naval Aeronautical And Astronautical University, Yantai
264001, China
2. Department of Systems Science, School of Management and Center for Complexity
Research, Beijing Normal University, Beijing 100875, China
Abstract
Rock–paper–scissors (RPS) game is a nice model to study the biodiversity in
ecosystem. However, the previous studies only consider the nearest- neighbor-
interaction among the species. In this paper, taking the long range migration into
account, the effects of the interplay between nearest-neighbor-interaction and
hl
(-0.3<h<-0.1) in
long-range-interaction of Levy flight with distance distribution
spatial RPS game is investigated. Taking the probability of long range Levy flight and
the power exponent as parameters, the coexistence conditions of three species are
found. The critical curves for stable coexistence of three species in the parameters
space are presented. It is also found that long-range-interaction with Levy flight has
interesting effects on the final spatiotemporal pattern of the system. The results reveal
that the long-range-interaction of Levy flight exhibit pronounced effects on
biodiversity of ecosystem.
Key Words: Rock–paper–scissors game, Levy flight, biodiversity, spatial pattern
PACS: 87.23.Cc Population dynamics and ecological pattern formation, 89.75.Da Systems
obeying scaling laws, 89.75.Kd Patterns
1. Introduction
Rock–paper–scissors (RPS) game describes a basic kind of cyclic, non-transitive
interaction among competing populations [1-4]. Experimental and theoretical
investigations have shown that diversity of some species in ecosystem is maintained
by this type of mechanism.
The previous theoretical works are mainly based on nearest-neighbor-interactions
(NNI) [5-10]. Recently RPS game, provided with the interplay between the interaction
range and mobility on coexistence, was studied [11, 12]. In reality, individuals in
∗ Author for Correspondence: [email protected]
populations may go long distance before predation or reproduction. For example,
birds fly through long distance apart from its’ habitat and prey, in special seasons
some kinds of birds migrate over very long distance, other animals do so in different
territories in some season, bacteria swim and tumble in the air or water, and so on. So
long-range-interaction (LRI) is universal in the process of biological evolution.
Empirical studies on the pattern of many living organisms’ moving have revealed
the existence of Levy flight. Now observations of Levy flight have been extended to
many species [13-21]. Levy flight is characterized by many small steps connected by
longer relocations; the same sites are revisited scarcely. The probability density
function for the traveling distance has a power-law tail in the long-distance regime, as
fellow
h
P l
l
− ≤ < − ,
( ) ~ ( 3
1)
h
i
i
il is the flight length of step i , h is the power-law exponent. Levy-flight foraging
hypothesis [22-24] predicts that Levy flight have higher search efficiencies in
environments where prey is scarce, nevertheless a Brownian walk take place more
likely where prey is abundant. Hence there exist NNI and LRI among species, such as
selection, reproduction and migration. Based on above ideas, we explore the effects of
long-range-interactions upon species diversity in the framework of RPS game in
spatially extended ecological system in this paper.
The model is described in Section 2. Section 3 gives some simulation results.
Taking the probability of long range Levy flight and the power exponent as
parameters, the coexistence conditions of three species are found. The critical curves
for stable coexistence of three species in the parameters space are presented. They
display that the higher the probability of long range interaction is and the larger the
exponent h is, the easier the coexistence state of three species is destroyed. More
interestingly, compared with the only nearest–neighbor -interaction, a small fraction
of long-range-interaction with Levy flight generate more fragments in the patterns,
lead to spatial heterogeneity and patchiness which are common in ecosystem.
However, the final steady pattern forms larger dense spiral wave when there is only
long-range-interaction. The results reveal that the long-range-interaction of Levy
flight exhibit pronounced effects on biodiversity of ecosystem.
2. The spatial RPS game with Levy flight
Consider a spatial and stochastic model of cyclically competing populations.
Three species A , B and C populate a square lattice of
NN ×
sites with periodic
boundary condition. Each site is occupied by one individual or left empty, φ
denotes empty site. Agent can interact with each other by cyclic competition,
following the rules
C
,
φ
)1(
AB
)2(
B
,
φ
CA
CC
,
BC
A
,
φ
μ
→
δ
BAA
,
φ
→
μ
→
δ
CBB
,
φ
→
μ
→
δ
A
φ
→
e
XY
YX
YX
φCBA
,
})
,
(
,{
,
,
)3(
→
∈
Equations (1) reflect cyclic selection: A can kill or prey B, yielding an empty site. In
the same way, B outperforms C, and C in turn defeats A. Selection occurs at rate μ.
Equations (2) describe reproduction of individuals at rate δ, which is only allowed
on empty site. In Equation (3), individual is able to swap position with each other at
exchange rate e , including empty sites.
Our interest is focused on the interplay between NNI and LRI in spatial RPS
game. We choose Levy flight as the pattern of LRI. In stochastic lattice simulations
with periodic boundary conditions, the initial densities of three species and empty
sites are always fixed as 25℅equally. At each Monte Carlo simulation step, firstly a
random individual is chosen，then the interaction pattern is decided randomly. NNI is
0p and LRI of Levy flight is chosen with probability 1-
0p .
chosen with probability
When the interaction is NNI, the individual previously chosen will interact with one
of its four nearest neighbors, which is also randomly determined. When the
interaction is LRI, the individual previously chosen will interact with another site
which is decided by Levy flight using the algorithm of Monte Carlo simulation. We
first chose a distance l (manhatton distance in this paper) from the power law
P l
l
( ) ~ h
distribution
, and then randomly chose a site from all the sites with the
i
i
same distance. Then, according to the algorithm of Gillespie [25,26], the probabilities
of the three possible reactions in equation (1)、(2) and (3), are computed by using the
μ , reproduction with the
reaction rates. Selection occurs with the probability
e++δμ
e
δ , and exchange occurs with the probability
e
e++δμ
++δμ
every individual has reacted on average once, one generation was set as the unit of
time.
Taking account of species in realistic situation, individual walks or flies under the
constraints of its bio-energy and territory. So we propose that each individual’s step
l
distributed from 1 to a maximal length (denoted by max
) when Levy flights is chosen.
Concisely selection rate μ and reproduction rate δ are set equal to 1 in the
simulation. In our stochastic lattice simulations, we mainly studied coexistence
condition of three species. We define the condition of species extinction so far as one
of the three species extinct in the simulations.
probability
. When
3. The simulation results
(1) Critical curves of
0p vs. h for given e and max
l
2.0=e
, then study the coexistence condition of three
We set the exchange rate
0p and power-law exponent h with the
species: the critical curves of parameter
l
maximal step length max
of Levy flight changes on the lattice with 100×100 sites.
In the stochastic lattice simulations, we employ 105 generations of each realization.
Fig. 1(a-b) show the results. Each curve means critical and stable coexistence of three
species in Fig. 1(a-b).
Proposing else condition are unchanged. Likewise we find the LRI effects of Levy
0p is reduced (means the probability of
flight are pronounced. As the parameter
choosing Levy flight increases), power-law exponent h must decrease for stable
l
increases
coexistence of three species. In Fig.1(a), when maximal step length max
from 5 to 50, the area which represent stable coexistence of three species on the graph
l
is closer to a margin. Along with the
diminish gradually. The curve of
50
max =
l
increment of maximal step length max
from 50 to 105 in Fig.1(b), we could find that
the area of stable coexistence of three species vary a little. Furthermore the area of
stable coexistence state tends to a minimal limit along with increment of the maximal
l
in Fig. 1(a-b). These results show that the long-range interaction
step length max
with Levy flight affect the stability of the coexistence state strikingly.
(a) (b)
0p and power-law exponent h in the context of different
Fig.1. Relation between parameter
l
l
,
of Levy flight. Colors represent different maximal step length of max
maximal step length max
in (a): 5 (blue circle), 10 (magenta circle), 20 (black circle), 30 (green circle), 50 (red flake); in
(b): 50 (blue triangle), 65 (green circle), 85 (black circle),105 (red flake). The horizontal ordinate
0p , vertical coordinate represents power-law exponent h . Each point
represents the parameter
is averaged over 30 random realizations of stochastic simulations employing same initial densities
100
100 ×
square lattice. Each realization consists of 105 generations. Above each curve is
on a
extinction region of three species, correspondingly below (and including ) each curve is stable
coexistence region of three species.
(2) The results of the small system
In order to uncover whether lattice scale have effect on stable coexistence
condition of three species, we reduce the square lattice into
for further
50 ×
50
simulations. The result is displayed in Fig.2. Each curve represents critical and stable
coexistence of three species. The same as the above simulations, each realization is
the results of 105 generations. It could be found that the area of stable coexistence of
three species become much lesser than that of
square lattice, although the
100
100 ×
trend of variation is the same. Similarly the effects of Levy flight are displayed
0p is reduced, power-law exponent h must be
obviously. When the parameter
decreased for stable coexistence of three species. With the increase of maximal step
l
from 2 to 25, the area representing stable coexistence of three species on
length max
the graph shrinks gradually. Furthermore in Fig.2, with the further increment of
l
from 25 to 55, the area of stable coexistence of three
maximal step length max
species only changes a little. It also appears that the area of stable coexistence have a
l
. Compared with the
minimal limit with the increase of maximal step length max
results in the last subsection, we can see the size of the system affects the results
obviously.
0p and power-law exponent h with different maximal
Fig.2. Critical curves of parameter
l
50 ×
50
of Levy flight. The square lattice is
. Colors represent different
step length max
l
: 2(green circle), 10 (red circle), 20 (magenta circle), 25 (red flake),
maximal step length of max
30 (blue circle), 50 (black triangle). The horizontal and vertical coordinate represent the same
parameters as Fig.1. Each point is averaged over 30 random realizations of stochastic
simulations of same initial densities. Each realization contains 105 generations.
(3) The spatiotemporal pattern under Levy flight
In stochastic lattice simulations, we find the LRI of Levy flight also changes the
spatial pattern in RPS game. Fig.3 and Fig.4 show snapshots of the patterns for
different values of parameters, the lattice sizes are
and
100
100 ×
500
500 ×
respectively. Individuals of three competing species A (red), B (yellow), C (light blue)
and empty site (dark blue) occupy the sites of the lattice.
The patterns in Fig.3 and Fig.4 display some characters different from the
previous work [5-12]. Furthermore, we can see that some individuals of three species
move with Levy flight behavior clearly. On condition that other parameters are
→=p
unchangeable. Firstly, as
0.1
0.0
(means that the probability of Levy flight
0
increases), the number of species patches abruptly increase and then succession
−→−=h
decrease gradually. Secondly, when
, the number of species patches
0.1
0.3
increase inchmeal. Compared with NNI, the introduce of LRI of Levy flight generate
more fragments in the spatial patterns initially, lead to spatial heterogeneity and
patchiness. Those results are similar to the characters of species patches in ecosystem.
l
As maximal step length max
increases, the number of species patches decrease, and
(a) (b) (c)
(d) (e) (f)
100 ×
100
Fig.3. Snapshots of patterns. The size of square lattice is
. For each realization, the
l
simulation consists of 104 time steps. The maximal step length max
of Levy flight equals 50.
h
e
p
e
p
,0.2
,1.0
9.0
,1.0
0.1
=
0 =
0 =
=
−=
.
. (b)
(a)
e
h
p
p
e
h
,1.0
,0.2
5.0
9.0
,1.0
,0.3
−=
=
0 =
−=
0 =
=
. (d)
.
(c)
h
e
p
h
e
p
,0.3
,1.0
0.0
,0.3
,1.0
5.0
=
0 =
=
0 =
−=
−=
.
. (f)
(e)
(a)
(b) (c) (d)
(e) (f) (g)
500
500 ×
Fig.4. Snapshots of patterns. The size of square lattice is
. For each realization, the
l
simulation consists of 104 time steps. The maximal step length max
of Levy flight equals 50.
e
p
0.1
,2.0
0 =
=
.
(a)
h
e
p
p
,5.1
,2.0
9.0
0 =
=
0 =
−=
. (c)
p
p
e
h
0.0
,2.0
,0.2
0 =
0 =
=
−=
. (e)
h
e
p
p
,0.2
,2.0
5.0
=
0 =
0 =
−=
. (g)
,5.1
−=
,5.1
−=
,0.2
−=
,2.0
,2.0
,2.0
(b)
(d)
(f)
h
h
h
5.0
9.0
0.0
.
.
.
e
e
e
=
=
=
→=p
bigger patches grow in size gradually. On the condition that
,
0.1
0.0
0
, and e unchanged, the patterns vary from spiral wave to
h = − → −
3.0
1.0
fragmentized patches, and then to spiral wave again with bigger size of patches. In
order to see the change of spatiotemporal patterns clearly, we have done the stochastic
lattice simulations in the lattice with
sites. The results demonstrate the
500 ×
500
same phenomena. The effects of lone-range interaction on the final spatiotemporal
patterns are interesting and are needed for further investigation.
4. Concluding remarks
In this paper, we have mainly studied the interplay between NNI and LRI of Levy
flight in spatial rock–paper–scissors game. Our aim is to find the effects of the
long-range-interactions on the biodiversity in ecosystem. The coexistence condition of
0p , power-law
three species, such as the critical curve with the change of parameter
l
exponent h , and maximal step length max
of Levy flight are presented. The curves
also show the change of the area representing stable coexistence of three species in
l
approximately equals half
the parameter space. When the maximal step length max
of lattice’s length, the area of stable coexistence of three species tend to a minimal
0p , maximal step length
limit. All the simulation results show that the probability
l
, and power-law exponent h of Levy flight make distinct impacts on the stability
max
and the forms of the spatiotemporal patterns. Long-range-interactions with Levy flight
are common in the real ecosystems. It needs further investigation to reveal the effects
and the corresponding mechanisms of Levy flight on the biodiversity.
Acknowledgments
This work was partially supported by the NSFC under the Grant Nos. 61174150
and 60974084, the Program for New Century Excellent Talents in University of
Ministry of Education of China (No. NCET-09-0228), fundamental research funds for
the Central Universities of Beijing Normal University, and HSCC of Beijing Normal
University.
References
[1] J. B. C. Jackson and L. Buss, Proc. Natl. Acad. Sci. USA 72, 5160 (1975).
[2] C. E. Paquin and J. Adams, Nature (London) 306, 368 (1983).
[3] B. Sinervo and C. M. Lively, Nature (London) 380, 240 (1996).
[4] B. Kerr, M. A. Riley, M. W. Feldman, and B. J. M. Bohannan, Nature (London) 418, 171
(2002).
[5] T. Reichenbach, M. Mobilia, and E. Frey, Nature (London) 448, 1046 (2007).
[6] T. Reichenbach, M. Mobilia, and E. Frey, Phys. Rev. Lett. 99, 238105 (2007).
[7] T. Reichenbach, M. Mobilia, and E. Frey, J. Theor. Biol. 254, 368 (2008).
[8] T. Reichenbach and E. Frey, Phys. Rev. Lett. 101, 058102 (2008).
[9] Szabo, G., Fath, G. Phys. Rep. 446, 97–216(2007).
[10] Hongjing Shi, Wen-Xu Wang, Rui Yang, and Ying-Cheng Lai. Phys. Rev. E 81,
030901R(2010).
[11] Xuan Ni, Wen-Xu Wang, Ying-Cheng Lai, and Celso Grebogi. Phys. Rev. E 82,
066211(2010).
[12] Wen-Xu Wang, Xuan Ni, Ying-Cheng Lai, and Celso Grebogi. Phys. Rev. E 83,
11917(2011).
[13] Metzler R. and Klafter J., Phys. Rep. 339, 1(2000).
[14] Metzler R. and Klafter J., J. Phys. A: Math. Gen. 37, R161 (2004).
[15] Dieterich P., Klages R., Preuss R. and Schwab A., Proc. Natl. Acad. Sci. USA 105,
459(2008).
[16] Viswanathan G. M., Raposo E. P. and da Luz M. G. E., Phys. Life Rev. 5, 133(2008).
[17] Nossal R., J. Stat. Phys. 30, 391(1983).
[18] Levandowsky M., White B. S. and Schuster F. L., Acta Protozool. 36, 237(1997) .
[19] Ramos-Fernandez G., Mateos J. L., Miramontes O., Cocho G., Larralde H. and Ayala-Orozco
B., Behav. Ecol. Sociobiol. 55, 223(2004).
[20] Sims, D. W. et al., Nature (London) 451, 1098–1102 (2008).
[21] Humphries, N. E. et al., Nature (London) 465, 1066–1069 (2010).
[22] Bartumeus F., Peters F., Pueyo S., Marrase C. and Catalan J., Proc. Natl. Acad. Sci. USA 100,
12771(2003).
[23] Bartumeus, F. Fractals 15, 151–162 (2007).
[24] Viswanathan, G. M., Raposo, E. P.&da Luz, M. G. E., Phys. Life Rev. 5, 133–150 (2008).
[25] Gillespie, D.T., J. Comput. Phys. 22, 403–434 (1976).
[26] Gillespie, D.T., J. Phys.Chem. 81, 2340–2361 (1977).
|
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.